Datasets:

Hendrik-a

/

arxiv

like 0

a theoretical framework is proposed for the understanding of verbal perception the conversion of words into meaning , modeled as a compromise between lexical demands and contextual constraints and the theory is tested against experiments on short - term memory . the observation that lists of short words are recalled better than lists of long ones has been a long - standing subject of controversy , further complicated by the apparent inversion of the effect for mixed lists . in the framework here proposed , these behaviors emerge as an effect of the different level of localization of short and long words in semantic space . events corresponding to the recognition of a nonlocal word have a clustering property in phase space , which facilitates associative retrieval . the standard word - length effect arises directly from this property , and the inverse effect from its breakdown . an analysis of data from the peers experiments ( healey and kahana , 2016 ) confirms the main predictions of the theory . further predictions are listed and new experiments are proposed . finally , an interpretation of the above results is presented .

1607.05530

in 1894 , with their pioneering work on free - recall experiments , binet and henry introduced a key tool for the controlled investigation of short - term memory ( binet and henry , 1894 ) . in its traditional form , a free - recall experiment is performed by presenting the subject with a list of words and then requesting him or her to recall it in any order ( murdock , 1960 , 1962 ; roberts , 1972 ; standing , 1973 ) . several types of effects have been reported : \1 ) effects depending on the lexical properties of individual words . in particular , lists of short words are recalled better than lists of long ones , a fact known in the literature as the word - length effect ( baddeley et al . , 1975 ; russo and grammatopoulou , 2003 ; tehan and tolan , 2007 ; bhatarah et al . , 2009 ) . \2 ) effects in which the recall probability depends on the absolute position of words in the list . it has been observed that the first and last words in the list are recalled more easily ( `` primacy '' and `` recency '' effects ) . \3 ) effects depending on the relative position of words with respect to each other . most notably , the recall probabilities of contiguous words correlate positively , a fact known as the contiguity effect ( murdock , 1960 , 1962 ) . the need to understand serial - position effects led to the devising of retrieved - context models , such as the temporal context model ( howard and kanaha , 2002 ) . in these models the recall process , rather than retrieving a word directly , retrieves the context associated to the word first . within this scenario , recency effects appear because the context at the time of the `` memory test '' is most similar to the context associated with recent items . when an item is retrieved , it reinstates the context active when that item was presented . because this context overlaps with the encoding context of the items neighbors , a contiguity effect results . through these models , serial - position effects have been substantially understood over the past fifteen years ( howard and kahana , 2002 , 2002b ; sederberg , howard , and kahana , 2008 ; polyn , norman , and kahana , 2009 , 2009b ; lohnas , polyn , and kahana , 2015 ; kahana , 2012 ) . the same can not be said , however , about the word - length effect . the word - length effect ( wle ) has been a traditional testing ground for models of short - term memory ( campoy , 2011 ; jalbert et al . , 2011 ) , and it has played a key role in establishing the working - memory paradigm and the phonological loop hypothesis ( baddeley and hitch . , 1974 ) . the standard account of the effect ( baddeley , 2007 ) relies on a trade - off between memory decay ( in the phonological store ) and subvocal rehearsal via an articulatory control process . because shorter words take less time to rehearse , more decaying traces of them can be refreshed than decaying traces of long items , and , therefore , more short items can be recalled . this picture , however , is not able to account for all experimental observations concerning this effect , and has been repeatedly called into question . in ( neath et al . , 2003 ) , it was shown that with words having the same number of syllables but different pronunciation times , no unambiguous wle arises . this result ( extended in jalbert et al . 2011 ) suggests that the effect depends on the number of syllables , and not on the time it takes to pronounce them . experiments have also been performed in conditions where there was a delay between lists , making subvocal rehearsal possible in the interval . no appreciable difference in recall probabilities was found ( campoy , 2008 ) . in the same study , experiments were performed in which subvocal rehearsal was prevented by a high presentation rate . no delay was allowed between the presentation of word lists and the memory test . yet , the wle occurred unperturbed . in the 2011 paper i just cited , jalbert et al . concluded : `` the wle may be better explained by the differences in linguistic and lexical properties of short and long words rather than by length per se '' . in the meanwhile , within the fields of experimental and computational linguistics , progress has been made in understanding the role of word length in verbal processing . over the years , it has emerged that words with different lengths tend to have different semantic properties . the idea was first put forth in pedagogical studies ( klare , 1988 ) . in ( elts , 1995 ) , a correlation coefficient of 0.96 was found between a noun s length and its average tendency to be used as a technical term ( `` terminologicality '' ) . mikk et al . ( 2000 ) , using data on the human - assessed complexity of a large sample of words , found a correlation coefficient 0.86 between words length and their semantic complexity . pinning down the precise semantic property that correlates to word length has proven difficult . already in ( greenberg , 1966 ) it was argued that a word s length correlates positively to its conceptual `` markedness '' of meaning . various notions of markedness have subsequently been discussed in the literature ( haspelmath , 2006 ) . piantadosi et al . ( piantadosi et al . , 2011 , 2011b ) and later mahowald et al . , ( mahowald et al . , 2012 ) reported that the length of words correlates positively with their contextual information rate . more recently , lewis and frank ( 2016 ) have carried out a comprehensive experimental study across 80 languages . they found that , in all the languages considered , judgments of conceptual complexity for a sample of real words correlate highly with their length , and they even control for frequency , familiarity , imageability , and concreteness . their conclusion is : `` while word lengths are systematically related to usage @xmath0 both frequency and contextual predictability @xmath0 our results reveal a systematic relationship with meaning as well '' . in the light of these findings , it would be a natural step to attempt an explanation of the wle in terms of the semantic differences among words . however , no such approach seems to have been attempted in the literature . recently , new aspects of the wle have emerged through the analysis of a large set of data from experiments by miller and al ( miller et al . , 2012 ) . the data analysis was performed by katkov et al . ( katkov et al . , 2014 ) , who found no negative correlation between total length of presented items and number of recalled words , thus disproving both rehearsal - time theories and hypotheses based on the increasing complexity of longer items . moreover , they reported an inversion of the effect in mixed lists , that is , lists where words are selected irrespectively of their length . they observed that , in this type of lists , the mean values of recall probabilities allow to establish an increasing trend . long words are recalled better than short ones . an `` inverse '' wle had been previously reported by at least two groups , but in somewhat less general circumstances : one of them ( hulme et al . , 2006 ) embedded strictly pure lists with a single word of a different type , while the results of xu et al . ( xu et al . , 2009 ) , may not bear direct comparison with data in languages other than chinese . if the inversion of the wle for mixed lists will be confirmed by further experiments , it will have to be taken into account by every general theory of the standard wle . let us consider , therefore , what requirements a model should fulfill to explain both phenomena simultaneously . call @xmath1 the fraction of long words in the list ; call @xmath2 the probability of recalling successfully a given long word from a list in which a fraction @xmath1 of words are long ; and let @xmath3 be the probability of recalling successfully a short word , from a list with a fraction @xmath1 of long words . obviously , the function @xmath2 is only defined for @xmath4 , and the function @xmath3 only for @xmath5 . for @xmath60 , 1[$ ] , both functions are defined . theorists would have to reconcile two observations on the curves @xmath7 : \1 . @xmath8 \2 . @xmath9 for @xmath60 , 1[$ ] . the only way these two inequalities can be simultaneously satisfied is if _ both _ @xmath10 and @xmath11 are , on the whole , decreasing functions of @xmath1 . the simplest choice of these curves compatible with experiments is one where both are monotonously decreasing , that is : @xmath12 [ fig1 ] ( -225,5)@xmath13 ( -50,5 ) @xmath14 ( -200,146 ) @xmath10 ( -200,97 ) @xmath11 this means that , whenever we replace a short word of the list with a long word , we are lowering the recall probability of all the words in the list , both long and short . a higher number of long words makes every single word in the list harder to recall . it is difficult to imagine how this could ensue from the different @xmath15 of long and short words . the question is then : can equation ( [ derivatives ] ) result directly from the different semantic properties of long and short words ? in this paper , i will show that the answer is positive as long as one models carefully the process of verbal perception . a suitable way of doing so is demonstrated in the next section . in section iii , i employ a retrieved - context description of verbal recall to derive both wles ( standard and inverse ) . in section iv , i test two key predictions of the theory against data from the peers experiment of kahana et al . ( lohnas and kahana , 2013 ; healey and kahana , 2016 ) . in section v , i list five experiments designed to test further predictions of the theory . in the conclusions , i sketch a possible interpretation of the results . verbal perception is the conversion of words into meaning , and any theory of the phenomenon must begin by defining the space in which the mental trajectory takes place . i will refer to this as `` semantic space '' and will represent `` semantic states '' as integers belonging to a segment @xmath16 , with @xmath17 . measurable quantities are to be computed in the limit @xmath18 , and a generalization to higher dimensions will be introduced below . since @xmath19 , it follows that , between two discrete times @xmath20 and @xmath21 , the psychological trajectory can be represented by a sequence of integers @xmath22 . as memory plays a crucial role in thought process , the laws of motion governing the trajectory will be in general highly non - markovian . [ fig2 ] .,title="fig:",scaledwidth=100.0% ] call @xmath23 the vocabulary available to the system . naturally , a given word @xmath24 may be appropriate to describing more than one state , though with a varying degree of appropriateness . for each word @xmath25 , thus , one can define its semantic range @xmath26 as the set of all states described by that word . meanings should be seen as distributed in an universal way over @xmath27 , as in figure 2 . the function @xmath28 matches each word with locations associated to meanings suitable for that word . not all states are equally fit to be verbalizable . hence , there will be a varying probability @xmath29 that the state @xmath30 will match the word describing it . the value of @xmath29 is the degree of verbalizability of state @xmath30 ; or conversely , given @xmath31 , @xmath29 is the fitness of word @xmath25 as a descriptor of state @xmath30 . if the sets @xmath32 do not overlap , we can define a verbalization function @xmath33 such that @xmath34 whenever @xmath31 . when the system tries to verbalize state @xmath35 , it produces word @xmath36 with probability @xmath29 , and the silent word @xmath37 with probability @xmath38 . we will define @xmath29 so that it is only null for nonverbalizable states . finally , we can take the trajectory to become markovian whenever it is driven by a specific verbal input : at the instant in which a new word is presented , the system hops into the nearest state described by that word . accordingly , let the function @xmath39 be defined by the condition : [ fig3 ] @xmath40 satisfy this condition , @xmath41 is chosen by flipping a coin . given a verbal input @xmath42 , with the @xmath43 s @xmath44 , and a starting point @xmath45 , the trajectory of the system while perceiving the input will be given by : @xmath46 @xmath47 . obviously , the ordering of words in the input is essential . suppose for instance that the vocabulary contains two words , @xmath48 , with @xmath49 and @xmath50 , and the starting point is @xmath51 . the verbal input @xmath52 yields the trajectory @xmath53 , while the input @xmath54 yields the trajectory @xmath55 . word @xmath56 corresponds to two different states in the two inputs . @xmath57 [ fig4 ] the verbalizable states associated to word @xmath25 can be seen as seeds of a voronoi partition of @xmath27 into cells @xmath58 , where states located on the boundaries are understood to belong to each of two cells with probability @xmath59 . [ fig5 ] during the perception of verbal inputs , the distance travelled by the system is controlled by the function latexmath:[$d(x , w):= state @xmath30 . this is just the distance of @xmath30 from the closest state belonging to @xmath61 . the level of nonlocality of a word @xmath25 may be measured by its reaching distance averaged over all starting points : @xmath62 obviously @xmath63 does not necessarily imply @xmath64 . consider two words @xmath65 , @xmath66 such that @xmath67 and @xmath68 , with @xmath56 even and @xmath69 . we have @xmath70 , but @xmath71 . a direct relation can be shown to exist between a word s average reaching distance and the size of its voronoi cells . call @xmath72 the size of the @xmath73-th voronoi cell of word @xmath25 . assuming @xmath74 , we may neglect boundary effects and write @xmath75 thus , the average reaching distance of a word depends solely on the first two moments of its voronoi length distribution . if @xmath76 for a word @xmath25 , its word structure contains strong fluctuations , so one may separate @xmath27 into regions were states described by @xmath25 are denser , and regions were they are sparser . word @xmath25 is effectively `` localized '' inside the regions were such states are dense , that is , it expresses those semantic areas better than those where its states are sparse . the degree of word localization has arguably a strong effect on the dynamics , as shown in figure 6 . [ fig6 ] in most experiments , the relevant number of word - lengths is four , since words with more than four syllables are rare in english . for simplicity , here we will consider the existence of only two word - lengths , short and long . call @xmath77 ( with @xmath78 ) the reaching distance averaged over space and over all words of the same length ( short or long ) . from eq . ( [ voronoi ] ) , we see that the space - averaged reaching distance is the product of two factors , involving respectively the first and second moment of the voronoi length distribution . let us consider how these two moments depend on the word s length . the average voronoi length @xmath79 may be related to the frequency of a given word in a corpus of the language . indeed , if the system , in its speaking mode , explores semantic space ergodically and uniformly , the frequency of a word is @xmath80 . in a typical corpus of the english language , the frequency @xmath81 of words with @xmath82 syllable is monotonously decreasing . as a consequence , we expect the average of the voronoi length @xmath83 to be larger for short words than for long words . while this is correct for most languages , notice that there are exceptions , such as turkish and arabic , where the function @xmath81 is peaked at @xmath84 ( fucks , 1956 ; grzybek , 2007 ) . let us now look at the relative fluctuations of the voronoi length , described by @xmath85 . we will surmise their magnitude through a qualitative argument . as mentioned in the introduction , various approaches have been taken to prove that long words are on average more technical , specialized , distinctive or marked than short words . several claims made in ( elts , 1995 ) , ( mikk et al . , 2000 ) , and ( lewis and frank , 2016 ) may be rephrased as the statement that long words are , on average , conceptually more specific . a word is conceptually specific if it is localized in certain areas of semantic space . a correlation exists , therefore , between word - length and semantic localization . localization , in turn , will occur if the scale over which the voronoi length fluctuates is comparable or greater than its average value . this leads to the conclusion that the relative fluctuations of the voronoi length will be larger for longer words . thus , both factors in eq . ( [ voronoi ] ) take a greater value if the word is long . it follows that @xmath86 . a one - dimensional modeling of semantic space is of course unrealistic , and may not suffice for every application of the theory . if @xmath27 is taken to be a connected subset of @xmath87 , the definition we have given for the word structure @xmath88 applies all the same . the points in space are now vectors , and equations ( [ hopping ] ) and ( [ reachingdistance ] ) are still valid , the distance in eq . ( [ reachingdistance ] ) being the euclidean distance in @xmath89 dimensions . the voronoi cells , however , become less simple to treat as they can be arbitrary polyhedra ( for a complete treatement , see aurenhammer et al . , formula ( [ voronoi ] ) for the reaching distance must be modified , and it takes a geometry - dependent form . the voronoi structure , yet , affects directly only the process of verbal perception , not the process of memory retrieval , which will be the subject of the next section . thus , while in the figures i will refer to the one - dimensional case , the mathematical results will apply to any number of dimensions . we have seen that the rules of motion of the system become markovian during the perception of verbal input . in retrieved - context models of short - term memory , the rules of motion are also markovian during the search for memories , which is conducted through the principle of free association . in these models , the retrieval of memories per se is not a measurable phenomenon . what can be observed is the retrieval of words describing those memories . therefore , a recall experiment must be seen as the composition of two processes : a processes of memory retrieval , and a process of memory verbalization . the markovian process of memory retrieval must be described here , in keeping with the spirit of context - driven models , as a random walk on @xmath27 that effects a retrieval whenever it meets a state corresponding to the experience to be recalled . the verbalization process , on the other hand , depends on the verbalizability of memories : a memory @xmath30 , once retrieved , has a probability @xmath29 of leading the system to produce the word describing it . [ fig7 ] ( -135,420)@xmath90 ( -216,253)@xmath45 the following mathematical problem arises . supposing one is given \1 ) the structure @xmath91 of the vocabulary ; \2 ) a word list @xmath92 presented to the system ; \3 ) the state of the system when the word list begins to be presented , that is , a probability distribution @xmath93 $ ] on its position @xmath90 ; \4 ) the state of the system when the retrieval process begins , that is , a probability distribution @xmath94 $ ] on the new position @xmath45 ; one wants to predict the probability that the @xmath73-th word will be among those recalled by the system . in the next section , this program will be carried out for the particular case of a vocabulary containing words of two different lengths . we will begin by defining the probability @xmath95 that a memory placed at distance @xmath96 from @xmath45 is met by the retrieving random walk for the first time after a time @xmath97 . in one dimension , this is given by @xmath98 where @xmath99 and @xmath100 . the probability that a memory will be retrieved is therefore @xmath101 where the cutoff @xmath102 is needed to obtain meaningful results in one or two dimensions , and can otherwise be let to infinity . suppose a list @xmath103 has been presented . call @xmath104 the probability of retrieving the memory created by the @xmath73-th word in the list , given a certain initial position @xmath90 for the trajectory during presentation , and a certain initial position @xmath45 for the trajectory during retrieval . we have : @xmath105 where @xmath106 . our goal is to compute the average recall probability for an arbitrary list composed by @xmath82 short words and @xmath107 long words arranged into a given order . this amounts to averaging eq . ( [ piv ] ) over all lists of the same type @xmath108 where @xmath109 for @xmath110 : @xmath111 where @xmath112 is the type of word @xmath25 and @xmath113 ( @xmath114 ) is the number of short ( long ) words in the vocabulary . we will perform the averaging over lists through a mean - field approach . mean - field approaches , widely employed in physics , consist in inverting the order of the two steps : the computation of observables and the averaging . instead of averaging the final probabilities , one averages an intermediate , non - observable quantity usually called the `` field '' . in this case , we may average the functions @xmath115 themselves . from section iic , we know that the average displacement induced by the function @xmath115 is equal to @xmath116 with @xmath117 , while no constraints emerged concerning the direction of this displacement as a function of word type . therefore , to average the functions @xmath118 over all word lists of one type , we replace @xmath115 with the function @xmath119 defined by @xmath120 , where @xmath121 is a unity vector randomly chosen from a suitable distribution @xmath122 $ ] . we can thus write @xmath123 , and eq . ( [ piv2 ] ) becomes @xmath124 where @xmath125 $ ] . the initial positions @xmath90 and @xmath45 are not measurable quantities . in eq . ( [ meanfield ] ) , therefore , they must be averaged over through two suitable distributions @xmath126 and @xmath127 , yielding @xmath128 where @xmath129 and @xmath130 . the function we need to average may be rewritten as @xmath131 + \sum_{j \neq i } \mathbb{p}\big[i_1=j\big ] \ p(y_i | y_j)\ ] ] where @xmath132 $ ] is the probability that the @xmath73-th memory , @xmath133 , will be the first one to be found , and @xmath134 < 1 $ ] . substituting eq . ( [ compound ] ) into eq . ( [ meanfield2 ] ) , we find @xmath135 \big\rangle_{\chi , \psi } + \sum_{j \neq i } \big \langle \mathbb{p}\big[i_1=j\big ] \big\rangle_{\chi , \psi } \ p(y_i | y_j ) \bigg \rangle_{\omega}\ ] ] the distribution @xmath136 refers to the state @xmath45 of the system after the so - called retention interval , during which the subject freely elaborates the information gathered during presentation . a full understanding of such elaboration would require modeling the free motion of this system , but we have only been able to markovianize the equations of motion during a progressive searching task or in the presence of a driving input . thus , a reasonable ansatz is necessary . neglecting recency effects , we can suppose @xmath136 to contain @xmath137 similar peaks at the locations @xmath138 explored during presentation . in this picture , the dependence of @xmath139 \big\rangle_{\chi , \psi}$ ] on the index @xmath73 will be negligeable , so we can approximate @xmath140 \big\rangle_{\chi , \psi}$ ] with a constant value @xmath141 . substituting this into eq . ( [ meanfield4 ] ) , and rewriting the argument of @xmath142 , we find @xmath143\ ] ] where @xmath144 is now the product of @xmath145 copies of the distribution @xmath146 , and the value of the summand depends solely on the number of long and short words located between word @xmath73 and word @xmath147 . using @xmath148 to include the @xmath149 term into the sum , and noticing that the labels of the @xmath150 s are interchangeable , we can finally rewrite eq . ( [ meanfield5 ] ) as @xmath151 where we have introduced the quantities @xmath152 and the unit vectors @xmath153 , @xmath154 are independently distributed according to @xmath155 copies of the distribution @xmath146 . for pure lists , eq . ( [ meanfield6 ] ) becomes @xmath156\ ] ] as mentioned above , the probability of recalling the @xmath73-th word of the list is equal to the probability of retrieving the @xmath73-th _ memory _ , multiplied by the factor @xmath157 . in the previous section , we averaged the retrieval probability over all words of the same type ; similarly , we must now average the verbalizability @xmath29 , which we take to be distributed independently of the word structure @xmath158 . defining @xmath159 as the average verbalizability of words of type @xmath160 , we obtain the full recall probability @xmath161 as mentioned in the introduction , the classical wle is the experimental fact that pure lists made of shorter words are easier to remember . of course such behavior may always be prevented , in principle , by making the verbalizability ratio @xmath162 sufficiently low . yet , if the reported inversion of the wle exists within this model , it must rely on the opposite requirement namely , that the verbalizability ratio is sufficiently high . the questions we are therefore supposed to answer are : first , whether there exists for this model a range of values of @xmath162 where both the classical and the inverse wle occur ; second , under what conditions on the parameters this may happen , and whether such conditions are relevant to current experiments . let us describe the typical experimental situation . in the experiments , word lists are generated by drawing words at random from a vocabulary @xmath23 . in a double - word length scenario , this vocabulary will contain @xmath113 short words and @xmath113 long ones . we may consider , therefore , an ensemble of lists of length @xmath137 , where each word has a probability @xmath163 of being long , and a probability @xmath164 of being short . the recall probability for the @xmath73-th word of the list can be averaged over all lists whose @xmath73-th word is of type @xmath160 , yielding @xmath165 . substituting eq . ( [ meanfield6 ] ) , this becomes : @xmath166\ ] ] in the first sum , which is the contribution from @xmath167 in eq . ( [ final ] ) , @xmath82 and @xmath107 stand for the number of short or long words in positions @xmath168 such that @xmath169 ; in the second sum ( the contribution from @xmath170 ) @xmath82 and @xmath107 stand for the number of short or long words in positions @xmath168 such that @xmath171 . in eq . ( [ random ] ) , the notation @xmath172 has come to denote an averaging over @xmath82 , @xmath107 performed for each separate value of @xmath173 and summed together . this is done through the binomial distribution @xmath174 given that the total recall probability is @xmath175 , the standard wle effect ( @xmath176 ) will occur if @xmath177 , where @xmath178 , and the inverse effect will occur if @xmath179 , that is , if @xmath180 , where @xmath181 . the two effects coexist if @xmath182 , which can only happen if @xmath183 . this condition can be rewritten as @xmath184 now , it can be seen that @xmath185 a decreasing function of @xmath1 , because by increasing @xmath1 one transfers weight from the first to the second argument of @xmath186 inside both terms of eq . ( [ random ] ) , which reduces the value of the function @xmath186 . hence , we have @xmath187 and @xmath188 , from which it follows that the inequality ( [ inequality ] ) is identically satisfied . we conclude that , in this model , the wle can undergo an inversion for any @xmath60 , 1[$ ] , as sketched in figure 1 . consider a list of the type @xmath189 , where @xmath190 , containing @xmath107 words and @xmath82 short ones . the trajectory during presentation is illustrated in figure 8 . the system begins from a random position @xmath90 , and at each new word @xmath43 of type @xmath191 , its position is shifted forth by the operator @xmath192 . the distance travelled at step @xmath73 is , in the mean field approach , equal to @xmath193 . so a memory produced by the presentation of a short word will be formed in the vicinity of the latest memory ( that is , within a distance of order @xmath194 ) whereas a memory produced by the presentation of a long word will be formed at a longer distance ( of order @xmath195 ) from the memory preceding it . consequently , memories are divided into clusters separated by a distance of order @xmath195 from each other . each cluster spreads over a width of order @xmath194 . [ fig8 ] , where @xmath196 stands for long words and @xmath197 for short ones . the longer jumps correspond to the presentation of long ( localized ) words.,title="fig:",scaledwidth=100.0% ] these memory clusters correspond to different `` segments '' of the list . call @xmath198 the index values corresponding to long words within a given list . if @xmath199 , the segments are @xmath200 for @xmath201 , and @xmath202 . if @xmath65 is short , there is an additional segment @xmath203 , and the number of segments is @xmath204 . call @xmath205 the length of segment @xmath206 . all the @xmath205 s must be positive except @xmath207 , which may be null , and @xmath208 . the direction in which the trajectory moves at each step is defined by the unknown distribution @xmath146 , which depends on the details of the word structure @xmath158 . for a generic choice of @xmath146 , clusters formed at longer time intervals from each other will lie further apart in semantic space . hence , the trajectory during presentation is the composition of two processes : a clustering process and a diffusion process . notice that in eq . ( [ meanfield5 ] ) the argument of @xmath142 is of the order of @xmath209 . if the list is short enough ( that is , if @xmath210 , @xmath211 and @xmath212 ) , the summand has two orders of magnitude : one of the order of @xmath213 and one of the order of @xmath214 . this corresponds to the fact that the diffusion process is slow on the scale of the trajectory during presentation . in this regime , formula ( [ meanfield5 ] ) for @xmath215 may be estimated by replacing the summand with @xmath216 whenever there is at least one long word between @xmath217 and @xmath218 , and with @xmath219 otherwise . this is equivalent to approximating the matrix elements @xmath220 of eq . ( [ pimatrix ] ) with @xmath221 if @xmath222 and @xmath223 , and with @xmath224 if @xmath225 , thus ignoring the dependence of the retrieval process on the distance between segments of the list . call @xmath226 the first memory to be retrieved . the conditional retrieval probability for memory @xmath227 is @xmath228 . in the slow - diffusion limit , this is of the order of @xmath219 if memories @xmath226 and @xmath229 belong to the same cluster , and is of the order of @xmath216 otherwise . averaging over the first retrieval , we find @xmath230\ ] ] where @xmath231 is the length of the segment of the list to which word @xmath73 belongs . the average recall probabilty @xmath232 for words of type @xmath160 will thus be equal to : @xmath233\\ p_l = \frac{q_l p_0}{l } \sum_{i=1}^{l } \big [ 1 + ( a_i -1 ) p_s + ( n - a_i ) p_l \big]\end{aligned}\ ] ] defining @xmath234 and @xmath235 , we can rewrite this as @xmath236 \\ \label{pl } p_l = q_l p_0 \big [ n p_l - p_s + 1 + \mu ( p_s - p_l ) \big]\end{aligned}\ ] ] all the dependence on the ordering of words in the list , therefore , enters the recall probabilities through the parameteres @xmath237 and @xmath238 . the values of these parameters are shown in table i for simple lists . [ table3 ] c c c c list structure & @xmath237 & @xmath238 & @xmath239 + + @xmath240 & @xmath241 & @xmath241 & + @xmath242 & @xmath241 & @xmath243 & @xmath244 + @xmath245 & @xmath246 & @xmath247 & @xmath248 + @xmath249 & @xmath250 & @xmath251 & @xmath252 } $ ] + @xmath253 & @xmath254 & @xmath255 & @xmath256 + + for a pure list , entirely composed of words of type @xmath160 , eqs . ( [ ps ] ) and ( [ pl ] ) become : @xmath257 \end{aligned}\ ] ] data for mixed lists may be interpolated with formulas ( [ ps ] ) and ( [ pl ] ) to test the theory and fix the values of the internal parameters . finally , let us look at the range of occurrence of the wles in the slow - diffusion regime . the classical wle ( @xmath258 ) emerges for @xmath259 where @xmath260 the ratio @xmath261 for mixed lists can be estimated as follows . given a fixed value of @xmath262 , @xmath237 ranges between @xmath263 ( for @xmath264 ) and @xmath265 ( for @xmath266 ) . the minimum value of @xmath238 is obtained by starting the list with a short word and having @xmath267 segments of @xmath268 words and @xmath269 segments of @xmath270 words . the maximal @xmath238 is obtained by setting @xmath271 and lumping all the short words into one segment : @xmath272 . substituting @xmath273 and @xmath274 in ( [ ps ] ) , ( [ pl ] ) , one obtains a strict upper bound on the ratio @xmath275 : @xmath276 , where @xmath277}\ ] ] once again , the inverse wle ( @xmath278 ) will emerge under the sufficient condition @xmath180 ; hence , the possibility that the classical and inverse effects may coexist requires @xmath279 . if we rewrite this inequality by means of eq . ( [ thetaclslow ] ) and ( [ thetainvslow ] ) , the `` microscopic '' probability parameters @xmath219 and @xmath216 cancel out and we obtain the general condition @xmath280 , which is identically satisfied for any @xmath281 . it follows that , in the slow - diffusion regime , the system displays an inversion of the wle for _ all _ mixed lists containing more than two long words . this statement , unlike the conclusions of the previous section , is not only true on average , but holds true regardless of the order in which short and long words are arranged within the list . in the previous section we have considered the case where the diffusion process was much slower than the clustering process which translates into upper bounds on the order of magnitude of @xmath82 and @xmath107 . notice that we defined these bounds in terms of the parameters @xmath194 and @xmath195 , whose value may vary from subject to subject . therefore , the very same list may be experienced in a fairly stationary regime by one subject , and in a regime of fast diffusion by a more easily distracted one . here , we will consider the opposite case of very fast diffusion , defined as the regime where @xmath282 goes down fast on the length scale of @xmath194 , so the matrix elements of @xmath283 decay quickly as the value of the indices grows , and the recall dynamics is dominated by the contiguity effect . if a long word causes a lesser diffusion than two short words , we may neglect all but the top - left elements of the @xmath283-matrix : @xmath284 , @xmath285 , and @xmath286 . otherwise , we can work in the lowest approximation by neglecting also @xmath216 . ( [ random ] ) becomes @xmath287\ ] ] where i dropped the pedix @xmath73 because all dependence on @xmath73 vanishes as long as @xmath73 is neither @xmath241 nor @xmath137 . the thresholds for the verbalizability ratio are @xmath288 and for @xmath289 , eq . ( [ fast ] ) is nothing but a linear version of figure 1 . when experiments yield a near - linear curve @xmath290 , thus , it may be taken as a sign that the system is operating in a very fast diffusion mode . when the observed curve is nonlinear , one must infer that terms of the type @xmath291 for @xmath292 are relevant , and a polynomial interpolation can be performed , by truncating the sums in eq . ( [ random ] ) . the crossover toward linearity of the curves @xmath290 may be explored by tuning experimentally the amount of diffusion in the system , as we will see in section v. the foregoing analysis has proven that , within this model of verbal perception , the recall process can display consistently both the classical and the inverse wle , as observed in the experiments . there may be , however , other mechanisms leading to a similar prediction . katkov et al , for instance , propose a lexical explanation for both the direct and inverse wle , based on possible differences in the long - term neural representation of long and short words ( katkov et al . , 2014 ) . in order to tease out which mechanism is really responsible for the effects , one needs to extract further information from the data , going beyond the computation of mere recall probabilities . i will do so by using data from peers ( penn electrophysiology of encoding and retrieval study ) , a large study conducted at the university of pennsylania and devoted to assembling a database on the electrophysiological correlates of memory ( lohnas and kahana , 2013 ) . the sample i have considered corresponds to experiment i of peers . it includes data from trials on @xmath293 college students ( age range : 18@xmath29430 ) and on @xmath295 older adults ( age range : 61@xmath29485 years ) . in each trial , @xmath296 words were presented one at a time on a computer screen . each word was drawn from a pool of @xmath297 words with length varying between @xmath298 and @xmath299 syllables . the word - length distribution was peaked at @xmath300 , reflecting that of a typical english lexicon ( but not of a typical english corpus ) . each item was kept on the screen for @xmath301 ms , followed by an interstimulus interval of 800@xmath2941200 ms . after the last item in the list , there was a delay of 1200@xmath294 1400 ms , after which the participant was given @xmath302 s to attempt to recall aloud any of the just - presented items . multiple trials were performed on each subject , summing up to @xmath303 trials for the students sample and to @xmath304 for older adults . for more details on the experimental procedure , see healey and kahana , 2016 . because the word lengths involved are more than two , the formulas we derived in the previous section may not be applied verbatim . nonetheless , two key consequences of the theory afford a direct comparison with the data . we say that a word is recalled `` by contiguity '' when its recall occurs immediately after the recall of a word contiguous to it within the list . in the model we have developed , no matter how high the amount of diffusion , recall by contiguity will be more frequent for short words than for long words . this follows from the fact that two consecutive short words belong necessarily to the same segment , whereas a short word and a long word , though contiguous , may belong to different segments , and two long words are sure to belong to different segments . while this prediction has been derived in a two - length model , it generalizes immediately to models with multiple lengths . we have proven in section ii that shorter words are characterized by a shorter reaching distance ; hence , memories formed by shorter words have a higher chance of being located in the proximity of memories formed by contiguous words . contiguity will play therefore a stronger role in the recall of a shorter word : the shorter the words involved , the higher the chances of recall by contiguity . [ contiguity ] syllables ( blue dots ) ; probability of recall by contiguity ( black ) ; and mean probability of recall by contiguity ( red ) . all values have been computed from peers data.,title="fig : " ] ( -370,255)@xmath305 ( -47,5 ) @xmath82 let us call @xmath306 the probability that a given word @xmath25 , if recalled , will be recalled by contiguity . the theory predicts that @xmath305 should be larger for shorter words . if so , computing @xmath305 from the data and averaging it over all words of the same length would yield a decreasing curve @xmath307 . in figure 9 , @xmath305 is shown for all words having the same number of syllables ( blue dots ) . the data have been aggregated from all trials . the distribution of @xmath305 ( black line ) is wide for all word lengths ; nevertheless , the mean probability of recall by contiguity ( shown in red ) decreases monotonically with the number of syllables , in agreement with the theory . the @xmath297-element wordpool used in peers contains only four @xmath308 five - syllable words , and a single @xmath309-syllable word ( `` encyclopedia '' ) . hence , the statistics for these two lengths may not be considered reliable . while figure 9 refers to experiment 1 of peers , the robustness of the effect has been checked for by repeating the analysis on the databases from experiments 2 and 3 ( see lohnas et al . , 2015 ) . the decreasing trend has proven invariant across databases . let @xmath310 be the serial positions of the words recalled by the subject during a certain trial . at each step @xmath89 in the recall process , the system performs a serial - position jump of a magnitude @xmath311 . in our terminology , longer jumps will require the exploration of a larger portion of semantic space . therefore , the distribution of jumps should be monotonously decreasing as a function of their size ( as appears indeed from the peers data , figure 10 ) . [ distribution ] , computed from peers data . in colors , distribution of jumps for a fixed size @xmath312 of the word - length barrier , shown in the legend . as the lists contain @xmath296 words , @xmath313 ranges between @xmath241 and @xmath314.,title="fig : " ] ( -320,225)@xmath315 ( -33,5 ) @xmath313 from our analysis of the two - length scenario , we know that these jumps cover greater distances in semantic space if the memory created by the word of departure and the memory created by the word of arrival belong to different clusters . since it is long words that have the ability to break clusters , a long word located in a position @xmath168 such that @xmath316 plays effectively the role of a recall barrier , hindering direct transitions between @xmath43 and @xmath317 . [ clusters ] -word list . word lenght is encoded in colors as shown in the vertical sidebar . at each level of segmentation , clusters are displayed as horizontal bars covering the given portions of the list . to the right , a plot of the mean cluster length ( blue ) and of the number of clusters per list ( red ) , as a function of the hierarchical level @xmath82 of clusters , computed from peers data by averaging over all trials . , title="fig : " ] ( -10,-10 ) @xmath82 similarly , in experiments with multiple word lengths , longer words create memories at longer distances . as clustering in semantic space will occur then over multiple length scales , clusters assume a hyerarchical structure . so does the segmentation of the list , with moderately long words marking off smaller segments within the larger segments delimited by longer words . this is shown in figure 11 , together with the average length of clusters as computed from the data , and the average number of clusters per list . the probability for the transition @xmath318 ( a jump of size @xmath319 ) will be affected mainly by the hyerarchical level of the clustering involved . since longer words break clusters at a higher level , we conclude that the effect of clustering on such a jump will be controlled by the length of the longest word located between the positions @xmath73 and @xmath147 : @xmath320 where @xmath321 is the length of word @xmath25 and , once again , the lowest index has been excluded for the same reasons why it was not counted in eq . ( [ meanfield5 ] ) of section iii . [ distribution ] to word @xmath322 during the recall of a nine - word list ( with words from the peers wordpool ) . the red bars describe word length : the `` barrier word '' for this transition is marked in yellow.,title="fig : " ] we will call @xmath323 the `` barrier size '' for the @xmath324 transition . the role played by this quantity is depicted in figure 12 . if the clustering mechanism does exist , it will enhance the transition probability for small values of @xmath323 , and suppress it for higher values . to test this prediction , we must first study the problem in the absence of clustering , calculating the transition probability @xmath325 for jumps of size @xmath326 . the resulting formula must be evaluated using the experimental values of parameters , and can then be compared with the values of @xmath325 extracted from the data . if the clustering occurs , we will observe an enhancement of @xmath315 for small values of the barrier size @xmath312 and a suppression at high values . let us suppose that the clustering does not occur . the system then will not `` feel '' the word - length barriers , and the probability @xmath315 of a jump will be independent on the size of the barrier to be overcome . conversely , the probability distribution of barriers over the jumps performed by the system will depend only on the jumps size , and on the probability distribution of word length with the list . the probability that a jump of size @xmath313 involves a word - length barrier @xmath312 is equal to the probability that the longest word out of a random sequence of @xmath313 words has length @xmath312 . this is equal to @xmath327 where @xmath328 is the word - length distribution with the lists . formula ( [ no_clustering ] ) describes the no - clustering barrier - size distribution , that is , the distribution of barrier sizes for jumps of a given length if the jumps are not influenced by the barriers . i have evaluated formula ( [ no_clustering ] ) using values of @xmath328 computed directly from the lists recorded in the peers database . in figure 13 ( see the end - page ) , the results are compared with the values of @xmath329 extracted directly from data . the no - clustering curves are dashed , the experimental curves solid . the effect of clustering is apparent . all experimental curves reveal an enhancement of the transition probability for small values of @xmath312 , and a suppression at large values . morevoer , the effect persists across the whole range of possible jump sizes . we would expect that the influence of clustering should be small for jumps shorter than the typical size of the clusters , and grow as the jumps get longer . this is indeed what we notice in figure 13 . the influence of clustering can be observed already for jumps of size one , with an enhancement of probability for small barriers and suppression for larger barriers . yet , the clustering becomes more prominent as the jumps get longer . a noticeable increase in the importance of clustering is observed after @xmath330 . this may be due to the fact that the @xmath331 clusters are becoming important , as the jumps have gotten longer than the average length of the clusters of the third level ( see figure 11 ) . since the peers lists have length @xmath296 , there is no available statistics for values of @xmath313 larger than @xmath314 . nonetheless , we can predict that this growth will saturate when the probability of meeting a maximal bareer approaches near - certainty . for jumps longer than that , @xmath312 is no longer a controlling parameter ; from that point on , the continued suppression of recall will derive from the need to find the next memory multiple clusters away . [ jumps ] of serial - position jumps in the recall process , plotted as a function of the size @xmath312 of word - length barriers , for all possible values of jump size @xmath313 . the red curves have been calculated in the absence of clustering , as per formula ( [ no_clustering ] ) . the blue curves are extracted from peers data . clustering manifests as an enhancement of transition probability for low barriers ( small @xmath312 ) and as a suppression where larger barriers are encountered . , title="fig:",scaledwidth=100.0% ] ( -340,490)@xmath332 ( -340,425)@xmath333 ( -340,360)@xmath334 ( -340,295)@xmath335 ( -340,235)@xmath336 ( -340,170)@xmath337 ( -340,110)@xmath338 ( -340,43)@xmath339 ( -55,490)@xmath340 ( -55,425)@xmath341 ( -55,360)@xmath330 ( -55,295)@xmath342 ( -55,235)@xmath343 ( -55,170)@xmath344 ( -55,110)@xmath345 most free - recall experiments performed so far have employed either a wordpool with an arbitrary number of syllables , as in peers , or a pool characterized by a fixed length that is usually @xmath346 syllables ( as in the toronto wordpool , see friendly et al . , 1982 ) . further tests on this theory , however , would be easiest to perform through experiments in which the words presented for recall are drawn from either of two sets : a set of very long words ( e.g. number of syllables @xmath347 ) and a set of very short ones ( @xmath298 ) . when the lists are sufficiently short , the system operates in a slow - diffusion ( or `` clustering '' ) regime . in this case , three parameters should be kept track of in the experiment : @xmath238 , @xmath237 , and the number @xmath107 of long words in each list ( see section iii ) . while the recall probability of individual words depends on the details of the list , the average recall probability for short or long words should be affected by the list structure only through those three parameters , which completely characterize the list for the purposes of this type of experiment . it is then feasible to test qualitatively the following predictions : \1 . @xmath348 correlates positively with @xmath238 and negatively with @xmath237 . @xmath349 correlates positively with @xmath237 . these are straightforward consequences of ( [ ps ] ) , ( [ pl ] ) . in experiments with words of two lengths , data may be interpolated with formulas ( [ ps ] ) and ( [ pl ] ) to test the theory and to fix the values of the internal parameters . serial position effects related to word - length can be isolated experimentally by using word lists constructed on two or more basic templates of segment - structure @xmath350 , with @xmath351 . the experimenter would present equivalent lists to a number of participants and would correlate the recall probability of words with their positions within each template structure . large data sets from such experiments should be able to corroborate or to rule out the mechanisms i have described . competing effects of familiar types , such as primacy and recency , may be easily subtracted . two simple predictions can be made , easier to check in a slow - diffusion regime but equally valid with faster diffusion : \1 . words from longer segments have a higher recall probability than words of the same type from shorter segments ; \2 . successful recall of words from one segment hinders the recall of words from others . the first prediction follows from eq . ( [ pi ] ) , and should be easy to test . the second prediction must be compared with data by computing the correlation functions @xmath352 , that is , the joint probability for the recall of the @xmath73-th and @xmath147-th words in the list . from there , it is straightforward to obtain the in - segment and cross - segment correlation functions : @xmath353 and to verify whether @xmath354 , as the theory suggests . we may add to these predictions a third one namely , the fact that the long word of each segment is the easiest one to recall . this results directly from the bounds we derived in the previous section on the ratio @xmath162 . a further prediction of the theory regards inter - response intervals that , is the time elapsing between one recalled item and the next whose measurement in free - recall experiments dates back to ( murdock and okada , 1970 ) . during retrieval from long - term memory , it was shown ( gruenewald and lockhead , 1980 ) that clusters occur due to stable semantic associations between objects : a subject who is asked to list some animals , for instance , may recall first a set of farm animals , then a number of house pets , then several birds . the inter - response intervals are shorter within clusters than between clusters . the retrieval of examples in the experiment of gruenewald and lockhead depends entirely on the long - term representation of items . in the situation we have described , on the contrary , short - term memory is at play , and the retrieval process has to locate the vanishing traces of a recent experience . nonetheless , it is easy to see that the time interval elapsing between the retrieval of two memories will be longer between two memories belonging to different clusters , and shorter for memories belonging to the same cluster . hence , the same is true among items of the list that are successfully verbalized . the inter - response intervals will be longer for the consecutive recall of two words belonging to different segments of the list , and shorter for the consecutive retrieval of two words belonging to the same segment . as mentioned in the introduction , the wle has been reported in experiments where the time interval between the presentation of consecutive items was a controlled parameter . experiments on the wle with rapid presentation of the stimuli were first performed by coltheart and langdon ( 1998 ) , who found the wle by presenting an item every 114 ms , every 157 ms , and every 243 ms . in ( campoy , 2008 ) , somewhat lower presentation rates were used ( between 300 and 400 ms ) , and again the persistence of the effect was proven over different rehearsal times . here , i will argue that such experiments may offer an ideal tool to study the crossover between the `` diffusive '' and the `` clustering '' regime . indeed , if we modify the foregoing computation to allow the system to random - walk on its own for a time @xmath355 between the presentation of the @xmath73-th and @xmath356-th items , this will be equivalent to increasing the average distance @xmath357 travelled between the memory @xmath133 generated by word @xmath43 and the memory @xmath358 generated by word @xmath358 . this amounts to rescaling time while replacing @xmath195 and @xmath194 with larger effective distances . the matrix elements of @xmath283 will decay faster as their indices grow . therefore , the system will move closer to the diffusive regime . on the other hand , the theory predicts ( section iiie ) that the curve @xmath359 will become linear in the fast - diffusion regime . hence , by reducing the presentation rate , one should see the two curves in fig . 1 becoming progressively linearized , at least up to values of @xmath355 so large that not only the clustering , but also the contiguity effect breaks down . in this limit , moreover , eq . ( [ fast ] ) predicts that the curves @xmath3 and @xmath2 will become parallel . other ways of controlling the crossover between clustering and diffusing regime ( e.g. by pharmacological means , or through distractor tasks such as those of bjork and whitten , 1974 ) can be similarly applied . we have just examined some ways of testing the theory that are going to require ad - hoc experiments . let us conclude by mentioning one type of test that may be performed on already available databases namely , data from the experiments performed so far on the classical wle . it is known that different people employ different strategies in order to exploit the structure of semantic space during memorization and recall ( healey et al . , 2014 ) . when it comes to the clustering described above , we expect its strength to be just as dependent on the subject considered . on the other hand , the strength of the classical wle is also a function of the particular subject . the difference between the recall probabilities of all - short lists and all - long lists , while being positive on average , will vary in magnitude from subject to subject . if the mechanism at the root of the of the wle is indeed a clustering phenomenon , we expect that it will be stronger for subjects for whom the clustering is stronger . let @xmath310 be again the serial positions of the words that have been recalled by the subject during a certain trial . the average serial - position jump between consecutively recalled words for this trial is @xmath360 . for each subject , we can define @xmath361 as the average of @xmath362 over all trials performed on him or her . the parameter @xmath361 may serve as a simple measure of the subject s inclination toward clustering . indeed , @xmath361 is close to unity for subjects strongly inclined toward clustering , and @xmath363 if the subject s tendency toward clustering is low . one way of gaining insight into the likelihood of the theory consists in correlating @xmath361 with the strength of the classical wle effects , controlled by @xmath364 , where @xmath219 and @xmath365 are the recall probabilities calculated for pure lists of short and long words . a negative correlation between these two variables would be a strong confirmation that the wle is indeed due to a clustering phenomenon . i have proposed a theory of verbal perception , extracted some of its properties , and validated it through a comparison with free - recall data . the theory is based on the notion that a word does not have , in general , a single meaning . a human subject exposed to a stream of verbal input will decide on the meaning of each new word on the basis of both the structure of its vocabulary and the meaning he / she has given to the words preceding it . this also applies to a list of random words , because our mind strives to interpret them as parts of a meaningful discourse . it may be instructive to think of such discourses as `` narratives '' . common experience tells us that a two - word list is already capable of creating a strong narrative sense ( e.g. : picnic , lightning ) . when a word in the list has no semantic connection to the context created by the words preceding it , the mind perceives a `` change of scenery '' and assumes that a _ new _ narrative is beginning . a list of words is thus perceived as a collection of distinct `` stories '' . when prompted to recall the list , the subject remembers each story as a separate experience , and needs to re - experiences a given story before retrieving the words responsible for creating it . words that have specific meanings have obviously less probability of fitting into a randomly generated story . otherwise said , the words most likely to break the narrative are those with the highest level of localization in semantic space . we have argued that this correlates positively with word length . hence , a list of @xmath137 long words is likely to break into as many one - word stories , whereas a list of short words is more likely to be perceived as a single continuous narrative . since a single narrative is easier to recall than many unrelated ones , the standard word length effect ensues . the clustering property of short words is at play in mixed lists as well . but its effect is hindered by the presence of long words breaking the narrative . as our analysis has shown , this can lead to an inversion of the wle , encountered in experimental observations . in this scenario , the behavior depicted by figure 1 becomes quite logical . by replacing a short word with a long word , one splits the list into a larger number of narratives , which makes every single word in the list ( whether short or long ) harder to reach during the retrieval process . the interplay between the trajectory of the system during the presentation of lists and the trajectory during the memory test produces a nontrivial spectrum of behaviors , highly dependent both on the structure of lists and on the amount of `` diffusion '' that interferes with clustering . a telltale symptom of these mechanisms is that a short word is more likely than a longer one to be recalled right after a word contiguous to it within the list . an analysis of data from the peers experiments ( healey and kahana , 2016 ) confirms this prediction . another prediction of the theory is that the serial - position jumps performed during recall will be enhanced by clustering for shorter words , and suppressed for longer ones . this is also confirmed by an analysis of peers data , across the whole available spectrum of jump lengths . several directions stand open for experimental and theoretical work on this model . experimentally , further tests may be obtained by performing the five types of measurements listed in section v. open directions for theoretical work include : 1 . generalizing the recall probability formulas to the case where the word lengths available are more than two ; 2 . including possible competition between words for the verbalization of a given state ; 3 . singling out extra effects from the fluctuations around the mean field behavior ; 4 . accounting for primacy and recency effects ; 5 . applying the same technique to predicting the genesis of false memories . finally , by positing a suitable mechanism for spontaneous language production , it would be useful to derive equations linking the underlying word structure to emerging verbal patterns , thus providing a direct link between the hidden variables and the observables of the model . i am grateful to michael j. kahana and his university of pennsylvania research group for making their raw data publicly available . i would like to thank heartily people at the neurotheory center of columbia university ( ken miller , misha tsodyks ) for hosting me there in the fall of 2015 and for responding with enthusiasm to this theory . thanks also to yashar ahmadian of the university of oregon , for reading an early draft of the paper and suggesting several improvements . aurenhammer f. , klein r. , and lee d.t . , voronoi diagrams and delaunay triangulations . singapore : world scientific publishing company . bhatarad p. , ward g. , smith j. hayes l. , 2009 . examining the relationship between free recall and immediate serial recall : similar patterns of rehearsal and similar effects of word length , presentation rate and articulatory suppression . memory and cognition 37 : 689 - 713 . friendly m. , franklin p.e . , hoffman d. , rubin d.c . , 1982 . the toronto word pool : norms for imagery , concreteness , orthographic variables , and grammatical usage for 1,080 words . behavior research methods and instrumentation , vol . 14(4 ) , 375 - 399 . lohnas , l. j. and kahana , m. j. ( 2013 ) . parametric effects of word frequency effect in memory for mixed frequency lists . journal of experimental psychology : learning , memory , and cognition , 39 , 19431946 . j. , heli u. , and elts j. , 2001 . word length as an indicator of semantic complexity , in _ text as a linguistic paradigm : levels , constituents , constructs _ , festschrift in honour of ludek hrebcek . trier , 187 - 195 . neath i. , brown g. d. a. , 2006 . simple : further applications of a local distinctiveness model of memory , in the psychology of learning and motivation , ed . ross b. h. , editor . ( san diego , ca : academic press ) , 201 - 243 .

neutron star magnetic fields may have polar caps ( pc ) that are offset from the dipole axis , through field - line sweepback near the light cylinder or non - symmetric currents within the star . the effects of such offsets on electron - positron pair cascades are investigated , using simple models of dipole magnetic fields with small distortions that shift the pcs by different amounts or directions . using a monte carlo pair cascade simulation , we explore the changes in the pair spectrum , multiplicity and energy flux across the pc , as well as the trends in pair flux and pair energy flux with spin - down luminosity , . we also give an estimate of the distribution of heating flux from returning positrons on the pc for different offsets . we find that even modest offsets can produce significant increases in pair multiplicity , especially for pulsars that are near or beyond the pair death lines for centered pcs , primarily because of higher accelerating fields . pair spectra cover several decades in energy , with the spectral range of millisecond pulsars ( msps ) two orders of magnitude higher than for normal pulsars , and pc offsets allow significant extension of all spectra to lower pair energies . we find that the total pc pair luminosity is proportional to , with for normal pulsars and for msps . remarkably , the total pc heating luminosity for even large offsets increases by less than a factor of two , even though the pc area increases by much larger factors , because most of the heating occurs near the magnetic axis .

1111.1668

electron - positron pair production by rotation - powered pulsars is believed to be essential to the generation of charged particles in their magnetospheres and winds , as well as to the generation of their coherent pulsed radio emission . pairs can be produced through electromagnetic cascades initiated by accelerated particles and several sites of acceleration have been proposed . near the neutron star surface above the magnetic polar caps ( pcs ) , electrons may be accelerated either in vacuum gaps ( ruderman & sutherland 1975 ) or in space - charge limited flow gaps ( arons & scharlemann 1979 ) . the gamma - ray photons radiated by the electrons will convert to electron - positron pairs by one - photon pair production in the strong magnetic field near the surface , and the pairs radiate synchrotron photons that produce further generations of pairs ( daugherty & harding 1982 ) . in outer gap vacuum accelerators , extending to near the light cylinder , seed photons can produce pairs through interaction with thermal x - ray photons from the hot neutron star surface to create a first generation of particles that accelerate through the gap radiating gamma - ray photons that produce further pairs ( cheng , ho & ruderman 1986 ) . to date , all of the simulations of pulsar pair cascades have assumed pure dipole magnetic fields both to derive the accelerating electric field parallel to the magnetic field and to calculate the spectrum and multiplicity of the pairs . the configuration of neutron star ( ns ) magnetic fields is currently an unsolved problem . the original solution for a rotating magnetic dipole in vacuum ( deutsch 1955 ) shows that retardation causes the field lines near and outside the light cylinder to sweep back in a direction opposite to that of rotation , transitioning to an electromagnetic wave at large distances . the magnetic field sweepback causes the boundary of the open field volume , and the footpoints of field lines on this boundary that define the pc , to shift relative to the magnetic axis opposite to the rotation direction ( arendt & eilek 1998 , dyks & harding 2004 ) . the amount of the shift depends on the magnetic field inclination to the rotation axis , with the largest offsets occurring for oblique rotators . in the last decade , numerical solutions of force - free magnetosphere models assuming ideal mhd ( @xmath4 ) ( contopoulos et al . 1999 ) show an even larger sweepback of the magnetic field near the light cylinder as it transitions to an mhd wind ( spitkovsky 2006 , timokhin 2006 ) , as currents cause significant distortions of the magnetic field . the open field boundary and the pc of a force - free magnetosphere thus have a larger offset than the vacuum magnetosphere . since the vacuum magnetosphere can accelerate particles but is not loaded with charges or currents and the ideal - mhd magnetosphere has charges and currents but can not accelerate particles , a real pulsar magnetosphere lies between these two solutions . in addition to the guaranteed distortions of the dipole field caused by retardation and currents , it is possible that there are also distortions due to multipolar components near the neutron star surface . the emission from millisecond pulsars in particular shows evidence of non - dipolar field structure . the thermal x - ray pulse profiles of some millisecond pulsars ( msps ) show asymmetries which have been successfully modeled by including offsets of the emitting hot spot on the neutron star surface . since the emission likely originates from pc heating , it is argued that msps such as psr j0437 - 4715 ( bognanov et al . 2007 ) and psr j0030 + 0451 ( bogdanov & grindlay 2009 ) have either offset dipoles or offset pcs . modeling of the x - ray pulsations of the nss in low - mass x - ray binaries , the likely progenitors of millisecond pulsars , show possible evidence of even more extreme magnetic field distortions ( lamb et al , 2009 ) , that could result from distortion of the global magnetic field by e.g. the crustal plate tectonics ( ruderman 1991 ) . we have recently shown ( harding & muslimov 2011 , hm11 ) that distortions of a magnetic dipole field that produce offset pcs lead to an increase in the multiplicity of pc pair cascades and to significant shifts of the death line for pair production of curvature radiation photons . we found that even offsets that are small fractions of a neutron star radius can enable high - multiplicity pair cascades in older pulsars that with pure dipole fields would be pair starved . there were a number of previous studies of how multipolar or non - dipolar neutron star fields could increase pair production ( ruderman & sutherland 1975 ; arons 1983 , 1997 ; asseo & khechinashvili 2002 ) , but all of these focussed on the decrease in field line radius of curvature as the primary effect . arons ( 1997 ) , considering the decrease in field line radius of curvature provided by an offset dipole , found that offsets that are large fractions of a stellar radius , ( 0.7 - 0.8)@xmath5 , are required to move the pair death line enough to include the entire radio pulsar population . medin & lai ( 2010 ) proposed an even larger offset of 0.95 @xmath5 , again only considering the decreased field line radius of curvature . kantor & tsygan ( 2003 , 2004 ) considered the space - charge limited flow ( sclf ) @xmath6 solution of a dipole field with the addition of a displaced dipole of smaller magnitude , that effectively produces narrow , bent flux tubes . they take into account both the increase in @xmath6 and decrease in radius of curvature to derive lower death lines for pair production , and find that displacements of @xmath7 for the smaller dipole component can significantly change the death lines for both curvature radiation ( cr ) and inverse compton scattering ( ics ) pair production . our study also takes into account the increase in @xmath6 and particle acceleration energy , which is by far the strongest effect on pair multiplicity of offset dipoles or pcs . the study of hm11 was based on derivation of the electric field parallel to the magnetic field near the pc assuming there is free flow of particles from the neutron star surface , the sclf flow boundary condition ( arons & scharlemann 1979 ) . the resulting current density of primary charges in this case is equal to @xmath8 where @xmath9 is the goldreich - julian charge density . this current distribution differs from the current density distribution demanded by the global ideal - mhd models , so it is presently not clear whether the steady sclf acceleration model will be compatible with a global pulsar magnetosphere . however , more realistic global magnetosphere models with resistivity and dissipation are in development ( li , spitkovsky & tchekhovskoy 2011 ) . the current density distributions of such pulsar magnetospheres with dissipation are significantly different from those with ideal - mhd ( kalapotharakos et al . 2011 ) , with the ability to possibly adjust to the @xmath10 of the pc cascades they rely on for their charge supply . furthermore , timokhin & arons ( 2011 ) have performed particle - in - cell simulations of pc pair cascades and show that sclf acceleration is stable for currents within a 10% range of @xmath11 . in this paper we extend the study of hm11 to include a more general expression for the distorted magnetic field , treating two cases in which the caps from opposite magnetic poles are offset in the same ( symmetric ) or opposite ( asymmetric ) directions . in section 2 , we give the expressions for the magnetic fields in these two cases , as well as the corresponding goldreich - julian charge density , field line equations and parallel electric fields . section 3 describes the pair cascade simulation used to derive the results for pair multiplicity and flux distributions over the pc , and pair spectra given in section 4 . we also examine the changes in the distribution of pc heating luminosity with offset pcs as well as the total pair and heating luminosity . derivation of a general non - dipolar magnetic field of a neutron star using a multipole expansion gives quite complex expressions ( e.g. asseo & khechinashvili 2002 ) . because of the mathematical complexity , this kind of model representation of the magnetic field is impractical . since our main purpose in this paper is to derive an expression for the accelerating electric field near the neutron star surface , we require field expressions from which we can derive analytic formulae for charge density for input to poisson s equation for the electric potential . we therefore propose the following heuristic model of a non - dipolar magnetic field that provides distortions leading to offset pcs . we introduce an azimuthal asymmetry to the field lines of an originally symmetric dipole such that the field lines over half of the pc have relatively smaller radius of curvature and over the other half of the pc have larger radius of curvature . this can be done by adding an azimuthal dependence , controlled by the parameter @xmath12 , to the magnetic colatitude of @xmath13 and @xmath14 , as in equations ( 1 ) and ( 2 ) . then , from the solenoidality condition , @xmath15 , we can derive the azimuthal component of the distorted magnetic field , which is proportional to the parameter @xmath16 . the resulting exact expression for the azimuthal component is rather cumbersome and can be simplified by expanding the appropriate terms in @xmath16 . in our study , for the sake of simplicity , we retained only the lowest - order ( linear ) terms in @xmath16 . note that equations ( 1 ) and ( 2 ) can be used to achieve a significant distortion of the field lines while keeping the dipole axis inside the offset polar cap boundary , which significantly facilitates the subsequent analytical treatment . thus , equations ( 1 ) and ( 2 ) represent global open field lines of a centered dipole that are pushed and bent toward the dipole axis on one side and pushed and bent away from the dipole axis on the other side of the polar cap , respectively ( see figure 1 ) . consequently , one side of the pc is larger and the pc is effectively shifted from the center of symmetry . _ case a : ( the magnetic field is symmetric w.r.t . the magnetic equator ) _ in this case , in magnetic spherical polar coordinates ( @xmath17 , @xmath18 , @xmath19 ) the magnetic field reads . [ b1 ] _ case b : ( the magnetic field is asymmetric w.r.t . the magnetic equator ) _ in this case the magnetic field can be presented as , [ b2 ] where @xmath20 is the surface magnetic field strength at the magnetic pole , @xmath21 is the dimensionless radial coordinate in units of stellar radius , @xmath5 , @xmath22 is the parameter characterizing the distortion of polar field lines , and @xmath23 is the magnetic azimuthal angle defining the meridional plane of the offset pc . we must point out that @xmath24 as presented by formula ( [ b2 ] ) is solenoidal only approximately , since in the @xmath25-component we have neglected corrections of order and higher than @xmath26 and in @xmath13- and @xmath27-components we have neglected the corrections of order and higher than @xmath28 and @xmath26 , respectively . figure 1 shows the field lines projected in the x - z plane for cases a and b. the corresponding goldreich - julian charge density is _ case a : _ @xmath29 \sin \chi \sin \theta \cos \theta \right\ } , \label{rhogj1}\end{aligned}\ ] ] _ case b : _ @xmath30\sin \chi \sin \theta \right\ } , \label{rhogj2}\end{aligned}\ ] ] where @xmath31 is the rotation rate of the ns , @xmath32 is the pulsar obliquity , @xmath33 ( @xmath34 g@xmath35@xmath36 , @xmath37 cm , @xmath38 is ns moment of inertia ) is the parameter accounting for the general - relativistic frame dragging ( muslimov & tsygan 1992 ) . if the magnetic axis lies in the x - z plane , @xmath39 to account for an effective offset of the dipole axis in the x - z ( @xmath40 ) plane and @xmath12 for an effective offset of the dipole axis in the y - z plane ( @xmath41 ) , where @xmath42 . by using formulae ( [ b1 ] ) and ( [ b2 ] ) we can get the equations of the magnetic field line , _ case a : _ [ thetaa ] = ^-1 ( x^12(1+a ) ) , [ fl1 ] _ case b : _ [ thetab ] = 1 ^-1 ( f x^12(1+a ) ) , [ fl2 ] respectively , where @xmath43 is the radial distance in units of the light cylinder radius , @xmath44 ; @xmath45 is the colatitude of a footpoint of polar field line normalized by the colatitude of the pc boundary . also , in formula ( [ fl2 ] ) @xmath46 is a parameter ( @xmath47 ) that can be more reliably determined via numerical tracing of the last open field line . for a rough estimate one can derive the analytic expression , f = ^12(1+a ) , [ f ] that is applicable for @xmath48 and can be used to plot the field lines with @xmath49 . the field lines with @xmath50 can be plotted by using the coordinate reflection , @xmath51 and @xmath52 . the field line radii of curvature ( in units of @xmath53 ) are _ case a : _ x_c = 4 x^12(1-a ) ^ 3/2 , [ xc1 ] _ case b : _ x_c = 4 x^12(1-a ) ( 1 - 3f^2/4 ) ^ 3/2 , [ xc2 ] respectively , where = ^2 x^1+a . neglecting the static general - relativistic corrections , we can write the accelerating electric field , assuming the boundary conditions of sclf ( cf . harding & muslimov 1998 ; hereafter hm98 ) . for @xmath54 , where @xmath55 $ ] is the standard pc angle , _ case a : _ [ ell_1 ] e_|| - 12 e_0 x^a , [ epar1 ] _ case b : _ e_|| - 12 ( f ) ^2 e_0 x^a , [ epar2 ] where @xmath56 , and e_1a = e_1b = 1 + 13 a(^3 -1 ) , [ e1 ] e_2a = \ { ( 1+)+ } [ e2a ] and e_2b = f , [ e2b ] where = x^a/2 and = . [ alpha - beta ] now we shall present the expressions for @xmath57 that are applicable at the small altitudes , @xmath58 ( cf . formula [ 19 ] of harding & muslimov 2001 ; hereafter hm01 ) , @xmath59 , \label{epar_small_z}\end{aligned}\ ] ] where @xmath60 and @xmath61 ; @xmath62 and @xmath63 $ ] in case a and b , respectively ; @xmath64\kappa /2 $ ] ; @xmath65 and @xmath66 are the bessel functions , and @xmath67 and @xmath68 are the positive zeros of @xmath65 and @xmath66 , respectively ; @xmath69 and @xmath70 ; and @xmath71 $ ] , where @xmath72 . the expression for the parallel electric field in eqn ( [ ell_2 ] ) above corresponds to the expression for a pure dipole magnetic field given in eqn ( 19 ) of hm01 . the main differences here are inclusion of the azimuthal asymmetry incorporated in the @xmath73 parameter , and the neglect of the static gr corrections . so if one takes @xmath74 , eqn ( [ ell_2 ] ) will be equivalent to eqn ( 19 ) in hm01 with @xmath75 and @xmath76 . note that in formula ( [ epar_small_z ] ) @xmath77 and are related via the field - line equation , ( - _ 0 ) = + z + z ^2 [ ( 1)-_0 ] . [ phi ] we see that the magnetic field configuration of eqns ( 1 ) and ( 2 ) produces an @xmath78 that is significantly larger on the offset side of the pc , by the factor @xmath79 , than the @xmath78 for the case of a pure dipole field . this increase in @xmath78 results from the strong dependence on the field - line curvature of the goldreich - julian charge density in the solution of poisson s equation for the electric potential ( see also kantor & tsygan 2003 ) . this dependence is embodied in the factor @xmath80 in the pure dipole case , and in the factor @xmath81 in the offset pc case , so that the pc boundary is dependent on azimuthal angle . for the case of an offset pc , the effective pc angle , @xmath81 , is larger on one side of the pc , producing a larger @xmath78 , and smaller on the opposite side , producing a smaller @xmath78 . we model the electromagnetic pair cascades above a pulsar pc using a hybrid steady - state acceleration / monte carlo pair cascade simulation . the first stage of the calculation , based on the self - limited acceleration model of hm98 , follows a primary electron that starts at the neutron star surface at magnetic colatitude @xmath82 and azimuth @xmath83 with lorentz factor @xmath84 . an initial guess is made for the height @xmath85 of the pair formation front ( pff ) above the surface and this distance is divided into 500 equal linear steps of length @xmath86 . the pff is the height where the first pair is produced and we assume that the @xmath78 is completely screened above this point . this is a good assumption for several reasons . first , the electric field arises due to a small imbalance between the actual charge density and the local , rotation - induced , goldreich - julian charge density . it therefore does not require much additional charge to short - out this field . second , the onset of pair cascading occurs very quickly ( daugherty & harding 1996 ) , so that the number of pairs produced per primary particle increases rapidly over small distances . thus , as found in arons ( 1983 ) , the width of the pff ( the screening distance of the electric field ) is very small compared to other dimensions of the problem . the electron advances in steps of @xmath86 , gaining energy through electrostatic acceleration and losing energy through radiative losses . the pff results from pairs produced by @xmath87-rays of energy @xmath88 ( in units of @xmath89 ) radiated by particles of energy @xmath90 . the height of the pff is then : [ sc ] s_c = min[s_a(_min ) + s_p(_min ) ] where @xmath91 is the distance required to accelerate the particle until it can radiate a photon of energy @xmath92 , and @xmath93 is the pair attenuation length of the photon . the acceleration distance , @xmath91 , is determined by first integrating the equation of motion of the particle to determine its energy as a function of its pathlength @xmath94 : [ dgamma ] cdds = emce _ - _ _ ic - _ _ cr , where @xmath78 is the electric field induced parallel to the magnetic field , @xmath95 and @xmath96 are the loss rates for ics and cr . expressions for @xmath95 and @xmath96 and details of their derivation are given in hm98 . the pair production attenuation length of photons radiated by the particle through either ics or cr is then determined . the attenuation length @xmath97 , defined to be the path length over which the optical depth is unity , is given by [ tau ] ( ) = _ 0^s_p ( ) t_pp(_kb , ) ds = 1 , where is the pathlength differential along the photon momentum vector , @xmath98 is the attenuation coefficient for one - photon pair production and @xmath99 is the angle between and the local magnetic field direction . the method we use to compute the electron - positron pair attenuation length of the photons , @xmath97 , has been described in detail in harding , baring & gonthier ( 1997 ) . using equation ( [ tau ] ) , @xmath97 is computed by integrating the pair production attenuation coefficient of the photon along its path through the dipole field . the photon is assumed to pair produce at the point where @xmath100 , and @xmath97 is then set to that path length . the two main inputs needed are the energy of the photon and its angle to the magnetic field , @xmath99 . the energies of the radiated photons for the purpose of computing the pff are taken to be ( following hm98 ) @xmath101 for @xmath102 , and @xmath103 for @xmath104 , where @xmath105 is the critical cr energy , @xmath106 is the electron classical radius , @xmath107 g , and @xmath108 is the field line radius of curvature ( eqn [ [ xc1 ] ] and [ [ xc2 ] ] ) . for the angle of the radiated photons at the emission point , we assume that @xmath109 for compton scattered photons ( dermer 1990 ) and @xmath110 for cr photons . the cr and ics photons are assumed to be completely polarized in the parallel mode . due to the curvature of the field lines , @xmath99 will grow as the photon propagates , roughly as @xmath111 . to evaluate the pair production attenuation coefficient at each point along the path of the photon , we lorentz - transform the photon energy and local magnetic field to the frame in which the photon propagates perpendicular to the local field . this is the center - of momentum frame for the created pair , where the attenuation has its simplest form and the photon energy is @xmath112 . the one - photon pair attenuation coefficient is considered in two regimes . for @xmath102 , photon pair produce far above threshold , where the asymptotic expression in the limit of large numbers of kinematically available pair landau states ( tsai & erber 1974 , daugherty & harding 1983 [ dh83 ] ) can be used : [ eq : ppasymp ] t^pp _ , = 12 b _ , ( ) , _ , ( ) \ { lr ( 0.31 , 0.15 ) & 1 + + ( 0.72 , 0.48 ) ^-1/3 & 1 . [ eq : ppratlim ] where @xmath113 , @xmath114 is the fine - structure constant , @xmath115 is the dipole field strength at point @xmath94 along the photon path . when @xmath104 , pair production will occur near threshold , where the above expression is not accurate . we thus include the factor @xmath116 in @xmath117 , introduced by dh83 , as an approximation to the near - threshold attenuation coefficient . in this paper we compute the attenuation length averaged over photon polarization . the path of each input photon is traced through the magnetic field , accumulating the survival probability for pair production , @xmath118 : @xmath119 each photon may pair produce or escape , based on a combination of the running survival probability for pair production . for this simulation , we have neglected photon splitting since pair production dominates the photon attenuation in magnetic fields @xmath120 g ( baring & harding 2001 ) . the first guess " value of @xmath85 , and thus also of @xmath121 , sets the initial acceleration length . equation ( [ dgamma ] ) is integrated in discrete steps upward from the starting point , computing @xmath78 from equation ( [ ell_2 ] ) , @xmath95 and @xmath96 at each step . at each step , the pair attenuation lengths , @xmath97 , of cr test photons radiated by the particle of energy @xmath122 are computed from equation ( [ tau ] ) . the pair attenuation length , and thus the value of @xmath121 , also computed at every step , is initially infinite , because the energy of the photons is small , but decreases with distance as the energy of the radiated photons increases . although the photon attenuation length continues to decrease , the particle acceleration length is increasing and @xmath121 has a minimum . this minimum value of @xmath123 is adopted as the new height of the electron pff . the electron is accelerated again with the new value of @xmath85 , producing a new pff at the next value of @xmath123 . the process is repeated , converging to a self - consistent value of @xmath85 . once the pff height is established , the full pair cascade spectrum is computed using the monte carlo simulation , adopting the electron lorentz factor as a function of height , @xmath124 , from the self - consistent pff calculation for the acceleration region from @xmath125 to @xmath126 . above the pff , it is assumed that @xmath127 due to pair screening ( hm01 ) , so that the electrons lose energy every step due to cr and ics . the step size for @xmath128 is limited to less than the distance over which they would lose 10% of their energy . each step , the electron radiates a cr spectrum , divided into a number of logarithmically spaced bins . the number of cr photons in each energy bin , @xmath129 , is determined by the energy loss rate and average energy in that bin . a representative photon from each bin having the average bin energy is followed to its pair conversion or escape point , as described above . the pair produced by the photon , or the escaping photon number , is then weighted by @xmath129 . if a photon pair produces , the total energy , landau state and parallel momentum of the electron and positron are determined . each member of the pair is assumed to have half the energy and the same direction of the parent photon . each member of the pair occupying an excited state emits a sequence of cyclotron or synchrotron photons . the method used to simulate the cyclotron / synchrotron emission is similar to that of daugherty & harding ( 1996 ) . if the particle landau level is larger than 20 , the high - energy limit of the quantum synchrotron transition rate ( sokolov & ternov 1968 ) is used , in which case we assume that the photons are emitted perpendicular to the magnetic field in the particle rest frame ( high - energy limit ) . when the particle landau level is smaller than 20 , the exact qed cyclotron transition rate ( harding & preece 1987 ) is used , in which case the angles of the emitted photons are sampled from a distribution . in both cases , the emitted photon polarizations are sampled from the corresponding polarization distributions . each emitted photon is propagated through the magnetic field from its emission point until it pair produces or escapes . the cyclotron / synchrotron emission sequence continues until each particle reaches the ground state . by use of a recursive routine that is called upon the emission of each photon , we can follow an arbitrary number of pair generations . the cascade continues until all photons from each branch have escaped . the cascade pairs are binned in energy and the magnetic colatitude @xmath82 and azimuth @xmath83 on the pc of the initial test electron . for a given set of input parameters that include pulsar period @xmath130 , surface magnetic field @xmath20 , inclination angle @xmath117 , and offset degree @xmath16 , direction @xmath23 and symmetry ( case a or b ) , the cascade calculation produces the pair spectrum per primary electron as a function of @xmath82 and surface @xmath83 , @xmath131 , where we will express the pair energy @xmath132 in units of @xmath89 . the pair multiplicity , the number of pairs per primary electron , as a function of @xmath82 and @xmath83 across the pc is [ m+ ] m_+ ( , ) = _ e_min^e_max dn_+(e , , ) de de while the total pair multiplicity from each pc is [ m+pc ] m^pc_+ = _ 0 ^ 2 d_0 ^ 1 m_+ ( , ) _ pc^2r_ns^2d . here , the pc angle varies with @xmath83 , [ sinthetapc ] _ pc = ( r_ns / c)^(1+a)/2 . which comes from setting @xmath133 and @xmath134 in eqn ( [ thetaa ] ) . the total flux of pairs emerging from each pc is [ n+ ] n_+ ( , ) = _ e_min^e_max dn_+(e , , ) de de where [ dotdnde ] dn_+(e , , ) de = dn_+(e , , ) de n_p ( , ) and where @xmath135 is the primary flux over the pc , [ np ] n_p ( , ) = n_gjc r_ns^2 ^2_pc . where @xmath136 is the goldreich - julian number density at the ns surface . in eqn ( [ np ] ) , we use the goldreich - julian charge density given in eqns ( [ rhogj_a ] ) and ( [ rhogj_b ] ) . similarly , the energy flux of pairs from each pc is e_+ ( , ) = _ e_min^e_max dn_+(e , , ) de ede we have computed @xmath131 for a range of parameters to explore the distributions of @xmath137 , @xmath138 and @xmath139 across the pc . our results are dependent on neutron star equation of state ( e0s ) through the @xmath6 dependence on the frame - dragging , specifically through the @xmath140 parameter , and on the neutron star radius , through the pc angle . there have been a number of measurements of neutron star mass for msps that give values well above the canonical @xmath141 and as high as @xmath142 ( demorest et al . we have therefore assumed different eos for normal ( non - recycled ) pulsars and msps . for non - recycled pulsars , we take @xmath143 and radius @xmath144 km , which give a moment of inertia @xmath145 ( lattimer & prakash 2007 ) . we take a rotating ns model ( friedman et al . 1986 ) for msps with @xmath146 , @xmath147 km and @xmath148 . examples of the pair multiplicity and pair flux distributions over the pc are shown in figures 2 and 3 , for a pulsar with @xmath149 s and @xmath150 g , and for case b with @xmath151 for different values of the offset degree @xmath16 . a pulsar with these parameter values lies above but near the cr pair death line for @xmath152 , in which case the cascade produces a modest pair multiplicity . as shown in figure 2 , the maximum pair multiplicity @xmath137 occurs for @xmath153 and is nearly symmetric in @xmath83 , with only a small asymmetry caused by the @xmath83 dependence of @xmath6 . as @xmath16 increases , the peak pair multiplicity increases but only in a small region of the pc toward the offset , in this case at @xmath154 , while the multiplicity decreases in the other side of the pc in the @xmath155 direction . for larger offsets , the region of high multiplicity expands as the pc boundary expands in the direction of the offset . the distribution of pair flux for this same case , shown in figure 3 , is somewhat different from the distribution of pair multiplicity . for @xmath76 , @xmath139 peaks in a small ring around the center of the pc and decreases outside the ring , again with a small asymmetry due to the @xmath83 dependence of @xmath6 . the concentration of @xmath139 closer to the pc center is due to the fact that the lorentz factor of the primary electron is maximum near the magnetic axis because the decrease in radius of curvature of the field lines raises the altitude of the pff ( see figures 4 and 5 of hm98 ) . for @xmath156 , the pair energy flux increases and becomes asymmetric , but with a greater concentration toward the pc center than the multiplicity . figure 4 shows the distribution of pair multiplicity @xmath137 for the case of a millisecond pulsar with @xmath157 ms , @xmath158 g , and for case a with @xmath159 for @xmath16 between 0 and 0.6 . these parameters place the pulsar below the cr pair death line with very small pair multiplicity for @xmath76 , although the distribution of @xmath137 can be seen to show a more pronounced asymmetry toward the ` favorably curved ' field lines at @xmath160 than in the case of the longer period pulsar shown in figure 2 . for pulsars with shorter periods , the second term in @xmath6 depending on @xmath161 is relatively larger since the value of @xmath162 is larger . as @xmath16 increases , the peak pair multiplicity again increases in a small region of the pc toward the offset , but with a stronger asymmetry toward the @xmath160 side of the pc . for millisecond periods there is also less increase of ellipticity of the pc with increasing @xmath16 . the period dependence of the pc ellipticity can easily be seen from theh expression eqn ( [ sinthetapc ] ) for the pc angle . if we define the elongation factor @xmath163 , then @xmath164 is higher for smaller @xmath162 . figure 5 illustrates the effect of inclination angle on the distribution of pair multiplicity and pair flux for a millisecond pulsar with @xmath165 ms , @xmath166 g , and for case a with @xmath159 and @xmath167 . @xmath137 peaks in roughly the same region of the pc for @xmath168 and @xmath169 , but the asymmetry toward favorably curved field lines is is stronger in the case of @xmath169 . both the maximum multiplicity and flux are higher for @xmath169 . the values of peak multiplicity , @xmath170 , for the radio pulsar population in the @xmath130-@xmath171 plane are shown in the contour plots in figures 6 - 8 . figure 6 shows contours of @xmath172 in the case of a centered pc ( @xmath76 ) , where we have assumed the different neutron star eos for normal ( non - recycled ) and msps described above . it can be seen that @xmath173 is high for short periods and higher @xmath171 , reaching a maximum of a few times @xmath174 , and drops sharply towards the death line even on a log scale . about half of the normal pulsar population and most of the msp population lies below the death line , a result that has been noted in a number of previous papers ( e.g. arons 1997 , zhang , harding & muslimov 2000 , hibschman & arons 2001 , harding & muslimov 2002 ) . in figure 7 and 8 , we show contours of @xmath172 for offset pcs with @xmath167 having symmetric ( case a ) and asymmetric ( case b ) offsets . in both symmetric and asymmetric offsets , the regions of high pair multiplicity spread to lower @xmath171 and longer periods . the increases in @xmath173 are most dramatic for that part of the population near and below the pair death line having very low @xmath173 in the centered pc case . the pair death line thus moves down through nearly the entire population of both recycled and non - recycled pulsars for an offset of @xmath167 . for pulsars with high @xmath171 , @xmath173 changes much less with even a large offset , with multiplicity saturating below @xmath175 . this saturation of @xmath173 , noted by hm11 , is caused by several effects . even in the centered pc case , when the magnetic field increases above @xmath176 g pairs are increasingly produced in the low - lying landau states ( baring & harding 2001 ) , which results in fewer synchrotron photons and fewer generations in the cascade . for increasing pc offsets , the resulting increase in accelerating field and particle energy initially produces higher @xmath173 pair cascades , but as the increasingly energetic particles produce higher energy cr photons , the pairs are produced at smaller angles to the magnetic field . this results in higher average pair energies with pairs in lower landau states , further reduces the number of synchrotron photons . there are some significant differences in the @xmath172 contours for cases a and b. the death line in the normal pulsar population is lower for case b and overall values of @xmath173 are a bit higher . for msps , the case b death line moves up for very short periods . the death lines for both @xmath76 and @xmath167 are flatter for the shorter period msps and the death lines at all periods are flatter as for @xmath167 . as noted by hm11 , particle acceleration in pulsars that require a large fraction of the open field voltage to produce pairs is limited by cr reaction . in this regime , the electron lorentz factors to not reach the full voltage drop between the neutron star and the pff , but instead reach a lower steady - state lorentz factor as the acceleration is balanced by the cr loss rate . even for centered pcs , acceleration in msps operates completely in the cr reaction limit ( luo et al . 2000 , harding & al . 2002 ) . the total spectrum of cascade pairs , integrated over the whole pc , is [ n+pc ] dn_+(e)de = _ 0 ^ 2 d_0 ^ 1 dn_+(e , , ) de _ pc^2r_ns^2d . figure 9 shows the integrated pc pair spectra for @xmath177 g and a range of periods typical of non - recycled pulsars , for centered and offset pcs . the spectra exhibit turnovers at low and high energies that depend on period . for centered pcs , the low - energy turnovers occur around lorentz factor @xmath178 and the spectra extend up to @xmath179 , with the range increasing for shorter @xmath130 . for offset pcs , the spectra extend to lower energies , by about a decade for @xmath167 . thus the total pc pair flux increases because pairs can be produced at lower energies . pair spectra for parameters typical of msps are shown in figure 10 . for both centered and offset pcs , the msp pair energies are much higher that those of normal pulsars , by a factor of about 100 . the spectra extend from a low - energy turnover at @xmath180 to a high energy cutoff around @xmath181 for @xmath76 , and from @xmath182 to @xmath183 for @xmath184 . thus the highest - energy pairs are nearly as energetic as the primary electrons ! the large difference between normal and msp pair spectra is due to the difference in field strengths . in lower fields , photons must pair produce at higher energies , decreasing the relative @xmath173 , even though the larger pc sizes of msps give smaller curvature radii . but as is the case for normal pulsars , the increase in @xmath173 for offset pcs results from lower possible pair energies . the pair flux emerging from each pc is [ ndotpair ] n_pair = _ e_min^e_max dn_+(e)de de which we examine as a function of pulsar spin - down luminosity , @xmath0 . figure 11 plots the pc pair flux vs. @xmath0 for the non - recycled pulsar population for different degrees of offset . the lines are least - squares fits to the points for each case , resulting in the following approximate expressions : @xmath185 where @xmath186 . the pair flux is closely proportional to spin - down luminosity for @xmath76 , with the dependence somewhat flattening for offset pcs due to a saturation at high @xmath0 . figure 12 shows the equivalent plot for msps with the least - squares fits giving the following relations between @xmath187 and @xmath0 : @xmath188 the total pair luminosity from each pc is [ lpair ] l_pair = _ e_min^e_max dn_+(e)de ede plots of the pc pair luminosity vs. @xmath0 for non - recycled pulsars are shown in figure 13 , with the least - squared fits yielding the relations : @xmath189 the pair luminosity seems to also be roughly proportional to spin - down luminosity , with the dependence flattening for higher @xmath16 , with the efficiency for pair luminosity being @xmath190 . the @xmath1 vs. @xmath0 for msps is shown in figure 14 , with the corresponding relations from the least - squares fits : @xmath191 the pair luminosity efficiency for msps is much higher than for non - recycled pulsars , with @xmath192 , about an order of magnitude higher . since msps have much lower surface magnetic field strengths , they need to use a higher fraction of the total open - field line voltage to produce pairs , resulting in a higher @xmath193 . msps also have higher @xmath87-ray luminosity efficiencies ( abdo et al . 2009 ) , because the primary particles which radiate the highest energy emission reach higher energies before pairs screen the accelerating field . pair cascades from particle accelerators near the pc will inevitably produce backflowing particles in the process of pair screening of the electric field . the backflowing particles will accelerate through the same potential drop as the primary particles and deposit this energy on the neutron star surface , increasing the surface temperature of the pc . the radiation from the hot pc is predicted to be in the x - ray band ( arons 1981 , hm01 ) , and observations have provided evidence for emission from hot pcs , especially in the case of middle - aged and millisecond pulsars ( zavlin 2007 ) . the predicted x - ray luminosity from heated pcs in the case of sclf accelerators with centered pcs ( hm01 ) roughly agrees with observed luminosities . with the possibility of offset pcs increasing the pair multiplicity near the neutron star surface , it is important to check that the returning positron luminosity does not overheat the pc and violate the x - ray luminosity constraints . for this estimate , we do not perform a detailed screening calculation with pair dynamics to derive the screening scale length , as in hm01 . rather we will compute the maximum returning positron fraction which will give a conservative limit on the positron heating luminosity . the maximum fraction of returning positrons from the pff can be estimated as ( see formula [ 33 ] in hm01 ) = 12 , [ rho_plus ] where @xmath194 is the pff dimensionless altitude and @xmath195 , the primary charge density at the pff , is derived by inserting @xmath196 inside the curly brackets in eqns ( [ rhogj_a ] ) and ( [ rhogj_b ] ) and @xmath197 outside the brackets . the heating flux over the pc from returning positrons can then be estimated as c _ + ( z__pff , , ) ( z__pff , , ) where @xmath198 is the primary electron energy at the pff on the field line with surface @xmath82 and @xmath83 . the reason that we use the the primary energy instead of the potential drop between the surface and the pff to estimate the energy of the returning positrons at the surface is that for some cases the particles become radiation - reaction limited before reaching the pff , so that the potential drop would be an overestimate . figure 15 shows the distribution of positron heating flux over the pc , @xmath199 , for the case of a non - recycled pulsar with @xmath200 s and @xmath150 g , for different @xmath16 values , while figure 16 shows results for a msp with @xmath157 ms and @xmath201 g. it is evident that most of the heating flux is concentrated at the center of the pc near the magnetic axis near @xmath202 , the contrast between the pc center and edges being about a factor of ten . the returning positron fraction is maximum both near @xmath202 and @xmath133 , where the pff is at the highest altitude . however , the @xmath198 is maximum near @xmath202 , causing a maximum heating flux near the pc center . for @xmath156 , most of the offset side of the pc is heated with lower flux than near the pc center . there is also an asymmetry such that the favorably curved field lines have higher heating flux , also because of higher returning positron energies . the substantially lower heating in the outer part of the offset side of the pc , which has a much larger area , keeps the increase in total pc heating due to the offset to a minimum . in fact in figure 15 and 16 , the physical size of the hottest area is comparable for all values of @xmath16 that we examined . thus the most intensely heated area is several times smaller than the canonical pc size , even for large @xmath16 values . in the case of the msp in figure 16 , this effectively smaller heated spot is shifted from the pc center by a smaller fraction of the canonical pc radius than the shift of the whole pc . for the non - recycled pulsar in figure 15 the heated area remains nearly centered while the whole pc is shifted by a larger fraction . the total heating power ( of precipitating positrons ) can be estimated as ( cf . formula [ 61 ] of hm01 ) l_+ c _ s(z__pff ) _ + ( z__pff , , ) ( z__pff , , ) ds , [ ] where the integration is over the area of a sphere cut by the polar flux tube at the radial distance @xmath203 . figure 17 shows the dependence of @xmath204 , as a fraction of spin down luminosity , on pulsar characteristic age , @xmath205 , for different period and offsets . comparing to the results in fig . 7 of hm01 for the @xmath76 case , the @xmath206 here is higher by factors of 2 - 5 since the values we have given here are the upper limits on positron heating rate , while the hm01 results are from numerical computations of the @xmath6 screening . our estimates here nevertheless serve to test the heating of offset pcs against the observational constraints . in the case of both normal pulsars and msps , the increase in heating for @xmath207 is very minimal . even for @xmath167 , the heating luminosity increases by no more than a factor of 2 , and for msps there is almost no increase except in cases where there are different degrees of pair screening for increasing offsets ( e.g. @xmath157 ms at large ages ) . thermal components have been detected from both middle - aged pulsars and msps in the soft x - ray band . in the case of middle - aged pulsars like psr b1055 - 52 and psr b06556 + 14 , typically both hot and cool thermal components appear in the spectra . the hot components , with luminosity @xmath208 , may be due to pc heating and the cool components from neutron star cooling , but in any case hot components provide a limit to any theoretical heating of the pc . for these pulsars , the pc heating efficiency @xmath209 for b0656 + 14 , and @xmath210 for b1055 - 52 ( deluca et al . 2005 ) , which are within a factor of 2 of the our estimated maximum heating efficiencies . in the case of msps , several have measured thermal components , including psr j0437 - 4715 , j2124 - 3358 and j0030 + 0451 . since these pulsars are too old to have significant cooling , their thermal emission is very likely from pc heating . their measured pc heating efficiencies , @xmath211 ( zavlin 2007 ) are a factor of 2 - 3 lower than the maximum estimated efficiencies in figure 17 . we have examined the change in magnetic flux over the pc as @xmath16 increases . for an offset of @xmath207 for a 0.1 s pulsar , the magnetic flux on the offset side of the pc is typically a factor of 2 higher than for the centered pc , while on the side opposite the offset the flux is a factor of 3 lower , and for larger offsets the flux ratio can be several orders of magnitude . for msps , the contrast in flux is much smaller with ratios of only a factor of 4 - 5 in flux across the pc for offset as high as @xmath184 . we have investigated the effects of offset polar caps on electron acceleration and pair cascades near the neutron star surface , including the cases where the pcs of the two hemispheres are offset symmetrically by the same amount or asymmetrically . the asymmetric offset case would apply to the pcs shifted by retardation and/or currents of the global magnetosphere , while the symmetric case could apply to neutron stars with some interior current distortions that produce multipolar components near the surface . the asymmetric pc offsets have now been shown to be standard in pulsar magnetosphere global geometry . the pcs in the retarded vacuum magnetosphere are shifted in a direction opposite to that of the rotation by an amount that varies from a maximum of 20% of the standard pc radius at @xmath212 to no shift for @xmath213 ( dyks & harding 2004 ) . in terms of our offset parameter @xmath16 , the ratio of offset to standard pc angle is @xmath214 , so that for the vacuum case @xmath215 , where the large values of @xmath216 apply to msps with large @xmath162 . in the force - free ideal - mhd magnetosphere , the offset percentages range from 45% at @xmath212 to 30% at @xmath168 ( bai & spitkovsky 2010 ) , giving a range @xmath217 , again with the larger values of @xmath218 applying to msps . from our calculations , we find that the pair multiplicity and pair flux for offset pcs is distributed very asymmetrically over the pc , with higher values on the side of the pc toward the offset . the regions of peak pair flux and multiplicity occur for @xmath76 at about half the pc radius , symmetrically around the magnetic axis . as @xmath16 increases , small regions of higher peak flux and multiplicity appear toward the offset and grow larger with increasing offset . the increase in peak multiplicity moves the pair death lines downward in @xmath130-@xmath171 space , to encompass nearly the entire radio pulsar population for @xmath167 . since such offset values are higher than those resulting from retardation and currents in pulsar magnetosphere models , they would likely require interior currents that produce large - scale non - dipolar fields near the neutron star surface . examination of the pair spectra reveal that the increase in multiplicity in pulsars with offset pcs comes from the extension of the spectra to lower pair energies . the discovery of pulsed gamma - ray emission from a large number of millisecond pulsars by the large area telescope ( lat ) on the @xmath219 gamma - ray space telescope ( abdo et al . 2009 ) has revealed light curves that are best modeled by narrow radiation gaps in the outer magnetosphere ( venter et al . 2009 , abdo et al . such narrow gaps require screening of the accelerating electric field over most of the magnetosphere by a pair multiplicity that is orders of magnitude larger than standard models of pc acceleration with no offsets are able to produce ( harding & muslimov 2002 ) . since pc offsets that are only small fractions of a stellar radius result in large increases in pair multiplicity , and there is evidence for such offsets in msps , offset pcs may be a viable explanation for the larger - than - predicted pair activity in msps . we find that the integrated pair flux and luminosity from each pc is roughly proportional to the pulsar spin down luminosity , both in the case of non - recycled pulsars and msps with msps being somewhat more efficient in converting spin - down to pair luminosity . for pulsars with offset pcs , both pair flux and luminosity increase with offset but to a greater degree for low spin down power . due to a saturation of the pair multiplicity , pulsars with high @xmath0 have only modest increases in pc pair flux even for large offsets . we estimate that the crab pulsar produces a pair flux from each pc of about @xmath220 in the case of no offset , @xmath221 for @xmath207 and @xmath222 for @xmath167 . the flux from both pcs is still more than an order of magnitude smaller than the pair flux required to account for the radiation from the nebula , which is estimated to be about @xmath223 ( dejager et al . the trend of pair flux and pair luminosity proportional to spin down power may have interesting implications for explaining observed trends in pulsar pulsed and un - pulsed x - ray luminosity that are also proportional to spin down power ( vink et al . such a trend contrasts that of observed @xmath87-ray luminosities which are proportional to @xmath224 ( thompson et al . 1997 , abdo et al . the @xmath87-ray luminosity trend can be understood if the emission is produced by the primary pc current ( proportional to @xmath224 ) accelerated by a constant voltage of about @xmath225 v ( harding 1981 , arons 1996 ) . the x - ray luminosity trend may be understood if both the pulsed emission from the magnetosphere and the unpulsed emission from the pulsar wind nebula ( pwn ) are produced by pairs . it is generally believed that pairs are producing the emission from pwne , and a number of high - energy emission models argue that secondary electron - positron pairs produce the pulsed x rays high in the magnetosphere through synchrotron and/or ics ( cheng et al . 1986 , romani 1996 , takata et al . 2007 , hirotani 2008 , harding et al . 2008 ) . the asymmetric distribution of particle acceleration and pair multiplicity that results from offset pcs should produce asymmetries in observed pulsar emission . the structure and energetics of the proposed slot gaps ( sg ) that form between the boundary of the open magnetic field and the upward curving pff ( arons & scharlemann 1979 ) , and can accelerate particles to high altitude ( muslimov & harding 2004 ) , could be strongly affected by a an offset pc . the particle lorentz factor @xmath87 in the sg , which is expected to reach curvature radiation - reaction limit such that @xmath226 , will be larger on one side of the pc , producing cr emission power proportional to @xmath6 that is larger than for a dipole field . in older pulsars that do not produce enough pair multiplicity and screening to form sgs in centered pcs , sgs may form on only one side of an offset pc . an azimuthal asymmetry of both the radiation power and width of the sg would change both the @xmath87-ray luminosity and the sharp @xmath87-ray peaks that are due to caustics formed by radiation from trailing edge field lines ( dyks & rudak 2003 ) . since the pc offsets that result from retardation and currents in pulsar magnetospheres occur toward the trailing side of the pc , the @xmath87-ray peaks should be enhanced relative to the off - peak emission that is due to emission along the leading - edge field lines . the ratio of @xmath6 , and thus @xmath87-ray flux , between trailing and leading edges of the sg is predicted to be @xmath227 , which can be larger than an order of magnitude for pulsars having short periods . sg model light curves assuming the emission asymmetry predicted in vacuum or non - ideal mhd magnetospheres fit the pulsar light curves measured by @xmath219 significantly better that those of symmetric sgs ( harding et al . 2011 , decesar et al . the asymmetry in pair flux in offset pcs should produce asymmetries in pulsar radio emission if the radio flux is proportional somehow to the pair flux . in fact , evidence for such asymmetries in pulsar radio emission have been observed in the form of partial cone emission ( mitra & rankin et al . 2011 ) . in a future study , we will explore the effect of an offset pcs on the acceleration of particles in the sg at high altitudes and the change to the shapes of @xmath87-ray light curves . we acknowledge support from the nasa astrophysics theory and fundamental physics program , the @xmath219 guest investigator program and the universities space research association . akh also thanks the aspen center for physics . abdo , a. a. et al . 2009 , science , 325 , 848 . abdo , a. a. et al . 2010 , apj , 712 , 957 abdo , a. a. et al . 2010b , apjs , 187 , 460 . arendt & eilek , j. a. 1998 , ( arxiv : astro - ph/9801257 ) arons , j. 1981 , apj , 248 , 1099 . arons , j. 1983 , apj , 266 , 215 . arons , j. 1996 , astron . astrophys . , 120 , 49 . arons , j. 1997 , in neutron stars and pulsars : thirty years after the discovery , eds . n. shibazaki , n. kawai , s. shibata and t. kifune , university academy press , inc . , tokyo , japan , p. 339 . arons , j. & scharlemann , e. t. 1979 , apj , 231 , 854 asseo , e. & khechinashvili , 2002 , mnras , 334 , 743 . bai , x - n . & spitkovsky , a. 2010 , apj , 715 , 1282 . baring , m. g. & harding , a. k. 2001 , apj , 547 , 929 . bogdanov , s. , rybicki , g. b. & grindlay , j. e. 2007 , apj , 670 , 668 . bogdanov , s. & grindlay , j. e. 2009 , apj , 703 , 2259 . cheng , k. s. , ho , c. , & ruderman , m. a. 1986 , apj , 300 , 500 . contopoulos , i. , kazanas , d. & fendt , c. 1999 , apj , 511 , 351 . daugherty , j. k. & harding , a. k. 1982 , apj , 257 , 603 . daugherty , j. k. & harding , a. k. 1983 , apj , 273 , 761 . daugherty , j. k. & harding , a. k. 1996 , apj , 458 , 278 . de jager , o. c. , a. k. harding , michelson , p. f. , nolan , p. l. , sreekumar , p. & thompson , d. j. 1996 , apj , 457 , 253 . decesar , m. e. , harding , a. k. , miller , m. c. , kalapotharakos , c. & contopoulos , i. 2011 , 2011 fermi symposium proceedings , econf c110509 ( arxiv : astro - ph/0349176 ) . de luca , a. , caraveo , p. a , mereghetti , s. , negroni , m. & bignami , g. 2005 , apj , 623 , 1051 . demorest , p. b. , pennucci , t. , ransom , s. m. , roberts , m. s. e. & hessels , j. w. t. 2010 , nature , 467 , 1081 . dermer , c. d. 1990 , apj , 360 , 197 . deutsch , a. j. , 1955 , ann . dastrophys . , 18 , 1 . dyks , j. & harding , a. k. 2004 , apj , 614 , 869 . dyks , j. & rudak , b. 2003 , apj , 598 , 1201 . friedman , j. l. , ipser , j. r. , & parker , l. 1986 , apj , 304 , 115 . harding , a. k. 1981 , apj , 245 , 267 . harding , a. k. , baring , m. g. & gonthier , p. l. 1997 , apj , 476 , 246 . harding , a. k. , & muslimov , a. g. 1998 , apj , 508 , 328 [ hm98 ] . harding , a. k. , & muslimov , a. g. 2001 , apj , 556 , 987 [ hm01 ] . harding , a. k. , & muslimov , a. g. 2002 , apj , 568 , 862 . harding , a. k. , muslimov , a. g. & zhang , b. 2002 , apj , 576 , 366 . harding , a. k. , j. v. stern , j. dyks & m. frackowiak , 2008 , apj , 680 , 1376 . harding , a. k. , & muslimov , a. g. 2011 , apj , 726 , l10 [ hm11 ] . harding , a. k. , decesar , m. e. , miller , m. c. , kalapotharakos , c. & contopoulos , i. 2011 , 2011 fermi symposium proceedings , econf c110509 ( arxiv : astro - ph/0349902 ) . harding , a. k. & preece , r. 1987 , apj , 319 , 939 . hibschman , j. a. & arons , j. 2001 , apj , 560 , 871 . hirotani , k. 2008 , open astronomy ( arxiv:0809.1283 ) kalapotharakos , k. , kazanas , d. , harding , a. k. & contopoulos , i. 2011 , apj , submitted . kantor , e. m. & tsygan , a. i. 2003 , astronomy reports , 47 , 613 . kantor , e. m. & tsygan , a. i. 2004 , astronomy reports , 48 , 1029 . contopoulos , i. , kazanas , d. & fendt , c. 1999 , apj , 511 , 351 . lamb , f. k. et al , 2009 , apj , 706 , 417 . lattimer , j. m. & prakash , m. 2007 , phys . , 442 , 109 . luo , q. ; shibata , s. ; melrose , d. b. 2000 , mnras , 318 , 943 . manchester , r. n. , hobbs , g. b. , teoh , a. & hobbs , m. 2005 , astron . j. , 129 , 1993 medin , z. & lai , d. 2010 , mnras , 406 , 1379 . mitra , d. & rankin , j. 2011 , apj , 727 , 92 . muslimov , a. g. & tsygan , a. i. 1992 , mnras , 255 , 61 muslimov , a. g. & harding , a. k. 2004 , apj , 606 , 1143 romani , r. w. 1996 , apjj , 470 , 469 . ruderman , m. 1991 , apj , 366 , p. 261 . ruderman , m.a . & sutherland , p. g. 1975 , apj , 196 , 51 sokolov , a. a. & ternov , i. m. 1968 , radiation from relativistic electrons , ed . c. w. kilmister ( aip : new york ) spitkovsky , a. apj , 648 , l51 ( 2006 ) li , j. , spitkovsky , a. & tchekhovskoy , a. 2011 , apj , submitted ( arxiv:1107.0979 ) . takata , j. ; chang , h .- k . ; cheng , k. s. 2007 , apj , 656 , 1044 . thompson , d. j. , harding , a.k . , hersen , w. & ulmer , m.p . 1997 , in proc . of the _ 4th compton symposium _ dermer , m.s . strickman & j.d . kurfess ( aip 410 : new york ) , 39 . timokhin , a. mnras , 36 , 1055 ( 2006 ) timokhin , a. & arons , j. 2011 , apj , in prep . tsai , w. y & erber , t. 1974 , phys . d , 10 , 492 . venter , c. ; harding , a. k. ; guillemot , l. 2009 , apj , 707 , 800 . vink , j , mamba , a. & tamazaki , r. 2011 , apj , 727 , 131 . zavlin , v. e. 2007 , astrophysics space sci . , 308 , 297 . zhang , b. , harding , a. k. & muslimov , a. g. 2000 , apj , 531 , l135 . for the symmetric ( case a , top ) and asymmetric ( case b , bottom ) polar cap offsets.,title="fig:",width=529 ] for the symmetric ( case a , top ) and asymmetric ( case b , bottom ) polar cap offsets.,title="fig:",width=529 ] ) as a function of radial distance from the magnetic axis ( in units of neutron star radius ) and magnetic azimuth for @xmath149 s , @xmath150 g and @xmath169 , for different values of offset parameter @xmath16 . , width=793 ] ) as a function of radial distance from the magnetic axis ( in units of neutron star radius ) and magnetic azimuth for @xmath165 ms , @xmath166 g and @xmath229 , for @xmath167 and different values of inclination angle @xmath117 . , width=793 ] and period derivative ( @xmath171 ) , for a centered pc and inclination angle @xmath169 for normal pulsars and @xmath229 for msps . the ns radius @xmath5 , mass @xmath230 ( in solar mass units ) and moment of inertia @xmath231 refer to different ns equations of state used for normal and millisecond pulsars and are described in the text . radio pulsars with measured @xmath171 from the atnf catalog ( manchester et al . 2005 , http://www.atnf.csiro.au/research/pulsar/psrcat ) are plotted as black dots . , width=793 ] ) as a function of radial distance from the magnetic axis ( in units of neutron star radius ) and magnetic azimuth for @xmath200 s , @xmath150 g and @xmath169 , for different values of offset parameter @xmath16 . , width=793 ] ) as a function of radial distance from the magnetic axis ( in units of neutron star radius ) and magnetic azimuth for @xmath157 ms , @xmath232 g and @xmath229 , for different values of offset parameter @xmath16 . , width=793 ]

in this paper , we give a natural construction of mixed tate motives whose periods are a class of iterated integrals which include the multiple polylogarithm functions . given such an iterated integral , we construct two divisors and in the moduli spaces of-pointed stable curves of genus , and prove that the cohomology of the pair is a framed mixed tate motive whose period is that integral . it generalizes the results of a. goncharov and yu . manin for multiple-values . then we apply our construction to the dilogarithm and calculate the period matrix which turns out to be same with the canonical one of deligne .

math0610670

the multiple polylogarithm functions were defined in goncharov s paper @xcite as the following power series : @xmath7 where the @xmath8 are in the unit disk of the complex plane for @xmath9 and @xmath10 are positive integers . for @xmath11 , we get the classical @xmath3-th polylogarithm which was first introduced by leibniz @xcite in 1696 : @xmath12 and for @xmath13 , we obtain the multiple @xmath6-values which were first studied by euler @xcite : @xmath14 moveover , multiple polylogarithms can be represented as iterated integrals . recall that iterated integrals are defined as follows . let @xmath15 be smooth one - forms on a manifold @xmath16 and @xmath17 \rightarrow m$ ] be a piecewise smooth path . then we define inductively as follows : @xmath18 where @xmath19 is the restriction of @xmath20 on @xmath21 $ ] and @xmath22 is a function of @xmath23 on @xmath24 $ ] . more explicitly , it can be computed in the following way : @xmath25 where @xmath26 are the pullback one - forms on @xmath24 $ ] , @xmath27 . for example , @xmath28 in ( * ? ? ? * chap2 ) , the following formula was proved : @xmath29 @xmath30 this formula also provides the analytic continuation of multiple polylogarithms . we denote by @xmath31 the moduli space of @xmath32-labeled pointed stable curves of genus @xmath4 , where @xmath32 is a finite set . it s been studied by grothendieck @xcite , deligne , mumford , knudsen @xcite and many others . it is defined over @xmath33 . roughly speaking , a complex point of @xmath31 is a tree of complex projective lines with @xmath34 distinct smooth points marked by the set @xmath32 . here @xmath34 denotes the cardinality of the set @xmath32 . we know that it is a smooth irreducible projective variety of complex dimension @xmath35 . moreover , @xmath36 provides a natural compactification of the space @xmath37 of @xmath34 distinct points on @xmath38 modulo automorphisms of @xmath38 . by cross - ratio , @xmath37 is isomorphic to @xmath39 the boundary @xmath40 is a normal crossing divisor . it can be described by the combinatorial data of the set @xmath32 . for more detail , see section @xmath41 . now given any subset @xmath42 , with @xmath43 , there is a contraction morphism : @xmath44 which contracts stably all sections but those marked by @xmath45 . in particular , for any @xmath46 , let @xmath47 , the contraction morphism @xmath48 is the universal @xmath49-labeled curve with universal sections @xmath50 , for each @xmath51 . for more information and proofs of @xmath31 , we refer to @xcite . if the set @xmath52 , we ll denote this space by @xmath2 . and @xmath31 is non - canonically isomorphic to @xmath53 . from now on , we ll take @xmath54 + @xmath55 and fix the cyclic order @xmath56 on @xmath32 unless otherwise stated . in @xcite , for each multiple @xmath6-value ( [ mz ] ) , the authors construct two divisors @xmath0 and @xmath1 of @xmath31 and then show that @xmath57 is a framed mixed tate motive whose period is this value . in the end of that paper , they suggest to generalize their results to the following convergent iterated integral : @xmath58 where @xmath17 \rightarrow \mathbb{c}$ ] is a piecewise smooth simple path from @xmath4 to @xmath59 and @xmath60 , @xmath61 . in particular , by the formula ( [ mul - li ] ) , multiple polylogarithms are of this type . in this paper , we show that the analogous results hold for the iterated integral ( [ int1 ] ) . in section @xmath41 , we review the basic combinatorial facts about the boundary divisors of @xmath31 and the stable @xmath41-partitions of the set @xmath32 . next , we briefly recall the divisor @xmath62 in @xmath31 which was introduced in @xcite , and then prove some interesting combinatorial properties of @xmath62 . in the end , we proceed to study in detail some non - boundary divisors of @xmath31 which we ll use later on . in section @xmath63 , for the integral ( [ int1 ] ) , we define a meromorphic differential form @xmath64 of @xmath36 . let @xmath65 be its divisor of singularities in @xmath36 . we explicitly determine the divisor @xmath65 . then we use it to prove the key proposition that _ the divisor @xmath65 does not contain any @xmath66-dimensional face of the divisor @xmath62 , @xmath67_. in section @xmath68 , we review the definitions of framed hodge - tate structure and its period , and discuss their basic properties . finally combining with all the information of @xmath65 and @xmath62 in section @xmath63 and @xmath68 , we can prove : [ main ] let @xmath69 . for the iterated integral @xmath70 of ( [ int1 ] ) , @xmath71 , there exists two divisors @xmath65 and @xmath62 in @xmath31 , @xmath72 , such that @xmath73 carries an @xmath3-framed hodge - tatestructure with two canonical frames @xmath74}\in gr_{2n}^{w}h^{n}(\mathcal{\overline{m}}_{0,s}-a_s(\vec{a } ) ) ; \ ; { [ \delta_{b}(\gamma)]}\in ( gr_{0}^{w}h^{n}(\mathcal{\overline{m}}_{0,s},b_n))^{\vee}.\ ] ] and the period with respect to these frames is exactly the iterated integral @xmath70 , where @xmath75 $ ] is a meromorphic @xmath3-form on @xmath76 , and @xmath77 is a relative @xmath3-cycle . furthermore , if the @xmath78 are elements of a number field @xmath79 , @xmath61 , then @xmath80 is a framed mixed tate motive over @xmath79 . in section @xmath81 , we apply our construction to the dilogarithm . namely , we consider the following integral : @xmath82 by the theorem [ main ] above , @xmath83 carries a @xmath41-framed hodge - tate structure . we calculate it and prove that : [ thm2 ] the mixed hodge structure given in theorem [ main ] for the dilogarithm , that is , @xmath83 , is isomorphic to the one given by p. deligne . and for our case the period matrix is the following : + if @xmath84 , it equals @xmath85\ ] ] if @xmath86 , it is @xmath87\ ] ] with above theorem [ thm2 ] , the interesting question is that for the classical @xmath3-th polylogarithm ( @xmath88 ) , whether or not our construction is isomorphic to the canonical one given by deligne . the general situation is more delicate than the dilogarithm case . we have some results and it seems that they are not isomorphic if @xmath88 . more detail will appear in @xcite . another construction of the multiple polylogarithm motives has been given by a.goncharov in @xcite where he uses another sequence of blowups . as * framed * mixed tate motives , the two constructions should be equivalent . but our construction is canonical and more natural . first recall that there is a one - to - one correspondence between the boundary divisors of @xmath31 and the stable unordered @xmath41-partitions of the set @xmath32 . let @xmath89 and @xmath90 . we ll denote by @xmath91 the corresponding boundary divisor . [ def1 ] let @xmath92 with the given cyclic order @xmath93 . a subset @xmath0 of @xmath94 is called strictly ordered if there exists some @xmath95 and @xmath96 a positive integer , such that @xmath97 ( the subscripts are counted mod @xmath66 ) . that is , its elements are in consecutive order with respect to @xmath93 . given a @xmath41-partition @xmath98 of @xmath94 , @xmath99 , we say that @xmath98 is strictly ordered with respect to @xmath93 if one of the @xmath100 is a strictly ordered subset of @xmath94 . for example , take @xmath101 , then @xmath102 is strictly ordered , but @xmath103 is not . now consider the open @xmath3-simplex @xmath104 . as mentioned in the introduction , via cross - ratio , @xmath37 is identified with the subset @xmath105 of @xmath106 . under this identification , @xmath107 is a subset of @xmath37 . thus we have a natural map @xmath108 which embeds @xmath107 into @xmath36 . let @xmath109 be the zariski closure of the boundary of the closure of @xmath110 in @xmath36 , which is called algebraic stasheff polytope ( see @xcite ) . then we have the following : [ prop1 ] @xmath109 is a union of boundary divisors indexed by the stable @xmath41-partitions of @xmath32 which are strictly ordered with respect to the cyclic order @xmath93 . that is , they correspond to breaking a circle into two connected arcs . furthermore , @xmath109 is an `` algebraic stasheff polytope '' . i.e. , there is a bijection between the irreducible components @xmath111 of @xmath109 and the codimension one faces @xmath112 of the stasheff polytope @xmath113 such that a subset of @xmath111 s has a non - empty intersection of expected codimension if and only if the respective subset of @xmath112 s has this property . see @xcite , proposition 2.1 . for two unordered stable @xmath41-partitions @xmath114 clearly , @xmath115 or @xmath68 , and @xmath116 if and only if @xmath117 , which implies that the boundary divisors @xmath118 . [ lemma1 ] ( 1 ) @xmath119 if and only if @xmath120 and @xmath121 . + ( 2 ) @xmath122 if and only if @xmath123 . \(1 ) by definition , @xmath119 if and only if one of the following holds : @xmath124 by the fact 4 in @xcite , this is exactly the sufficient and necessary condition for that @xmath120 and @xmath125 . \(2 ) by definition , @xmath122 means that there are distinct elements @xmath126 such that @xmath127 , @xmath128 , @xmath129 , @xmath130 . then it follows from the proof of the fact 4 in @xcite . a family of stable @xmath41-partitions @xmath131 of @xmath32 is called good if @xmath132 , for @xmath133 . a family of boundary divisors @xmath134 is called compatible , if the corresponding partitions @xmath135 are good . [ lemma2 ] given @xmath136 pairwise distinct boundary divisors @xmath137 , then @xmath138 if and only if @xmath131 is good . first , if @xmath139 is not good , then @xmath140 for some @xmath133 . since @xmath141 and @xmath142 are distinct , by lemma [ lemma1 ] , @xmath143 hence , @xmath144 now suppose that @xmath139 is good . the by ( * ? ? ? * proposition 3.5.2 ) , there exists a stable @xmath32-tree @xmath145 with @xmath136 ( internal ) edges @xmath146 such that @xmath147 , where @xmath148 is the stable @xmath41-partition of @xmath32 corresponding to the one edge @xmath32-tree obtained by contracting all ( internal ) edges but @xmath149 . denote by @xmath150 the closure of the stratum parametrizing the stable curves of the combinatorial type @xmath145 . then for each @xmath151 , we have @xmath152 . hence , @xmath153 . and since the boundary divisors meet transversely , we can conclude that in this case @xmath154 . now fix an @xmath155 , and let @xmath156 with the induced cyclic order @xmath157 . let @xmath158 be the contraction morphism which contracts the section marked by @xmath159 . similarly , we have the natural embedding of @xmath160 into @xmath161 . and we also have the corresponding algebraic stasheff polytope @xmath162 . now we have the following key property : [ prop3 ] under the contraction morphism @xmath163 , the vertices of @xmath109 are mapped to the vertices of @xmath162 . let @xmath164 be a vertex of @xmath62 . then by proposition [ prop1 ] , there exists exactly @xmath3 compatible boundary divisors of @xmath62 , say , @xmath165 such that @xmath166 . consider the images @xmath167 in @xmath161 , for @xmath168 . there are two possible cases for the @xmath169 s : case 1 : @xmath170 ; or @xmath171 ; + case 2 : @xmath172 , with @xmath43 and @xmath173 . in case 1 , we get @xmath174 and in case 2 , we have @xmath175 where @xmath176 is a stable partition of @xmath49 , and @xmath177 is an irreducible component of @xmath162 . it s clear that if @xmath178 belong to case 2 , then @xmath179 . now we claim that among the @xmath180 , @xmath181 , @xmath182 , there exists exactly one of them belonging to case 1 . indeed , first suppose all of them belong to case 2 . then we have @xmath3 pairwise distinct boundary divisors of @xmath183 and their intersection is empty due to the dimension consideration . but @xmath184 is contained in this intersection , contradiction . secondly , suppose there are two of them in case 1 . say they are @xmath185 and @xmath186 . but then @xmath187 , thus @xmath188 . it contradicts that @xmath166 . hence the claim is proved . by the claim , we see that @xmath184 is contained in the intersection of @xmath189 distinct compatible divisors of @xmath162 . by proposition [ prop1 ] , @xmath184 is a vertex of @xmath162 . using proposition [ prop3 ] , we can prove the following : [ prop4 ] let @xmath190 be the contraction morphism which forgets all sections but the ones marked by @xmath191 . then the images of the vertices of @xmath109 are contained in the set @xmath192 . under the identification of @xmath193 with @xmath194 , @xmath195 is the interval @xmath24 $ ] . thus the vertices are @xmath4 , @xmath59 . the map @xmath196 is a composition of @xmath189 contraction morphisms , each of which forgets only one section of those labeled by the subset @xmath197 . now we can apply proposition [ prop3 ] repeatedly to those @xmath189 contraction morphisms . therefore , the images of the vertices of @xmath109 are either @xmath4 or @xmath59 . next , we ll introduce some non - boundary divisors in @xmath31 and prove some properties which we need later on . for each @xmath198 , consider the morphism @xmath199 which contracts all sections but those labeled by @xmath200 . now let @xmath201 , that is , it lies in the open stratum @xmath202 . then the inverse image @xmath203 is a divisor of @xmath31 . the following proposition describes how @xmath203 intersects the boundary divisors of @xmath31 . [ nonb1 ] let @xmath204 be a stable @xmath41-partition of @xmath32 and @xmath91 be the corresponding boundary divisor . then : + ( 1 ) if @xmath205 and @xmath206 , @xmath207 and @xmath208 , or @xmath209 and @xmath210 belong to different parts of @xmath98 , then @xmath211 ( 2 ) otherwise , we can assume that @xmath212 . then:@xmath213 where @xmath214 is some fixed element of @xmath180 and @xmath215 is the map contracting all sections but those marked by @xmath216 ; and if @xmath217 , @xmath23 is some fixed element of @xmath218 ; otherwise , @xmath219 , then @xmath23 is the only element of @xmath220 not in @xmath180 . for the first case , it s easy to check that the image of @xmath91 under @xmath221 is contained in the set @xmath222 . but @xmath223 , thus @xmath224 for the second case , consider the morphism:@xmath225 which is the product of contracting all sections labeled by @xmath218 but @xmath23 and contraction of all sections labeled by @xmath180 but @xmath214 . by the fact 2 of @xcite , the restriction of @xmath226 on @xmath227 is an isomorphism which is independent of the choices of @xmath214 and @xmath23 . then we have the following commutative diagram : @xmath228 where @xmath229 and @xmath230 are the restrictions of the maps @xmath226 and @xmath231 on the divisor @xmath91 respectively , and @xmath232 is the projection on the first factor . thus we obtain that : @xmath233 since @xmath229 is an isomorphism , we have @xmath234 via the map @xmath226 . next we have the following description of @xmath203 : [ nonb2 ] ( 1 ) . @xmath203 is an irreducible , reduced , smooth divisor of @xmath31 . it is a tate variety , that s , its motive is a direct sum of pure tate motives . we ll use the induction on @xmath3 . when @xmath235 , @xmath236 and @xmath237 is the identity map . and the proposition is obvious . let s suppose that it s true for @xmath238 , @xmath239 , and consider @xmath240 . now @xmath241 . let @xmath242 . then we have the following commutative diagram : @xmath243 where @xmath244 is the map forgetting the section @xmath245 , @xmath20 is the product of @xmath244 and the morphism from @xmath31 to @xmath246 contracting all sections but those marked by @xmath247 ; and @xmath248 is the projection on the first factor . moreover , we have:@xmath249 hence,@xmath250 it s known that @xmath251 is the universal curve on @xmath183 , so @xmath244 has geometrically reduced fibers , and by induction,@xmath252 is reduced , thus @xmath240 is reduced . in @xcite , keel proved that the map @xmath20 in the above diagram is isomorphic to a sequence of blowups of @xmath253 along all the boundary divisors of @xmath183 . from the above diagram , we see that : @xmath254 by the explicit blowups given in @xcite , @xmath240 is the composition of blowups of @xmath255 along the intersections of @xmath203 with the boundary divisors . by the proposition [ nonb1 ] , these intersections are either empty or of the form : @xmath256 where @xmath257 and @xmath258 . by the induction assumption , @xmath259 is irreducible and smooth , and @xmath255 is an irreducible smooth divisor of @xmath253 . altogether we see that @xmath240 is the composition of blowups of an irreducible smooth variety along an irreducible smooth subvariety . therefore , it is irreducible and smooth . and this proves the first part of the proposition . for the second part , we need the following formula of the motive of the blowup of a variety along a subvariety . this was proved by manin in ( * ? ? ? * corollary , page 463 ) . let @xmath260 be the blowup of a variety @xmath261 along a subvariety @xmath262 of codimension @xmath263 . then we have : @xmath264 where @xmath265 , @xmath266 and @xmath267 denote the motives of @xmath260 , @xmath261 and @xmath262 respectively ; @xmath268 is the tate motive , @xmath269 . therefore , if both @xmath261 and @xmath262 are tate varieties , by the formula above , the blowup @xmath260 is also a tate variety . because the product of tate varieties is also a tate variety , and by induction , @xmath240 is the composition of blowups of a tate variety along a tate subvariety . thus it s a tate variety . this proves the second part . now we consider the intersections of the divisors @xmath203 . let @xmath270 , and @xmath271 . then we have : [ nonb3 ] the intersection @xmath272 is a smooth irreducible subvariety ; it s also a tate variety . first let s look at the case @xmath273 . notice for any irreducible boundary divisor @xmath91 , where @xmath98 is some stable @xmath41-partition of @xmath32 , we have @xmath274 by the proposition [ nonb1 ] , @xmath275 is either empty or of the form : @xmath276 here we use the notations in proposition [ nonb1 ] . now by induction on @xmath3 , we see that @xmath277 is obtained by the blowups of smooth irreducible tate varieties along the tate subvarieties . hence it s also smooth , irreducible and tate . when @xmath278 , similarly for any irreducible boundary divisor @xmath91 , we have @xmath279 then similarly , use proposition [ nonb1 ] and induction on @xmath3 , it s true for any @xmath136 . [ nonb4 ] let @xmath280 be stable @xmath41-partitions of @xmath32 . for each @xmath281 , @xmath141 is the corresponding boundary divisor . then the intersection @xmath282 is either empty or a smooth irreducible tate variety . by the combinatorial description of the intersection @xmath283 , ( for @xmath284 , see ( * ? ? ? * fact 4 ) , and for general @xmath66 , it can be done inductively ) we see that the non - empty intersection in the corollary can be written as the product of irreducible smooth tate varieties . thus , it s also an irreducible smooth tate variety . let @xmath286 and consider the following @xmath3 contraction morphisms : @xmath287 which forget all sections but those labeled by @xmath288 . we define the meromorphic @xmath3-form @xmath285 of @xmath31 as follows : @xmath289 furthermore , we have another useful description of @xmath285 . let @xmath290 be the birational proper morphism of the products of @xmath291 , @xmath61 . and let @xmath292 be the affine coordinates of @xmath293 , then by definition , we have @xmath294 we define @xmath295 to be the divisor of singularities of @xmath285 . following @xcite , for @xmath296 , we define @xmath297 by @xmath298 and @xmath299 we say that a @xmath41-partition of @xmath32 has _ type @xmath300 _ if one part of it is of form @xmath301 where @xmath94 is a non - empty subset of @xmath297 . now we have the explicit description of the divisor @xmath295 : [ divisor - a ] the divisor @xmath295 of singularities of @xmath64 on @xmath31 consists of the following irreducible components : + ( a ) the boundary divisors @xmath91 corresponding to those stable @xmath41-partitions of @xmath32 which have some type @xmath300 , @xmath302 ; + ( b ) the non - boundary divisors @xmath303 , for each @xmath304 ; where @xmath305 is the morphism contracting all sections but those labeled by @xmath288 . moreover , @xmath295 is a normal - crossing divisor . we ll prove it by induction on @xmath3 . for the case @xmath235 , @xmath306 and @xmath307 . then @xmath308 , where @xmath23 is the affine coordinate of @xmath309 . thus @xmath310 , and the proposition is clear in this case . assume that it s true for @xmath3 . now @xmath311 and @xmath312 . let @xmath313 , @xmath314 , and @xmath315 the divisor of singularities of the meromorphic form @xmath316 in @xmath183 . consider the morphism @xmath226 of the product of two contraction maps : @xmath317 then we obtain that @xmath318 let @xmath319 be the divisor of singularities of the form @xmath320 in @xmath321 . clearly , the divisor of singularities of the meromorphic form @xmath322 in @xmath323 is @xmath324 , where @xmath248 and @xmath325 are the projections on the first and second factors respectively . and it s a normal crossing divisor . by @xcite , @xmath226 is isomorphic to a sequence of blowups of @xmath253 along all the boundary divisors of @xmath183 . by the following lemma [ blowup ] on blowups , and proposition [ nonb1 ] and [ nonb2 ] , we can conclude that @xmath326 is a normal crossing divisor in @xmath31 . @xmath295 is contained in it . thus it s also a normal crossing divisor . on the other hand , the meromorphic form @xmath285 does not necessarily have as singularities all the exceptional divisors of @xmath226 in @xmath326 . in fact , @xmath295 equals @xmath326 minus those spurious divisors . by ( * ? ? ? * lemma 3.8 ) , the spurious divisors are those irreducible boundary divisors in @xmath31 which get blown down by @xmath226 to a subvariety of the product which is not a stratum of the divisor @xmath324 . by ( * ? ? ? * lemma 1 , page 554 ) , the exceptional divisors of @xmath226 are exactly those boundary divisors corresponding to the following @xmath41-stable partitions : @xmath327 , with @xmath328 , @xmath329 , and @xmath330 . by induction , the divisor @xmath315 consists of the following four types of irreducible divisors ( we identify the boundary divisors with the corresponding @xmath41-stable partitions here ) : @xmath331 where @xmath332 is a non - empty subset of @xmath297 , @xmath333 . first , notice the pullback of the divisor @xmath303 equals @xmath334 which is not an exceptional divisor of @xmath226 , thus it s a component of the divisor @xmath295 . for each @xmath335 , by ( * fact 3 , page 552 ) , the pullback of the boundary divisor @xmath336 in the list above is equal to the sum of the following two components : @xmath337 by the condition @xmath338 , @xmath339 is not an exceptional divisor of @xmath226 , hence it s a component of the divisor @xmath295 ; but @xmath340 is an exceptional divisor . we need to check if it s a spurious divisor . notice that @xmath341 now clearly @xmath342 is a stratum of the divisor @xmath324 if and only if @xmath300 is in the divisor @xmath319 , therefore , @xmath340 is a component of the divisor @xmath295 if and only if @xmath300 is in the divisor @xmath319 . for @xmath343 , it depends upon whether or not @xmath344 . and @xmath345 is always in the divisor @xmath319 , hence @xmath346 is a component of @xmath295 . now there are only two more components of @xmath295 . one is @xmath347 . it s not an exceptional divisor and is contained in the pullback of @xmath348 . the other one is @xmath349 , when @xmath350 . and it is contained in the pullback of @xmath319 . thus we have all the components of @xmath295 listed in the proposition . this concludes the proof . now let s prove a lemma on blowups which is used in the proof of the preceding proposition . [ blowup ] let @xmath16 be a smooth complex variety of dimension @xmath3 , @xmath1 an irreducible smooth subvariety of @xmath16 with codimension @xmath351 , and @xmath352 a normal - crossing divisor of @xmath16 with smooth irreducible components @xmath353 , @xmath354 . suppose that for each @xmath353 , only one of the following cases happens : @xmath355 , @xmath356 , or : @xmath357 is an irreducible smooth subvariety with codimension @xmath358 . let @xmath359 be the blowup of @xmath16 along @xmath1 . let @xmath360 be the strict transform of @xmath353 , @xmath361 , and @xmath362 the exceptional divisor . then the divisor @xmath363 is a normal - crossing divisor in @xmath364 . let @xmath365 . if @xmath366 , the @xmath244 is locally at @xmath367 an isomorphism . so we only need to consider the case @xmath368 . then @xmath369 . by rearranging the indices of the @xmath353 , if necessary , we may assume that @xmath355 for @xmath370 , @xmath371 for @xmath372 , and @xmath373 for @xmath374 . here @xmath375 , @xmath376 and @xmath377 . since @xmath378 is a normal - crossing divisor , there exists an open neighborhood @xmath379 of @xmath380 with local coordinates @xmath381 such that @xmath382 and @xmath383 , for @xmath370 , @xmath384 ; for @xmath372 , @xmath385 . and @xmath386 . here $ ] is the homogenous coordinate of @xmath388 . then a point @xmath389 in the inverse image of @xmath390 has the property that @xmath391 . now in @xmath392 , suppose that @xmath393 for @xmath394 , and @xmath395 for @xmath396 . then @xmath397 . now since @xmath398 , it can not happen that all @xmath391 . then there exists some @xmath399 such that @xmath400 . consider the open neighborhood @xmath401 of @xmath402 . and the restriction of the blowup @xmath244 on the @xmath403 has the following formula:@xmath404 where @xmath405 , @xmath406,@xmath407 . therefore for @xmath408 , @xmath409 and @xmath410 . thus @xmath392 is a normal - crossing divisor in @xmath364 . next we ll prove the main theorem in this section : [ thm1 ] the divisor @xmath65 of singularities @xmath64 does not contain any @xmath66-dimensional face of the algebraic stasheff polytope @xmath109 , @xmath411 . here @xmath412 with @xmath413 and @xmath414 . first observe that if @xmath65 contains some @xmath66-dimensional face of @xmath109 , then it also contains the vertices of this face . therefore it suffices to show that @xmath65 does not contain any vertex of @xmath109 . we ll prove that each irreducible component of @xmath65 listed in the preceding proposition [ divisor - a ] ca nt contain any vertex of @xmath109 . first consider the non - boundary divisor @xmath303 , for @xmath415 . clearly , its image under the contraction morphism : @xmath416 is @xmath417 . on the other hand , by the proposition [ prop4 ] in section 2 , the image of a vertex of @xmath62 under @xmath231 is @xmath4 , or @xmath59 . since @xmath415 , @xmath303 does nt contain any vertex of @xmath62 . let s consider the boundary divisor in @xmath65 . it has one of the following types @xmath418 by the proposition [ prop1 ] in section 2 , we know that a vertex of @xmath62 is contained in the intersection of exactly @xmath3 irreducible components of @xmath62 . notice that the intersection of any @xmath419 irreducible boundary divisors of @xmath31 is empty , therefore , it s enough to show that none of the boundary divisors in @xmath65 appears as an irreducible component of @xmath62 . by the proposition [ prop1 ] in section 2 , the irreducible components of @xmath62 correspond to the stable @xmath41-partitions of @xmath32 which are strictly ordered with respect to the cyclic order @xmath56 . the partition @xmath420 is not strict with respect to @xmath93 because @xmath413 then @xmath421 and @xmath422 separate any element of @xmath423 from @xmath4 ; similarly , @xmath424 is not strict with respect to @xmath93 because @xmath414 and @xmath425 and @xmath422 block any element of @xmath426 to @xmath59 ; finally @xmath427 is not strict with respect to @xmath93 because @xmath59 and @xmath4 separate @xmath422 from any element of @xmath428 . hence , none of the boundary divisors in @xmath65 contains a vertex of @xmath62 . this concludes the proof . we ll continue to use the notations in the previous sections . now consider the convergent iterated integral : @xmath429 where @xmath17 \rightarrow \mathbb{c}$ ] is a piecewise smooth simple path from @xmath4 to @xmath59 and @xmath430 , @xmath61 , @xmath431 is an open @xmath3-simplex in @xmath432 , and @xmath433 . as mentioned in the introduction , we can identify the open stratum @xmath37 with the set @xmath434 . then @xmath435 is a subset of @xmath37 , thus we have a natural map @xmath436 which embeds @xmath435 into @xmath36 . clearly we have the following equality : @xmath437 where @xmath285 is the meromorphic @xmath3-form in @xmath31 defined in section 2 . let @xmath438 be the closure of @xmath435 in @xmath31 . then the zariski closure of the boundary of @xmath438 is @xmath62 . this follows from the fact that @xmath62 is the zariski closure of the boundary of the closure of @xmath439 in @xmath31 . hence we have a relative homology class:@xmath440\in h_n(\mathcal{\overline{m}}_{0,s}(\mathbb{c}),b_n(\mathbb{c});\mathbb{q})\ ] ] the meromorphic @xmath3-form @xmath285 gives rise to a cohomology class : @xmath441\in h^{n}_{dr}(\mathcal{\overline{m}}_{0,s}(\mathbb{c})-a_s(\vec{a})(\mathbb{c } ) ; \mathbb{c}).\ ] ] now given the iterated integral ( [ ite1 ] ) , we ll define : @xmath442 then we have the following main theorem : [ mt ] @xmath443 carries an @xmath3-framed hodge - tate structure with the frames coming from @xmath444 and @xmath438 . the period is just the iterated integral @xmath445 . moreover , if all the @xmath417 are elements of a number field @xmath79 , then @xmath446 is a framed mixed tate motive over @xmath79 . before we prove the theorem , first let s briefly recall the definitions of framed hodge - tate structure and its period . a mixed hodge @xmath447-structure consists of the following data : + 1 ) a finite - dimensional @xmath447-vector space @xmath448 with a finite increasing filtration @xmath449 called the weight filtration ; + 2 ) a finite decreasing filtration @xmath450 of @xmath451 called the hodge filtration ; + these data satisfy the following conditions : + for each associated graded piece @xmath452 , one has the decomposition:@xmath453 with @xmath454 here `` '' means the complex conjugation of @xmath455 with respect to @xmath456 . the hodge numbers are the integers @xmath457 . by definition , hodge - tate structure _ @xmath458 is a mixed hodge @xmath447-structure @xmath458 with the hodge numbers @xmath459 unless @xmath460 . this means that for the weight filtration , @xmath461 and @xmath462 is a finite direct sum of @xmath463 . it also implies that , for each @xmath464 , the natural map @xmath465 is an isomorphism . an _ @xmath3-framed hodge - tate structure _ @xmath458 is a hodge - tate structure @xmath458 equipped with a nonzero vector in @xmath466 and a nonzero functional on @xmath467 , that is , we have two nonzero morphisms : @xmath468 to define the period of an @xmath3-framed hodge - tate structure , we need to choose a map of @xmath447-vector spaces @xmath469 which lifts @xmath470 , that is , @xmath471 . now let @xmath472 . consider the composition : @xmath473 where the first one is @xmath164 and the second is provided by ( [ ht ] ) . it gives rise to a vector @xmath474 . the period is the number @xmath475 . a different choice of the lifting @xmath79 will change this period by @xmath476 `` weight @xmath189 period '' . in this sense , periods are multi - valued . for more information about periods of framed mixed tate motives or periods of framed hodge - tate structures , we refer to ( * ? ? ? * chapter 5 ) and ( * ? ? ? * section 3.2 ) . now let s prove some lemmas about hodge - tate structures . [ h - t1 ] the category of hodge - tate structures is abelian . it s closed under subquotients and extensions . it s straightforward ( c.f . * thorm 2.3.5 ) ) . [ h - t3 ] let @xmath0 be a complex algebraic variety . suppose that @xmath477 where @xmath478 are closed subvarieties of @xmath0 , @xmath479 . let @xmath480 be a fixed positive integer . for any non - empty subset @xmath481 , denote by @xmath482 . suppose that all the cohomology groups of @xmath483 carry the hodge - tate structures for all @xmath484 . then @xmath485 also carries a hodge - tate structure . we ll use the mayer - vietoris exact sequence and the induction on @xmath66 . if @xmath486 , then it s obvious . assume that it s true for @xmath487 , and consider the case @xmath488 . we can write @xmath0 as : @xmath489 where @xmath490 . by the mayer - vietoris exact sequence for cohomology , we have : @xmath491 it s known that the mayer - vietoris sequence is also an exact sequence of mixed hodge structures . therefore , we get the short exact sequence of mixed hodge structures : @xmath492 notice that @xmath493 therefore , by the induction assumption , both @xmath494 and @xmath495 are hodge - tate structures . now by the lemma [ h - t1 ] , @xmath485 is a hodge - tate structure . [ h2 ] let @xmath261 be a complex algebraic variety , and @xmath0 be a subvariety of codimension @xmath59 . then if both @xmath496 and @xmath497 are hodge - tate , so are the relative cohomology @xmath498 and @xmath499 . consider the long exact sequence of cohomologies for the pair @xmath500 and we know that it s also a long exact sequence of mixed hodge structures . similar to lemma [ h - t3 ] , by the lemma [ h - t1 ] , we can conclude that the relative cohomology @xmath498 carries a hodge - tate structure , then by duality , so does @xmath499 . now let s prove the theorem : proof of theorem [ mt ] . by deligne @xcite , there is a canonical mixed hodge structure on the relative cohomology @xmath501 . first we ll show that @xmath502 is a hodge - tate structure . by the proposition [ divisor - a ] and corollary [ nonb4 ] , we know that the intersections of the components of @xmath65 are either empty or tate varieties . by lemma [ h - t3 ] and [ h2 ] , @xmath503 carries a hodge - tate structure . for the same reason , @xmath504 also has a hodge - tate structure . by the lemma [ h2 ] again , @xmath501 carries a hodge - tate structure . now we need to show that @xmath444 and @xmath438 give the frames . first , there are natural non - zero morphisms of pure tate structures by @xmath444 and @xmath438 : @xmath505 : \mathbb{q}(-n)\rightarrow gr^w_{2n}h^{n } ( \mathcal{\overline{m}}_{0,s}-a_s({\vec{a}})),\ ] ] @xmath440:\mathbb{q}(0)\rightarrow gr_0^{w}h_n ( \mathcal{\overline{m}}_{0,s},b_n)\ ] ] then composing with the canonical isomorphisms : @xmath506 and @xmath507 we get the frame morphisms : @xmath508^{\prime } : \mathbb{q}(-n)\rightarrow gr^w_{2n } h^{n}(\mathcal{\overline{m}}_{0,s}-a_s({\vec{a } } ) , b_n - b_n\cap a_s({\vec{a}}))\ ] ] @xmath440^{\prime}:\mathbb{q}(0)\rightarrow gr_0^{w}h_{n } ( \mathcal{\overline{m}}_{0,s}-a_s({\vec{a } } ) , b_n - b_n\cap a_s({\vec{a}})).\ ] ] since @xmath430 , for @xmath61 , by the description of @xmath509 and theorem [ thm1 ] in section 3 , we have @xmath510 therefore , @xmath438 provides an element of the relative betti homology @xmath440\in h_{n}(\mathcal{\overline{m}}_{0,s}-a_s({\vec{a } } ) , b_n - b_n\cap a_s({\vec{a}}))\ ] ] which is a lift of the frame morphism @xmath511^{\prime}$ ] . the form @xmath444 obviously gives an element of the relative de rham cohomology class @xmath508\in h^{n}(\mathcal{\overline{m}}_{0,s}-a_s({\vec{a } } ) , b_n - b_n\cap a_s({\vec{a}})).\ ] ] and their pairing is exactly the iterated integral in the theorem . now suppose that @xmath417 , @xmath512 are elements of a number field @xmath79 . we ll show that @xmath513 is a framed mixed tate motive over @xmath79 . for this , we ll follow the proof of ( * ? ? ? * theorem 4.1 ) very closely . in @xcite , goncharov constructed the abelian category of mixed tate motive over any number field @xmath79 , where he used the theory of triangulated category of mixed motives by voevodsky in @xcite . let s apply it to our case . first , consider the standard cosimplicial variety : @xmath514 here @xmath515 , and @xmath516 is the disjoint union of the codimension @xmath66 strata of the divisor @xmath517 . according to the standard procedure , we can get a complex @xmath518 of varieties from the above cosimplicial variety with @xmath519 at the degree @xmath4 . then it gives an object in voevodsky s triangulated category of mixed motives over @xmath79 . in fact it also belongs to the triangulated subcategory @xmath520 of mixed tate motives over @xmath79 . and there is a canonical @xmath23-structure @xmath23 on @xmath520 . then @xmath521 is our mixed tate motive . similarly , by using the construction above for the hodge - tate structures and the fact that the hodge realization is a fully faithful functor on the category of pure tate motives , we obtain the frame morphisms for our motive coming from @xmath522^{\prime}$ ] and @xmath511^{\prime}$ ] defined above . the theorem is proved . in this section , we ll apply our preceding construction to the dilogarithm and show that the corresponding period matrix is the same as the one given by p. deligne . hence they are isomorphic . recall that @xmath82 now let s consider the moduli space @xmath523 . it s known that @xmath523 is the blow - up of @xmath524 at three points @xmath525 on the diagonal . let @xmath526 denote this blow - up . and let @xmath527 be the affine coordinates on @xmath524 . we list the ten boundary divisors of @xmath523 as follows ( see figure 1 ) . moreover we can describe the divisors @xmath538 and @xmath539 explicitly . by definition , @xmath538 is the divisor of singularities of the pullback meromorphic @xmath41-form @xmath540 of @xmath523 . we can check the following lemma directly and leave it to the reader . first we state the results . in the following statements , we only list the ( relative ) betti homology groups because by the comparison theorem the derham cohomology groups are isomorphic to the corresponding betti cohomology groups tensor with @xmath548 . [ m5.2 ] ( a ) if @xmath549 then : @xmath554 furthermore , there are natural bases such that the corresponding period matrix is @xmath85.\ ] ] hence , it coincides with the one given by deligne and they are isomorphic as hodge - tate structures . now we ll prove these two theorems . before we do it , let s give a more explicit description of the affine variety @xmath556 . we know @xmath557 is the blow - up of @xmath558 at the points @xmath559 , @xmath560 , @xmath561 , hence @xmath562 in which @xmath563 denotes the blow - up of @xmath564 at the points @xmath565 ( see figure 1 ) . we ll use the same letter @xmath566 for this blow - up . proof of theorem [ m5.1 ] : the idea is to decompose @xmath556 into the union of two subspaces , then use the mayer - vietoris exact sequences . + * case ( a ) : @xmath549 . * in this case we have : @xmath567 where @xmath568 denotes the unique intersection point of @xmath569 and the proper transform of @xmath570 . let @xmath571 , then it s open and isomorphic to @xmath572 which denotes the blow - up of @xmath573 at the point @xmath574 . let @xmath575 in @xmath564 . define @xmath576 , then it s clear that @xmath577 is the deformation retract of @xmath578 by the line contraction : @xmath579 . so we have the decomposition : @xmath580 . we claim that @xmath581 deformation retracts to @xmath582 , where @xmath583 is the @xmath66-sphere of radius @xmath584 . indeed , notice that @xmath585 , hence it is isomorphic to @xmath586 which deformation retracts to @xmath582 under the map @xmath587 , where @xmath588 and @xmath589 . hence , @xmath590 . by the alexander duality , we have @xmath591 the natural generator of @xmath592 is a simple loop around the divisor @xmath593 . since @xmath594 deformation retracts to @xmath595 , we have @xmath596 and @xmath597 . now @xmath598 , so it follows from ( * ? ? ? * page 473 - 474 ) that : @xmath599 now apply mayer - vietoris exact sequence , we obtain : + @xmath600 ; + @xmath601 @xmath602 + we also see that the generator for @xmath603 is a small simple loop around @xmath543 . * case ( b):@xmath86*. we see in this case : @xmath604 , where @xmath568 is the intersection point of @xmath569 and @xmath593 ( the proper transform of @xmath570 ) ; @xmath605 is the intersection point of @xmath606 and @xmath607 ( the proper transform of @xmath608 ) . let @xmath609 . it s open and isomorphic to @xmath610 . as we did in case ( a ) , we can construct a small open neighborhood @xmath611 of @xmath577 and @xmath612 of @xmath613 such that : ( 1 ) @xmath611 deformation retracts to @xmath577 , and @xmath612 deformation retracts to @xmath614 ; ( 2 ) @xmath615 ; ( 3 ) @xmath616 deformation retract to @xmath617 for some small @xmath584 . we define @xmath618 . then we have the decomposition : @xmath619 . and @xmath620 deformation retracts to the disjoint union of two copies of @xmath582 . now we can proceed as in case ( a ) and find @xmath621 , @xmath622 , other homology groups vanish . theorem [ m5.1 ] is proved . proof of theorem [ m5.2 ] : now we ll use the results of theorem [ m5.1 ] to compute the relative homology and cohomology groups . + * case ( a):@xmath549 . * first , let s look at the divisor @xmath623 ( see figure 2 ) . it consists of five irreducible components . let s label them as follows ( see the figure 1 and figure 2 ) : + @xmath624 , @xmath625 , @xmath626 , @xmath627 , @xmath628 + we have five intersection points of these components : + @xmath629 let @xmath630 be the disjoint union of the irreducible components @xmath631 , and @xmath632 be the disjoint union of the points @xmath633 ; @xmath634 . then we can compute @xmath635 by the following bicomplex @xmath636 : @xmath637 where @xmath638 denote the vector space of singular @xmath151-chain on @xmath261 with coefficients in @xmath447 . the vertical differential @xmath639 is the differential of the singular chain complex of @xmath261 and the horizontal differential @xmath584 is defined as following : @xmath640 where each @xmath641 is a @xmath642-chain of @xmath643 , @xmath644 . and @xmath645 where for each pair @xmath646 , @xmath647 is a @xmath642-chain of @xmath648 and @xmath649 represents the image of @xmath647 in @xmath643 under the incusion @xmath650 . that is , we have the following diagram : @xmath651 consider the spectral sequence with @xmath652 terms given as : @xmath653 the differential @xmath654 is induced by the horizontal differential @xmath584 . use the explicit generators of the homology groups in the theorem [ m5.1 ] , we immediately get the @xmath655 terms and @xmath656 : @xmath657 hence the spectral sequence degenerates at the @xmath658 terms . so , @xmath659 and the filtration induced by this spectral sequence coincides with the weight filtration . now let s turn to the relative de rham cohomology group @xmath660 . it is defined as the @xmath151-th cohomology of the total complex of the following bicomplex @xmath661 : @xmath662 where @xmath663 denote the vector space of @xmath664 complex @xmath151-forms on @xmath261 , the vertical differential @xmath665 is the exterior differentiation of forms and @xmath584 is defined as follows:@xmath666 where @xmath667 is a @xmath642-form of @xmath556 and @xmath668 is the restriction of @xmath667 on @xmath643 . @xmath669 where @xmath670 is a @xmath642-form of @xmath195 and for each pair @xmath646 , @xmath671 is the restriction of @xmath672 on @xmath648 . we have the following diagram : @xmath673 similar to the case for homology , we see that the corresponding spectral sequence degenerates at the @xmath658 terms . so we obtain : @xmath674 , and other relative cohomology groups vanish . for the case @xmath86 , the proof is similar and we leave it to the reader . we have the following three cochains in the total complex associated to the derham bicomplex @xmath661 : @xmath675 where @xmath676 is a function on @xmath677 satisfying @xmath678 and @xmath679 is a @xmath59-form on the component @xmath680 of @xmath195 . since @xmath679 and @xmath681 are holomorphic on @xmath680 and @xmath550 respectively , they are closed forms . hence @xmath682 are cocycles and represent elements in @xmath683 . + next , let s consider the cycles in the betti bicomplex @xmath684 . + let @xmath685 , where @xmath686 denotes the boundary of @xmath687 in @xmath195 . it s just the @xmath81 sides of the dotted pentagon in figure @xmath41 . let @xmath688 , where @xmath689 , and @xmath690 is a small positive number , @xmath691 . it s clear that the boundary @xmath692 consists of @xmath41 small cycles @xmath693 and @xmath694 . since @xmath695 does not contain the blow - up points , @xmath696 is isomorphic to @xmath695 and the boundary @xmath697 . the key point is that @xmath698 is contained in @xmath630 , which means exactly that @xmath699 is a cycle in the total complex . finally , let @xmath700 , where @xmath701 is the inverse image of a torus and since the three blow - up points are not on the torus , @xmath94 is isomorphic to its image . clearly the boundary of @xmath94 is zero . hence @xmath702 is a cycle . now we can calculate the period matrix @xmath703 between @xmath704 and @xmath705 , here @xmath706 , @xmath707 . it s straightforward that @xmath708 , @xmath709 + @xmath710 , and @xmath711 , @xmath712 . since @xmath713 is equal to the integral of the @xmath59-form @xmath714 over a small circle with center @xmath543 , we have @xmath715 . and @xmath716 therefore , we get the period matrix : @xmath717.\ ] ] since it s nonsingular , it follows that @xmath718 and @xmath719 are bases . for part(b ) , it s clear that : for the cocycles , we have @xmath720 ; for the cycles , we have @xmath721 . therefore , its period matrix is + @xmath722.\ ] ] a. a. beilinson , a. b. goncharov , v. v. schechtman and a. n. varchenko , _ aomoto dilogarithms , mixed hodge structure and motivic cohomology of pairs of triangle on the plane _ , in the grothendieck festschrift , i , prog . in math . 86 , birkhuser , boston , 1990 . p. deligne , _ theorie de hodge , i _ , actes cong . 1970 , t. 1 . 425 - 430 . p. deligne , _ theorie de hodge , ii _ , publ . 40,(1971),5 - 58 . deligne , _ theorie de hodge , iii _ , inst . hautes . 44,(1974),5 - 77 . p. deligne , _ resum des premiers exposs de a.grothendieck_ , sga 7 , lnm 288 , 1 - 24 . p. deligne , ; d. mumford , _ the irreduciblity of the space of curves of given genus _ , publ . math . inst . hautes . 36,(1969),75 - 109 . p. deligne and a. b. goncharov , _ groupes fondamentaux motiviques de tate mixte _ , ann . , @xmath723 srie , t. 38 , 2005 , 1 - 56 . l. euler , _ opera omnia _ , ser.1 , vol xv , teubner , berlin 1917 , 217 - 267 . p. griffiths and j. harris , _ principles of algebraic geometry _ , wiley and sons , 1978 . a. b. goncharov and yu . i. manin , _ mutiple @xmath6-motives and moduli spaces @xmath2 _ , compositio math . 140 ( 2004 ) , 1 - 14 . a. b. goncharov , _ polylogarithms in arithmetic and geometry _ , proc . of icm . vol . 1 , 2 ( zurich , 1994 ) , 374 - 387 , birkhauser , basel , 1995 . a. b. goncharov , _ polylogarithms and motivic galois groups _ , sympos . pure math . motives , vol . 55 , part 2 , ( 1994 ) , 43 - 97 . a. b. goncharov , _ volume of hyperbolic manifolds and mixed tate motives _ , j. amer . 12 ( 1999 ) , n2 , 569 - 618 . a. b. goncharov , _ multiple polylogarithms and mixed tate motives _ , math.ag/0103059 . a. b. goncharov , _ periods and mixed motives_. math.ag/0202154 . r. hain , _ classical polylogarithms _ , motives , sympos . pure math . vol . 55 , part 2 , ( 1994 ) , 1 - 42 . r. hain and r. macpherson , _ higher logarithms _ , ill . 34 , ( 1990 ) . c. gerhardt ( ed ) and g. leibniz , _ mathematische schriften iii/1 _ , georg olms verlag , hildescheim and new york , 1971 , pp . 336 - 339 . l. lewin , _ dilogarithms and associated functions _ , north - holland , 1981 . s. keel , _ intersection theory of moduli space of stable n - pointed curves of genus zero _ , trans.ams , 330 , n2 ( 1992 ) , 545 - 574 . m. kontsevich and yu . i. manin , _ quantum cohomology of a product _ , invent . math.,124 , 313 - 339 ( 1996 ) . f. f. knudsen , _ the projectivity of the moduli space of stable curves ii the stacks @xmath724 _ , math . scand . 52 ( 1983 ) , 163 - 199 i. manin , _ correspondencs , motifs and monoidal transformations _ , math.ussr-sbornik 6 ( 1968 ) , 439 - 470 . i. manin , _ frobenius manifolds , quantum cohomology , and modulis spaces _ , vol . 47 , colloquim publications , ams , 1999 . v. voevodsky , _ triangulated category of motives over a field _ , in : cycles , transfers , and motivic homology theories , 188 - 238 , ann . of math . , 143 , princeton univ . press , nj , 2000 . q. wang , _ multiple polylogarithm motives and moduli spaces @xmath2 _ , ph.d . dissertation , brown university , may 2004 . q. wang , _ a geometric construction of motivic classical polylogarithm _ , in preparation .

we formulate a simple model for a gas of extended hadrons at zero chemical potential by taking inspiration from the compressible bag model . we show that a crossover transition qualitatively similar to lattice qcd can be reproduced by such a system by including some appropriate additional dynamics . under certain conditions , at high temperature , the system consist of a finite number of infinitely extended bags , which occupy the entire space . in this situation the system behaves as an ideal gas of quarks and gluons .

0812.1044

the phase transition of strongly interacting matter has been intensively studied for many years . as early as 1960s , before the discovery of qcd , there was speculation about a possible new phase of strongly interacting matter , based on studies of the thermodynamics of a hadron gas . particularly , in the statistical bootstrap model @xcite , the asymptotically exponential mass spectrum of hadrons implied the existence of a limiting temperature of about @xmath0 mev ( the hagedorn temperature ) above which hadrons can not exist . after the discovery of qcd and in particular asymptotic freedom , the discussion focused on the ideas of a quark gluon plasma ( qgp ) , a system of weakly interacting quarks and gluons , and a possible transition between a pion gas and a quark gluon plasma . the physical picture for such a transition was that at the critical temperature the additional degrees of freedom carried by the quarks and gluons would be released leading to a rapid increase of the entropy , energy - density and pressure . this was supported by first lattice qcd ( lqcd ) calculations @xcite . lately , however , the hadron resonance gas experienced a renaissance mainly due to its successful description of hadron yields in heavy ion and also elementary particle collisions @xcite . in addition it was realized that the sharp rise of the entropy density near the transition temperature , as observed in lattice qcd calculations , could be accounted for by a hadron resonance gas as well @xcite . the success of the hadron resonance gas below the transition temperature @xmath1 , however , changes the physical interpretation for the qcd phase transition . instead of releasing additional degrees of freedom at @xmath1 , the system has to reduce the number of active degrees of freedom at the transition temperature , because a hadron gas has many more than a quark gluon plasma . in case of a hagedorn exponential mass spectrum , the entropy density actually diverges at the hagedorn temperature . but even if the number of hadrons is restricted to those needed for a successful description of hadronic final states in heavy ion and elementary particle collisions , the entropy quickly exceeds that observed on the lattice . meanwhile many papers have addressed this issue ranging from qcd - based approaches ( see for example @xcite ) to various generalizations of the hadron gas @xcite . in this article we will study the transition region , and we will provide a simple and intuitive modification to the hadron gas model in order to qualitatively reproduce the crossover transition observed in lattice qcd @xcite . our calculations are based on the ideas proposed in @xcite , where it was shown that under certain circumstances a gas of extended hadrons could produce phase transitions of the first or second order , and also a smooth crossover transition that might be qualitatively similar to that of lattice qcd . we first observe that for the typical particle density at the transition the size of hadrons needs to be taken into account as it leads to a considerable suppression of the available phase space . as a result the number of effective degrees of freedom is reduced . this suppression alone , however , does not explain the ideal gas behavior of lattice qcd at high temperatures , above @xmath1 . additional model assumptions have to be made . for example one could explicitly introduce a deconfined phase and then match the two different phases . for this exercise to work , however , rather detailed assumptions about the intrinsic nature of hadrons have to be made in order to avoid the occurrence of an actual phase transition @xcite . in this paper , we follow a different route . we adopt the philosophy that the same partition function should describe both `` phases '' . to this end we need to introduce appropriate dynamics in order to model the crossover observed on the lattice . this approach is similar in spirit to @xcite . we find that the mit bag model @xcite of the hadrons is well suited for our purposes , because it embodies confined and deconfined phases from the very beginning . thus , we will describe hadrons as extended bags of qgp and we will assume an infinite mass spectrum of the hagedorn s type . the additional dynamics needed to describe the transition are simply the elastic interactions between hadrons . they give rise to a _ kinetic _ pressure that in turn `` squeezes '' the bag - like hadrons . we find that the behavior of our system depends sensitively on the choice of parameters for the mass spectrum . one can obtain either a real phase transition @xcite or , as we shall show , a crossover . in addition , in the latter case the specific parametrization for the mass spectrum affects the microscopic structure of the gas of `` compressible '' hadrons at high temperatures ( @xmath2 ) . one finds either a high temperature phase that is populated by one or few infinitely large bags , consistent with the usual picture of a qgp . or , for a different choice of parameters , one obtains a system of many , densely packed heavy hadrons , which nonetheless exhibit the thermodynamic properties of a qgp of massless quarks and gluons . on first sight , our approach appears to be similar to the ideas of percolation models @xcite . the finite size corrections to the statistical ensemble remove all the configurations with overlapping hadrons , resulting in large hadronic states dominating the partition function . this is similar to percolation . however , there are quite some differences in the specific implementation . first , in our model , we do not consider an explicit coupling between the bags as it is done in the percolation model of @xcite . instead , we take the effect of the kinetic pressure onto the bag sizes into account , resulting in a self - consistency relation for the effective bag pressure . this is more in the spirit of a mean field description , although we do not introduce an additional interaction but simply consider the kinetic pressure . second , in contrast to purely geometric percolation , in our model the number of hadrons and their sizes are not independent quantities . in a given multihadron state , melting two or more hadrons together ( to form a bigger one ) results in a different kinetic pressure and , in turn , in the rearrangement of the sizes of all the hadrons . throughout this paper , we will maintain a simple schematic approach to highlight the main features of the model leaving a more detailed quantitative analysis and further generalizations to future work . we will confine ourselves to the simplest case of nonrelativistic boltzmann particles and we will neglect subtleties such as surface effects and van der waals type residual interaction among the bags . this paper is organized as follows : in section [ pf ] we introduce the main ideas of the model and derive the grand - canonical partition function for the gas of compressible hadrons . in section [ isobpf ] we will use the corresponding isobaric partition function to perform a comprehensive numerical analysis . we will study the pressure , the energy density and the entropy density of the system . we will further analyze particles number , the filling fraction and the average mass of particles in the system . in this section we will set up the general formalism for our model . let us start with the partition function @xmath3 of an ideal gas of boltzmann particles of mass @xmath4 and degeneracy @xmath5 in the nonrelativistic limit . for the subsequent discussion it is advantageous to express the partition function @xmath3 in a multiplicity expansion , i.e. as a sum of partition functions @xmath6 for fixed particle numbers @xmath7 : @xmath8 with @xmath9}= \exp{\left[-\frac{m}{t}\right ] } \left(\frac{m t}{2 \pi}\right)^{3/2}\ ; . \label{1.2}\ ] ] here @xmath10 and @xmath11 are the volume and the temperature of the system , respectively . the function @xmath12 is the _ grand - canonical _ partition function with vanishing chemical potentials . in the context of this paper we shall refer to @xmath13 as the _ canonical _ partition function keeping in mind that this notation deviates from the conventions for relativistic hadron gases , where the canonical ensemble has fixed abelian charges ( such as electric charge , strangeness , baryon number ) , but no constraints on the number of particles . ( [ 1.1 ] ) can be easily generalized to a multispecies gas of particles . if we label with @xmath14 the various particle species we have : @xmath15 = \exp{\left[v \sum_{j=1}^{k } g_j \phi(m_j , t)\right]}\ ; . \label{1.3}\ ] ] in case of @xmath16 , it is convenient to replace the discrete index @xmath17 with a continuous spectrum density @xmath18 so that the number of species in the mass interval @xmath19 $ ] is given by @xmath20 . formally , we make the substitution : @xmath21 by expanding the exponential in eq . ( [ 1.3 ] ) the partition function can then be written as : @xmath22 \ ; . \label{1.5}\ ] ] because a hadron gas , or , more precisely , a hagedorn gas is characterized by an exponential mass spectrum , we set @xmath23 where the parameters @xmath24 and @xmath25 will be determined from empirical data . in the case of gas - of - bags models , which also have an exponential mass spectrum , the parameters @xmath24 and @xmath25 will have to be determined from the underlying ( dynamical ) model parameters , such as bag pressure , and so on . we note that @xmath24 has dimensions of @xmath26^{\alpha-1}$ ] and @xmath25 typically ranges from @xmath27 to @xmath28 depending on the model . in eq . ( [ 1.7 ] ) , @xmath29 simply parametrizes the mass spectrum . in the context of the mit bag model , @xmath29 can be interpreted as the effective `` temperature '' inside the bag , as will be discussed in section [ rpf ] . by substituting eq . ( [ 1.7 ] ) in eq . ( [ 1.5 ] ) we arrive at the following partition function for a hadron gas @xmath30 \\ \nonumber & & \times \exp{\left[\frac{\sum_{i=1}^n m_i}{t_0}-\frac{\sum_{i=1}^n m_i}{t}\right ] } \ ; .\end{aligned}\ ] ] this partition function @xmath3 ( and also @xmath6 ) is divergent for @xmath31 as already pointed out by hagedorn . although an upper limit in the mass spectrum @xmath18 regulates the divergences , it will not prevent the system from having a much higher entropy density than that observed on the lattice . an exponential mass spectrum without any cutoff may certainly be an oversimplification and a more realistic calculation may take into account discrete states as well as a mass spectrum that grows less than exponential above a certain mass . however , empirically the known hadronic states do indeed grow exponentially up to a mass of @xmath32 . above that , very few states are known and it is not clear if this is an indication of a saturating density of states or simply the lack of experimental data on higher mass resonances . therefore , working with an exponential mass spectrum without any cutoff appears to be an approximation as good as any other . furthermore , because we are interested only in bulk thermodynamic quantities such as energy density and pressure , the use of a continuous mass spectrum should be a reasonable approximation as all these quantities represent integrals / sums over the mass spectrum . therefore , in this paper we will assume that the mass spectrum is of the hagedorn type and will discuss a dynamical scenario in the framework of the mit bag model , which will regulate the partition function . to develop the partition function of our model , we need to recall some of the basic features of the mit bag model @xcite . in its simplest formulation , hadrons can be considered as _ bags _ of partonic fields confined in a spatial region with a constant potential energy per unit volume @xmath33 , where @xmath33 is commonly referred as the bag constant or bag pressure . the total energy , i.e. the mass @xmath4 of a bag with volume @xmath34 is then given by @xcite : @xmath35 where @xmath36 is the internal energy of the field inside the bag . when its linear extension is larger than the wavelengths of the partons ( the quanta of the inner field ) , we can approximate the bag by a gas of free massless particles confined to its volume @xcite . for a sufficiently large @xmath34 , the internal energy is then given by the relation : @xmath37 where @xmath38 is the pressure of the gas . for a single hadron the stability condition requires : @xmath39 which then gives @xmath40 the effective temperature of the bag is related to the bag pressure by the relation @xmath41 , where @xmath42 is a dimensionless constant whose value depends on the number of internal degrees of freedom of the gas inside the bag . for large @xmath34 , the entropy @xmath43 of a bag , is the entropy of a massless gas with internal energy @xmath36 and pressure @xmath33 @xcite , therefore ) one should subtract a constant @xmath44 which corresponds to the entropy at @xmath45 . here , this term has been omitted as it is immaterial for our purposes . ] , @xmath46 from eq . ( [ entrobag ] ) one can derive the level density @xmath47 note that the generic spectrum introduced in eq . ( [ 1.7 ] ) , has an additional contribution : @xmath48 . this factor can be interpreted as a logarithmic correction to the entropy of the bag . in what follows , we will retain the spectrum in eq . ( [ 1.7 ] ) and we will analyze different values of @xmath25 . of course , eq . ( [ baglevdens ] ) will correspond to the case @xmath27 . once the temperature @xmath11 of the gas of bags approaches @xmath29 , the average masses and hence the volume ( see eq . ( [ massbagm ] ) ) of the bags grow very fast . therefore , the bag - like hadrons tend to occupy more and more of the available space and , eventually , they will overlap . to avoid multiple counting of the phase space , configurations with overlapping bags need to be excluded from the partition function . for a finite system of volume @xmath10 this can be achieved with an excluded volume correction , where the total volume @xmath10 is replaced by the available volume @xmath49 , where @xmath50 is the volume of the @xmath51th particle . in addition the volume of all bags @xmath52 should not exceed the total volume @xmath10 . following ref . @xcite , this leads to the modified @xmath7-particle phase space integral @xmath53 \rightarrow \left[\prod_{i=1}^n \frac{1}{(2 \pi)^3 } \int \d^3 { { \rm p}}\right ] \left(v-\sum_{i=1}^n v_i\right)^n \theta{\left ( v-\sum_{i=1}^n v_i \right ) } \ ; . \label{nphsp}\ ] ] this modified phase space integral results in a well - defined and finite partition function at every temperature @xmath54 \\ \nonumber & & \times \exp{\left[\frac{\sum_{i=1}^n m_i}{t_0}-\frac{\sum_{i=1}^n m_i}{t}\right]}\left(v-\sum_{i=1}^n v_i\right)^n \theta{\left ( v-\sum_{i=1}^n v_i \right ) } \ ; .\end{aligned}\ ] ] it can be shown that such a system of extended hadrons leads to a constant value for the energy - density @xmath55 , in contradiction with lattice qcd , where the energy density is found to increase with the fourth power of the temperature , @xmath56 . in fact , for @xmath57 , the most favorite configurations are those where the hadrons occupy all the available space . in this case , the energy density of the system correspond to @xmath58 , i.e. the inner density of the hadrons ( see eq . ( [ massbagm ] ) ) . the underlying reason for this behavior is that the system is not able to pick up additional kinetic energy once the entire volume is filled with bags , as the bags have no more room to move . obviously some additional dynamics needs to be included to allow for the system to pick up more energy as the temperature is increased . to this end we adopt the idea of compressible bags @xcite . more precisely , we will allow the volume of the hadrons to vary under the effect of the pressure generated by their own thermal motion in a self - consistent way . consequently , as the temperature and hence the pressure increase , the bags will be compressed and acquire a higher internal mass / energy density . we will show that the system does not exhibit any limiting value of the energy density , and , under appropriate conditions , exhibits the desired increase of the energy density and entropy . to illustrate the underlying mechanism , let us consider a gas of many hadrons . for small temperatures , @xmath59 , the system is dilute ( @xmath60 ) and behaves like a gas of noninteracting point - particles . with increasing temperature , the average mass , and hence the spatial extent of the hadrons , increases and as @xmath11 approaches @xmath29 the dilute - gas approximation seizes to be valid . the pressure exerted by the other particles becomes sizable and its effect on the hadrons properties , such as the size , can no longer be ignored . in other words , in addition to the bag pressure @xmath33 , every particle in the system will feel an additional _ kinetic _ pressure @xmath61 that is generated by the thermal motion of the other hadrons in the gas . in this situation , the stability condition , eq . ( [ dum ] ) , needs to be modified by taking onto account the contribution of the kinetic pressure @xmath61 . microscopically , the pressure @xmath61 can be interpreted as the consequence of elastic collisions . the effect of inelastic collisions , which are certainly present in a hadron gas , in our approach are accounted by the infinite mass spectrum of hadrons without enforcing any constraint on the number of particles @xmath7 . in this way all the possible configurations with few large hadrons or many small ones are included . this is analogous to the hadron - resonance gas model , where a large part of the inelastic interaction is taken into account by adding resonances as free particles in the gas . neglecting any surface effect , the simplest generalization of the stability condition , eq . ( [ dum ] ) , is @xmath62 a pictorial illustration of the pressure balance in the last equation is given in fig . ( [ bp ] ) . each hadron in the gas is characterized by the same internal pressure @xmath38 . but instead of being a constant ( as in the case of a single hadron in the vacuum where @xmath63 ) , @xmath38 now depends on @xmath11 and @xmath10 , and must be evaluated in a self - consistent fashion from the partition function itself . it is clear that the number of hadrons , their sizes and the kinetic pressure are all connected by the above self - consistency relation . accordingly , as we have already pointed out , if we split or combine two or more hadrons , the pressure , and thus the volume of all the hadrons in the system , changes . this new state corresponds to a distinct state of the ensemble that can not be obtained by a simple geometrical clustering procedure as usually done in percolation models @xcite . to account for this additional dynamics , we need to generalize the partition function , eq . ( [ 1.7b ] ) . to this end , we write the explicit dependence of volumes and the masses of the bags on the pressure @xmath38 @xmath64 we also rewrite the exponential mass spectrum , @xmath65 in terms of the general expression for the entropy @xmath66 leading to the substitution @xmath67 finally , because the bag masses depend on @xmath38 , instead of integrating on @xmath68 as in eq . ( [ 1.7b ] ) we will perform the integral over the internal energies , i.e. we substitute @xmath69 the factor @xmath70 in the previous formula ensures that we recover the partition function , eq . ( [ 1.7b ] ) in the limit of @xmath71 , i.e. in the dilute gas limit . with the replacements in eq . ( [ replacements ] ) to ( [ intvar ] ) , the modified grand - canonical partition function can now be written on the basis of eq . ( [ 1.7b ] ) and reads : @xmath72 \exp{\left\{\left [ \frac{4}{3kp_r^{1/4}}-\frac{1}{t}\left(1+\frac{b}{3p_r}\right ) \right]\sum_{i=1}^n u_i\right\}}\\ \nonumber \ ; & & \times \left(v-\sum_{i=1}^n \frac{u_i}{3p_r } \right)^n \theta{\left ( v-\sum_{i=1}^n\frac{u_i}{3p_r } \right)}\ ; .\end{aligned}\ ] ] the new partition function @xmath3 is identical ( by construction ) to eq . ( [ 1.7b ] ) when @xmath63 . conversely , for a given set @xmath73 , the effect of a finite kinetic pressure @xmath74 is to squeeze each bag to a smaller size ( see eq . ( [ replacements ] ) ) , and as a result , the effective bag temperature @xmath75 increases . the eq . ( [ gcpartfunupr ] ) can be made more familiar by substituting @xmath76 leading to @xmath77\\ \nonumber \ ; & \times & \exp{\left\{\left [ \frac{1}{kp_r^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4p_r}\right ) \right]\sum_{i=1}^n \eta_i\right\ } } \\ \nonumber \ ; & & \times \left(v-\sum_{i=1}^n \frac{\eta_i}{4p_r } \right)^n \theta{\left ( v-\sum_{i=1}^n\frac{\eta_i}{4p_r } \right)}\ ; .\end{aligned}\ ] ] obviously , in the dilute gas limit , @xmath71 @xmath78 . we further introduce a lower bound @xmath79 for the integrals over @xmath80 . this is needed because , for @xmath81 , the spectrum in eq . ( [ 1.7 ] ) has a pole in @xmath82 , resulting in a divergent partition function for @xmath83 . because there are no hadrons lighter than the pion , we will set @xmath84 gev . because the pressure @xmath61 is thermally generated , it must be calculated from the partition function itself , resulting in a self - consistency relation . this is best achieved by introducing the isobaric partition function , which is defined as the laplace transform of @xmath3 over the variable @xmath10 : @xmath85 } \ ; . \label{ibpf1}\ ] ] the quantity @xmath86 in eq . ( [ ibpf1 ] ) plays the role of a constant external pressure . accordingly , the equilibrium condition requires @xmath87 the integral in eq . ( [ ibpf1 ] ) can be solved analytically ( see appendix a ) and gives : @xmath88^n=\frac{1}{s - f(t , s ) } \label{ibpf2}\ ] ] with @xmath89 } \ ; .\end{aligned}\ ] ] in the limit @xmath90 the asymptotic behavior of @xmath3 is defined by the singularity of @xmath91 with the largest real part @xcite . we have two distinct cases : @xmath92 and @xmath93 . for @xmath92 the transform @xmath91 in eq . ( [ ibpf2 ] ) has two kinds of singularities : the first , @xmath94 , is given by the pole of @xmath95 , i.e. @xmath96 whose solution is the pressure of the system according to eq . ( [ pkst ] ) . the second singularity , @xmath97 , corresponds to a divergence of the function @xmath98 itself . this happens if the exponent of the integrand of @xmath99 in eq . ( [ ibpf3a ] ) vanishes , i.e. @xmath100 the situation is schematically represented by the leftmost curve in fig . ( [ fts ] ) . the solid line represents the function @xmath99 for a given temperature and the @xmath101 s dashed line corresponds to @xmath102 . the intersection between the dashed and the solid line corresponds to the solution @xmath103 of eq . ( [ ibpf4 ] ) and is denoted by a black dot . the function @xmath99 , is a positive function of @xmath102 that goes to infinity for @xmath104 and tends to zero as @xmath105 . consequently , there is always a solution for eq . ( [ ibpf4 ] ) and the pole @xmath106 corresponds to the rightmost singularity . thus the pressure has always a solution . for @xmath93 the situation is different , however . in this case the function @xmath98 has an essential discontinuity at @xmath97 : it is finite at @xmath104 , and diverges for @xmath107 . since @xmath97 increases with temperature ( see eq . ( [ sing ] ) ) and @xmath108 as @xmath105 , for sufficiently large @xmath11 , @xmath109 , and , consequently , eq . ( [ ibpf4 ] ) does not have a solution . this situation is illustrated by the rightmost curve in fig . ( [ fts ] ) where @xmath110 ( denoted by a x ) lies below the diagonal since the singularity @xmath111 for @xmath112 ( see eq . ( [ sing ] ) ) . ] . both these cases , have been discussed in @xcite where the absence of the solution @xmath94 was identified with the onset of a phase transition . here , we analyze in detail the case of a crossover transition , i.e. @xmath92 . before we proceed , let us fix the model parameters . in what follows , we set @xmath113 and we keep the product @xmath114 fixed . this yields a bag pressure @xmath115 gev@xmath116 ( i.e. @xmath117 mev ) which is a plausible value for the bag model parameter . the constant @xmath42 has been chosen in order to obtain roughly the same value for @xmath118 as lqcd for large @xmath11 ( see fig . ( [ isob1 ] ) ) . its value can also be estimated by counting the degrees of freedom of a thermal system of independent quarks and gluons . our choice lies between the values for the lightest quark doublet u , d ( @xmath119 ) and for u , d and s quark ( @xmath120 ) . the values for the remaining parameter @xmath24 have been fixed by fitting the shape of the actual hadron mass spectrum over the mass range of @xmath121 . they are given in table [ tab1 ] . . [ cols="<,^,^,^,^,^,^",options="header " , ] of course a different mass range or a different choice for @xmath29 would affect these fits . however , for our schematic considerations here , a fine tuning of the model parameters is rather meaningless . in the same spirit we also ignore the @xmath122 deviation of the lqcd result for @xmath118 from the free gas ( stefan - boltzmann ) limit @xcite . in fig . ( [ isob1]a ) we plot the pressure @xmath123 evaluated with the isobaric partition function , eq . ( [ ibpf4 ] ) . as one can see , the results depend on the choice of @xmath25 . for @xmath27 and @xmath124 , the curves grow with the temperature with larger slopes for smaller @xmath25 s . instead , for @xmath125 , they settle onto constant asymptotic values ( as we have verified numerically up to @xmath126 gev ) . as @xmath25 change from @xmath127 to @xmath128 the asymptotic value converges very fast to @xmath129 ( the stefan - boltzmann limit . accordingly , the energy density is given by @xmath130 . ] ) from above . as shown in the plot , the curves practically coincide with the stefan - boltzmann limit already for @xmath131 . this behavior can be understood by inspecting the solution @xmath132 for the pressure . because @xmath94 is always larger than @xmath97 , we have @xmath133 and for @xmath134 @xmath135 the pressure @xmath136 converges to @xmath137 only for sufficiently large values of @xmath25 , when the solutions @xmath94 and @xmath97 get closer and closer as @xmath134 , to obtain the pressure of an ideal gas the difference @xmath138 must decrease faster than @xmath139 . for the ratio @xmath140 it is sufficient that the difference @xmath138 grows slower than @xmath141 . ] . in fig . ( [ isob1]b ) , we plot the ratio @xmath118 , where @xmath55 has been evaluated numerically by using the relation : @xmath142 again this quantity converges to a finite asymptotic limit only for @xmath125 and , as before , coincides with the massless gas limit @xmath143 for @xmath131 and @xmath144 . the overall behavior is roughly the same as lqcd except for a small `` horn '' right above @xmath29 . a closer look reveals that this is due to the constant bag pressure @xmath33 that produces a contribution @xmath145 to the energy density of the system . being a constant term its contribution to @xmath146 becomes negligible at high temperatures . in fig . ( [ isob2]a ) , the ratio @xmath147 is plotted up to @xmath148 gev . this quantity corresponds to the trace of the energy - momentum tensor @xmath149 , which is actually the fundamental quantity calculated in lqcd @xcite . in fig . ( [ isob2]b ) , we plot the ratio @xmath150 , where @xmath102 is the entropy density . again , for @xmath27 and @xmath124 , the curves do not converge to a constant value , as expected from the previous results for @xmath55 and @xmath61 . in our scheme , the pressure @xmath61 ( and thus the energy density and the entropy ) exceeds the corresponding stefan - boltzmann limit , except for @xmath151 and @xmath128 where it converges to it . to obtain a lower pressure ( at least at finite temperatures ) one needs to further reduce the effective degrees of freedom of the system . this could be possibly achieved by introducing a surface energy term . in addition to the fact that part of the energy of the system would be spent to create the bags surface , such a contribution would favor spherically shaped bags , resulting in a further suppression of the accessible phase space . probably , a similar effect could be also obtained including some residual repulsive interaction of the van der waals type . we stress that the general behavior of our model for the pressure , energy- and entropy - density can not be obtained by simply introducing an upper mass - cutoff on the exponential spectrum in eq . ( [ 1.7a ] ) : the entropy - density would considerably exceed that obtained from lqcd even if we kept only masses up to @xmath152 gev . in addition , the system would reach the massless gas limit only at temperatures much higher than the cutoff itself . on the other hand , in our model it is absolutely essential to assume an infinite mass spectrum . otherwise the flat behavior in fig . ( [ isob1 ] ) and on fig . ( [ isob2]b ) would be spoiled and all these quantities would decrease with the temperature . it is also interesting to plot the average bag effective `` temperature '' @xmath153 . as shown in fig . ( [ tb ] ) , for @xmath131 and @xmath144 this converges to @xmath11 very quickly above @xmath29 . for @xmath154 or smaller the effective bag temperature is always larger than the system temperature . this fact is a direct consequence of the inequality in eq . ( [ presslim ] ) that gives @xmath155 notice also that in the region @xmath112 , @xmath156 for any value of @xmath25 . in other words , the pressure @xmath61 is negligible with respect to @xmath33 and the system behaves as a standard hadron gas . it is worth mentioning that in our framework the compressible hadrons can exchange energy only through a mechanical work ( compression ) . they are , therefore , thermally insulated from the rest of the system and the bag temperature , as well as the entropy , must be understood as quantities that measure the degeneracy of the hadronic states . finally , for @xmath125 , our model seems to produce a smooth crossover transition toward a new regime whose features are very similar to those of a gas of massless particles , even though no deconfined states are included in the partition function . to better understand this behavior it is useful to study the particles density @xmath157 ( fig . ( [ new1]a ) ) and the _ filling fraction _ ( @xmath158 ) ( fig . ( [ new1]b ) ) which is defined as : @xmath159 where @xmath160 is the average volume occupied by the hadrons ( for a rigorous definition and formula see appendix b ) . the particles density can be calculated from the isobaric partition function by introducing a fictitious fugacity @xmath161 ( to be set to 1 afterward ) for each particle in the system , i.e. replacing @xmath3 with @xmath162 accordingly , the isobaric partition function in eq . ( [ ibpf2 ] ) becomes @xmath163 and the corresponding solution for the pressure @xmath164 . in the infinite volume limit , the particles density can be then obtained as : @xmath165 where in the last equality we have used the known relation @xmath166 . as shown in fig . ( [ new1]a ) , the particles density grows very rapidly as @xmath11 approaches @xmath29 from below . this is qualitatively what one expects for the hadron gas , where the average number of particles shows a monotonically growing behavior . conversely , for @xmath167 this quantity depends very strongly on the choice of the parameter @xmath25 . for @xmath27 , @xmath124 and @xmath127 we observe a change in the slope , but the curves still grow monotonically . for @xmath168 and @xmath152 the particles density has a local minimum at @xmath169 gev and @xmath170 gev , respectively ( the latter lies outside the plotted region ) , and then it starts growing again with smaller slopes for larger values of @xmath25 . for @xmath144 the local minimum has disappeared , and after a sharp maximum at @xmath171 gev the particles density goes to zero . this is the effect of the finite size of hadrons , which tends to saturate the available system volume . in fact , as shown in fig . ( [ new1]b ) , when @xmath172 the filling fraction , eq . ( [ 3.5a ] ) , is relatively small , whereas for higher temperatures , the system is almost totally filled by extended particles , i.e. @xmath173 . in this scenario , the space and the phase space available is strongly suppressed , and the system tends to be populated by a smaller number of heavy particles . this effect is strongest for @xmath144 ( the filling fraction converges very fast to 1 ) and therefore @xmath174 . for smaller values of @xmath25 this saturation effect becomes slightly less pronounced . a closer inspection reveals that for @xmath175 the filling fraction has a maximum at very high temperature and then decreases with a very small slope , the rate of decrease of the filling fraction for large temperatures is maximum , but still only @xmath176 going from @xmath177 gev to @xmath178 gev ] , whereas for @xmath179 it seems to settle to a constant value @xmath180 . the phase space , therefore , is never entirely suppressed . the behavior of the system changes continuously by varying @xmath25 . a numerical analysis indicates that there exist a value @xmath181 between @xmath182 and @xmath183 such that , at high @xmath11 , the particles density vanishes for any @xmath184 . in this range , the system is populated by one or few infinitely extended hadrons that occupy the entire space , filling the system with their inner qgp matter , which is a possible scenario for the deconfined phase . conversely , for @xmath185 , we find many , rather heavy,``squeezed '' hadrons , which nonetheless mimic an ideal gas of massless particles . the number of particles , however , might be affected by the introduction of a surface energy term in the spectrum in eq . ( [ 1.7 ] ) . such a term would result in an energy cost associated with the splitting of a large hadron into many smaller ones and might then widen the range of values for @xmath25 that lead to @xmath186 at temperatures above @xmath29 . note that , even though @xmath187 can have a minimum or even vanish , the entropy of the system always increases monotonically with the temperature ( see fig . ( [ isob2]b ) ) . this is due to the fact that the dominant contribution ( already at @xmath188 ) comes from the bags entropies . this has been checked by using the classical expression for a gas with @xmath189 particles ( with the excluded volume correction ) and degeneracy @xmath5 : @xmath190 the contribution @xmath191 corresponds to the sum of the bag entropies in eq . ( [ entropy ] ) , and it is dominant at high @xmath11 . another interesting quantity is the average hadron mass @xmath192 , shown in fig . ( [ new2 ] ) that has been evaluated according to @xmath193 as one can see , smaller values of @xmath25 correspond to higher @xmath192 at low temperature , whereas the situation is reversed at high @xmath11 . this follows from eq . ( [ amass ] ) as a direct consequence of the behavior of @xmath187 ( fig . ( [ new1]a ) ) . for @xmath144 the average mass @xmath192 grows with the temperature with the maximum slope and diverges when @xmath174 , i.e. when the system is populated by a finite number of infinite `` hadrons '' . in the vicinity of the transition region @xmath192 varies from @xmath194 gev ( for @xmath27 ) to @xmath195 gev ( for @xmath144 ) at @xmath196 gev , and from @xmath197 gev to @xmath198 at @xmath199 gev , for @xmath27 and @xmath128 , respectively . these values will somewhat depend on the choice of the model parameters . they would depend even more on an eventual upper mass cutoff . in fact , the high value of @xmath192 for @xmath27 results from our assumption of an infinite mass spectrum and already at @xmath196 gev it falls outside the region of the known hadrons . our estimates are , however , lower than what one obtains for an exponential spectrum of point - like hadrons as in eq . ( [ 1.7a ] ) . in such a case , for @xmath27 the average mass is @xmath200 gev at @xmath196 gev and infinity at @xmath201 gev ( the partition function itself is divergent ) . a mass cutoff at @xmath202 gev would reduce these numbers to @xmath203 gev and @xmath204 gev at @xmath196 gev and @xmath199 gev , respectively . as a final remark , we want to discuss the consistency of the model . to check this point we must make sure that quantities such as the energy density ( that we have evaluated from the pressure by using eq . ( [ isobeps ] ) ) correspond to a thermal average of the form @xmath205 a first hint in this direction , is given , _ a posteriori _ , by the results shown in this section , particularly , by quantities such as the filling fraction . the @xmath158 has been evaluated by making use of the relation in eq . ( [ amass ] ) ( see appendix b ) that implicitly relies on a form like eq . ( [ concheck1 ] ) for the energy density . indeed , a wrong thermodynamical interpretation of @xmath55 would likely have lead to dramatic consequences on the filling fraction , which contrarily assumes only physical values in the interval @xmath206 $ ] . however , to make a more direct test , we will provide an approximate expression for the grand - canonical partition function and we will compare @xmath55 in fig . ( [ isob1]b ) with its corresponding value obtained as in eq . ( [ concheck1 ] ) . this can be done for the case @xmath154 . for this particular value of @xmath25 the multiple integrals over @xmath80 in eq . ( [ partfunmpr ] ) can be reduced to an unidimensional integral ( see appendix c ) that greatly facilitates the numerical treatment . although , as demonstrated in this section , results do depend on the choice of @xmath25 , for the range of @xmath207 , one can hope that the following arguments will be still valid . using @xmath208 in eq . ( [ partfunmpr ] ) we get for the partition function @xmath209 with @xmath210^n \frac{1}{n ! } \left [ \prod_{i=1}^n \int_{m_c}^{\infty } \d \eta_i \right]\nonumber \\ \ ; & & \times \exp{\left\{\left [ \frac{1}{kp_{r , n}^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4p_{r , n}}\right ) \right]\sum_{i=1}^n \eta_i\right\}}\\ \nonumber \ ; & & \times \left(v-\sum_{i=1}^n \frac{\eta_i}{4p_{r , n } } \right)^n \theta{\left ( v-\sum_{i=1}^n\frac{\eta_i}{4p_{r , n } } \right)}\ ; .\end{aligned}\ ] ] here we have added the suffix @xmath7 to @xmath38 to indicate the dependence of the pressure @xmath211 on the particle number in the canonical ensemble . the pressure @xmath211 is now just a parameter of the model . to determine its value , we need to find the value of @xmath211 that maximizes the logarithm of the integrand in eq . ( [ gcpartfun ] ) : @xmath212\sum_{i=1}^n \eta_i+n \log \left(v-\sum_{i=1}^n \frac{\eta_i}{4p_{r , n } } \right ) \ ; , \label{integrand}\ ] ] where ( for the moment ) we omit the @xmath213 function . the above expression depends on @xmath214 . to simplify the following derivation , we will introduce an approximation . instead of solving @xmath215 for a generic set @xmath216 , we find a solution for the most important configurations defined by the value @xmath217 that maximizes the integrand . in other words , we solve the system of equations : @xmath218 note that the second condition ensures that @xmath219 has the form of eq . ( [ concheck1 ] ) as any implicit dependence on @xmath11 of @xmath211 do not contribute to @xmath220 . in fact , if we denote with @xmath221 the derivative performed only on the explicit @xmath11 dependence of @xmath222 we have @xmath223 where the second term vanishes because of the second condition in eq . ( [ system ] ) . ( [ system ] ) results in ( see appendix d ) : @xmath224 the interpretation of the above expression is straightforward . writing @xmath225 one obtains the equivalent equation : @xmath226 here , the right - hand side of eq . ( [ eq2a ] ) is simply the canonical pressure of an ideal gas of @xmath7 particles with the total volume @xmath10 replaced by the available volume @xmath227 . it is then natural to identify @xmath228 with the kinetic pressure in the canonical ensemble . of course , once averaged over @xmath7 and over @xmath229 , for a sufficiently large @xmath10 , this pressure must coincide with @xmath136 evaluated with the isobaric partition function . ( [ eq2a ] ) has the form of a self - consistency relation , as @xmath230 appears also in the right - hand side in the excluded volume term . ( [ eq2a ] ) is a quadratic form and has two solutions : a negative and positive one , and the negative corresponds to the situation where volume of hadrons exceeds the total volume , @xmath231 . the positive solution on the other hand ensures @xmath232 and therefore the condition for the @xmath213 function in eq . ( [ gcpartfun ] ) is always fulfilled . in what follows , we will adopt the positive solution of the eq . ( [ eq2a ] ) for any @xmath233 ( not only for the most probable set @xmath234 ) and we will integrate numerically @xmath6 in the variables @xmath235 . the partition function @xmath3 is then evaluated by summing over the particle number @xmath7 up to a cutoff @xmath236 @xmath237 where @xmath236 is sufficiently large to ensure the accuracy of our calculations . finally we test the consistency of our picture by comparing the ratios @xmath140 and @xmath118 with the results obtained with the isobaric partition function . in fig . ( [ check ] ) , the ratios @xmath140 ( left panel ) and @xmath118 ( right ) have been evaluated with the isobaric partition function ( solid line ) and with the grand canonical partition function ( dashed line ) for a volume @xmath238 gev@xmath239 , which is , as we checked , a good approximation of the infinite volume limit . as one can see , in both cases the two curves are very close , actually for the energy density they are practically coincident . the small ( expected ) difference between the isobaric and the grand - canonical result reflects the quality of our approximation . the same test has been also performed on the entropy density , the particles density and the filling fraction . for all these quantities we have observed an equally good agreement . in this article , we have studied the crossover transition of the gas of bags @xcite . we have found that the behavior of the system depends sensitively on the parameter @xmath25 of the hagedorn - like mass spectrum @xmath240 . the system exhibits a crossover transition for @xmath207 and an actual phase transition for larger values . in the range @xmath207 we made a coarse scan of @xmath25 , setting @xmath27 , @xmath124 , @xmath127 , @xmath241 , @xmath152 and @xmath128 . for @xmath242 the gas of bags undergoes a sharp ( yet , continuous ) transition qualitatively similar to lattice qcd . in this range , the asymptotic values of @xmath243 , @xmath118 , @xmath244 , coincide with the stefan boltzmann limit for @xmath131 and @xmath144 , whereas they settle to slightly larger values for @xmath154 and @xmath127 . for @xmath245 and @xmath124 , these quantities grow indefinitely with the temperature with small , decreasing , slopes going from @xmath27 to @xmath246 . we have also studied the ( strong ) dependence of the particles density @xmath247 ( where @xmath248 ) on @xmath25 . we have found that there exist a limiting value @xmath181 between @xmath182 and @xmath183 such that for @xmath249 the particles density vanishes at high temperature . the system is then populated by one ( or few ) infinite bag(s ) that occupies the entire volume . conversely , for @xmath250 , @xmath187 grows with the temperature . in the range @xmath251 , the ideal gas behavior is mimicked by a number of heavy extended bags that saturates the phase space forming a dense system . a pictorial representation summarizing the various high - temperature phases of the model is given in fig . ( [ schem ] ) . in this work we have explored a simple , intuitive model for a gas of hadrons in the vicinity of the transition at vanishing baryochemical potential . to this end , we have adopted the idea of the mit bag model and we have described hadrons as extended qgp bags . we have shown that , in the vicinity of @xmath29 ( which is directly related to the transition temperature of our model ) , elastic interactions among hadrons play a fundamental role . in our schematic approach , they are quantified by the thermal pressure @xmath61 . the effect of @xmath61 is to squeeze the hadrons , and for a certain set of model parameters , @xmath125 , the ideal gas behavior at high temperature can be reproduced . at the same time , the effective inner `` temperature '' , or rather degeneracy parameter , @xmath252 , of the bags increases with @xmath61 , resulting in a temperature - dependent mass spectrum . above @xmath29 , the physical picture of the qgp phase corresponds to a number @xmath253 of infinite bags that occupy the entire space . this is indeed the situation for @xmath249 . on the other hand , a large number of independent qgp bags ( such as for @xmath250 ) would contradict the lattice findings of vanishing flavor - flavor correlations @xcite . however , in order to define precisely the range of values of @xmath25 that lead to a consistent qgp scenario , it is fundamental to study the effect of a surface energy term . this contribution disfavors configurations with a large number of particles ( they are more `` expensive '' in terms of surface energy ) and might reduce the number of bags in the high - temperature phase , widening the range of possible values for @xmath25 . future work will likely concentrate on the effect of surface energy terms in addition to the study of van der waals type residual interactions . it would be also interesting to analyze in detail the behavior of the system at finite baryochemical potential . this could be done starting from the formalism developed in @xcite . the authors thank m. i. gorenstein for a critical reading of the manuscript and for useful suggestions . this work is supported by the director , office of energy research , office of high energy and nuclear physics , divisions of nuclear physics , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 . by using the equations ( [ simplass ] ) and ( [ pkst ] ) , the isobaric partition function in eq . ( [ ibpf1 ] ) can be conveniently written as : @xmath254 h_n(\{\eta_i \},t , s ) \\ \nonumber \ ; & & \times \int_{0}^{\infty}\d v \exp{\left[-sv\right ] } \left(v-\sum_{i=1}^n \frac{\eta_i}{4 ( b+st ) } \right)^n \theta{\left ( v-\sum_{i=1}^n\frac{\eta_i}{4(b+st ) } \right ) } \end{aligned}\ ] ] where @xmath255 \nonumber\\ \ ; & & \times \exp{\left\{\left [ \frac{1}{k(b+st)^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4(b+st)}\right ) \right]\sum_{i=1}^n \eta_i\right\ } } \;.\end{aligned}\ ] ] the integral over @xmath256 can be carried out @xmath257 } \left(v-\sum_{i=1}^n \frac{\eta_i}{4 ( b+st ) } \right)^n \theta{\left ( v-\sum_{i=1}^n\frac{\eta_i}{4(b+st ) } \right ) } \\ \nonumber & & = \frac{n!}{s^{n+1 } } \exp{\left\ { -\sum_{i=1}^n \eta_i \frac{s}{4(b+st ) } \right\ } } \ ; .\end{aligned}\ ] ] and eq . ( [ appa1 ] ) becomes : @xmath258 \frac{h_n(\{\eta_i \},t , s)}{s^n } \exp{\left\ { -\sum_{i=1}^n \eta_i \frac{s}{4(b+st ) } \right\ } } \ ; .\ ] ] the multiple integral in the last equation , can be factorized as : @xmath259 h_n(\{\eta_i \},t , s ) \exp{\left\ { -\sum_{i=1}^n \eta_i \frac{s}{4(b+st ) } \right\ } } \equiv f(t , s)^n\ ; .\ ] ] one then obtains eq . ( [ ibpf2 ] ) with @xmath99 given by eq . ( [ ibpf3a ] ) . by using eq . ( [ replacements ] ) , we write the average volume occupied by hadrons as @xmath260 where we used the stability condition @xmath261 . in turn , the average mass @xmath192 reads ( see also eq . ( [ amass ] ) ) : @xmath262 the filling fraction @xmath158 can then be expressed in terms of @xmath55 , @xmath61 and the particles density @xmath187 as : @xmath263 the partition function of @xmath7 particles can be reduced to a unidimensional integral when @xmath154 as the factor @xmath264 reduces to 1 . let us begin by making the substitution : @xmath265 that gives for @xmath13 in eq . ( [ gcpartfun ] ) : @xmath266^{n } \frac{1}{n ! } \left [ \prod_{i=1}^n \int_{0}^{\infty } \d y_i \ ; y_i \right]\left(v-\frac{n m_c}{4 p_{r , n}}- \sum_{i=1}^n \frac{y^2_i}{4 p_{r , n } } \right)^n \nonumber \\ \ ; & & \times \exp{\left\{\left [ \frac{1}{kp_{r , n}^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4p_{r , n}}\right ) \right]\left(\sum_{i=1}^n y^2_i+n m_c\right)\right\}}\ ; . \label{a1}\end{aligned}\ ] ] it is now convenient to rewrite our integral by using @xmath7-dimensional hyperspherical coordinates by setting : @xmath267 and @xmath268 with @xmath269 we now note that , apart from @xmath270 , the integrand in eq . ( [ a1 ] ) depends only on @xmath271 . this is true also for the pressure @xmath272 , where @xmath228 is given by the equations ( [ eq2a ] ) and , in the new variables , reads : @xmath273 where @xmath274 eq . ( [ a1 ] ) can now be written as : @xmath275^{n}\nonumber \\ \nonumber \ ; & & \times \frac{1}{n ! } \int_{0}^{\infty } \d r \ ; r^{n-1 } \int_{0}^{\frac{\pi}{2 } } \d \phi_1 \ldots \int_{0}^{\frac{\pi}{2 } } \d \phi_{n-1 } \sin^{n-2}{\phi_1}\sin^{n-3}{\phi_2 } \ldots \sin{\phi_{n-2}}\\ \nonumber \ ; & & \times r^n \left [ \prod_{i=1}^{n-1 } \cos{\phi_i } \right ] \sin^{n-1}{\phi_1}\sin^{n-2}{\phi_2 } \ldots \sin{\phi_{n-1 } } \left(v-\frac{n m_c}{4 p_{r , n}}- \frac{r^2}{4 p_{r , n } } \right)^n \nonumber \\ \ ; & & \times \exp{\left\{\left [ \frac{1}{kp_{r , n}^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4p_{r , n}}\right ) \right]\left(r^2 + n m_c\right)\right\}}\ ; .\end{aligned}\ ] ] next we separate eq . ( [ a6 ] ) into an angular integral , @xmath276 , and a radial integral , @xmath277 , such that : @xmath278^{n}\frac{1}{n ! } a_n i_n \label{a7}\ ] ] where : @xmath279 \sin^{n-1}{\phi_1}\sin^{n-2}{\phi_2 } \ldots \sin{\phi_{n-1 } } \ ; , \end{aligned}\ ] ] and @xmath280\left(r^2 + n m_c\right)\right\}}\ ; .\end{aligned}\ ] ] the angular integral @xmath276 can be solved by applying the following recursion relation : @xmath281 yielding @xmath282 accordingly , by using @xmath283 one obtains : @xmath284 the eq . ( [ a6 ] ) then reduces to : @xmath285^{n}\frac{1}{n!}\frac{1}{2^{n-1}(n-1 ) ! } \int_{0}^{\infty } \d r \ ; r^{2n-1 } \\ \nonumber \ ; & & \times\left(v-\frac{n m_c}{4 p_{r , n}}- \frac{r^2}{4 p_{r , n } } \right)^n \exp{\left\{\left [ \frac{1}{kp_{r , n}^{1/4}}-\frac{1}{t}\left(\frac{3}{4}+\frac{b}{4p_{r , n}}\right ) \right]\left(r^2 + n m_c\right)\right\}}\ ; .\end{aligned}\ ] ] performing the derivatives , the system in eq . ( [ system ] ) reads @xmath286 by subtracting the first equation from the second equation , one easily obtains eq . ( [ cccheck ] ) . r. hagedorn , in _ erice 1972 , proceedings of the study institute on high energy astrophysics _ ( cambridge university press , cambridge , 1974 ) , pp . 255 - 296 . m. creutz , l. jacobs and c. rebbi , phys . d * 20 * , 1915 ( 1979 ) . braun - munzinger , j. stachel , j. p. wessels and n. xu , phys . b * 365 * , 1 ( 1996 ) . j. cleymans , d. elliott , h. satz and r. l. thews , z. phys . c * 74 * , 319 ( 1997 ) . w. broniowski , a. baran and w. florkowski , acta phys . polon . b * 33 * , 4235 ( 2002 ) . f. becattini , j. manninen and m. gazdzicki , phys . c * 73 * , 044905 ( 2006 ) . a. andronic , p. braun - munzinger , k. redlich and j. stachel , j. phys . g * 35 * , 104155 ( 2008 ) . j. manninen and f. becattini , phys . c * 78 * , 054901 ( 2008 ) . f. becattini , z. phys . c * 69 * , 485 ( 1996 ) . f. becattini , in _ proceedings of xxxiii eloisatron workshop on universality features in multihadron production and the leading effect _ edited by a. zichichi ( 1996 ) , pp . 74 - 104 , hep - ph/9701275 f. becattini and u. w. heinz , z. phys . c * 76 * , 269 ( 1997 ) . f. becattini and g. passaleva , eur . j. c * 23 * , 551 ( 2002 ) . f. karsch , k. redlich and a. tawfik , eur . j. c * 29 * , 549 ( 2003 ) . m. cheng _ et al . _ , phys . rev . d * 77 * , 014511 ( 2008 ) . d. h. rischke , h. stoecker , w. greiner and b. l. friman , j. phys . g * 14 * , 191 ( 1988 ) . m. xu , m. yu and l. liu , phys . lett . * 100 * , 092301 ( 2008 ) . r. hagedorn and j. rafelski , _ from hadron gas to quark matter . 1 _ cern th 2947 ( 1980 ) . j. rafelski and r. hagedorn , _ from hadron gas to quark matter . 2 _ cern th 2969 ( 1980 ) . j. i. kapusta , phys . d * 23 * , 2444 ( 1981 ) . m. i. gorenstein , v. k. petrov and g. m. zinovev , phys . b * 106 * , 327 ( 1981 ) . s. kagiyama , a. minaka and a. nakamura , prog . phys . * 89 * , 1227 ( 1993 ) . m. i. gorenstein , w. greiner and s. n. yang , j. phys . g * 24 * , 725 ( 1998 ) . s. kagiyama , s. kumamoto , a. minaka , a. nakamura , k. ohkura and s. yamaguchi , eur . j. c * 25 * , 453 ( 2002 ) , and references therein . m. i. gorenstein , m. gazdzicki and w. greiner , phys . rev . c * 72 * , 024909 ( 2005 ) . i. zakout , c. greiner and j. schaffner - bielich , nucl . phys . a * 781 * , 150 ( 2007 ) . k. a. bugaev , phys . c * 76 * , 014903 ( 2007 ) . i. zakout and c. greiner , phys . c * 78 * , 034916 ( 2008 ) . y. aoki , g. endrodi , z. fodor , s. d. katz and k. k. szabo , nature * 443 * , 675 ( 2006 ) . a. chodos , r. l. jaffe , k. johnson , c. b. thorn and v. f. weisskopf , phys . rev . d * 9 * , 3471 ( 1974 ) . f. becattini and l. ferroni , eur . phys . j. c * 35 * , 243 ( 2004 ) . d. stauffer , a. aharony , _ introduction to percolation theory _ , taylor & francis , london 1994 . a. coniglio and w. klein , j. phys . a * 13 * , 2775 ( 1980 ) . s. fortunato and h. satz , phys . b * 475 * , 311 ( 2000 ) . s. fortunato and h. satz , nucl . a * 681 * , 466 ( 2001 ) . s. fortunato , f. karsch , p. petreczky and h. satz , nucl . . proc . suppl . * 94 * , 398 ( 2001 ) . s. fortunato , f. karsch , p. petreczky and h. satz , phys . b * 502 * , 321 ( 2001 ) . v. koch , a. majumder , and j. randrup _ phys . _ * 95 * 182301 ( 2005 ) . c. r. allton _ * d71 * 054508 ( 2005 ) . r. v. gavai and s. gupta _ phys . rev . _ * d73 * 014004 ( 2006 ) .

the process of opinion formation through synthesis and contrast of different viewpoints has been the subject of many studies in economics and social sciences . today , this process manifests itself also in online social networks and social media . the key characteristic of successful promotion campaigns is that they take into consideration such opinion - formation dynamics in order to create a overall favorable opinion about a specific information item , such as a person , a product , or an idea . in this paper , we adopt a well - established model for social - opinion dynamics and formalize the campaign - design problem as the problem of identifying a set of target individuals whose positive opinion about an information item will maximize the overall positive opinion for the item in the social network . we call this problem campaign . we study the complexity of the campaign problem , and design algorithms for solving it . our experiments on real data demonstrate the efficiency and practical utility of our algorithms .

1301.7455

individuals who participate in social networks form their opinions through synthesis and contrast of different viewpoints they encounter in their social circles . such processes manifest themselves more strongly in online social networks and social media where opinions and ideas propagate faster through virtual connections between social - network individuals . opinion dynamics have been of considerable interest to marketing and opinion - formation agencies , which are interested in raising public awareness on important issues ( e.g. , health , social justice ) , or increasing the popularity of person or an item ( e.g. , a presidential candidate , or a product ) . today , social - networking platforms and social media take up a significant amount of any promotion campaign budget . it is not uncommon to see activist groups , political parties , or corporations launching campaigns primarily via facebook or twitter . such campaigns interfere with the opinion - formation process by influencing the opinions of appropriately selected individuals , such that the overall favorable opinion about the specific information item is strengthened . the idea of leveraging social influence for marketing campaigns has been studied extensively in data mining . introduced by the work of domingos and richardson @xcite , and kempe et al . @xcite the problem of _ influence maximization _ asks to identify the most influential individuals , whose adoption of a product or an action will spread maximally in the social network . this line of work employs probabilistic propagation models , such as the _ independent - cascade _ or the _ linear - threshold _ model , which specify how actions spread among the individuals of the social network . at a high level , such propagation models distinguish the individuals to either _ active _ or _ inactive _ , and assume that active individuals influence their neighbors to become active according to certain probabilistic rules , so that activity spreads in the network in a _ cascading _ manner . in this paper , we are interested in the process of how individuals form opinions , rather than how they adopt products or actions . we find that models such as the independent cascade and the linear threshold are not appropriate for modeling the process of forming opinions in social networks . first , opinions can not be accurately modeled by binary states , such as being either active or inactive , but they can take continuous values in an interval . second , and perhaps more importantly , the formation of opinions does not resemble a discrete cascading process ; it better resembles a social game in which individuals constantly influence each other until convergence to an equilibrium . accordingly , we adopt the opinion - formation model of friedkin and johnsen @xcite , which assumes that each node @xmath0 of a social network @xmath1 has an internal opinion @xmath2 and an expressed opinion @xmath3 . while the internal opinions of individuals are fixed and not amenable to external influences , the expressed opinions can change due to the influence from their neighbors . more specifically , people change their expressed opinion @xmath3 so that they minimize their _ social cost _ , which is defined as the disagreement between their expressed opinion and the expressed opinion of their social neighbors , and the divergence between the expressed opinion and their true internal belief . it can be shown that the process of computing expressed opinion values by repeated averaging leads to a nash equilibrium of the game where the utilities are the node social costs @xcite . armed with this opinion - formation model , we introduce and study a problem similar to the influence - maximization problem , introduced by kempe et al . we consider a social network of individuals , each holding an opinion value about an information item . we view an opinion as a numeric value between zero ( negative ) and one ( positive ) . we then ask to identify @xmath4 individuals , so that once their expressed opinions becomes 1 , the opinions of the rest of the nodes at an equilibrium state are , on average , as positive as possible . we call this problem campaign . note that there is a fundamental qualitative difference between our framework and existing work on influence maximization : in contrast to other models used in influence - maximization literature , our work views the nodes as rational individuals who wish to optimize their own objectives . thus , we assume that the opinions of individuals get formed through best - response dynamics of a social game , which in turn is inspired by classical opinion - dynamics models studied extensively in economics @xcite . furthermore , from a technical point of view , the maximization problem that results from our framework requires a completely different toolbox of techniques than the standard influence - maximization problem . to this end , we exploit an interesting connection between our problem and absorbing random walks in order to establish our technical results , including the complexity and the approximability of the problem . interestingly , as with the influence - maximization problem , we show that the objective function of the campaign problem is submodular , and thus , it can be approximated within factor @xmath5 using a greedy algorithm . in addition , motivated by the properties of the greedy algorithm , we propose scalable heuristics that can handle large datasets . our experiments show that in practice the heuristics have performance comparable to the greedy algorithm , and that are scalable to very large graphs . finally , we discuss two natural variants of the campaign problem . our discussion reveals a surprising property of the opinion formation process on _ undirected _ graphs : the average opinion of the network depends only on the internal opinions of the individuals and not on the network structure . to the best of our knowledge , we are the first to formally define and study the campaign problem . however , our work is related to a lot of existing work in economics , sociology and computer science . in his original work in 1974 , degrout @xcite was the first to define a model where individuals organized in a social network @xmath1 have a starting opinion ( represented by a real value ) which they update by repeatedly adopting the average opinion of their friends . this model and its variants , has been subject of recent studies in social sciences and economics @xcite . most of these studies focus on identifying the conditions under which such repeated - averaging models converge to a consensus . in our work , we also adopt a repeated averaging model . however , our focus is not the characterization or the reachability of consensus . in fact , we do not assume consensus is ever reached . rather , we assume that the individuals participating in the network reach a stable point , where everyone has crystallized a personal opinion . as has been recently noted by social scientist davide krackhardt @xcite , studies of such non - consensus states are much more realistic since consensus states are rarely reached . key to our paper is the work by bindel et al . in fact , we adopt the same opinion - dynamics model as in @xcite , where individuals selfishly form opinions to minimize their personal cost . however , bindel et al . focus on quantifying the social cost of this lack of central coordination between individuals , i.e. , the _ price of anarchy _ , and they consider a network - design problem with the objective of reducing the social cost at equilibrium . our work on the other hand , focuses on designing a promotion campaign so that , at equilibrium , the overall positive inclination towards a particular opinion is maximized . recently , there has been a lot of work in the computer - science literature on identifying a set of individuals to advertise an item ( e.g. , a product or an idea ) so that the spread of the item in the network is maximized . different assumptions about how information items propagate in the network has led to a rich literature of targeted - advertisement methods ( e.g. , see @xcite ) . although at a high level our work has the same goal as all of these methods , there are also important differences , as we have already discussed in the introduction . we consider a social graph @xmath1 with @xmath6 nodes and @xmath7 edges . the nodes of the graph represent people and the edges represent social affinity between them . we refer to the members of the social graph by letters such as @xmath0 and @xmath8 , and we write @xmath9 to denote the edges of the graph . with each edge @xmath9 we associate a weight @xmath10 , which expresses the strength of the social affinity or influence from person @xmath0 to person @xmath8 . we write @xmath11 to denote the _ social neighborhood _ of person @xmath0 , that is , @xmath12 . unless explicitly mentioned , we do not make an assumption whether the graph @xmath13 is directed or undirected ; most of our results and our algorithms carry over for both types of graphs . the directed - graph model is more natural as in many real - world situations the influence @xmath14 from person @xmath0 to person @xmath8 is not equal to @xmath15 . following the framework of bindel et al . @xcite we assume that person @xmath0 has a _ persistent internal opinion _ @xmath2 , which remains unchanged from external influences . person @xmath0 has also an _ expressed opinion _ @xmath3 , which depends on their internal opinion @xmath2 , as well as on the expressed opinions of their social neighborhood @xmath11 . the underlying assumption is that individuals form opinions that combine their internal predisposition with the opinions of those in their social circle . we model the internal and external opinions @xmath2 and @xmath3 as real values in the interval @xmath16 $ ] . the convention is that 0 denotes a negative opinion , and 1 a positive opinion . the values in - between capture different shades of positive and negative opinions . given a set of expressed opinion values for all the people in the social graph , represented by an _ opinion vector _ @xmath17 , and the vector of internal opinions @xmath18 , we consider that the _ personal cost _ for individual @xmath0 is @xmath19 this cost models the fact that the expressed opinion @xmath3 of an individual @xmath0 is a `` compromise '' between their own internal belief @xmath2 and the opinions of their neighbors . as the individual @xmath0 forms an opinion @xmath3 , their internal opinion @xmath2 and the opinions @xmath20 of their neighbors @xmath21 may have different importance . the relative importance of those opinions is captured by the weights @xmath14 . now assume that , as a result of social influence and conflict resolution , every individual @xmath0 is selfishly minimizing their own cost @xmath22 . if the internal opinions are _ persistent _ and can not change ( an assumption that we carry throughout ) , minimizing the cost @xmath23 implies changing the expressed opinion @xmath3 to the weighted average among the internal opinion @xmath2 and the expressed opinions of the neighbors of @xmath0 . in other words , @xmath24 in fact , it can be shown that if every person @xmath0 iteratively updates their expressed opinion using equation ( [ equation : averaging ] ) , then the iterations converge to a unique _ nash equilibrium _ for the game with utilities of players expressed by equation . that is , the stationary vector of opinions @xmath25 is such that no node @xmath0 has an incentive to change their opinion to improve their cost @xmath23 . the selfish optimization process we described above guarantees that we reach an equilibrium , but it does not guarantee that we minimize globally the overall disagreements within the social network . given an opinion vector @xmath25 , we define the _ global cost _ @xmath26 to be the sum of the individual costs @xmath22 . a natural question addressed by bindel et al . @xcite is the relationship between the cost @xmath27 achieved by the nash - equilibrium opinion vector @xmath25 , and the globally optimum cost @xmath28 . bindel et al . show that for undirected graphs @xmath29 , indicating that the selfish - optimization cost is very close to the optimum cost . on the other hand , for directed graphs the nash cost @xmath27 can be as large as @xmath30 times the optimum cost @xmath28 . the goal of a promotion campaign is to improve the overall opinion about a product , person , or idea in a social network . given an opinion vector @xmath25 , we define the _ overall opinion _ @xmath31 as @xmath32 which is also proportional to the _ average expressed opinion _ of the individuals in @xmath13 . the goal of a campaign is to maximize @xmath31 . following the paradigm of kempe et al . @xcite , we assume that such a campaign relies on selecting a set of _ target nodes _ @xmath33 , which are going to be convinced to change their expressed opinions to @xmath34 . for the rest of the discussion , we will use @xmath35 to denote the overall opinion in the network , when vector @xmath25 is the nash - equilibrium vector obtained under the constraint that the expressed opinions of all nodes in @xmath33 are fixed to 1 . given this notation , we can define the campaign problem as follows . [ problem : expressed ] given a graph @xmath1 and an integer @xmath4 , identify a set @xmath33 of @xmath4 nodes such that fixing the expressed opinions of the nodes in @xmath33 to @xmath34 , maximizes the overall opinion @xmath35 . we emphasize that fixing @xmath36 for all @xmath37 means that equation is only applied for the @xmath20 s such that @xmath38 , while for the nodes @xmath37 the values @xmath3 remain @xmath34 . the definition of the campaign problem reflects our belief of what constitutes a feasible and effective campaign strategy . expressed opinions are more amenable to change , and have stronger effect on the overall opinion in the social network . thus , it is reasonable for a campaign to target these opinions . we note that other campaign strategies are also possible , resulting in different problem definitions . for example , one can define the problem where the campaign aims at changing the fundamental beliefs of people by altering their internal opinions @xmath2 . it is also conceivable to ask whether it is possible to improve the overall opinion @xmath31 by introducing a number of new edges in the social graph , e.g. , via a link - suggestion application . we discuss both of these variants at the end of the paper . it turns out that from the algorithmic point of view , both these problems are relatively simple . for instance , we can show that for undirected graphs , surprisingly , it is not possible to improve the overall opinion @xmath31 by introducing new edges in the graph . we now show the connection between computing the nash - equilibrium opinion vector @xmath25 and a random walk on a graph with absorbing nodes . this connection is essential in the analysis of the campaign problem . * absorbing random walks : * let @xmath39 be a graph with a set of @xmath40 nodes @xmath41 , and a set of edges @xmath42 . the graph is also associated with the following three @xmath43 matrices : ( @xmath0 ) the _ weight matrix _ @xmath44 with entries @xmath45 denoting the weight of the edges ; ( @xmath46 ) the _ degree matrix _ @xmath47 , which is a diagonal matrix such that @xmath48 ; ( @xmath49 ) the _ transition matrix _ @xmath50 , which is a row - stochastic matrix ; @xmath51 expresses the probability of moving from node @xmath0 to node @xmath8 in a random walk on the graph @xmath52 . in such a random walk on the graph @xmath52 , we say that a node @xmath53 is an _ absorbing node _ , if the random walk can only transition into that node , but not out of it ( and thus , the random walk is absorbed in node @xmath54 ) . let @xmath55 denote the set of all absorbing nodes of the random walk . the set of the remaining nodes @xmath56 are non - absorbing , or _ transient _ nodes . given this partition of the states in @xmath41 , the transition matrix of this random walk can be written as follows : @xmath57 in the above equation , @xmath58 is an @xmath59 identity matrix and @xmath60 a matrix with all its entries equal to @xmath61 ; @xmath62 is the @xmath63 sub - matrix of @xmath64 with the transition probabilities between transient states ; and @xmath65 is the @xmath66 sub - matrix of @xmath64 with the transition probabilities from transient to absorbing states . an important quantity of an absorbing random walk is the expected number of visits to a transient state @xmath8 when starting from a transient state @xmath0 before being absorbed . the probability of transitioning from @xmath0 to @xmath8 in exactly @xmath67 steps is the @xmath9-entry of the matrix @xmath68 . therefore , the probability that a random walk starting from state @xmath0 ends in @xmath8 without being absorbed is given by the @xmath9 entry of the @xmath63 matrix @xmath69 which is known as the _ fundamental matrix _ of the absorbing random walk . finally , the matrix @xmath70 is an @xmath66 matrix , with @xmath71 being the probability that a random walk which starts at transient state @xmath0 ends up being absorbed at state @xmath72 . assume that each absorbing node @xmath73 is associated with a fixed value @xmath74 . if a random walk starting from transient node @xmath75 gets absorbed in an absorbing node @xmath72 , then we assign to node @xmath0 the value @xmath74 . the probability of the random walk starting from node @xmath0 to be absorbed in @xmath8 is @xmath71 . therefore , the expected value of @xmath0 is @xmath76 . if @xmath77 is the vector with the expected values for all @xmath75 , and @xmath78 keeps the values @xmath79 for all @xmath72 , then we have that @xmath80 a fundamental observation , which highlights the connection between our work and random walks with absorbing states , is that the expected value @xmath81 of node @xmath75 can be computed by repeatedly averaging the values of the neighbors of @xmath0 in the graph @xmath52 . therefore , the computation of the nash - equilibrium opinion vector @xmath25 can be done using equation on an appropriately constructed graph @xmath52 . we discuss the construction of @xmath52 below . more details on absorbing walks can be found in the excellent monograph of doyle and snell @xcite . * the augmented graph . * we will now show how the theory of absorbing random walks described above can be leveraged for solving the campaign problem . this connection is achieved by performing a random walk with absorbing states on an _ augmented graph _ @xmath39 , whose construction we describe below . given a social network @xmath1 where every edge @xmath82 is associated with weight @xmath14 , we construct the augmented graph @xmath39 of @xmath13 as follows : * the set of vertices @xmath41 of @xmath52 is defined as @xmath83 , where @xmath84 is a set of @xmath6 new nodes such that for each node @xmath85 there is a copy @xmath86 ; * the set of edges @xmath42 of @xmath52 includes all the edges @xmath87 of @xmath13 , plus a new set of edges between each node @xmath88 and its copy @xmath86 . that is , @xmath89 , and @xmath90 ; * the weights of all the new edges @xmath91 are set to 1 , i.e. , @xmath92 . for @xmath93 , the weight of the edge @xmath94 is equal to the weight of the corresponding edge in @xmath13 , i.e. , @xmath95 . for the augmented graph @xmath52 , we can instantiate the matrices @xmath64 , @xmath96 , and @xmath97 defined in the previous paragraph . the augmented graph @xmath52 is a construction that we are going to use throughout the paper . our main observation is that we can compute the opinion vector @xmath25 that corresponds to the nash equilibrium defined by equation by performing an absorbing random walk on the graph @xmath52 . in this random walk , we set @xmath98 and @xmath99 , that is , we make all copy nodes in @xmath84 to be absorbing . we also set @xmath100 , that is , we assign value @xmath2 to each absorbing node @xmath101 . the nash - equilibrium opinion - vector @xmath25 can be computed using equation , that is , @xmath102 . the opinion @xmath103 is the expected internal opinion value at the node of absorption for a random walk that starts from node @xmath88 . given the vector @xmath25 we can compute the overall opinion @xmath31 . the campaign problem can be naturally defined in this setting . selecting a set of nodes @xmath33 is equivalent to adding the nodes in @xmath33 into the set of absorbing nodes @xmath104 , and assigning them value 1 . that is , we have @xmath105 and @xmath106 . for the vector @xmath78 , we have @xmath107 for all @xmath108 , and @xmath109 for all @xmath110 . we use equation to compute vector @xmath25 and using this @xmath25 , we can then compute the overall opinion @xmath35 . hence , the campaign problem becomes the problem of selecting a set of @xmath4 nodes @xmath111 to make absorbing with value 1 , such that @xmath35 is maximized . in this section , we establish the complexity of the campaign problem by showing that it is an @xmath112-hard problem . we also discuss properties of the objective function @xmath35 , which give rise to a constant - factor approximation algorithm for the campaign problem . [ theorem : np ] problem campaign is @xmath112-hard . we prove the theorem by reducing an instance of the vertex cover on regular graphs problem ( vcrg ) @xcite to an instance of the decision version of the campaign problem . we remind that a graph is called regular if all its nodes have the same degree . given a regular graph @xmath113 and an integer @xmath114 the vcrg problem asks whether there exists a set of nodes @xmath115 such that @xmath116 and @xmath117 is a vertex cover ( i.e. , for every @xmath118 it is @xmath119 or @xmath120 ) . an instance of the decision version of the campaign problem consists of a social graph @xmath1 , internal opinions @xmath121 , an integer @xmath4 and a number @xmath122 . the solution to the decision version is `` yes '' iff there exists a set @xmath123 such that @xmath124 and @xmath125 . given an instance of the vcrg problem , we will construct an instance of the decision version of campaign by setting @xmath1 to be equal to @xmath113 , @xmath126 for every @xmath88 , @xmath127 and @xmath128 . then , we show that @xmath123 is a solution to the campaign problem with value @xmath129 if and only if @xmath33 is a vertex cover of the input instance of vcrg . in order to show this , we use the absorbing random walk interpretation for the campaign problem . recall that by the definition of the campaign problem , every node @xmath37 becomes an absorbing node with value @xmath36 . for every other node @xmath38 we compute a value @xmath20 . since @xmath130 for all @xmath131 and @xmath132 for all @xmath133 , the value @xmath20 represents the probability that the random walk starting from @xmath8 will be absorbed in some node in @xmath33 . now suppose that @xmath33 is a vertex cover for @xmath134 . then , for every non - absorbing vertex @xmath135 , for each one of the @xmath136 edges @xmath137 incident on @xmath8 , it must be that @xmath133 ; otherwise edge @xmath138 is not covered , and @xmath33 is not a vertex cover . therefore , in the augmented graph @xmath52 , node @xmath8 is connected to @xmath136 nodes in @xmath33 , and to node @xmath139 , all of them absorbing . a random walk starting from @xmath8 will be absorbed and converge in a single step . the probability of it being absorbed in a node in @xmath33 is @xmath140 . there are @xmath141 nodes in @xmath142 , therefore , @xmath143 . if @xmath33 is not a vertex cover for @xmath134 , then there is an edge @xmath144 , such that @xmath145 . as we noted before , @xmath20 is the probability of being absorbed in some node in @xmath33 . the transition probability of the edge @xmath146 is @xmath147 , therefore , node @xmath8 has probability at least @xmath147 of being absorbed in @xmath139 . since there is a path from @xmath8 to @xmath148 with non - zero probability , the probability of @xmath8 being absorbed in node @xmath148 is strictly greater than zero . therefore , the probability of being absorbed in some node not in @xmath33 is strictly greater than @xmath147 , and thus @xmath149 . it follows that @xmath150 . the proof of the theorem appears in the appendix [ appendix : np - hard ] . the proof relies on a reduction from the vertex cover on regular graphs problem ( vcrg ) @xcite . since the campaign problem is @xmath112-hard , we are content with algorithms that approximate the optimal solution in polynomial time . fortunately , we can show that the function @xmath151 is monotone and submodular , and thus a simple greedy heuristic yields a constant - factor approximation to the optimal solution . [ theorem : submodular ] the function @xmath151 is monotone and submodular . we only give here a proof sketch . a detailed proof is given in appendix [ appendix : submodularity ] . recall that @xmath152 . in the absorbing random walk interpretation of the opinion formation process , we have shown that the expressed opinion of node @xmath0 is the expected opinion value at the point of absorbtion for a random walk that starts from node @xmath0 . that is , @xmath153 , where @xmath104 is the set of absorbing nodes , @xmath154 is the probability of the random walk starting from node @xmath0 to be absorbed at node @xmath54 , and @xmath155 the opinion value at node @xmath54 . when we add a node @xmath156 to @xmath33 , and hence to the set @xmath104 , some of the probability mass of the random walk will be absorbed at @xmath156 . since @xmath156 has the maximum possible opinion value , @xmath157 , if follows that @xmath3 can only increase , and thus @xmath151 is monotone . furthermore , the less competition there is for @xmath156 ( i.e. , the smaller the size of @xmath104 ) , the more mass of the random walk will be absorbed in @xmath156 , and the larger the increase of @xmath151 . hence @xmath151 is submodular . recall that @xmath152 , where the values of vector @xmath25 are computed using equation . as we have already described , we can view the computation of the vector @xmath25 as performing a random walk with absorbing nodes on the augmented graph @xmath158 . assume that @xmath104 is the set of absorbing nodes , and that each @xmath159 is associated with value @xmath155 . let @xmath160 be the probability that a random walk that starts from @xmath0 gets absorbed at node @xmath54 , when the set of absorbing nodes is @xmath104 . the expressed opinion of node @xmath161 is @xmath162 since this value depends on the set @xmath104 , we will write @xmath163 to denote the value of @xmath3 when the set of absorbing nodes is @xmath104 . initially , the set of absorbing nodes is @xmath164 and @xmath165 for all @xmath166 . when we select a subset of nodes @xmath111 such that their expressed opinions are fixed to 1 , we have that @xmath167 , @xmath165 for all @xmath168 and @xmath169 for all @xmath170 . since the set of nodes @xmath84 is always part of the set @xmath104 , and the parameter that we are interested in for this proof is the target set of nodes @xmath33 , we will use @xmath171 to denote @xmath163 where @xmath167 . we thus have @xmath172 note that the summation is over all nodes in @xmath173 including the nodes in @xmath33 . if @xmath133 , then @xmath174 , and @xmath175 for all @xmath176 . therefore , @xmath177 for all @xmath133 . we now make the following key observation . let @xmath8 be a node in @xmath142 . we have that @xmath178 the equation above follows from the observation that we can express the probability of a random walk starting from @xmath0 to be absorbed in some node @xmath166 as the sum of two terms : ( i ) the probability @xmath179 that the random walk is absorbed in @xmath54 , while avoiding passing through @xmath8 ( thus we add @xmath8 in the absorbing set ) ; ( ii ) the probability @xmath180 that the random walk is absorbed in @xmath8 while avoiding the nodes in @xmath104 ( that is , the probability of all paths that go from @xmath0 to @xmath8 of arbitrary length , without passing through @xmath8 or @xmath104 ) , times the probability @xmath181 of starting a new random walk from @xmath8 and getting absorbed in @xmath54 ( being able to revisit @xmath8 and any node in @xmath142 ) . if we add node @xmath8 into the set @xmath33 we have that @xmath182 hence , @xmath183 therefore , we can conclude that function @xmath151 is monotone with respect to the set of target nodes @xmath33 . we now need to show that @xmath151 is submodular , that is for any @xmath184 such that @xmath185 , and for any node @xmath186 , we have that @xmath187 . for this we will use two random walks : one with absorbing states @xmath188 and the other with absorbing states @xmath189 . following reasoning and notation similar to the one we used for monotonicity we have that @xmath190 from the monotonicity property we have that @xmath191 also , as the number of absorbing nodes increases , the probability of being absorbed in a specific node @xmath8 decreases , since the probability of the random walk to be absorbed in a node other than @xmath8 increases . therefore , we also have that @xmath192 combining these last two observations we conclude that @xmath187 which shows that the function is submodular . interestingly , the i - campaign problem , where the advertising campaign aims to change the internal rather than the expressed opinions of individuals , is solvable in polynomial time . we prove this fact below . [ theorem : poly ] there is a polynomial time algorithm for problem i - campaign . let @xmath1 be the input to the i - campaign problem . we form the augmented graph @xmath158 and compute @xmath25 using equation for @xmath98 and @xmath99 . in this case , the vector @xmath194 has values corresponding to the input @xmath2 s , and the vector @xmath195 has values corresponding to @xmath3 s . thus , we have @xmath196 where @xmath197 is a fixed matrix . if @xmath198 is the @xmath8-th column sum of the matrix @xmath199 , then our objective function becomes @xmath200 since we want to maximize the previous quantity by selecting a set @xmath201 of size @xmath4 and setting @xmath202 for every @xmath203 , it is sufficient to select the @xmath4 nodes in @xmath173 that have the largest column sums @xmath204 . note that a property similar to the one used in the above proof has also been used by even - dar and shapira @xcite.for the rest of the discussion , we will refer to the above algorithm as the ` i - campaign ` algorithm . a central component in all the algorithms presented in this section is the estimation of the the opinion function @xmath35 . in section [ sec : background ] , we have already discussed that this can be done by evaluating equation . for appropriately defined sets @xmath206 and @xmath104 , this requires computing the matrix @xmath207 . hence , this calculation involves a matrix inversion , which is very inefficient . the reason is that despite the fact that the social graph is typically sparse , matrix inversion does not preserve sparseness . thus , it may be too expensive to even store the matrix @xmath199 . instead , we resort to the power - iteration method implied by equation : at each iteration we update the opinion @xmath3 of a node @xmath88 by averaging the opinions of its neighbors @xmath21 and its own internal opinion @xmath2 . during the iterations we do not update the values of opinions that are fixed . this power - iteration method is known to converge to the equilibrium vector @xmath25 , and it is highly scalable , since it only involves multiplication of a sparse matrix with a vector . for a graph with @xmath6 nodes , @xmath7 edges , and thus , average degree @xmath208 , the algorithm requires @xmath209 operations per iterations . therefore , the overall running time is @xmath210 , where @xmath58 is the total number of iterations . in our experiments we found the the method converges in around 50 - 100 iterations , depending on the dataset . our algorithms for the campaign problem , include a constant - factor approximation algorithm as well as several efficient and effective heuristics . * the ` greedy ` algorithm . * it is known that the greedy algorithm is a @xmath211-approximation algorithm for maximizing a submodular function @xmath212 subject to cardinality constraints , i.e. , finding a set @xmath213 that maximizes @xmath214 such that @xmath215 @xcite . consequently , the ` greedy ` algorithm constructs the set @xmath33 by adding one node in each iteration . in the @xmath216-th iteration the algorithm extends the set @xmath217 by adding the node @xmath0 that maximizes @xmath218 when setting @xmath36 . the computational bottleneck of ` greedy ` is due to the fact that we need to compute the nash - equilibrium opinion vector @xmath205that results from setting @xmath219 for all @xmath220 , and we need to do such a computation for all candidate nodes @xmath221 . overall , for a solution @xmath33 of size @xmath222 ` greedy ` needs to perform @xmath223 computations of finding the optimal vector @xmath205 . as we saw , each of these computations is performed by a power - iteration in time @xmath210 , yielding an overall running time @xmath224 . such a running time is super - quadratic and therefore the algorithm is not scalable to very large datasets . one way to speedup the algorithm is by storing , for each node that it is not yet selected , its marginal improvement on the score , at the last time it was computed . this speedup , which is commonly used in optimization problems with submodular functions @xcite , is not adequate to make the greedy algorithm applicable for large data , at least for the version of the algorithm described here . the reason is that in the very first iteration there is no pruning and therefore we need to make @xmath225 power - iteration computations , yielding again a quadratic algorithm . to overcome these scalability limitations , we present a number of scalable heuristics . to characterize the nodes selected by ` greedy ` we execute the algorithm on small datasets , and we compute a number of measures for each node selected by ` greedy ` . in particular , for each node we compute measures such as its degree , the average degree of its neighbors , the maximum degree of its neighbors , the value of its internal opinion @xmath2 , the average value of @xmath2 over its neighbors , and so on . three of the features with the most clear signal are shown in figure [ figure : newman - order ] for the * karate club * dataset ( described in detail in section [ section : experimental - evaluation ] ) . we obtain similar behavior on all the datasets we tried . in figure [ figure : newman - order ] we plot measures of nodes in the order selected by the ` greedy ` . a good measure would be one that is monotonic with respect to this order . in the first panel , we show the degree of a node in the selection order of ` greedy ` , and we see that ` greedy ` tends to select first high degree nodes . as shown in the second panel , this dependence is even more clear for the _ free degree _ , i.e. , the number of neighbors that are not already selected by ` greedy ` . finally , in the third panel of figure [ figure : newman - order ] we see the internal opinion @xmath2 of nodes in the order selected by the ` greedy ` . we see that the ` greedy ` tends to select first nodes with low internal opinion . there are a few exceptions of nodes with high internal opinion @xmath2 selected at the initial steps of greedy . such nodes are nodes with high degree , connected to many nodes with small values of @xmath2 . armed with intuition from this analysis we now proceed to describe our heuristics . * the ` degree ` algorithm . * this algorithm simply sorts the nodes of @xmath1 in decreasing order of their in degree and forms the set of target nodes @xmath33 by picking the top-@xmath4 nodes of the ranking . the running time of ` degree ` is @xmath226 , i.e. , the time required for sorting . * the ` freedegree ` algorithm . * this algorithm is a `` greedy '' variant of ` degree ` ; ` freedegree ` forms the set @xmath33 iteratively by choosing at every iteration the node with the highest free degree . the free degree of a node is the sum of the weights of the edges that are incident to it and are are connected to nodes not already in @xmath33 . when the set @xmath33 consists of @xmath4 nodes , the running time of ` freedegree ` is @xmath227 . * the ` rwr ` algorithm . * as we saw in figure [ figure : newman - order ] , a good choice for nodes to be added in the solution are not only the nodes of high degree but also the nodes of small value of internal opinion @xmath2 . the ` rwr ` algorithm combines both of these features : selecting nodes with high degree and with small @xmath2 . this is done by performing a random walk with restart ( ` rwr ` ) , where the probability of restarting at a node @xmath0 is proportional to @xmath228 , where @xmath229 , and ordering the nodes according to the resulting stationary distribution . the intuition is that a random walk favors high - degree nodes , and using the specific restart probabilities favors nodes with low value of @xmath2 . for the restart probability , we use the parameter @xmath230 , which has been established as a standard parameter of the pagerank algorithm @xcite . making one ` rwr ` computation can be achieved by the power - iteration method , which similarly to computing the optimal vector @xmath205 , has running time @xmath210 . therefore , the overall running time of the algorithm for selecting a set @xmath33 of size @xmath4 is @xmath231 . * the ` min - s ` and ` min - z ` algorithms . * the ` min - s ` algorithm simply selects the @xmath4 nodes with the smallest value @xmath2 . this heuristic is motivated by the observation that the ` greedy ` algorithm tends to select nodes with small value @xmath2 . for completeness , we also experiment with the ` min - z ` algorithm , which greedily selects and add in the solution set @xmath33 the node that at the current iteration has the smallest value of expressed opinion @xmath3 . the objective of our experiments is to compare the proposed heuristics against ` greedy ` , the algorithm with the approximation guarantee , and demonstrate their scalability . we experiment with a number of publicly available small networks . we evaluate our algorithms by reporting the value of the objective function @xmath31 as a function of the solution set size @xmath222 . our results for three small networks are shown in figure [ figure : newman ] . the datasets shown in the figure are the following : ( @xmath0 ) * karate club * : a social network of friendships between 34 members of a karate club at a us university in the 1970s @xcite ; ( @xmath46 ) * les miserables * : co - appearance network of characters in the novel _ les miserables _ @xcite ; and ( @xmath49 ) * dolphins * : an undirected social network of frequent associations between 62 dolphins @xcite . for this set of experiments we set the internal opinions @xmath2 to be a uniformly - sampled value in @xmath16 $ ] . we see that the difference between all the algorithms is relatively small but their relative performance is consistent . the ` greedy ` algorithm achieves the best results , while the three heuristics , ` degree ` , ` freedegree ` , and ` rwr ` come close together . ` min - s ` and ` min - z ` have the poorest performance , even though for larger values of @xmath4 they improve and slightly outperform some of the heuristics . between the two , ` min - s ` performs best , outperforming ` min - z ` , especially for small values of @xmath4 . both of those trends are expected : the three heuristics ` degree ` , ` freedegree ` , and ` rwr ` , are better motivated than ` min - s ` , which in turn , is better motivated than ` min - z ` . we obtain similar results for other small networks , although we do not provide the plots for lack of space . we also evaluate our algorithms on two large social networks , derived from bibliographic data . the first dataset , * bibsonomy * , is extracted from bibsonomy @xcite , a social - bookmarking and publication - sharing system . from the available data , we extract a social graph of @xmath232 nodes representing authors and @xmath233 edges representing co - authorship relations . for each author we also keep the set of tags that have been used for the papers of that author . the second dataset , * dblp * , is also a co - authorship graph among computer scientists extracted from the dblp site . the dataset is a large graph containing @xmath234 nodes and @xmath235 edges . again , for each author we keep the set of terms they have been used in the titles of the papers they have co - authored . in this experiment , we generate the internal opinion vectors by identifying keywords related to data mining ( e.g. , we picked _ data _ , _ mining _ , _ social _ , _ networks _ , _ graph _ , _ clustering _ , _ learning _ , and _ community _ ) . for each author @xmath88 we then set @xmath2 to be the fraction of the above keywords present in his set of terms . this setting corresponds to a hypothetical scenario of designing a campaign to promote the `` data mining '' topic among all computer - science researchers . performance of the algorithms on the * bibsonomy * network for the topic `` semantic web''.,scaledwidth=47.0% ] the results of the heuristics for the two datasets are shown in figure [ figure : bibsonomydblp ] . for the * bibsonomy * dataset ( left ) , there is a clear distinction between the three heuristics ; ` rwr ` clearly performs better than both ` degree ` and ` freedegree ` . this superior performance of ` rwr ` is expected as this algorithm takes into account both the degrees and the values of the internal opinions @xmath2 . also the better performance of ` freedegree`compared with the performance of ` degree ` is consistent with the results obtained for smaller networks . on the other hand , on the * dblp * dataset ( right part of figure [ figure : bibsonomydblp ] ) the behavior of the three heuristics is more surprisingly , as all three perform almost identical . finally , for both datasets , the difference of the three best heuristics ` degree ` , ` freedegree ` , and ` rwr ` with the other two heuristics , ` min - s ` and ` min - z ` is more pronounced . in fact , the performance of ` min - s ` and ` min - z ` is very poor . we investigate the difference on the relative performance of the best three heuristics , ` degree ` , ` freedegree ` , and ` rwr ` , on the two datasets by plotting the degrees of the nodes versus their @xmath2 value . this is shown in figure [ figure : degree - vs - si ] . recall that our intuition for selecting nodes that is to choose nodes that have large degree and small value of @xmath2 . figure [ figure : degree - vs - si ] demonstrates that in the * dblp * dataset , large degree correlates well with small @xmath2 values , while this is not the case for the * bibsonomy * dataset . therefore , for * dblp * all three heuristics pick high - degree nodes , which makes their performance almost identical . scatter plot of degrees vs. internal opinion values @xmath2 in the two datasets , * bibsonomy * and * dblp*.,title="fig:",scaledwidth=23.0% ] scatter plot of degrees vs. internal opinion values @xmath2 in the two datasets , * bibsonomy * and * dblp*.,title="fig:",scaledwidth=23.0% ] the campaign problem we studied in this paper focuses on campaigns that aim to alter the expressed opinions of individuals . however , other campaign strategies are also possible . for example one could aim at altering the internal opinions of individuals such that the overall opinion is improved as much as possible . formally , the goal would be to select a set of nodes @xmath236 , which are going to be convinced to change their _ internal opinions _ @xmath2 to @xmath34 , such that the resulting overall opinion @xmath237 is maximized . we call this problem the i - campaign problem . the difference between the campaign and i - campaign problems is that in the former we are asking to fix the expressed opinions @xmath3 for @xmath4 individuals , while in the latter we are asking to fix the internal opinions @xmath2 . even though the difference is seemingly small , the problems are computationally very different . algorithmically , the problem of selecting @xmath236 individuals to change their internal opinions , so that we maximize @xmath237 is much simpler . in fact , we can show that for an undirected social graph @xmath1 , where each node @xmath85 has internal opinion @xmath2 and expressed opinion @xmath3 , the following invariant holds , independently of the structure of the graph ( the set of edges @xmath87 ) : @xmath238 consequently , the goal of maximizing @xmath31 by modifying @xmath4 values @xmath2 can be simply achieved by selecting the @xmath4 smallest values @xmath2 and setting them to 1 . we note that the above observation does not hold once the expressed opinions of some individuals are fixed , as is the case in the campaign problem . the proof of the invariant , and its implications are discussed in the appendix [ appendix : invariant ] . the graph invariant has obvious implications for the other variant of the campaign problem , where we seek to maximize @xmath31 by adding or removing edges to the graph . from equation ( [ equation : invariant ] ) it follows that for undirected social graphs this problem variant is meaningless ; the overall opinion @xmath31 does not depend on the structure of the graph . this observation has important implications for opinion formation on social networks . it shows that although the network structure has an effect on the individual opinions of network participants , it does not affect the average opinion in the network . for the campaign problem , this says that you can not create more goodwill by altering the network . for the study of social dynamics , this implies that the collective wisdom of the crowd remains unaffected by the social connections between individuals . we considered a setting where opinions of individuals in a social network evolve through processes of social dynamics , reaching a nash equilibrium . adopting a standard social and economic model of such dynamics we addressed the following natural question : given a social network of individuals who have their own internal opinions about an information item , which are the individuals that need to be convinced to adopt a positive opinion so that in the equilibrium state , the network ( as a whole ) has the maximum positive opinion about the item ? we studied the computational complexity of this problem and proposed algorithms for solving them exactly or approximately . our theoretical analysis and the algorithm design relied on a connection between opinion dynamics and random walks with absorbing states . our experimental evaluation on real datasets demonstrated the efficacy of our algorithms and the effect of the structural characteristics of the underlying social networks on their performance . 10 benchmark folksonomy data from bibsonomy . technical report , knowledge and data engineering group , university of kassel , 2007 . d. acemoglu and a. ozdaglar . opinion dynamics and learning in social networks . , 2011 . d. bindel , j. m. kleinberg , and s. oren . how bad is forming your own opinion ? in _ focs _ , 2011 . s. brin and l. page . the anatomy of a large - scale hypertextual web search engine . in _ www _ , 1998 . n. chen . on the approximability of influence in social networks . in _ soda _ , 2008 . m. h. degrout . reaching consensus . , 1974 . p. m. demarzo , d. vayanos , and j. zweibel . persuasion bias , social influence , and unidimensional opinions . , 2003 . p. doyle and j. snell . . mathematical association of america , 1984 . e. even - dar and a. shapira . a note on maximizing the spread of influence in social networks . in _ wine _ , 2007 . u. feige . vertex cover is hardest to approximate on regular graphs . , 2003 . n. e. friedkin and e. johnsen . social influence and opinions . , 1990 . b. golub and m. o. jackson . naive learning in social networks : convergence , influence and the wisdom of the crowds . , 2010 . m. o. jackson . . princeton university press , 2008 . d. kempe , j. m. kleinberg , and . tardos . maximizing the spread of influence through a social network . in _ kdd _ , 2003 . d. e. knuth . . addison - wesley , 1993 . d. krackhardt . a plunge into networks . , 2009 . j. leskovec , a. krause , c. guestrin , c. faloutsos , j. m. vanbriesen , and n. s. glance . cost - effective outbreak detection in networks . in _ kdd _ , 2007 . d. lusseau , k. schneider , o. j. boisseau , p. haase , e. slooten , and s. m. dawson . , 54:396405 , 2003 . g. nemhauser , l. wolsey , and m. fisher . an analysis of the approximations for maximizing submodular set functions . , 1978 . m. richardson and p. domingos . mining knowledge - sharing sites for viral marketing . in _ kdd _ , 2002 . w. w. zachary . an information flow model for conflict and fission in small groups . , 33:452473 , 2003 . we prove the theorem by reducing an instance of the vertex cover on regular graphs problem ( vcrg ) @xcite to an instance of the decision version of the campaign problem . we remind that a graph is called regular if all its nodes have the same degree . given a regular graph @xmath113 and an integer @xmath114 the vcrg problem asks whether there exists a set of nodes @xmath115 such that @xmath116 and @xmath117 is a vertex cover ( i.e. , for every @xmath118 it is @xmath119 or @xmath120 ) . an instance of the decision version of the campaign problem consists of a social graph @xmath1 , internal opinions @xmath121 , an integer @xmath4 and a number @xmath122 . the solution to the decision version is `` yes '' iff there exists a set @xmath123 such that @xmath124 and @xmath125 . given an instance of the vcrg problem , we will construct an instance of the decision version of campaign by setting @xmath1 to be equal to @xmath113 , @xmath126 for every @xmath88 , @xmath127 and @xmath128 . then , we show that @xmath123 is a solution to the campaign problem with value @xmath129 if and only if @xmath33 is a vertex cover of the input instance of vcrg . in order to show this , we use the absorbing random walk interpretation of the campaign problem . recall that by the definition of the campaign problem , every node @xmath37 becomes an absorbing node with value @xmath36 . for every other node @xmath38 we compute a value @xmath20 , which is the expected value at the absorption point of a random walk that starts from @xmath8 . since @xmath130 for all @xmath85 and @xmath132 for all @xmath133 , the value @xmath20 represents the probability that the random walk starting from @xmath8 will be absorbed in some node in @xmath33 . now suppose that @xmath33 is a vertex cover for @xmath134 . then , for every non - absorbing vertex @xmath135 , for each one of the @xmath136 edges @xmath137 incident on @xmath8 , it must be that @xmath133 ; otherwise edge @xmath138 is not covered , and @xmath33 is not a vertex cover . therefore , in the augmented graph @xmath52 , node @xmath8 is connected to @xmath136 nodes in @xmath33 , and to node @xmath139 , all of them absorbing . a random walk starting from @xmath8 will be absorbed and converge in a single step . the probability of it being absorbed in a node in @xmath33 is @xmath140 . there are @xmath141 nodes in @xmath142 , therefore , @xmath143 . if @xmath33 is not a vertex cover for @xmath134 , then there is an edge @xmath144 , such that @xmath145 . as we noted before , @xmath20 is the probability of being absorbed in some node in @xmath33 . the transition probability of the edge @xmath146 is @xmath147 , therefore , node @xmath8 has probability at least @xmath147 of being absorbed in @xmath139 . since there is a path from @xmath8 to @xmath148 with non - zero probability , the probability of @xmath8 being absorbed in node @xmath148 is strictly greater than zero . therefore , the probability of being absorbed in some node not in @xmath33 is strictly greater than @xmath147 , and thus @xmath149 . it follows that @xmath150 . recall that @xmath152 , where the values of vector @xmath25 are computed using equation . as we have already described , we can view the computation of the vector @xmath25 as performing a random walk with absorbing nodes on the augmented graph @xmath158 . assume that @xmath104 is the set of absorbing nodes , and that each @xmath159 is associated with value @xmath155 . let @xmath160 be the probability that a random walk that starts from @xmath0 gets absorbed at node @xmath54 , when the set of absorbing nodes is @xmath104 . the expressed opinion of node @xmath161 is @xmath162 since this value depends on the set @xmath104 , we will write @xmath163 to denote the value of @xmath3 when the set of absorbing nodes is @xmath104 . initially , the set of absorbing nodes is @xmath164 and @xmath165 for all @xmath166 . when we select a subset of nodes @xmath111 such that their expressed opinions are fixed to 1 , we have that @xmath167 , @xmath165 for all @xmath168 and @xmath169 for all @xmath170 . since the set of nodes @xmath84 is always part of the set @xmath104 , and the parameter that we are interested in for this proof is the target set of nodes @xmath33 , we will use @xmath171 to denote @xmath163 where @xmath167 . we thus have @xmath172 note that the summation is over all nodes in @xmath173 including the nodes in @xmath33 . if @xmath133 , then @xmath174 , and @xmath175 for all @xmath176 . therefore , @xmath177 for all @xmath133 . we now make the following key observation . let @xmath8 be a non - absorbing node in @xmath142 . we have that @xmath178 the equation above follows from the observation that we can express the probability of a random walk starting from @xmath0 to be absorbed in some node @xmath166 as the sum of two terms : ( i ) the probability @xmath179 that the random walk is absorbed in @xmath54 , while avoiding passing through @xmath8 ( thus we add @xmath8 in the absorbing set ) ; ( ii ) the probability @xmath180 that the random walk is absorbed in @xmath8 while avoiding the nodes in @xmath104 ( that is , the probability of all paths that go from @xmath0 to @xmath8 of arbitrary length , without passing through @xmath8 or @xmath104 ) , times the probability @xmath181 of starting a new random walk from @xmath8 and getting absorbed in @xmath54 ( being able to revisit @xmath8 and any node in @xmath142 ) . if we add node @xmath8 into the set @xmath33 we have that @xmath182 hence , @xmath183 therefore , we can conclude that function @xmath151 is monotone with respect to the set of target nodes @xmath33 . we now need to show that @xmath151 is submodular , that is for any @xmath184 such that @xmath185 , and for any node @xmath186 , we have that @xmath187 . for this we will use two random walks : one with absorbing states @xmath188 and the other with absorbing states @xmath189 . following reasoning and notation similar to the one we used for monotonicity we have that @xmath190 from the monotonicity property we have that @xmath191 also , as the number of absorbing nodes increases , the probability of being absorbed in a specific node @xmath8 decreases , since the probability of the random walk to be absorbed in a node other than @xmath8 increases . therefore , we also have that @xmath192 combining these last two observations we conclude that @xmath187 which shows that the function is submodular . in this section we prove a graph invariant related to the sum of the values @xmath3 and @xmath2 . this invariant has repercussions in the following scenarios : * maximize @xmath31 by modifying only the internal opinions @xmath2 of the users ( problem i - campaign in section [ section : variants ] ) ; and * maximize @xmath31 by adding edges in the social graph , for instance , recommend friendships or certain accounts for users to connect and follow . we prove the invariant in a slightly more general setting than the one we consider in the paper . we then formulate the more special case of the invariant for our exact problem setting , and we discuss its implications in the above - mentioned scenarios ( @xmath0 ) and ( @xmath46 ) . consider an undirected graph @xmath239 . we use @xmath14 to denote the weight of edge @xmath9 and @xmath240 to denote the total weight of all edges incident on node @xmath0 . we assume that the vertices in @xmath173 are partitioned in two sets @xmath206 and @xmath104 . the nodes in @xmath104 are _ absorbing _ nodes for the random walk . each node @xmath73 is associated with a value @xmath241 . the value @xmath241 can be either the internal opinion of node @xmath8 ( in which case @xmath242 ) , or the fixed expressed opinion of node @xmath8 ( in which case @xmath243 ) . for each node @xmath244 we will compute a value @xmath3 which is the expected value at the node of absorption for a random walk that starts from node @xmath0 , as given by equation ( [ equation : equilibrium ] ) . we further make the assumption that the set of absorbing nodes can be partitioned into @xmath245 disjoint subsets @xmath246 , one for each node @xmath244 , such that the nodes in @xmath247 are connected _ only _ with the node @xmath0 . we can make this assumption without loss of generality , since in the case that a node @xmath73 is connected to @xmath4 nodes @xmath248 in @xmath206 , we can create @xmath4 copies of @xmath8 , each with value @xmath241 , and connect each copy with a single node in @xmath206 with an edge of the same weight , while removing the original node @xmath8 from the graph . in the resulting graph the @xmath3 values computed by the absorbing random walk are the same as in the original graph . to introduce some additional notation let @xmath249 denote the set of edges between non - absorbing nodes , and let @xmath250 denote the set of edges between nodes in @xmath206 and in @xmath104 . note that by the construction above there are no edges between the nodes in @xmath104 . such edges would not have any effect anyway , since the nodes in @xmath104 are absorbing . given a node @xmath244 , let @xmath11 denote the set of neighbors of @xmath0 , let @xmath251 denote the set of non - absorbing neighbors of @xmath0 , and let @xmath247 denote the set of absorbing neighbors of @xmath0 . from the definition of the absorbing random walk we have that @xmath252 and thus @xmath253 summing over all @xmath244 we get @xmath254 the left - hand side can also be written as : @xmath255 by equations ( [ equation : invariant_1 ] ) and ( [ equation : invariant_2 ] ) we obtain @xmath256 equation ( [ eq : general : z - s ] ) is the most general form of our invariant . the equation relates the values of @xmath3 and @xmath241 via the weights @xmath14 across the edges @xmath250 , i.e. , only the edges between absorbing and non - absorbing nodes . the set of edges @xmath249 between the non - absorbing nodes does not play any role . we now consider the special case in which equation ( [ eq : general : z - s ] ) is applied to graphs considered in this paper , that is , in augmented graphs of type @xmath257 , as defined in section [ section : notation ] . in that case , each node @xmath85 is connected to a single absorbing node @xmath258 , which has value @xmath107 . furthermore , for each edge @xmath259 , we have @xmath260 , namely , all edges to absorbing nodes have the same weight . in this case , the invariant becomes @xmath261 as already discussed in section [ section : variants ] , equation ( [ equation : invariant ] ) has the following implications . * regarding the problem i - campaign , that is , when we ask to maximize @xmath31 by modifying only the internal opinion values @xmath2 , it is easy to see that the maximum increase occurs when selecting the @xmath4 smallest values @xmath2 and setting them to 1 . this observation motivates the algorithm ` min - s ` described in section [ section : algorithms ] . * consider the following problem : we want to maximize @xmath31 by only adding or removing edges in the social graph and without modifying any of the values @xmath2 or @xmath3 . equations ( [ eq : general : z - s ] ) and ( [ equation : invariant ] ) provide an expression for @xmath31 that is independent on the structure of the graph defined by the edges in @xmath249 , and thus , show that it is not possible to change @xmath31 by adding or removing edges .

we present iso - sws spectroscopy of ngc 1068 for the complete wavelength range 2.4 to 45 m at resolving power . selected subranges have been observed at higher sensitivity and full resolving power . we detect a total of 36 emission lines and derive upper limits for 13 additional transitions . most of the observed transitions are fine structure and recombination lines originating in the narrow line region ( nlr ) and the inner part of the extended emission line region . we compare the line profiles of optical lines and reddening - insensitive infrared lines to constrain the dynamical structure and extinction properties of the narrow line region . the most likely explanation of the considerable differences found is a combination of two effects . ( 1 ) the spatial structure of the ngc1068 narrow line region is a combination of a highly ionized outflow cone and lower excitation extended emission . ( 2 ) parts of the narrow line region , mainly in the receding part at velocities above systemic , are subject to extinction that is significantly suppressing optical emission from these clouds . line asymmetries and net blueshifts remain , however , even for infrared fine structure lines suffering very little obscuration . this may be either due to an intrinsic asymmetry of the nlr , as perhaps also suggested by the asymmetric radio continuum emission , or due to a very high column density obscuring component which is hiding part of the narrow line region even from infrared view . we present detections and limits for 11 rotational and ro - vibrational emission lines of molecular hydrogen ( h ) . they arise in a dense molecular medium at temperatures of a few hundred kelvin that is most likely closely related to the warm and dense components seen in the near - infrared h rovibrational transitions , and in millimeter wave tracers ( co , hcn ) of molecular gas . any emission of the putative pc - scale molecular torus is likely overwhelmed by this larger scale emission . in companion papers we use the sws data to derive the spectral energy distribution emitted by the active nucleus of ngc1068 ( ) , to put limits on infrared emission from the obscured broad line region ( ) , and discuss the continuum and its features in conjunction with sws spectra of other galaxies ( ) .

astro-ph0002008

ngc1068 is one of the nearest and probably the most intensely studied seyfert 2 galaxy . observations in all wavelength bands from radio to hard x - rays have formed a uniquely detailed picture of this object . ngc 1068 has played a key role in the development of unified scenarios for seyfert 1 and seyfert 2 galaxies ( @xcite ) , in the study of molecular gas in the nuclear region of seyferts ( e.g. @xcite ; @xcite ) , and in elucidating the importance of star formation activity coexistent with the agn , both on larger ( e.g. @xcite ) and smaller ( @xcite ; @xcite ) scales . ngc 1068 hosts a prominent narrow line region ( nlr ) that is approximately cospatial with a linear radio source with two lobes ( @xcite ) . the narrow emission line region has been extensively characterized from subarcsecond clouds probed by hst ( @xcite ; @xcite ) , the @xmath3 arcseconds of the nlr hosting most of the line flux ( e.g. @xcite ; @xcite ; @xcite ) , and the ionization cone and extended emission line region ( @xcite ; @xcite ) extending to radii of at least 30 ( 1 = 72 pc at the distance of 14.4 mpc , @xcite ) . the velocity field is complex , with an ensemble of rapidly moving clouds dominating the inner arcseconds and a more quiescent rotation pattern prevailing at larger radii ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) . while most of the excitation of the narrow line region and the extended emission line region is likely through photoionization by the central agn ( @xcite ) , high resolution observations suggest kinematic disturbance and possibly shock excitation of regions close to the radio outflow ( e.g. @xcite ) . with esa s _ infrared space observatory _ iso , sensitive mid - infrared spectroscopy of agns became possible , with detections of a broad range of low- and high - excitation fine structure lines , recombination lines , and pure rotational lines from such sources ( @xcite ; @xcite ; @xcite ) ) . model predictions of the mid - ir spectra of agn had been obtained prior to iso ( e.g. @xcite ) , but observations were restricted by limited sensitivity and focussed primarily on the continuum emission and broad features rather than emission lines . in this paper , we present the iso - sws spectra of ngc1068 and draw conclusions on the structure of the narrow line region that can be obtained mainly from the comparison of optical and reddening - insensitive infrared lines , and discuss the nature of the mid - infrared molecular hydrogen emission . the iso - sws data of ngc1068 are analysed further in several companion papers . alexander et al . ( 2000 ) use photoionization modelling based on the iso fine structure line set and other nlr lines to model the shape of the agn s spectral energy distribution . lutz et al . ( 2000 ) analyze limits on emission from the obscured broad line region . finally , sturm et al . ( 2000 ) discuss continuum energy distribution and features of ngc1068 in conjunction with iso - sws spectra of other galaxies . our paper is organised as follows . in 2 we discuss the iso - sws observations and data reduction . 3 presents results and implications of the density of the narrow line region . 4 uses infrared line profiles in comparison to optical ones to constrain the structure of the narrow line region . we discuss the mid - infrared molecular hydrogen emission in 5 and summarize in 6 . we have used the short wavelength spectrometer sws ( @xcite ) on board the _ infrared space observatory _ iso ( @xcite ) to observe the nuclear region of ngc1068 . in table [ tab : listobs ] we present a log of our observations . we carried out observations in the sws01 mode which provides a full 2.4 45@xmath0 m scan at slightly reduced spectral resolving power , as well as observations in the sws02 and sws06 modes targeted at full resolution observations of individual lines or short ranges . because of the large width of the emission lines in the ngc1068 narrow line region , we have mostly relied on the sws06 mode which can be set up to provide wider continuum baselines than standard sws02 line scans . we supplement observations from our iso guaranteed and open time with serendipitous information on some fine structure lines obtained in another iso project ( pi g. stacey ) , the main results of which are to be presented elsewhere . table [ tab : listobs ] includes also the position angle of the long axis of the sws apertures for the various observations . our pointing was always centered on the nucleus of ngc 1068 , but the sws apertures range from 14@xmath420 at short wavelengths to 20@xmath433 at the longest wavelengths ( @xcite ) . at the long wavelengths , we partly include the @xmath115 radius ring of star forming regions encircling the nucleus of ngc 1068 . the apertures were always oriented in approximately north - south ( or south - north ) direction , with position angles between -11 and -23 . we have analyzed the data using the sws interactive analysis ( ia ) system ( @xcite ; @xcite ) and calibration files of july 1998 . a preliminary account of part of the observations is given by lutz et al . since then , calibration files have been updated for wavelength calibration and in particular with respect to the sws relative spectral response function , leading to more reliable intercalibration between the ` aot bands ' forming a full sws spectrum . our data reduction started in the standard way and continued with steps of ( interactive ) dark current subtraction and matching up- and downscans . we eliminated data from those detectors of band 3 that were most noisy during a particular revolution , and from interactively identified regions with single detector signal jumps in bands 1 and 2 , and simultaneous 12-detector signal jumps in band 3 . after relative spectral response correction and flux calibration , we ` flatfielded ' the 12 detectors of a band to a consistent level , corrected for the iso velocity , and extracted the aar data product . redundant scans of the same line were shifted to a consistent level . single - valued spectra were produced by kappa - sigma clipping the aar dot cloud and rebinning it with a resolution of typically 3000 which does not lead to significant smearing for ngc1068 linewidths . for those ranges affected by fringes , the single - valued spectra were defringed using the iterative sine fitting option of the aarfringe module within the sws interactive analysis . the large number of observations required special treatment of redundant data . in addition , observations from revolution 285 where apparently affected by a slight ( @xmath5 ? ) pointing problem , which occured occasionally in the earlier phase of the iso mission . since sws beam profiles in some aot bands are peaked and slightly offset with respect to the nominal pointing ( a. salama , 1999 , priv . comm . ) , modest pointing offsets can cause noticeable flux losses in some aot bands and resulting band mismatches . such mismatches were evident in revolution 285 band 3 data . since most of the ngc 1068 mid - infrared flux comes from a small region ( @xcite ; @xcite ; @xcite ) , we corrected for this problem and the small scatter between other observations by the following scaling procedure : for the sws01 full spectrum obtained in revolution 285 , the individual aot bands were scaled to obtain both good match at band limits , and good agreement with the overall flux level as estimated from our other sws data and ground - based photometry ( @xcite ; @xcite ) . at wavelengths below 10@xmath0 m photometry from different epochs should be used with great caution because of the known variability ( @xcite ) , our fluxes are however in good agreement with the photometry of glass for the iso epoch . all other data were then scaled to this sws01 spectrum by the ratio inferred from the continuum flux densities . we believe the final flux scale to be accurate within the 20 - 30% typical for sws data ( @xcite ) . accurate wavelength calibration of the sws grating spectrometer is central for part of our line profile analysis , since shifts between lines in ngc1068 tend to be of the order 300km / s or less ( @xcite ) . valentijn et al . ( 1996 ) deduce an accuracy of @xmath130km / s from extensive calibrations during the sws performance verification phase . since then , a slow secular drift in sws wavelength calibration has been calibrated to similar accuracy . we have tested the wavelength calibration of the ngc1068 data , using identical calibration files to analyze spectra of the planetary nebulae ngc7027 and ngc6543 taken close to revolution 633 where some of the most important ngc1068 lineprofiles were taken . we confirm the excellent accuracy from these observations , the largest error not exceeding the value given by valentijn et al . this test and the good internal consistency of velocities measured in ngc1068 for different lines from the same species spanning most of the sws wavelength range ( e.g. h@xmath2 , see table [ tab : lineflux ] ) leads us to adopt an upper limit of 50 km / s for any systematic errors in our wavelength scale , taking into account a margin for mispointing . two emission features , which were tentatively detected in the preliminary analysis of lutz et al . ( 1997 ) could not be confirmed with the larger observational database and the improved calibration . a broad emission feature near 19@xmath0 m , which might be interpreted as silicate emission , is not confirmed with the new spectral response calibration . an emission line at 28@xmath0 m was identified as an unusually strong h@xmath2 s(0 ) line . this identification was later found suspect because of the line s larger width compared to the other h@xmath2 lines observed in ngc 1068 . the line was not confirmed in deeper follow up observations . detailed inspection of the original data indeed shows that it is an artifact of a highly unlikely coincidence of detector ` glitches ' at the expected wavelength of the s(0 ) transition . the 2.4 - 45@xmath0 m full spectrum of ngc1068 is displayed in figure [ fig : fullspec ] . some solid state features are superposed on the strong agn - heated mid - infrared continuum . these include 3.4@xmath0 m c - h absorption , 9.6@xmath0 m silicate absorption , and 7.7 , 8.6 , 11.3@xmath0 m ` pah ' emission ( see sturm et al ( 2000 ) for a discussion in conjunction with other sws spectra of galaxies ) . bright fine structure emission lines , mainly originating in the narrow line region , are already visible in the full spectrum . figures [ fig : ionspec ] and [ fig : h2spec ] show individual emission lines . the display range is chosen to be @xmath62500km / s around systemic velocity for recombination and fine structure lines , and @xmath61000km / s for the much narrower molecular lines . throughout this paper , we adopt a systemic velocity of 1148km / s ( @xcite ) . good rest wavelengths are available for the observed transitions from the literature and from recent iso determinations ( @xcite ) . many lines were observed repeatedly , figures [ fig : ionspec ] and [ fig : h2spec ] show only the best quality data . table [ tab : lineflux ] lists the measured line fluxes and limits , presenting averages of independent measurements with higher weight given to better data . table [ tab : lineflux ] includes also upper limits for some transitions ( not shown in figure [ fig : ionspec ] ) that were observed with good enough signal - to - noise ratio . we list such limits for transitions from elements like na or ar where other ionization stages are detected . fluxes of relatively narrow lines , for example from h@xmath2 and [ ] , were measured by direct integration of the continuum subtracted line profiles . the large linewidth makes this procedure error prone for faint lines from the nlr , where continuum definition is the main source of measurement error . the relative constancy of nlr line profiles over a wide range of lower ionization potentials ( see below ) lead us to adopt a different procedure to measure fluxes for the nlr lines : we derived a simple two - gaussian template from the brightest nlr lines ( [ ] 25.89@xmath0 m ( note nearby [ ] ) , [ ] 24.32@xmath0 m , [ ] 7.652@xmath0 m ) and used fits of this template ( figure [ fig : allprof ] ) plus a linear continuum to measure the fluxes of fainter nlr lines . the two gaussian components have fwhm 333 and 1246km / s , peak ratio narrow / wide 1.34 , and the wider component is blueshifted by 100km / s . in fitting , we varied only continuum flux and slope , total line flux , and total velocity . the fluxes measured this way agreed very well ( @xmath710% ) with those determined by direct integration not only for the lines used to derive the template , but also for other bright nlr lines like [ ] 3.028@xmath0 m . the fit thus preserves the fluxes for bright lines and is preferable for faint lines where continuum subtraction is the dominant source of error . for the blended lines of [ ] and h@xmath2 ( 0 - 0 ) s(7 ) near 5.5@xmath0 m , we list the fluxes resulting from a tentative gaussian fit using two components for [ ] and one component for h@xmath2 . the uncertainty of the s(7 ) flux is particularly large , up to a factor 2 . the h@xmath2 excitation diagram ( figure [ fig : excit ] ) in fact suggests that it may be overestimated . similar caution has to be applied to the tentative flux listed for the [ ] 25.99@xmath0 m line . a slight shoulder appears in the long wavelength wing of the [ ] 25.89@xmath0 m transition , but its flux is very uncertain and may at best be good enough to serve for consistency checks with other [ ] lines . feuchtgruber et al . ( 1997 ) discuss evidence that the lines of [ ] and [ ] at 9.0@xmath0 m are blended at the resolution of sws . we list only a total flux in table [ tab : lineflux ] . we detect a weak unidentified feature at rest wavelength about 7.555@xmath0 m . if real , its width would suggest a nlr origin . an instrumental origin due to an imperfection of the relative spectral response function ( rsrf ) can not be excluded but is unlikely since the rsrf shows very little structure at that wavelength . also , instrumental ` ghosts ' to strong sws lines are not known at such a level . a possible interpretation of this feature is that it is a blueshifted ( @xmath13800 km / s ) component of the nearby strong [ ] line containing @xmath11.5% of the total line flux . residual instrumental fringing prevents us from looking for analogous components near other strong lines such as [ ] or [ ] . at this point , the feature must be considered as possibly real but without an obvious identification by a line that is potentially strong in agn spectra . also , no line is seen at this wavelength in archival iso spectra of the high excitation planetary nebula ngc6302 . we postpone a detailed discussion of the line profiles to section [ sect : prof ] . figures [ fig : ionspec ] and [ fig : h2spec ] already suggest , however , that we are dealing with three distinct components : fine structure transitions from species with lower ionization potential @xmath8 40 ev ( i.e. from [ ] upwards ) have similar wide profiles and apparently originate in the nlr . fine structure lines from lower ionization stages , in particular [ ] and [ ] , are narrower and are most likely contaminated by star formation within their beams . this limits their use in modelling of the agn - excited nlr spectrum ( @xcite ) to upper limits rather than measurements . inspection of figure [ fig : ionspec ] suggests that the starburst contribution strongly dominates the low excitation lines like [ ] 33.48@xmath0 m and [ ] 34.81@xmath0 m which are measured with the largest aperture . lines measured with intermediate apertures like [ ] 12.81@xmath0 m and in particular [ ] 18.71@xmath0 m still show strong wings and will have a considerable nlr contribution . the smallest line widths are measured for transitions of molecular hydrogen . the fine structure lines detected by sws can be used to determine the density and , in conjunction with optical forbidden lines , the electron temperature of the line emitting gas in the narrow line and coronal line regions . differences in infrared and optical lineprofiles ( [ sect : prof ] ) discourage a determination of electron temperatures from the integrated fluxes , which would not account for significant variations in extinction across the nlr . a reliable average density can be determined , however , from the mid - infrared lines alone which are insensitive to electron temperature and extinction variations . the contribution of starburst excitation to the density - sensitive forbidden lines can be estimated using the large line width variation between the nlr and the circumnuclear ring of star formation regions . the most suitable nlr density diagnostic is provided by the ratio of the [ ] transitions at 14.32 and 24.32@xmath0 m . these lines can not be diluted significantly by circumnuclear star formation since they are undetected in starburst galaxies ( @xcite ) . they were observed with the same sws aperture size and with good signal - to - noise . adopting the same atomic data as alexander et al . ( 1999 , see also their figure 3 for diagrams of several density sensitive ratios ) and an electron temperature of 10000k , the observed [ ] ratio of 1.39 corresponds to an electron density @xmath9@xmath10 in the region of the nlr where species with lower ionization potential near 100ev prevail . a seemingly discrepant result is obtained from the [ ] transitions at 18.71 and 33.48@xmath0 m the observed ratio of 0.73 is consistent with the low density limit and corresponds to @xmath11@xmath10 . but , figure [ fig : ionspec ] shows the line profiles of the two transitions to be quite different : [ ] 18.71@xmath0 m shows strong wide wings and is apparently nlr - dominated with small starburst contamination . in contrast , the larger aperture of [ ] 33.48@xmath0 m collects more emission from the starburst ring showing up as a strong narrow component of the profile . if _ all _ 18.71@xmath0 m emission were from a nlr at 2000@xmath10 , the nlr contribution to the 33.48@xmath0 m flux would be about 1/3 , consistent with the weaker wings of this line . a similar problem may affect , to a lesser degree , the density sensitive ratio of the [ ] transitions at 15.55 and 36.01@xmath0 m . the observed ratio of 8.9 is lower than but probably still consistent with the low density limit ( @xmath112 ) which applies up to the @xmath9@xmath10 derived from [ ] . the modest signal - to - noise ratio of the 36.01@xmath0 m line makes it impossible to use the line profile to assess starburst contamination in the large aperture . a crude estimate can be obtained assuming that the ratio of _ starburst _ [ ] 36.01@xmath0 m and [ siii ] 33.48@xmath0 m seen additionally in the large aperture is 0.030.04 as in the prototypical starburst m82 ( @xcite ) . then , @xmath810% of the 36.01@xmath0 m line would be starburst contamination , bringing the ratio closer to its low density limit value . the density in nlr regions dominated by lower excitation species like [ ] and [ ] hence appears consistent with that for the higher excitation region containing [ ] . with respect to the coronal line region , the observed ratio 0.55 of the [ ] lines at 2.584 and 3.936@xmath0 m is close to its low density limit which implies @xmath12@xmath10 . the same limit is found for the circinus galaxy ( @xcite ) and ngc 4151 ( @xcite ) . such a density limit is consistent with all popular scenarios for coronal line formation in agn except for origin in a very dense transition region between nlr and blr . integrated emission lines profiles are an indirect tool to constrain the dynamical structure and extinction properties of the narrow line region . different lines probe different parts of the nlr and the velocity field is generally far from uniform . with the advent of linear optical detectors , considerable effort was devoted to studies of both forbidden and permitted optical line profiles in seyfert galaxies . although there is still no full consensus among different studies of the nlr forbidden lines , the emerging picture is as follows . + ( 1 ) the forbidden lines in most cases show blue asymmetries in the sense of a sharper falloff to the red than to the blue . line centroids are blueshifted with respect to the systemic velocity , whereas line peaks in high resolution spectra are close to systemic velocity . this has been most thoroughly studied in moderate excitation species including [ ] 5007 ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) but holds also for the higher excitation coronal lines ( @xcite ) . + ( 2 ) line widths and blueshifts often vary between different species observed in the same source . line profiles appear to be correlated with the ionization potential and/or the critical density . there are indications , but no complete consensus , that the correlation with the critical density may be the fundamental one ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . various scenarios have been put forward to explain these trends . most of them invoke extinction to explain blue asymmetries , and the most popular ones assume outflow in a dusty nlr , with higher excitation species probably originating closer to the central source in regions of higher velocity and obscuration . it is obvious that observations of _ infrared _ nlr emission are a powerful independent method to test such scenarios : near- and mid- infrared lines suffer more than an order of magnitude less extinction than in the optical . the combination of optical and infrared data should hence elucidate the role of dust obscuration . comparison of recombination line profiles in the optical with infrared ones would be advantageous because of the relative insensitivity of recombination line emissivities to local gas conditions . the line - to - continuum ratio of recombination lines in the infrared is low , however , and better profiles are obtained for coronal and fine structure lines which additionally cover a wide range of excitations . sturm et al . ( 1999 ) have presented a first such analysis using iso - sws observations of ngc4151 . on the basis of the similarity of optical and infrared profiles , they ruled out the most simple scenario of an outflowing nlr with pervasive dust , and suggested either a geometrically thin but optically highly thick obscuring disk , or an intrinsic asymmetry of the nlr . because of the large flux and width of its ` narrow ' lines , ngc1068 is best suited for a line profile analysis at the modest resolving power ( @xmath12000 ) of iso - sws . optical line profiles have been observed at very high resolving power by various groups ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) and show complex , multi - peaked structure related to individual cloud complexes within the narrow line region of ngc1068 . marconi et al . ( 1996 ) have extended this work into the near infrared . at lower resolving power , they do not discriminate the fine details of the best optical profiles but derive the line centroids by gaussian fits for a large set of near - infrared and optical lines . they find that all optical and near - infrared high excitation lines are significantly blueshifted with respect to systemic velocity ( @xmath13200km / s for lower ionization potential @xmath8 20ev ) . they interpret this significant trend as a consequence of non - isotropic flows or ionization patterns rather than selective extinction effects . we extracted line profiles for five high signal - to - noise sws lines by subtracting a linear continuum fitted outside 2500km / s from the line center and normalizing to the peak of the line . these five lines originate in species spanning a wide range of excitation potentials ranging from 55 to 303ev and are shown in figure [ fig : allprof ] . the remaining uncertainty of these profiles is dominated by noise for the high excitation lines of [ ] and [ ] and by continuum uncertainties for the other lines . these could be both due to weak underlying real continuum features ( e.g. pah near [ ] ) and due to residual fringing ( [ ] and [ ] ) . we do not show low excitation lines with significant starburst contribution , and the [ ] 14.32@xmath0 m and [ ] 15.55@xmath0 m lines which are consistent with those shown in figure [ fig : allprof ] but more uncertain due to fringing . brackett @xmath14 has a much lower line to continuum ratio but still good signal - to - noise ratio , and a line profile similar to the fine structure lines , as discussed by lutz et al . ( 2000 ) in the context of putting limits on a broad line region contribution . for lines too faint to derive a good line profile , we fitted a single gaussian plus linear continuum to derive at least a centroid velocity ( table [ tab : lineflux ] ) . while such a gaussian is not a good approximation to the intrinsic nlr profile , we adopted it for simplicity and for consistency with the optical / near - infrared data of marconi et al . we also fitted the two - component nlr profile of figure [ fig : allprof ] but do not list the derived velocities since they agree with the simple gaussian fit except for an offset that is constant within the uncertainties . from repeated observations for some of these lines , we estimate an error of @xmath750km / s . for lines with no velocity listed in table [ tab : lineflux ] , we estimate that the uncertainty of deriving the centroid of a broad noisy line is too large to include it into an analysis of ngc1068 . none of them , however , is discrepant by more than @xmath15300km / s which would suggest misidentification . in the following subsections , we will derive a large aperture optical nlr line profile for comparison with the iso data , compare optical and infrared profiles and centroid velocities , and interpret the differences found . mismatch between the typically small optical apertures and the large mid - ir ones is important when attempting to compare optical and mid - ir line profiles . datacubes from imaging spectroscopy would be ideal to extract optical line profiles matching the iso apertures . at this point , however , published imaging spectroscopy of ngc1068 is either limited in field size ( @xcite ) or lacks wavelength coverage : the datacube of cecil et al . ( 1990 ) has been obtained with 2600km / s total coverage in the [ ] lines that are additionally heavily blended with h@xmath14 , making it difficult to determine the extent to which broad components are missing in their total line profile ( their fig . we make use of two auxiliary large aperture optical spectra to address the problem of aperture mismatch : a 4000 to 7800 spectrum from wise observatory ( wo , s. kaspi 1999 , priv . communication ) , providing good fluxes of the brightest lines in a @xmath16 aperture ( position angle 0 ) , and a high spectral resolution coud echelle spectrum from karl schwarzschild observatory tautenburg ( kso , e. guenther 1999 , priv . communication ) , providing a good [ ] line profile ( though not good fluxes ) in a @xmath17 aperture ( mean position angle -28 , varying during integration ) . in addition , we estimated relative emission line fluxes in our apertures by integrating over the corresponding regions of a narrow band [ ] map ( r. pogge , m.m . derobertis , 1999 , unpublished data ) . both the wise spectrum and the [ ] map confirm that iso line fluxes of the ngc1068 nlr can be sensibly compared to smaller aperture optical data , since those already sample most of the flux in the narrow line region . for example , a 4 diameter aperture will already get @xmath1570% of the flux in the iso aperture . the fluxes measured in the large wise aperture for the brightest optical lines agree within @xmath130% with published smaller aperture ones ( e.g. @xcite ; @xcite ; @xcite ) . figure [ fig : ksoveill ] displays our kso @xmath17 aperture [ ] 5007 line profile in comparison to its @xmath18 equivalent ( @xcite ) . the line profile changes induced by this more than tenfold increase in aperture area are relatively modest and fit expectations from high resolution longslit spectroscopy . while the major components of veilleux spectrum are well reproduced in the kso data , their ratios differ somewhat leading to an overall slightly wider profile . this is fully consistent with observations of relatively broad components over larger regions not sampled by veilleux aperture ( @xcite ; @xcite ; @xcite ) . the only feature in the kso profile not present in the veilleux profile is an additional narrow feature at or slightly redshifted from systemic velocity . this feature almost certainly corresponds to the ` velocity spike ' in the ne region of the nlr detected by many authors but seen perhaps most clearly in the data of meaburn & pedlar ( 1986 ) . this feature is missed by veilleux aperture but partly covered by the kso data . the kso aperture is still about three times smaller in area than the sws apertures through which the best fine structure line profiles have been taken . the drop in [ ] surface brightness with radius is so rapid ( e.g. fig . 1 of meaburn & pedlar 1986 ) that only modest differences in the total line profile are expected . an exception to this is the ne region of the nlr about 6 from the nucleus which was incompletely covered . the kso slit orientation can not be chosen freely and was approximately aligned with the iso apertures but not with the nlr ( pa -28 instead of pa @xmath1530 ) , missing part of the ne end of the nlr . long slit spectroscopy ( e.g. @xcite ) shows this ne region to be dominated by the narrow ` velocity spike ' near systemic velocity which is already seen in the comparison of kso and veilleux ( 1991 ) profiles . we hence expect this spike to be more prominent in an optical line profile fully equivalent to the iso aperture . integrating the [ ] map over the iso and kso apertures we estimate a need to add @xmath111% to the kso flux to account for the ne region and other low surface brightness emission near systemic velocity . we have taken this into account by adding such a narrow ( fwhm 150km / s ) component to the kso profile , and will use this in the following as basis of our optical line profile comparison ( see also figure [ fig : ksoveill ] ) . use of such a modified profile is supported by the spectrum of pelat & alloin ( 1980 ) which was obtained with a rotating longslit sweeping across the ne region of the nlr , and showing a similar narrow spike ( their component 5 ) . the optical and infrared line profiles in ngc1068 differ strongly . this is most evident in figure [ fig : opto4 ] which compares the profiles of [ ] 25.89@xmath0 m and [ ] 5007 . we chose [ ] as the representative infrared line since it was observed with very good s / n and is close to [ ] in lower ionization potential of the emitting species ( 55 vs. 35ev ) , ensuring origin in a similar region of the nlr . critical densities ( @xmath19@xmath10 vs. @xmath20@xmath10 ) match less well than [ ] or [ ] would , but this is less relevant given the low nlr density we have inferred ( see [ subsect : density ] ) . the optical [ ] line profile is both blueshifted and broader than the infrared [ ] line profile . there are however also significant similarities . shoulders near -900km / s and @xmath1350km / s are present in both profiles . adopting different relative scalings ( figure [ fig : opto4 ] ) , the impression arises that the two profiles in fact agree fairly well over parts of their extent if the scaling is set properly . the main difference lies in _ different relative strengths of blue wing , center , and red wing _ , the infrared profile having a stronger red wing and center . the four highest quality infrared profiles are compared in figure [ fig : procomp ] using a spread velocity scale . the most obvious and significant variation is in [ ] which is both broader and more blueshifted than the lower excitation lines . the same is observed for the more noisy [ ] line . as already noted , the [ ] , [ ] , and [ ] profiles with excitation energies ranging from 55 to 126ev are very similar but the overplot shows some variation in detail . there is a minor shift in the narrow core of the [ ] line which is , however , not significant compared to the quoted systematic uncertainty . comparing [ ] to [ ] taken from the same observation there is even less shift . concerning the broader wings , there are significant differences in addition to the possible presence of [ ] at @xmath151100km / s in the [ ] profile . there is a trend from [ ] to [ ] in the blue wing becoming stronger and the red wing fainter . such profile variations determine the line centroids derived from gaussian fits ( table [ tab : lineflux ] ) to which we add a centroid of 1015 km / s derived in the same way for our extrapolated large aperture optical [ ] profile . anticipating that the shifts may reflect several partially degenerate influences , we show in figure [ fig : centroids ] centroid velocities as a function of lower ionization potential , critical density , and extinction for the particular line . the extinction values are relative and based on a preliminary iso - based extinction curve for the center of our galaxy ( @xcite ) . the trend observed in the iso data for the velocity centroids as a function of ionization potential ( figure [ fig : centroids ] ) is markedly different from the equivalent dataset obtained in the optical and near - infrared ( figure [ fig : centroids ] , data taken from marconi et al . 1996 ) . in both data sets the velocity centroids of the lowest excitation non - nlr lines ( h@xmath2 , [ ] ) are close to systemic . however , none of the iso lines reach the large blueshifts observed consistently over a wide excitation range in the optical and near - infrared . the data sets are least discrepant at the high excitation end . here , [ ] 3.936@xmath0 m is common to both sets and agrees within the errors , though being somewhat less blueshifted in the iso data . the discrepancy is largest at intermediate excitation ( 20 - 200ev ) where the optical / nir lines are all strongly blueshifted while the iso lines only slowly deviate from systemic velocity as excitation increases . in addition to the cloud distribution and kinematics , line profiles reflect the emissivities of fine structure or forbidden lines in the narrow line region clouds , which are affected by many parameters : ionization equilibrium , density , electron temperature , and extinction . if any of these parameters vary among kinematically distinct structural components of the nlr , line profile variations will result that in turn can help elucidate the nlr structure . an important aspect is that some of these parameters are partially degenerate in infrared datasets : shorter wavelength ( 2 - 5@xmath0 m ) lines which suffer higher extinction are also typically high excitation coronal lines with high critical densities , whereas the longer wavelengths are dominated by ions of lower excitation with transitions with lower critical densities . here we have assumed that the observed lines are emitted locally , with no contribution of scattered light . this is not strictly correct for part of the ngc 1068 narrow line region , in particular the ne region ( @xcite ; @xcite ) . the impact of scattering on our analysis depends on the properties of the scatterer . if scattering is wavelength - independent , our profile comparison is unaffected since scattering would effectively only redistribute the line emission spatially within our large apertures . if ( dust ) scattering decreases with wavelength , shorter wavelength line profiles would be modified more strongly . we estimate the effect on the integrated line profiles is not large , since the [ ] polarisations measured by inglis et al . ( 1995 ) reach at most a few percent in regions that are in addition minor contributors to the total flux , and since the considerable spatial variation in line profiles is not suggestive of scattered radiation originating in a central source . it is unlikely that density variations play a direct role in the profile variations . our density estimate of @xmath21 is too far below the relevant critical densities , making much higher densities over a significant part of the nlr unlikely , which would be required to create the variations . strictly speaking however , this estimate applies only to the moderately excited ( 100ev ) gas , and our upper limit for the coronal region density is still consistent with densities higher than the critical densities of some lower excitation line . strong collisional suppression of part of some low critical density fine structure line profiles is also unlikely from the similarity of their line centroids to that of the recombination line brackett @xmath14 . we hence believe that the clear correlation of centroid velocities with fine structure line critical density ( figure [ fig : centroids]b ) is mostly a secondary consequence of the correlation with ionization potential . variations in electron temperature can also affect the line profiles . emissivities of optical forbidden lines like [ ] 5007 strongly vary with electron temperature while the infrared fine structure lines originate close to the ground state and are much less sensitive to temperature . hence , at least those profile variations seen among the infrared lines ( figures [ fig : procomp ] and [ fig : centroids ] ) must be unrelated to temperature fluctuations . for the optical / ir profile variations there is a basic ambiguity of an optical component being faint due to low electron temperature or due to high extinction . optical electron temperature determinations e.g. from [ ] 4363/5007 are not available for the various kinematic components of ngc1068 . the line profiles of the temperature insensitive recombination lines in the optical and ir do not resemble each other , but rather follow the shapes of the forbidden optical and ir lines , respectively ( @xcite , fig . [ fig : ionspec ] ) . this is inconsistent with temperature variations being the main origin of profile variations . the exclusion of other factors suggests that ionization structure and extinction are the main source of the observed variation of optical / ir line profiles . previous studies of the spatial and kinematical structure of the ngc1068 nlr ( e.g. @xcite , @xcite ) point to the existence of two spatial and kinematical components . the first is a strong ionization cone with associated blueshifted outflow , with the highest excitation species likely concentrated towards the central and fastest part of this cone . the second is an extended system of photoionized clouds closer to systemic velocity . the correlation between the iso centroid velocities and the ionization potential ( figure [ fig : centroids]a ) fits this picture well . if the relative importance of the fast outflow gradually increases towards high excitation lines , the gradual centroid shift is easily explained . however , the marked differences between optical [ ] and infrared [ ] profile ( figure [ fig : opto4 ] ) and the different optical and infrared centroid velocities at similar ionization potential ( figure [ fig : centroids]a , including data from marconi et al . 1996 ) show that this picture must be incomplete . we suggest that these remaining differences are due to extinction variations across the nlr , with a general trend of higher exctinction in the redshifted ( sw ) than in the blueshifted ( ne ) part . the optical profile can be explained by an intrinsic profile similar to that of the ir lines whose line center and red wing are reddened by a few magnitudes , leading to the obscuration of about half of the total line flux . a similar extinction pattern is suggested by the increase of polarization from the blue to the red wing of the _ narrow _ lines in the central arcseconds ( @xcite ; @xcite ; @xcite ) , attributed to absorption by aligned grains . considering that the optical profiles are modified by extinction , it is important to recognize that the less obscured near- and mid - infrared lines remain blueshifted with respect to systemic velocity , high excitation ones more strongly than lower excitation ones . also , coronal line emission observed in the little obscured near - infrared is still much stronger in the northeast cone than in the southwest ( @xcite ) . this might be due to heavy extinction of an intrinsically symmetric nlr , or due to a real asymmetry . no distinction can be made from the present data . a trend with ir extinction is not clear in figure [ fig : centroids]c and would be difficult to separate from the trends with ionization potential and critical density because of the mentioned degeneracy . an intrinsically asymmetric nlr with outflow preferentially towards the observer is fully consistent with the observations of ngc1068 . the difference in strength of the two radio lobes ( @xcite ) may be in support of such an asymmetric scenario , although the relation between radio lobe flux and nlr emission is certainly not simple . consistency with an asymmetric nlr was also noted for the iso spectra of ngc4151 ( @xcite ) . however , large randomly oriented agn samples should not show the preferential blueshift which is noted at least in optical samples . an alternative scenario of a geometrically small but optically highly thick screen ( disk or torus ) obscuring part of the receding nlr region was proposed by sturm et al . ( 1999 ) for ngc 4151 . such a scenario can also fit the ngc1068 data provided the screen obscures a relatively larger fraction of the coronal line region than of the larger nlr which is dominated by medium excitation species . this is not implausible given the similar ( arcsecond ) spatial scale of the central concentration of high column density molecular gas ( e.g. @xcite ) and the coronal line region as mapped in [ ] ( @xcite ; @xcite ) . while studies of few individual sources will remain ambiguous , a search for preferential shifts using high resolution near- or mid - infrared spectroscopy should address this issue provided the sample of seyferts is large enough and not biased by orientation of a putative asymmetric outflow , as may be the case when identifying seyferts in the optical . overall , the structure of the ngc1068 narrow line region seems to determine the optical / infrared line profiles via differences in weight of outflowing cone and extended components , and additional extinction variations that are significant for the optical wavelength range . the relative weight of the cone is larger for higher excitation species . a tantalizing ambiguity remains that can not be resolved from a single source study : is the nlr intrinsically one - sided or is there a very high obscuration screen blocking also part of the ir emission from our view ? ngc1068 was the first galaxy detected in the rovibrational transitions of molecular hydrogen ( @xcite ) and has been studied since in considerable detail both in these lines tracing fairly excited molecular material ( e.g. @xcite ; @xcite ; @xcite ) , and by millimeter wave interferometry tracing colder components ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) . the system of dense , warm cloud cores in the central few arcseconds inferred from the millimeter studies calls for observations in the pure rotational transitions of molecular hydrogen , which trace gas of typically a few 100k , intermediate between the near - infrared and millimeter wave tracers . our sws observations of these lines ( figure [ fig : h2spec ] , table [ tab : lineflux ] ) are summarized in the h@xmath2 excitation diagram of figure [ fig : excit ] . the diagram has a curved shape suggestive of a mixture of temperatures , as expected from other galaxies observed with iso in the rotational lines of molecular hydrogen ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . the location of the s(3 ) point slightly below the general trend suggests a moderate extinction towards this line whose wavelength is near the center of the silicate absorption feature ( @xmath22 ) . as noted earlier , the flux for the heavily blended s(7 ) line is very uncertain , so that the shorter wavelength rovibrational lines of similar excitation will probably be a more trustworthy representation of the excitation diagram at these upper level energies . while the higher rotational lines like s(5 ) and s(7 ) probe the same excited but low mass component as the near - infrared rovibrational lines , the bulk of the warm gas observed with iso will reside in the component traced by the s(1 ) line . in estimating its mass , we will assume a hydrogen ortho / para ( o / p ) ratio in equilibrium at the local temperature ( @xcite ) . combining the s(1 ) flux with the s(0 ) limit on one hand and the s(3 ) detection on the other , the temperature of the s(1 ) emitting gas is found to lie in the range 140k@xmath23t@xmath7375k , assuming the extinction correction for s(3 ) is modest and aperture effects are minor . because of the very steep temperature sensitivity of the rotational line emissivities , the corresponding mass varies drastically , between about @xmath24 for the 375 k case and @xmath25 for the 140 k case . for comparison with other mass estimates , we will adopt @xmath1200k and @xmath26 as a possible approximation to the curvature of the excitation diagram . this mass would be of the order 5% of the total gas mass estimated from the co interferometric map ( @xcite and priv . comm . ) within the 17@xmath0 m iso beam . compared to a similar estimate for the starburst galaxy ngc3256 ( @xcite , 3% for 150k warm gas temperature ) , the fraction of warm gas and/or its temperature must be higher , but not exceeding on average that inferred by kunze et al . ( 1996 ) for the highly active star forming region in the ` overlap region ' of the antenna galaxies ( 8% for 200k warm gas temperature ) . the ngc1068 molecular hydrogen emission sampled by sws may represent a mixture of relatively cool gas from the 15 radius molecular ring partly covered by the longer wavelength apertures , and a warmer component from the unusually dense and warm central few arcseconds ( @xcite ; @xcite ) . knowing from the s(3 ) measurement that the extinction to the h@xmath2 emitting region can not be very large , we can compare our 1 - 0 q(3 ) flux to 1 - 0 s(1 ) fluxes measured in smaller apertures ( 1.4 to 2.2@xmath27w@xmath28 ; @xcite ; @xcite ) . our q(3 ) line flux falls in the middle of that range . since the intrinsic q(3)/s(1 ) flux ratio is 0.7 , the implication is that the central few arcseconds dominate at least for the more highly excited hydrogen lines , though there may be some extended contribution . some of the molecular hydrogen emission in ngc1068 may originate in x - ray irradiated gas . we include in table [ tab : lineflux ] upper limits for two transitions of h@xmath29 which have been proposed as a signature of x - ray heated molecular gas ( @xcite ) . we note , however , that these limits are not at all stringent and are fully consistent with even the ` high ' end of h@xmath29 flux expectations for x - ray illumination . estimates are uncertain , see e.g. the much lower predictions of maloney et al . . the question can be raised whether the ngc1068 h@xmath2 spectra in fact contain a direct signature of a parsec - scale molecular torus . krolik and lepp ( 1989 ) have modelled molecular line emission of such an x - ray illuminated torus , predicting emission in some molecular hydrogen lines like ( 0 - 0 ) s(5 ) that may under favorable conditions be detectable at iso - sws sensitivities . for ngc1068 it is evident that larger scale emission may swamp any possible torus emission , as already cautioned by krolik and lepp ( 1989 ) . the iso data smoothly complement the near - infrared emission originating in larger scale ( @xmath1100pc ) clouds , following an excitation diagram plausibly ascribed to the same clouds . if any torus emission were present at lower level , it may be difficult to discriminate from the larger scale emission since , depending on black hole mass and spatial scale , its velocity width could be very similar to the larger scale emission . the widths of our h@xmath2 rotational lines are consistent with those for co observed on 100pc and larger scales , with no evidence for other kinematic components . however , even if the rotational emission does not trace a compact torus , it may still be excited to a significant fraction by uv radiation , x - rays or shocks related to the agn . iso - sws spectroscopy provides the first detailed census of the mid - infrared spectrum of the prototypical seyfert 2 galaxy ngc1068 . we have detected 36 emission lines on top of the strong agn - heated continuum . most lines originate in the nlr characterized by a density of @xmath12000@xmath10 . we have compared the mid - infrared iso line profiles with optical emission line profiles produced in the nlr . the line profiles are consistent with a model where the nlr is a combination of a highly ionized outflow and lower excitation extended emission , with extinction significantly affecting the optical line profiles . remaining blueshift and asymmetry of the least obscured lines may reflect either intrinsic asymmetry of the nlr or an additional very high column density obscuring component . we detect strong emission from warm molecular hydrogen , which most likely originates on the 100pc to kpc scale , and which is also probed by emission in near - infrared and millimeter wave tracers of molecular material . this emission masks any possible emission from a putative parsec - scale molecular torus . companion papers use the sws data to model the spectral energy distribution of the active nucleus , to put limits on emission from the obscured broad line region , and discuss the continuum and its features . we are grateful to eike guenther and shai kaspi for obtaining optical spectra that were invaluable in the interpretation of the iso spectra , and to richard pogge and m.m . de robertis for providing us with an unpublished [ ] image of ngc1068 . sws and the iso spectrometer data center at mpe are supported by dlr ( dara ) under grants 50 qi 8610 8 and 50 qi 9402 3 . we acknowledge support by the german - israeli foundation ( grant i-0551 - 186.07/97 ) . alexander t. , sturm e. , lutz d. , sternberg a. , netzer h. , genzel r. 1999 , , 512 , 204 alexander t. , et al . , submitted to apj alloin d. , pelat d. , boksenberg a. , sargent w.l.w . 1983 , , 275 , 493 antonucci r.r.j . , miller j.s . 1985 , , 297 , 621 appenzeller i. , streicher r. 1988 , , 95 , 45 axon d.j . , marconi a. , capetti a. , macchetto f.d . , schreier e. , robinson a. , 1998 , , 496 , l75 bailey j. , axon d.j . , hough j.h . , ward m.j . , mclean i. , heathcote s.r . 1988 , , 234 , 899 blietz m. , cameron m. , drapatz s. , genzel r. , krabbe a. , van der werf p. , sternberg a. , ward m. 1994 , , 421 , 92 bock j.j . , marsh k.a . , ressler m.e . , werner m.w . 1998 , , 504 , l5 braatz j.a . , wilson a.s . , gezari d.y . , varosi f. , wilson a.s . 1993 , , 409 , l5 brinks e. , skillman e.d . , terlevich r.d . , terlevich e. 1997 , , 248 , 31 cameron m. , storey j.w.v . , rotaciuc v. , genzel r. , verstraete l. , drapatz s. , siebenmorgen r. , lee t.j . 1993 , , 419 , 136 capetti a. , axon d.j . , macchetto f. , sparks w.b . , boksenberg a. 1995 , , 446 , 155 cecil g. , bland j. , tully r.b . 1990 , , 355 , 70 dahari o. , de robertis m.m . 1988 , , 331 , 727 davies r. , sugai h. , ward m.j . 1998 , , 300 , 388 de robertis m.m . , osterbrock d. 1986 , , 301 , 727 dietrich m. , wagner s.j . 1998 , , 338 , 405 draine b.t . 1989 , in : `` proc . 22nd eslab symposium on infrared spectroscopy in astronomy '' , esa sp-290 , ( noordwijk : esa ) , 93 draine b.t . , woods d.t . 1990 , , 363 , 464 evans i.n . , ford h.c . , kinney a.l . , antonucci r.r.j . , armus l. , caganoff s. , 1991 , , 369 , 27 feuchtgruber h. , et al . 1997 , , 487 , 962 frster - schreiber n. 1998 , phd thesis , ludwig - maximilians - universitt mnchen genzel r. , et al . 1998 , , 498 , 579 glass i.s . 1997 , , 248 , 191 de graauw th . , et al . 1996 , , 315 , l49 heckman t.m . , miley g.k . , van breugel w.j.m . , butcher h.r . 1981 , , 247 , 403 helfer t.t . , blitz l. 1995 , , 450,90 hummer d.g . , storey p.j . 1987 , , 224 , 801 inglis m.d . , young s. , hough j.h . , gledhill t. , axon d.j . , bailey j.a . , 1995 , , 275 , 398 kessler m.f . , 1996 , , 315 , l27 koski a.t . 1978 , , 223 , 56 krolik j.h . , begelman m.c . 1988 , , 329 , 702 krolik j.h . , lepp s. 1989 , , 347 , 149 kunze d. , et al . 1996 , , 315 , l101 lahuis f. , et al . 1998 , in astronomical data analysis software and systems vii , a.s.p . conference series , vol . 145 , eds . r. albrecht , r.n . hook and h.a . bushouse , p.224 lebofsky m.j . , rieke g.h . , kemp j.c . 1978 , , 222,98 lutz d. , et al . 1996 , , 315 , l269 lutz d. , sturm e. , genzel r. , moorwood a.f.m . , sternberg a. 1997 , , 248 , 217 lutz d. , et al . 1997 , in : `` first iso workshop on analytical spectroscopy '' , eds . a.m. heras et al . , esa - sp419 ( noordwijk : esa ) , 143 lutz d. , et al . 2000 , , in press macchetto f. , capetti a. , sparks w.b . , axon d.j . , boksenberg a. 1994 , , 435 , l15 maloney p.r . , hollenbach d.j . , tielens a.g.g.m . 1996 , , 466 , 561 marconi a. , van der werf p.p . , moorwood a.f.m . , oliva e. 1996 , , 315 , 335 meaburn j. , pedlar a. 1986 , , 159 , 336 moorwood a.f.m . , lutz d. , oliva e. , marconi a. , netzer h. , genzel r. , sturm e. , de graauw th . 1996 , , 315 , l109 myers s.t . , scoville n.z . 1987 , , 312 , l39 oliva e. , moorwood a.f.m . 1990 , , 348 , l5 pcontal e. , ferruit p. , binette l. , wilson a.s . 1997 , , 248 , 167 pelat d. , alloin d. 1980 , , 81 , 172 pelat d. , alloin d. , fosbury r.a.e 1981 , , 195 , 787 penston m.v . , fosbury r.a.e . , boksenberg a. , ward m.j . , wilson a.s . 1984 , , 208 , 347 pier e.a . , krolik j.h . 1992 , , 401 , 99 pogge r.w . 1988 , , 328 , 519 rieke g.h . , low f.j . 1975 , , 199 , l13 rigopoulou d. , et al . 1996 , , 315 , l125 roelfsema p.r . , et al . 1996 , , 315 , l289 schaeidt s.g . , et al . 1996 , , 315 , l55 schinnerer e. 1999 , phd thesis , ludwig - maximilians - universitt mnchen shields g.a . , oke j.b . 1975 , , 197 , 5 spinoglio l. , malkan m.a . 1992 , , 399 , 504 spoon h.w.w . , et al . 2000 , submitted to a&a sternberg a. , genzel r. , tacconi l. 1994 , , 436 , l131 sternberg a. , neufeld d. 1999 , , 520 , 137 sturm e. , et al . 1996 , , 315 , l133 sturm e. , alexander t. , lutz d. , sternberg a. , netzer h. , genzel r. 1999 , , 512 , 197 sturm e. , et al . 2000 , in preparation tacconi l.j . , genzel r. , blietz m. , cameron m. , harris a.i . , madden s. 1994 , , 426 , l77 telesco c.m . , decher r. 1998 , , 334 , 573 thatte n. , quirrenbach a. , genzel r. , maiolino r. , tecza m. 1997 , , 490 , 238 thatte n. et al . , 1999 , in preparation thompson r.i . , lebofsky m.j . , rieke g.h . 1987 , , 222 , l49 thompson r.i . , corbin m. 1999 , , in press ( proceedings of ringberg workshop on ultraluminous infrared galaxies ) tully r.b . 1988 , nearby galaxies catalog ( cambridge : cambridge university press ) unger s.w . , lewis j.r . , pedlar a. , axon d.j . 1992 , , 258 , 371 valentijn e.a . , et al . 1996 , , 315 , l60 valentijn e.a . , van der werf p.p . , de graauw th . , de jong t. 1996 , , 315 , l145 veilleux s. 1991 , , 75 , 357 vrtilek j.m . , carleton n.p . 1985 , , 294 , 106 walker m.f . 1968 , , 151 , 71 wieprecht e. , et al . 1998 , in astronomical data analysis software and systems vii , a.s.p . conference series , vol . 145 , eds . r. albrecht , r.n . hook and h.a . bushouse , p.279 whittle m. 1985a , , 213 , 1 whittle m. 1985b , , 216 , 817 wilson a.s . , ulvestad j.s . 1983 , , 275 , 8 cccrc revolution&date & aot & duration&slit position angle + & ut & & s & + 285 & 28-aug-1996&sws01&6538 & -11 + 285 & 28-aug-1996&sws02&3212 & -11 + 605 & 13-jul-1997&sws06&4080 & -23 + 633 & 10-aug-1997&sws06&8410 & -16 + 643 & 20-aug-1997&sws06&5710 & -13 + 788 & 12-jan-1998&sws06&7766 & 157 + 788 & 12-jan-1998&sws02&1990 & 157 + 792 & 15-jan-1998&sws06&14642 & 158 + 806 & 29-jan-1998&sws06&11810 & 162 + 818 & 11-feb-1998&sws06&2134 & 165 + lrrrccr line & @xmath30&e@xmath31 & flux & meth & aperture&velocity + & @xmath0 m & ev & 10@xmath32w@xmath28 & & @xmath33&km / s + h@xmath2 ( 1 - 0 ) q(3 ) & 2.424 & 0.0 & 1.7&i&14@xmath420&1140 + [ siix ] & 2.584 & 303.2 & 3.0&f&14@xmath420 & + br @xmath34 , h@xmath2 ( 1 - 0 ) o(2)&2.626&13.6&4.1&f&14@xmath420 & + [ mgviii ] & 3.028 & 224.9 & 11.0&f&14@xmath420&1000 + [ caiv ] & 3.207 & 50.9 & 1.3&f&14@xmath420 & + h@xmath2 ( 1 - 0 ) o(5 ) & 3.235 & 0.0 & 0.8&i&14@xmath420&1120 + pf @xmath35 & 3.741 & 13.6&@xmath364.0&u&14@xmath420 & + [ siix ] & 3.936 & 303.2 & 5.4&f&14@xmath420 & 950 + br @xmath14 & 4.052 & 13.6 & 6.9&f&14@xmath420&1110 + h@xmath37&4.350&0.0&@xmath363.0&u&14@xmath420 & + [ mgiv ] & 4.487 & 80.1 & 7.4&f&14@xmath420 & + [ arvi ] & 4.530 & 75.0 & 15.0&f&14@xmath420 & + pf @xmath34 & 4.654 & 13.6&@xmath3610.0&u&14@xmath420 & + [ navii ] & 4.685 & 172.1&@xmath3611.0&u&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(9 ) & 4.695 & 0.0&@xmath363.5&u&14@xmath420 & + [ feii ] & 5.340 & 7.9 & 5.0&i&14@xmath420&1120 + [ mgvii ] & 5.503 & 186.5 & 13.0&s&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(7 ) & 5.511 & 0.0&3.5::&s&14@xmath420 & + [ mgv ] & 5.610 & 109.2 & 18.0&f&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(5 ) & 6.910 & 0.0 & 5.9&i&14@xmath420&1180 + [ arii ] & 6.985 & 15.8 & 13.0&i&14@xmath420 & + [ naiii ] & 7.318 & 47.3 & 5.8&f&14@xmath420 & + pf @xmath14 & 7.460 & 13.6&@xmath363.0&u&14@xmath420 & + unidentified & @xmath17.555 & & 1.8&f&14@xmath420 & + [ nevi ] & 7.652 & 126.2&110.0&f&14@xmath420&1030 + [ fevii ] & 7.815 & 99.1 & 3.0&f&14@xmath420 & + [ arv ] & 7.902 & 59.8&@xmath3612.0&u&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(4 ) & 8.025 & 0.0 & 3.5&i&14@xmath420&1130 + [ navi ] & 8.611 & 138.4&@xmath3616.0&u&14@xmath420 & + [ ariii]/[mgvii]&8.991&27.6/186.5&25.0&f&14@xmath420 & + [ fevii ] & 9.527 & 99.1 & 4.0&f&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(3 ) & 9.665 & 0.0 & 6.0&i&14@xmath420&1130 + [ siv ] & 10.511 & 34.8 & 58.0&f&14@xmath420 & + h@xmath2 ( 0 - 0 ) s(2)&12.279 & 0.0&@xmath368.0&u&14@xmath427 & + [ neii ] & 12.814 & 21.6 & 70.0&i&14@xmath427&1080 + [ arv ] & 13.102 & 59.8&@xmath3616.0&u&14@xmath427 & + [ nev ] & 14.322 & 97.1 & 97.0&f&14@xmath427&1020 + [ neiii ] & 15.555 & 41.0&160.0&f&14@xmath427&1090 + h@xmath38&16.331&0.0&@xmath396.0&u&14@xmath427 & + h@xmath2 ( 0 - 0 ) s(1)&17.035 & 0.0 & 7.7&i&14@xmath427&1120 + [ feii ] & 17.936 & 7.9&@xmath3610.0&u&14@xmath427 & + [ siii ] & 18.713 & 23.3 & 40.0&i&14@xmath427&1120 + [ nev ] & 24.318 & 97.1 & 70.0&f&14@xmath427&1060 + [ oiv ] & 25.890 & 54.9&190.0&f&14@xmath427&1100 + [ feii ] & 25.988 & 7.9&8.0::&s&14@xmath427 & + h@xmath2 ( 0 - 0 ) s(0)&28.219 & 0.0&@xmath362.5&u&20@xmath427 & + [ siii ] & 33.418 & 23.3 & 55.0&i&20@xmath433&1110 + [ siii ] & 34.815 & 8.2 & 91.0&i&20@xmath433&1110 + [ neiii ] & 36.014 & 41.0 & 18.0&f&20@xmath433 & +

most quantum states have wavefunctions that are widely spread over the accessible hilbert space and hence do not have a good description in terms of a single classical geometry . in order to understand when geometric descriptions are possible , we exploit the ads / cft correspondence in the half - bps sector of asymptotically universes . in this sector we devise a coarse - grained metric operator " whose eigenstates are well described by a single spacetime topology and geometry . we show that such half - bps universes have a non - vanishing entropy if and only if the metric is singular , and that the entropy arises from coarse - graining the geometry . finally , we use our entropy formula to find the most entropic spacetimes with fixed asymptotic moments beyond the global charges .

0705.4431

[ intro ] the @xmath1-bps sector of @xmath2 @xmath3 yang mills theory provides a simple playground for understanding quantum gravity , as the semiclassical mapping between spacetime geometries and coherent field theoretic states is particularly simple @xcite . here we propose an extension of this mapping to the quantum level . using a second - quantized formalism , we define a ` metric ' operator in the yang - mills theory whose eigenstates map to universes with a single well - defined topology and geometry , while non - eigenstates do not have a good description in terms of single spacetimes . in the field theory , the half - bps states can be constructed by reducing the @xmath3 yang - mills lagrangian to a hermitian matrix model and studying its hilbert space @xcite . this in turn has a description in terms of @xmath4 fermions in a harmonic potential . approximate eigenstates of our metric operator can be constructed by placing the individual fermions in coherent states . ( see @xcite for a related discussion in the @xmath5 context . ) using this formalism , we can associate an entropy to any classical , asymptotically @xmath6 spacetime in the llm family given by @xcite , by counting how many microscopic configurations give the same coarse - grained metric . we find that only the singular spacetimes have a finite entropy ; thus spacetime singularities and the appearance of gravitational entropy are intimately tied together , at least within the class of llm geometries . by studying the partition sum over geometries we show that only the smooth spacetimes appear in the underlying configuration space , which is thus effectively discretized . this provides further evidence that singularities and their associated entropy arise only in passing to our long - distance effective description . our formalism can be used to determine the most entropic geometries with specified angular moments ; in essence , this extends the extremal `` superstar '' black hole geometry to black objects carrying higher angular moments . the data required for constructing the metric in these half - bps universes can be extracted from the effective single - particle phase space distribution of the fermionic representation of the half - bps matrix model . this procedure generically loses information about the complete state , which is fully represented in an @xmath4-particle phase space . hence the semiclassical geometry will generally lose information about the underlying quantum state . we show that in the semiclassical limit ( @xmath7 ) this sort of information loss occurs if and only if the metric is singular . while this paper was in preparation we received @xcite which has substantial overlap with some sections of this paper , and @xcite which carries out similar analyses of the entropy of coarse - grained bps spacetimes in other settings . [ [ gravity ] ] gravity : + + + + + + + + + type iib string theory in 10 dimensions has a family of asymptotically @xmath0 solutions that preserve 16 supercharges ; these are the @xmath1-bps states . of particular interest are the subset of states that preserve an additional @xmath8 symmetry . all states in this subset were written down in @xcite and are of the form @xmath9 where the coefficients are given by @xmath10 and are entirely specified by the function @xmath11 . the one - form field @xmath12 can also be given in terms of @xmath13 , but it will not be needed here . note that @xmath14 and @xmath15 have units of length squared . this function in turn solves a harmonic equation in @xmath16 and is determined by its boundary condition on the @xmath17 plane @xmath18 ( we will frequently drop the arguments and simply call this function @xmath13 . ) on this space of solutions it can be shown @xcite that the hamiltonian is @xmath19 the flux of @xmath20 passing through the asymptotic @xmath21 in such spaces is @xmath22 it turns out that the solutions ( [ llmsolution ] ) are singular unless @xmath23 everywhere . when @xmath24 or @xmath25 the solutions have closed timelike curves @xcite , so we will always take @xmath26 . [ [ field - theory ] ] field theory : + + + + + + + + + + + + + + in the dual @xmath27 @xmath3 yang - mills theory on @xmath28 the corresponding states preserve 16 supersymmetries . due to the bps condition these states are constant on the @xmath29 , leading to an @xmath30 symmetry . when one constructs states using only one of the three complex scalar fields , the internal @xmath31 r - symmetry is broken to @xmath30 . thus these states have the same symmetries as the gravity solutions described above . it has been shown @xcite that all of these states can be described in terms of a matrix model for the one homogenous mode of the complex adjoint scalar field : @xmath32 the dictionary relating that yang - mills theory to string theory ensures that @xmath33 where @xmath34 is planck s constant governing the semiclassical limit of ( [ ymaction ] ) and @xmath35 is the 10 dimensional planck length governing the semiclassical limit on the gravity side . the matrix model ( [ ymaction ] ) can be solved in the usual way by going to the eigenvalue basis , in which case it reduces to a system of @xmath4 non - interacting fermions in a harmonic potential . a basis for the hilbert space is specified by a set of non - decreasing integers @xmath36 which measure the excitation of each fermion above the vacuum . the data @xmath37 may be summarized in a young diagram in which the @xmath38 row has length @xmath39 . in the semiclassical limit , it has been proposed that these states are mapped to the dual solutions ( [ llmsolution ] ) with the identification into the definition of @xmath40 in comparison with @xcite in order to maintain consistency with sec . 4.2 of @xcite . ] @xcite @xmath41 where @xmath40 is the semiclassical density of fermions in the single particle harmonic oscillator phase plane @xmath42 . there are many different quantum mechanical density functions on phase space ; however , all well - defined phase space distributions share the same semiclassical limit . thus we will mostly work with the well - known wigner density function ( see appendix a ) and occasionally with the husimi distribution ( see sec . [ husimisec ] ) that is obtained by convolving the wigner distribution with a coherent state @xcite . [ [ integrable - charges ] ] integrable charges : + + + + + + + + + + + + + + + + + + + + in @xcite we pointed out that the half - bps sector of @xmath43 yang - mills theory is integrable and thus has a family of higher hamiltonians @xmath44 where @xmath45 is the hamiltonian in the free fermion hilbert space and @xmath46 is the energy of the @xmath47 fermion in units of @xmath34 . this can be written in terms of a phase space integral , and using the correspondence with spacetime ( [ uwcorrespond ] ) , becomes @xmath48 which is accurate to leading order in @xmath4 . any eigenstate of this tower of conserved charges can be completely identified by measurement of all the eigenvalues . in @xcite we showed that in the semiclassical limit these higher hamiltonians can be read off from the asymptotic angular moments of the spacetime . specifically , the k@xmath49 angular moment of the function @xmath11 that appears in ( [ llmsolution ] ) is proportional to the expected value of @xmath50 : @xmath51 where @xmath52 , @xmath16 is the transverse direction to the @xmath53 plane , @xmath54 is the radial direction in this plane and @xmath55 is the angle between the plane and the transverse direction measured by @xmath56 . is independent of @xmath57 because we are dealing with energy eigenstates . ] nevertheless , it turns out that in typical states with a given total energy ( @xmath58 ) , planck scale measurements are required to distinguish between the microstates . hence information about these states is lost in the semiclassical limit . [ quantum ] the naive correspondence ( [ uwcorrespond ] ) associates a geometry to every state in the hilbert space via the phase space fermion density in the free fermi description of the half - bps states . the correspondence can not be correct in this naive form while a suitably coherent state in the field theory can correspond to a semiclassical geometry , there will be many states which correspond to wavefunctions that have support on spacetimes of widely different geometry and topology . thus it is of interest to identify which field theory states have a description in terms of a single classical geometry vs. a superposition of geometries . in this section , we will define a ` metric ' operator in the yang - mills theory and propose that the eigenstates of this operator are the ones that can be mapped to semiclassical geometries using ( [ uwcorrespond ] ) , while the non - eigenstates can not be associated to a unique metric in this manner . [ superstar - example ] four cft states naively would be mapped to the superstar . ] to illustrate the difficulty with the naive correspondence ( [ uwcorrespond ] ) we will consider different ways in which the extremal half - bps black hole ( the `` superstar '' @xcite ) would be generated by applying ( [ uwcorrespond ] ) to field theory states . it was shown in @xcite that the metric ( [ llmsolution ] ) realizes the superstar geometry when the function @xmath59 ( [ udef ] ) takes a constant value between 0 and 1 within a circular droplet in the @xmath53 plane , and vanishes outside it . this @xmath13 function is depicted in the center of fig . [ superstars ] . it was also shown that a fermion basis state ( [ rdef ] ) described by a triangular young tableau ( upper left corner of fig . [ superstars ] ) has a phase space density of this form , and hence corresponds to a superstar geometry according to ( [ uwcorrespond ] ) . we will now examine other states in the field theory that have the same single particle wigner distribution . for definiteness , we choose , without loss of generality , to work with a droplet with @xmath60 inside the droplet . in terms of the triangular tableau , this corresponds to having an equal number of rows and columns ( @xmath61 ) @xcite , while for the superstar this means that the number of giant gravitons sourcing the geometry is equal to the flux of the five - form field @xcite . the radius of the droplet for this configuration is @xmath62 . and @xmath15 ( and therefore @xmath54 ) are ( length)@xmath63 . ] in terms of excitation numbers the triangular tableau state is given by @xmath64 another way of producing the superstar is to place the @xmath4 fermions in the field theory in coherent states that are randomly distributed in phase space inside the circle of radius @xmath62 ( upper right corner , fig . [ superstars ] ) . since coherent states are minimum uncertainty packets , as we will see in secs . [ coarsegrained ] and [ partition ] , they will cover half the area contained inside the circle . upon coarse graining at the @xmath34 scale , this configuration has a phase space distribution that is @xmath65 inside the circle and @xmath66 outside . upon using the correspondence ( [ uwcorrespond ] ) we once again recover the superstar geometry . thirdly , we can consider a superposition of two very similar triangles , one with even excitations and one with odd excitations ( fig . [ superstars ] ; bottom right ) : @xmath67 and finally , we can consider the superposition of the empty tableau corresponding to the @xmath68 vacuum , and a square tableau of size @xmath69 ( fig . [ superstars ] ; bottom left ) : @xmath70 the last two superpositions are also depicted in figure [ superstars ] . computing the corresponding phase space distributions ( following @xcite or ( [ eq : wigner ] ) later in this paper ) one can see that the phase space distributions corresponding to these states will indeed reproduce the superstar geometry upon coarse graining . the last two cases are both simple superpositions of basis states . however , the ` square + square ' example is more dramatic as it is a superposition of states of vastly different energies , unlike ` triangle + triangle ' . the energy associated to the @xmath68 vacuum is @xmath71 , while the energy of the @xmath72 tableau is @xmath73 for @xmath61 , yielding an average energy of @xmath74 . as soon as an observer performs even a rough measurement of the energy , the universe is projected into one of the two energy eigenstates . on the gravity side , the corresponding geometries differ from each other ( and from the superstar ! ) at scales that are easily measurable , since the relevant scale is roughly the @xmath68 radius @xmath75 . thus , ` square + square ' does not describe anything like a classical superstar , and can only be described as a superposition of distinct geometries . we wish to develop a formalism that can distinguish states of this sort from the ones that can be usefully mapped to a unique metric . it will be convenient to use a second quantized description of the fermion system . for each state @xmath76 in the 1-particle hilbert space @xmath77 , we define both a fermion creation operator @xmath78 and a fermion annihilation operator @xmath79 which satisfy @xmath80 where the inner product on the right is taken in @xmath77 . since we deal with free fermions , the anti - commutators on the left - hand side of equations ( [ acom ] ) are proportional to the identity . thus it makes sense to set them equal to the complex numbers on the right - hand side . we will in particular be interested in the case where @xmath76 is a coherent state labelled by a parameter @xmath81 . in phase space these states manifest themselves as gaussian wavepackets localized around @xmath82 , and are defined by @xmath83 throughout this section , the normalization conventions are as in @xcite . we shall abuse notation somewhat by taking @xmath84 to represent the creation and annihilation operators corresponding to the coherent state determined by @xmath81 . the simple form and well - known properties of the coherent state ( [ eq : coherent ] ) provide a convenient foundation upon which to base a semiclassical formalism . we make use of this opportunity in the rest of this work . however , in section [ discussion ] , we describe certain desirable improvements which , unfortunately , will come with increased technical complications . we wish to define a ` metric ' operator in the field theory with the property that its eigenstates are well described by a single classical metric which is in turn given by the eigenvalue of the operator . since any 1/2-bps metric may be expressed in terms of the function @xmath59 , it is enough to define an operator @xmath85 in terms of the fermion operators ; we can take all other metric operators to be given by the classical llm formulae with @xmath13 replaced by @xmath86 . we will see that , with our definition of @xmath86 below , such expressions will require no regularization . for @xmath87 , we make the definition @xmath88 and @xmath89 . thus , for each @xmath90 the operator @xmath91 is a fermion number operator , having eigenvalues 0 and 1 . our definition satisfies a number of useful properties . for example , we have @xmath92 these are precisely the relations which hold between the corresponding classical quantities . furthermore , one may show that these @xmath93 generate a complete set of operators which do not change fermion number . here the essential point is that any operator on the 1-particle hilbert space can be expressed in the form @xmath94 denotes the @xmath95 energy level of the 1-particle harmonic oscillator , we may define a complex function @xmath96 by @xmath97 it is then straightforward to show that @xmath98 it is clear that any operator which preserves particle number can be built by taking sums of products of @xmath99 , and thus that any such operator can be built from @xmath93 . we now wish to connect the operator @xmath100 to a semi - classical phase space distribution function by showing that the expectation value of @xmath100 in a general state @xmath101 is the husimi distribution on phase space : @xmath102 where @xmath103 is the husimi distribution corresponding to the one particle density matrix @xmath104 associated to @xmath101 . to show this , we write the general state as a superposition of basis states as @xmath105 where @xmath106 sums over the basis states in the superposition , the normalization is @xmath107 and @xmath108 is the set of excitation numbers characterising a basis state @xmath109 . the basis states in turn are given by @xmath110 using the commutators ( [ acom ] ) one can show that @xmath111 the expectation value on the right side of the first line can clearly be nonzero only when at least @xmath112 of the excitation numbers in @xmath108 and @xmath113 are the same . this allowed us to simplify the expression at the cost of introducing some new notation . in the above , @xmath114 denotes the set of pairs of states that differ from each other by one excitation , and for each such pair we define the excitation to be @xmath115 . finally , @xmath116 indexes the number of the differing excitation , i.e. if @xmath117 , then @xmath118 . next we need to evaluate the right hand side of ( [ husimi ] ) , and thus we recall the definition of the husimi distribution from @xcite : @xmath119 the one particle density matrix corresponding to @xmath120 can be easily found by tracing over @xmath112 of the particles . due to the orthogonality , the trace can also be nonzero only if at least @xmath121 of the excitations are the same , and after a little work we get @xmath122 ) . [ coarsegrained ] the metric operator @xmath100 is not well suited to a semiclassical observer , since it is very sensitive to details at the planck scale . thus , we wish to coarse grain it over some distance scale @xmath123 . to do this , we ll compute a convolution with a gaussian kernel : @xmath124 since this operator is essentially a weighted average of local values of @xmath100 , it will have a continuous spectrum of eigenvalues between 0 and 1 and it is not sensitive to details at the planck scale . this operator valued function on the single fermion phase space is our proposal for a `` metric operator '' , namely the operator whose eigenstates can be associated to semiclassical geometries using the llm prescription . to make this more precise , we next need to define exactly what me mean by eigenstates and the semiclassical limit . the semiclassical limit can be defined simply as @xmath125 we can see that the commutator @xmath126 $ ] does not vanish , though it approaches zero rapidly when @xmath90 and @xmath127 are separated by more than a distance @xmath123 . as a result , only states which are eigenstates of the @xmath128 with eigenvalue @xmath129 or @xmath130 for all @xmath90 ( i.e. states with an empty or completely filled phase plane ) can be exact eigenstates of @xmath131 . neither of these possibilities is of interest . other states are only approximate eigenstates . for example , consider empty ads space , the vacuum of the theory . in terms of young tableaux for fermion excitation energies , ads space is described as the empty tableau and the correspondence phase space density is the filled fermi sea , i.e. a filled disk of fermions in phase space . for @xmath132 well within the filled disk , the state is an approximate eigenstate of @xmath133 with eigenvalue one . when @xmath132 is well outside the black disk , the state is an approximate eigenstate of @xmath133 with eigenvalue zero . however , when @xmath90 is within @xmath123 of the boundary , the state is far from being an eigenstate of @xmath133 . thus we will define approximate eigenstates as follows : we say that a state @xmath134 is an approximate eigenstate of @xmath133 with eigenvalue function @xmath135 and a given accuracy @xmath136 , if and only if @xmath137 with this definition we can easily see that of the four aspiring superstars introduced earlier , only ` square + square ' is not an approximate eigenstate of @xmath131 thus should nt be associated with the superstar geometry . armed with the definition of eigenstates of the coarse - grained @xmath138 , such eigenstates can be associated with a slowly - varying density function which we have called @xmath135 , and it is in terms of this function that we shall now discuss the thermodynamics and entropy of these solutions . following sec . [ 2nd ] all eigenstates of the metric operator can be constructed by placing each fermion in the free - fermi description of half - bps states in a coherent state in phase space . it is known that coherent states placed on an @xmath34-spaced lattice in phase space also provide a complete basis for the wigner distributions in the single - particle phase space @xcite , and hence , by ( [ uwcorrespond ] ) for all semiclassical geometries . because of the coarse - graining , many different quantum states may have the same geometrical description . thus some geometries should be associated to an entropy , whose exponential counts the number of underlying states with the same geometric description . in this section we want to find a formula for the entropy of half - bps spacetimes and then use it to discuss the partition sum over these geometries . with this motivation we will construct the wigner distribution of the coherent state basis for half - bps states and coarse - grain that to find our formula for the entropy of half - bps semiclassical spacetimes . the semiclassical limit of half - bps states in yang - mills theory is most conveniently studied in terms of coherent states ( [ eq : coherent ] ) , rather than the energy eigenstates described above . each of the @xmath4 fermions in the harmonic potential can be placed separately in a coherent state , with the fermionic statistics imposed by a slater determinant . consider an @xmath4-particle ( unnormalized ) wavefunction given by antisymmetrizing 1-particle wavefunctions @xmath139 , @xmath140 : @xmath141 = \sum_{\sigma \in s_n } ( -)^\sigma \prod_i^n \psi_{\sigma(i)}(x_i ) \label{slater}.\ ] ] the 1-particle - reduced density matrix of this state is @xmath142 in terms of the quantities @xmath143 the wigner phase space distribution of @xmath120 is given by @xmath144 where @xmath145 . finally , if we define the overlap matrix @xmath146 then the 1-particle reduced density matrix takes the simple form @xmath147 where @xmath148 goes over @xmath149 subject to @xmath150 . we are interested in the case where each @xmath151 is a coherent state , namely @xmath152 . then , applying equations ( [ overlap]-[cohdefrho2 ] ) , we get : @xmath153 with @xmath154 according to the semiclassical holographic map ( [ wignercoh ] ) is identified with the function @xmath13 in spacetime as in ( [ uwcorrespond ] ) . in the semiclassical limit , the phase space distribution of fermions can be described as a droplet or set of droplets in the phase plane . coherent states were constructed to be minimum uncertainty droplets . they occupy an area of @xmath155 , the smallest possible quantum of phase space . this agrees with ( [ ndef ] ) and can be verified by close packing coherent states to form a fermi sea . in our conventions , @xmath4 closely packed coherent states form a fermi sea of area @xmath156 , covered with a wigner distribution which fluctuates around a mean value @xmath157 , consistent with ( [ uwcorrespond ] ) . in a similar way a length @xmath158 is a minimal allowed separation between coherent states . attempts to force the states closer than this distance cause them to delocalize into rings as illustrated in the example in fig . 2 . the observations above lead to the conclusion that for a semiclassical observer , it makes sense to think of states as inhabiting a lattice of unit cell area @xmath155 @xcite . in fact a semiclassical observer measures the phase plane at an area scale @xmath159 . at this scale , the observer is only sensitive to a smooth , coarse grained wigner distribution @xmath160 which erases many details of the precise underlying precise microstates . it lies between 0 and @xmath161 @xcite . ] we may view the region @xmath162 as consisting of @xmath163 lattice sites , a fraction @xmath164 of which are occupied by coherent states . then the entropy of the local region @xmath165 is @xmath166 the stirling approximation used in ( [ localent ] ) is valid when @xmath167 is reasonably far from 0 and 1 . for the total entropy this gives @xmath168 it is beautiful that thinking about @xmath164 as the probability of occupation of a site by a coherent state , this is simply shannon s formula for information in a probability distribution . these facts imply that in the semiclassical limit the function @xmath59 which completely defines a classical solution should effectively be defined on a lattice with each plaquette of area @xmath169 , and take values of 0 or 1 in each site . likewise ( [ entropyequation ] ) will be interpreted as an expression for the entropy of arbitrary half - bps asymptotically @xmath170 spacetimes . as an example it exactly reproduces the formula for the entropy of the typical states described in @xcite that correspond in spacetime to the `` superstar '' geometry @xcite . note that the entropy vanishes if and only if @xmath13 equals @xmath129 or @xmath130 everywhere . following the correspondence ( [ uwcorrespond ] ) such states map into geometries that are non - singular . we learn that semiclassical half - bps geometries that are smooth all have vanishing entropy ; and the presence of singularities @xmath171 also implies that the spacetime carries an entropy . thus , in this setting , entropy is a measure of ignorance of a part of the underlying state which is captured in classical gravity as a spacetime singularity . in the semiclassical limit the partition function over the half - bps sector of iib string theory ( @xmath172 ) with asymptotically @xmath0 boundary conditions can be written as @xmath173 we are able to write this as a functional integral over just @xmath13 because the entire classical solution can be derived from this function , as can the hamiltonian @xmath174 and the number of units of 5-form flux @xmath4 . the measure @xmath175 reflects not only the jacobian in transforming between the supergravity fields and @xmath13 , but also the number of underlying microscopic configurations that give rise to the same macroscopic spacetime . to derive the semiclassical measure @xmath176 recall first that the classical metric is a useful concept at scales @xmath177 and the semiclassical limit can be defined as ( [ semilimit ] ) . in this limit the smooth function @xmath59 arises by considering the limit of piecewise constant functions in the lattice of scale @xmath123 . to derive the continuum measure @xmath175 we first ask how many microscopic configurations can give rise to a given average value of @xmath13 in each @xmath178 lattice cell . from the previous section we have learned that we can think of @xmath13 as the average number of planck size cells that are populated by coherent states within an area of size @xmath178 . following ( [ localent ] ) this means that a coarse - grained plaquette of size @xmath123 with a given value of @xmath13 arises from @xmath179 underlying configurations . many of these configurations , representing different occupation numbers of planck cells , will have different energies because the energy contribution of a given planck cell depends upon its location as @xmath180 ( [ hdef ] ) . however , because @xmath181 in the semiclassical limit , configurations that populate a fixed fraction of planck sized cells will necessarily have very similar energies . in fact it can be shown that the energies are even similar in planck units in the strict semiclassical limit ( [ semilimit ] ) . this is because as @xmath182 , the number of planck cells in each plaquette is extremely large . one can use this to show that the energies of configurations that populate a fixed fraction @xmath13 of the planck cells have a standard deviation to mean ratio that vanishes as @xmath182 . putting everything together , the measure on semiclassical half - bps spacetimes is @xmath183 where @xmath184 is understood as the entropy of the spacetime . in the semiclassical limit , a spacetime is nonsingular if @xmath185 everywhere . in that case , @xmath186 and the measure @xmath176 is 1 . in other words , semiclassical half - bps spacetimes have an entropy if and only if they are singular . defining @xmath187 the partition function over semiclassical half - bps spacetimes becomes @xmath188 } \ ] ] evaluating this by the method of saddlepoints gives @xmath189 where @xmath190 and @xmath191 is a fermi - dirac function . it is worth emphasizing that we have summed over both singular and non - singular semiclassical spacetimes , but have included the correct degeneracy factor ( [ measure ] ) . a naive approach to summing over geometries would have failed to include this measure because there is no semiclassical horizon giving arise to a bekenstein - hawking entropy . in such a a naive approach the partition function would have been @xmath192 this integral diverges at the upper limit . a naive approach taking the unit measure over geometries would only reproduce ( [ partfn ] ) if we restricted the partition sum to be over smooth geometries ( @xmath185 ) with @xmath13 taking constant values within elementary cells at the planck scale . this would mimic in geometry the coherent state analysis given above , but its validity is more doubtful because at the planck scale the wavefunction over geometries is always relevant . the semiclassical partition function should reproduce the partition function obtained from first principles by coarse - graining the scale of the fundamental cells in the quantum mechanical phase space . this procedure is not ambiguous because the free fermion gauge theory description provides us with an honest quantum mechanical phase space . coarse - graining is defined as a renormalization group transformation in this space , and it is the 1/2 bps nature of the quantum states that allows us to trust this description as the value of the coupling is turned on to achieve a reliable gravitational description . consider then a lattice whose cells are @xmath193 ( in planck units ) . from the microscopic point of view , the energy of each distribution of populated planck scale cells is different , but as we argued before , in the limit @xmath194 , almost all distributions cluster close to a certain typical distribution in the @xmath195 cell , and thus observers at these scales will assign the same energy to all of them . in this case , the coarse - grained function @xmath13 will take values @xmath196 in the @xmath195 cells . this can also be seen from the flux quantization which is scale independent . in other words , discretising phase space and comparing two lattices of sizes @xmath193 and @xmath197 , in planck units , we find that @xmath198 where variables with superscript @xmath199 are defined in the @xmath200 lattice and in the second equality we are summing over all planck - scale lattice sites inside a single @xmath193 cell labelled by @xmath201 . this sum computes the fraction of populated sites in the coarse - grained cell . finally , the sum over @xmath202 at cell location @xmath203 becomes @xmath204 the factors in front of each exponential count how many ways a given value of @xmath202 in the coarse - grained lattice can be attained in terms of the planck scale lattice . the complete partition function becomes @xmath205 thus , coarse - graining the phase space lattice size rescales the temperature , i.e. @xmath206 . this derivation reproduces the semiclassical computation if we identify the temperature @xmath127 in as the rescaled one . we could view the computation as a derivation for the entropy formula . in the previous section we computed the partition function over the @xmath1-bps sector , fixing the energy @xmath207 and the five - form flux @xmath4 using lagrange multipliers @xmath208 . in this section , we shall analyse other interesting ensembles in which different charges are fixed in the same way . the typical states of fixed energy have a spectrum of the charges @xmath50 defined in ( [ mk ] ) that is fixed by the universal @xmath13 function that describes their classical limit @xcite . however , it is clear that classical spacetimes with different angular moments are easily constructed by varying the radial dependence of the @xmath13 function . hence it is of interest to ask what spacetime gives the universal classical description of the typical state with fixed moments @xmath50 . we can also use the measure ( [ measure ] ) on half - bps spacetimes to find out how many classically indistinguishable microstates are described by the same metric . we also see that classical spacetimes with atypical macroscopic moments are exponentially unlikely in the space of all states . a natural way to identify typical states carrying fixed values of these charges is to analyze ensembles of states that generalize the grand - canonical one by introducing as many chemical potentials @xmath209 as conserved charges @xmath50 we want to fix . as usual , we will fix the values of these charges by requiring the expected values to match our desired value , a step that determines the set of chemical potentials . while it is technically difficult to achieve this description in the basis of states provided by the young tableaux , it is very natural in the coherent state basis . indeed , the semiclassical expressions are linear in the phase space density @xmath210 , and so they are straightforward to integrate in the path integral that defines the semiclassical partition function : @xmath211 above we used @xmath212 and @xmath213 and relabelled their chemical potentials to reduce the notation . using the measure and evaluating the partition function by the method of saddlepoints , its logarithm becomes @xmath214 which is a natural extension of the usual fermi dirac statistics . the value of the phase space density @xmath13 is fixed by the saddlepoint : @xmath215 in this ensemble it is trivial to find an expression for the expected moments ; it is simply @xmath216 the eigenstates of the @xmath50 ( [ mk ] ) that were considered above have phase space densities that are rotationally invariant in the phase space and thus correspond to @xmath13 functions that are rotationally invariant in the @xmath217 plane . in particular , any half - bps state whose single particle phase space density is not rotationally invariant can not be an eigenstate of the hamiltonian . density functions @xmath13 with angular dependence can be efficiently parameterized by expanding them in a fourier series : @xmath218 to work out the kinds of states which give rise to @xmath13 functions with @xmath219 we work with the young tableaux basis of states , and consider superpositions of the form : @xmath220 above , @xmath106 indexes the basis states in the superposition , each given as a slater determinant of single particle states as @xmath221 , and @xmath108 is the set of excitation numbers characterizing the basis state @xmath222 . the normalization is such that @xmath223 . the states are @xmath4 particle states , whereas the phase space density @xmath224 which is mapped onto the spacetime data @xmath13 ( [ uwcorrespond ] ) is a one particle density . the method for calculating the one - particle wigner density function is given in @xcite ( also see appendix a ) . using this , it is easy to show that because the wavefunctions @xmath225 are orthogonal , the only interference terms between various pieces of the superposition ( [ eq : state ] ) that survive involve states @xmath106 and @xmath226 having @xmath112 of the fermion excitations in @xmath108 and @xmath113 equal . let us denote by @xmath114 the set of pairs of states of this kind . for each pair define the differing excitation @xmath227 by @xmath228 . with this definition we can now compute the one particle wigner distribution corresponding to ( [ eq : state ] ) @xmath229 the shorthand notation @xmath230 and @xmath231 was used above for the sake of clarity . we also defined @xmath232 to index the number of the differing excitation , i.e. if @xmath233 , then @xmath234 . . illustrates how interference between young tableau basis states in the superposition ( [ eq : state ] ) is responsible for breaking the @xmath235 invariance in the @xmath17 plane , i.e. all @xmath57 dependence in @xmath236 is due to interference between basis states . to quantify this , we match the fourier components of ( [ fourier ] ) with those of ( [ eq : wigner ] ) giving @xmath237 this shows that we have a simple relation between the fourier modes of @xmath13 and the states in the superposition : _ a given mode @xmath238 arises from interference between basis states that share @xmath121 of their excitations , but differ by @xmath239 units in the remaining excitation_. thus the energy difference between the two states is @xmath240 . while the above shows that breaking of @xmath241 invariance in the @xmath17 plane necessarily arises from interference in a young tableau basis , we should also recall that the geometric description is only valid if we have an approximate eigenstate of the metric operator . we have seen in sec . 4 that any given wigner distribution can be constructed out of coherent states for the fermions and that such a state will be a eigenstate of the metric operator . such states can also be written as complicated superpositions in a young tableau basis . we would like to identify what typical states correspond to which violations , and count how many of them there are . this is again difficult to achieve in the young tableaux basis , but it is straightforward in the coherent state basis by introducing the appropriate lagrange multipliers fixing the appropriate charges . to quantify the angular moments , define the quantities @xmath242 with a similar expression for @xmath243 . note that for @xmath244 the corresponding quantity is just the flux of the five - form field , @xmath245 . with this definition we can construct the ensemble by introducing lagrange multipliers @xmath246 and @xmath247 as @xmath248 from which we can again compute the partition function to be @xmath249 the logarithm can again be computed @xmath250 thus we see that fixing the fourier modes @xmath251 is very natural in this language , though the integrals are again difficult to compute and we are no longer working with fermi dirac statistics . [ [ exponential - suppression - of - moments ] ] exponential suppression of moments : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we shall now show that the number of states with non - zero moments @xmath252 is exponentially smaller than the number of states with @xmath253 . to show this , we shall fix one of the moments , chosen to be @xmath254 , and consider the density matrix of this ensemble : @xmath255 . from this it follows that the entropy is @xmath256 where the partition funtion is given by ( [ eq : z - moments ] ) , and @xmath257 and @xmath258 are determined in the standard way by differentiating the partition function with respect to the lagrange multipliers : @xmath259 analyzing these relations exactly is complicated and we need to specify a regime of parameters in which to work . to produce macroscopic angular moment , we choose to work with states of high energy or equivalently with a low density of fermions . this corresponds to working in the limit @xmath260 , such that @xmath261 . in this limit the fermi dirac functions can be approximated by @xmath262 , and the relations ( [ eq : thermo ] ) simplify considerably . the remaining integrals in ( [ eq : thermo ] ) can be computed using the calculus of residues , and after a short computation they yield @xmath263 to invert these relations let us choose the lagrange multiplier @xmath90 to be very small , which will correspond to choosing the moment to be small , but still macroscopic and measurable . truncating the sums to their first two terms , we can invert the relations above to give @xmath264 & \equiv & -xn \ln [ x(1+a^2 ) ] , \nonumber \\ \alpha = 2\frac{\mathcal{u}}{n } & \equiv & 2a,\end{aligned}\ ] ] where we have defined the scalings @xmath265 and @xmath266 . note that the assumption of high energy requires large @xmath267 , which immediately results in @xmath268 , which is consistent with our assumption . also , the angular moment is still measurable , since it scales linearly with @xmath4 , with @xmath90 small but fixed . inserting these relations into ( [ entropy - with - violation ] ) we get @xmath269 which shows that the number of states with macroscopic moments of size @xmath270 is exponentially suppressed from the rotationally invariant case : @xmath271 [ [ variances ] ] variances : + + + + + + + + + + + since fixing the macroscopic moment exponentially reduces the number of available states , it is necessary to check that the fluctuations in the previous computation are not so large as to render the computation invalid . the spread in the expectation value of @xmath258 is given by the standard deviation to mean ratio , which can be computed to give @xmath272 where the computation was carried out in the previous high energy regime , using the relations ( [ eq : thermo - sum ] ) . this vanishes as @xmath273 , and therefore this ensemble is statistically valid . [ [ measurability - of - interference ] ] measurability of interference : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + just as eigenstates of the hamiltonian generate non - trivial higher conserved charges at infinity , we can ask whether the presence of angular moments in the @xmath274 plane , arising from interference between young tableau components in the superposition ( [ eq : state ] ) , has any characteristic effect asymptotically . in @xcite we analysed how the integrable charges @xmath50 associated to young tableau states appear at different orders in an asymptotic expansion of @xmath13 , the function determining the metric . for the question at hand the same method turns out to be very cumbersome , as the young tableaux basis is not a convenient one to use when analysing the moments in the @xmath275 plane . for completeness , we still include the asymptotic expansion analogous to ( [ eq : wignerlimit ] ) for the superposition state ( [ eq : state ] ) : @xmath276 + 1}^l ( -1)^i ( i+1 ) \sin^{2i - l } \theta \sum_{p=0}^ { [ i-\frac{l+1}{2 } ] } 2^{\frac{l}{2}+p - i } \nonumber \\ & & \times \binom{i}{2i - l } \binom{2i - l}{p } \sum_{\substack{(w , w')\in \mathcal{c } \\ n - m=2(i - p)-l } } ( -1)^{n_m+n_n } \sqrt{\frac{n!}{m ! } } \textrm{re } ( c_m^ * c_n e^{i(n - m)\phi } ) \nonumber \\ & & \times \left(\frac{d}{da}\right)^{l - i+p } \left [ a^{i - p } ( a-1)^m \right]_{a=2}. \label{horror}\end{aligned}\ ] ] the first row of this expansion is a weighted sum of the single state contributions of ( [ eq : wignerlimit ] ) , while the last three rows arise from interference between the states in the superposition . as one might imagine , analysing interference directly from this expression would be very complicated , but we want to point out that all odd powers of @xmath277 in the expansion come from the interference terms . thus , if an asymptotic observer measures such a moment , she could infere that the underlying quantum state _ necessarily _ was in a linear superposition of young tableaux . to examine whether these moments are semiclassically measurable , it is more convenient to work with the coherent state picture of section ( [ sec : coherent ] ) . using coherent states we can engineer any moments @xmath278 we wish , by distributing the fermions in a suitable way . as an example , let us pick a very specific configuration with exactly one @xmath241 violating moment turned on : @xmath279 where @xmath280 is the heaviside step function , @xmath281 is a parameter controlling the size of @xmath241-violation and @xmath282 is the ads radius . in the argument of the step function is necessary to ensure the correct normalization : @xmath283 . ] one sees that for this distibution @xmath284 , and therefore for small @xmath281 the state is of the form analyzed earlier in this section . an asymptotic expansion of the function @xmath285 corresponding to this fermion distribution can be computed via the technique used in @xcite , and the first non - trivial terms turn out to be @xmath286 the @xmath287 dependence of the first odd @xmath56 negative power illustrates the finiteness of this moment in the semiclassical limit : @xmath288 thus , the interference between the young tableau components of some states is measurable by an asymptotic observer , even though for a typical state these effects are highly suppressed as shown in ( [ expsmall ] ) . [ infloss ] our work above has emphasized that individual quantum states do not generically correspond to individual geometries . in particular , we have shown how a geometry arising from a specified function @xmath59 as in ( [ udef ] ) can be the effective coarse - grained description of many quantum states . as noted in section [ coarsegrained ] , the function @xmath13 determines only the expectation value of the ( coarse - grained ) operator @xmath138 . furthermore , this function encodes only information present in the one - particle phase space density @xcite derived by projection of the full quantum state . below , we first characterize more precisely when this projection loses information . it will turn out that in the semiclassical limit , the 1-particle projection loses information if and only if the description as an llm geometry is singular . we then characterize the operators which probe _ beyond _ this one - particle projection from several points of view . ( see @xcite for a discussion of related issues . ) [ sec : density ] in sec . [ superstar - example ] we gave four examples of states that give rise to single particle phase space distributions that reproduce the metric of the ` superstar ' extremal black hole . two of these states , square + square " ( [ squaresquare ] ) and triangle + triangle " ( [ triangletriangle ] ) , have identical single particle distributions even though they are very different as @xmath4-particle states . in this case , the `` square + square '' state does not have a description as a single geometry since it is not an approximate eigenstate of the metric operator . but it is easy to devise examples of multiple states , all of which have good descriptions as a single coarse - grained geometry , and which have identical single particle phase space distributions . the projection to the single particle phase space has lost information about the underlying @xmath4-particle configurations . to see when this situation may occur , suppose that the support of the function @xmath13 on the @xmath53 plane has an area @xmath290 . as in section [ sec : coherent ] , we envision the fermions as coherent states inhabiting a lattice , and identify the support of @xmath13 with lattice sites @xmath291 . also , because there are @xmath4 fermions , we must have @xmath292 thus , the wavefunction is a sum of @xmath4-particle components , each of which is of the form @xmath293 , where @xmath294 . there are @xmath295 such @xmath4-particle components , so the superposition coefficients required to specify the general state involve @xmath296 real numbers ( where we subtract @xmath297 to account for the normalization and the unphysical overall phase ) . on the other hand , since the wigner distribution is nonzero at @xmath298 lattice sites , it contains @xmath299 real numbers ( after accounting for the normalization ( [ norm ] ) ) . , and , up to @xmath34 corrections , is diagonal for coherent states based on lattice sites that give rise to approximate metric eigenstates . thus it also only contains @xmath299 real numbers in the semiclassical limit . ] therefore , when @xmath300 , the 1-particle projection simply does not contain enough data to determine the underlying @xmath4-particle state . in any situation where the effective phase space distribution has support on more than @xmath4 lattice sites , the normalization ( [ norm ] ) will force the distribution to take values that are neither @xmath129 nor @xmath130 in at least some locations . in that case , following the correspondence ( [ uwcorrespond ] ) , the spacetime geometry is necessarily singular . since the density takes values other than @xmath129 or @xmath130 , we see that the fine - grained metric also fluctuates ; our state is not an approximate eigenstate of the fine - grained @xmath86 . in this sense then , singularities in the half - bps sector of @xmath301 gravity arise precisely when a single classical geometry is unable to encode some data concerning the underlying quantum mechanical configuration space . this is satisfactory because any attempt to do physics in these singular spacetimes forces us to specify boundary conditions on the singularity . as mentioned in @xcite , it would be very interesting to understand how these boundary conditions emerge from the effect of coarse - graining of wave functions applied to correlation functions in the gauge theory . [ [ second - quantized - formalism ] ] second - quantized formalism : + + + + + + + + + + + + + + + + + + + + + + + + + + + + the second - quantized formalism of section [ 2nd ] makes it particularly easy to exhibit operators sensitive to more than 1-particle structure . we simply string together creation and annihilation operators corresponding to different sites on the llm plane . one example is @xmath302 where @xmath303 stands for the second - quantized operator that destroys a particle in the @xmath304 level in the energy eigenbasis . then : @xmath305 where the two states above are specified in ( [ squaresquare],[triangletriangle ] ) . here the ` square + square ' state does not have a good description as a single geometry , but we could have equally given an example where the @xmath4-particle distinguished two states that have both have good descriptions in an identical geometry . [ [ wigners - formalism ] ] wigner s formalism : + + + + + + + + + + + + + + + + + + + + recall that the expectation value of a 1-particle operator @xmath306 in a state represented by the 1-particle reduced wigner distribution @xmath307 is @xmath308 where @xmath309 is the classical phase space distribution corresponding to @xmath306 . the formalism of wigner distributions extends to calculating expectation values of operators sensitive to multi - particle structure . the @xmath4-particle wigner distribution is given by : @xmath310 from this , the 1-particle reduced wigner distribution is recovered by tracing over @xmath112 particles : @xmath311 in analogy to the 1-particle case , expectation values of operators in states characterized by @xmath312 are given by : @xmath313 note that @xmath314 is necessarily symmetric in the @xmath315 s . from the above , it is clear that operators admitting a representation @xmath316 can be completely evaluated in the 1-particle reduced wigner distribution : @xmath317 on the other hand , any operator whose classical density is not of the form @xmath318 ( i.e. contains a monomial in @xmath90 s of mixed indices ) will generically probe the @xmath4-particle structure . the space of 1-particle operators with non - vanishing expectation values admits a basis given by the moments @xmath319 ( [ higherham ] ) with classical phase space densities @xmath320 this is intuitive as the knowledge of all the @xmath319 s is in principle necessary and sufficient for specifying @xmath321 . any operator which is not a linear combination of the @xmath319 s will in general probe the @xmath4-particle information . [ [ n - particle - operators - in - full - adscft ] ] @xmath4-particle operators in full ads / cft : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the recognition of 1-particle operators as the higher hamiltonians of the fermionic system afford a clean characterization of operators probing the @xmath4-particle structure in the full cft . in particular , any operator which may not be written as a linear combination of @xmath322 will belong to this family . @xmath323-charge conservation and gauge invariance translates this requirement into the following : operators probing the @xmath4-particle structure are multi - trace operators . not surprisingly , multi - trace operators correspond to multi - particle states on the gravity side @xcite . they have been investigated in @xcite . a simple example of such an operator is @xmath324 where the second term removes the 1-particle component . its classical phase space distribution is given by @xmath325 where @xmath326 denotes complete symmetrization over @xmath327 . [ discussion ] our definition of the metric operator @xmath85 was based on the use of standard coherent states , having equal dispersions in @xmath14 and @xmath15 . similarly , our notion of coarse - graining was based on gaussian smearing kernels having equal @xmath14- and @xmath15-dispersions . this choice was made for the sake of simplicity and , furthermore , for an observer who makes sufficiently rough measurements ( i.e. , who is sufficiently semi - classical ) of a fixed state , the detailed shape of @xmath34-sized phase space cells is irrelevant . however , consider a given `` semi - classical '' observer who is able to measure energies with some fixed accuracy @xmath328 . suppose that this observer studies one of the above coherent states centered on a point with @xmath329 . this state has an energy uncertainty @xmath330 . clearly for large @xmath54 we have @xmath331 and the observer s measurements will cause this state to quickly decohere . such measurements are not well - described by our formalism above . instead , such observers require an analogous formalism based on _ squeezed _ coherent states , where the squeezing reduces the radial dispersion to achieve @xmath332 at the expense of increasing the angular dispersion to the level @xmath333 , such that @xmath334 . they also require a correspondingly squeezed notion of coarse graining . it would be very interesting to study such a squeezed formalism in detail , but we have chosen the simpler ( unsqueezed ) formalism for our work above . it would also be interesting to analyze bulk semi - classical observers from first principles to verify that fixed energy resolution @xmath328 ( as opposed to a @xmath328 which changes with @xmath207 ) is in fact an appropriate description of their measurements . we thank jan de boer , tamaz brelidze , veronika hubeny , vishnu jejjala , thomas levi , rob myers , mukund rangamani , simon ross , slava rychkov and jung - tay yee for useful discussions . , v.b . and j.s . were supported in part by the doe under grant de - fg02 - 95er40893 . k.l . was supported in part by the netter fellowship from the university of pennsylvania , and by a fellowship from the academy of finland . b.c . was supported in part by a dissertation completion fellowship from the university of pennsylvania . thanks the theoretical physics group at berkeley and the organizers of the sowers workshop in theoretical physics at virigina tech for hospitality while this paper was finished . d.m . was supported in part by the national science foundation under grant no phy03 - 54978 , and by funds from the university of california . j.s . was supported in part by doe grant de - ac03 - 76sf00098 and nsf grant phy-0098840 . in this appendix we present the computation of the wigner distribution for a general state in the @xmath4-particle harmonic oscillator , deriving ( [ eq : wigner ] ) . a general state can be written as a superposition of basis states as ( [ eq : state ] ) , and the single particle wavefunctions appearing in the slater determinant are given by @xmath335 where @xmath336 is a hermite polynomial . we then compute the effective one particle density matrix as @xmath337 before explaining the notation above , some comments are in order . the right hand side of ( [ densitymatrix ] ) consists of two parts . the ` diagonal ' terms on the upper line are contributions of single basis states , as already computed in @xcite , only now weighted by @xmath338 . the second part , the ` cross terms ' , represent interference between basis states . due to the orthogonality of the 1-particle wavefunctions ( [ 1partpsi ] ) it is clear that a cross term between two basis states appears if @xmath121 of the excitations in these states are the same , i.e. the sets @xmath339 and @xmath340 have @xmath121 equal elements , while the remaining elements differ . note that when this happens , the crossed basis states necessarily have different energies and the superposition state is prohibited from being a hamiltonian eigenstate . the notation is as follows : let @xmath114 denote the set of pairs that give rise to interference terms as explained above . then for each excitation we can define the differing excitation @xmath227 as @xmath341 . we also define the index of the differing excitation such that if @xmath342 , then @xmath343 . the wigner distribution corresponding to a density matrix @xmath104 is defined as @xmath344 ) , ( [ densitymatrix ] ) and the identity @xmath345 it is now straightforward to verify ( [ eq : wigner ] ) . above the @xmath346 are associated laguerre polynomials . v. balasubramanian , m. berkooz , a. naqvi and m. j. strassler , `` giant gravitons in conformal field theory , '' jhep * 0204 * , 034 ( 2002 ) [ arxiv : hep - th/0107119 ] . s. corley , a. jevicki and s. ramgoolam , `` exact correlators of giant gravitons from dual n = 4 sym theory , '' adv . * 5 * , 809 ( 2002 ) [ arxiv : hep - th/0111222 ] . l. f. alday , j. de boer and i. messamah , `` the gravitational description of coarse grained microstates , '' jhep * 0612 * , 063 ( 2006 ) [ arxiv : hep - th/0607222 ] . k. larjo , `` on the existence of supergravity duals to d1d5 cft states , '' arxiv : hep - th/0705.4433 . l. derrico , w. muck and r. pettorino , `` stretched horizon and entropy of superstars , '' arxiv : hep - th/0703223 . shieh , g. van anders , m. van raamsdonk , `` coarse - graining the lin - maldacena geometries , '' arxiv : hep - th/0705.4308 . g. milanesi and m. oloughlin , `` singularities and closed time - like curves in type iib 1/2 bps geometries , '' jhep * 0509 * , 008 ( 2005 ) [ arxiv : hep - th/0507056 ] . m. hillery , r.f . oconnell , m.o . scully and e.p . wigner , `` distribution functions in physics : fundamentals '' , physics reports 106 , no . 3 , ( 1984 ) 121 - 167 . r. c. myers and o. tafjord , `` superstars and giant gravitons , '' jhep * 0111 * , 009 ( 2001 ) [ arxiv : hep - th/0109127 ] . v. bargmann , p. butera , l. girardello and j. r. klauder , `` on the completeness of the coherent states , '' rept . * 2 * , 221 ( 1971 ) ; a. m. perelomov , teor . * 6 * , 213 ( 1971) h. bacry , a. grossmann and j. zak , `` proof of completeness of lattice states in the k q representation , '' phys . b * 12 * , 1118 ( 1975 ) . o. aharony , m. berkooz and e. silverstein , `` multiple - trace operators and non - local string theories , '' jhep * 0108 * , 006 ( 2001 ) [ arxiv : hep - th/0105309 ] , `` non - local string theories on ads(3 ) x s**3 and stable non - supersymmetric backgrounds , '' phys . d * 65 * , 106007 ( 2002 ) [ arxiv : hep - th/0112178 ] . e. witten , `` multi - trace operators , boundary conditions , and ads / cft correspondence , '' arxiv : hep - th/0112258 . m. berkooz , a. sever and a. shomer , `` double - trace deformations , boundary conditions and spacetime singularities , '' jhep * 0205 * , 034 ( 2002 ) [ arxiv : hep - th/0112264 ] .

gaussian graphical models have become a well - recognized tool for the analysis of conditional independencies within a set of continuous random variables . from an inferential point of view , it is important to realize that they are composite exponential transformation families . we reveal this structure by explicitly describing , for any undirected graph , the ( maximal ) matrix group acting on the space of concentration matrices in the model . the continuous part of this group is captured by a poset naturally associated to the graph , while automorphisms of the graph account for the discrete part of the group . we compute the dimension of the space of orbits of this group on concentration matrices , in terms of the combinatorics of the graph ; and for dimension zero we recover the characterization by letac and massam of models that are transformation families . furthermore , we describe the maximal invariant of this group on the sample space , and we give a sharp lower bound on the sample size needed for the existence of equivariant estimators of the concentration matrix . finally , we address the issue of robustness of these estimators by computing upper bounds on finite sample breakdown points . ,

1207.5910

gaussian graphical models are popular tools for modelling complex associations in the multivariate continuous case . if the graph with vertex set @xmath0:=\{1,\ldots , m\}$ ] is complete , then the general linear group @xmath1 , consisting of all invertible @xmath2-matrices , acts on the space of concentration matrices in the model , as well as on the sample space . the maximum likelihood estimator ( mle ) of the concentration matrix is equivariant with respect to this group action , but many other equivariant estimators have been proposed , for example , by @xcite . for smaller graphs , only some proper subgroup of @xmath1 will act on the set of compatible concentration matrices . in this paper , we describe that subgroup explicitly , and pave the way for its use in designing invariant tests , ( robust ) equivariant estimators and improved inference procedures . having an explicit group acting on a statistical model has numerous advantages . this was first pointed out by @xcite in the context of the location and scale models , which then led to the notion of a transformation family , that is , a statistical model on which a group acts transitively . group actions give rise , for example , to the study of model invariants and distributional aspects of the maximum likelihood estimator ( mle ) or other equivariant estimators [ see @xcite ] . when a group acts on a model in a nontransitive manner , the model is sometimes called a composite transformation family [ see @xcite ] . in this case , the model can be decomposed into a family of transformation models each corresponding to a fixed value of some parameter . to set the stage , let @xmath3,e)$ ] be an undirected graph with set of vertices @xmath0 $ ] and set of edges @xmath4\choose2}$ ] . denote by @xmath5 the set of symmetric matrices in @xmath6 and by @xmath7 the cone of positive definite matrices . let @xmath8 denote the linear space of symmetric matrices whose @xmath9 off - diagonal entry is zero if @xmath10 , and by @xmath11 the cone of all positive definite matrices in @xmath12 . as a running example in this , we take @xmath13 to be the path @xmath14 . so @xmath15 consists of all symmetric matrices of the form @xmath16.\ ] ] let @xmath17}$ ] be a random vector with multivariate normal distribution @xmath18 . the _ gaussian graphical model _ is the statistical model for @xmath19 given by @xmath20 so @xmath21 is the space of concentration matrices compatible with the model [ see @xcite ] . the group @xmath22 acts on @xmath23 by matrix - vector multiplication , and this induces an action on @xmath24 and @xmath25 given by @xmath26indeed , note that this is the concentration matrix of @xmath27 if @xmath28 is the concentration matrix of @xmath29 . a leading role in this paper is played by the group @xmath30 this is a closed subgroup of the lie group @xmath1 ( see section [ secg ] ) . for example , if @xmath13 is the complete graph , then @xmath31 is all of @xmath1 . for _ any _ graph @xmath13 , the group @xmath31 contains the invertible diagonal matrices , which correspond to scaling the components of @xmath19 . furthermore , @xmath31 contains elements coming from graph automorphisms of @xmath13 . specifically , if @xmath32 \to[m]$ ] is such an automorphism , then the permutation matrix @xmath33 with ones on the positions @xmath34 $ ] lies in @xmath31 , since its action on @xmath24 stabilizes the zero pattern prescribed by @xmath13 . for our running example @xmath35 , the permutation matrix @xmath36\ ] ] lies in @xmath31 . our first result is an explicit description of @xmath31 in terms of @xmath13 , and requires the pre - order on @xmath0 $ ] defined by @xmath37 where @xmath38\dvtx \{i , j\ } \in e\}$ ] denotes the set of neighbors of @xmath39 in @xmath13 . so in our running example @xmath35 we have @xmath40 . consider the closed subset @xmath41 of @xmath1 defined by @xmath42 we show in section [ secg ] that this set is a subgroup of @xmath1 . for @xmath43 it consists of all invertible matrices of the form @xmath44.\ ] ] [ thmmg ] for any undirected graph @xmath3,e)$ ] , the group @xmath31 is generated by the group @xmath41 and the permutation matrices corresponding to the automorphism group of the graph @xmath13 . for @xmath35 , this theorem says that @xmath31 is the group of all matrices of the form above , together with all matrices of the form @xmath45.\ ] ] the two subgroups of @xmath31 in theorem [ thmmg ] can have a nontrivial intersection . for instance , when @xmath13 is the complete graph , the automorphism group of @xmath13 is contained in @xmath41 . in section [ secg ] , we state and prove a more refined statement that gets rid of that intersection . now that we know explicitly which matrix group @xmath31 acts on our graphical model @xmath46 , we can use this group to develop classical notions of multivariate statistics in the general context of graphical models . one of these notions is that of an equivariant estimator [ see , e.g. , @xcite ] . let @xmath47 denote the @xmath48 matrix , whose columns correspond to @xmath49 independent copies of the vector @xmath19 . then an equivariant estimator for the concentration matrix is a map @xmath50 , that is , a map from the space of @xmath49-samples @xmath47 to the parameter space of the model , that satisfies @xmath51 for all realisations @xmath52 of @xmath47 . the standard example is the maximum likelihood estimator ( mle ) . indeed , the likelihood of concentration matrix @xmath28 given an @xmath49-sample @xmath52 equals the likelihood of @xmath53 given @xmath54 , for any @xmath55 , and this implies that the mle is @xmath31-equivariant . other equivariant estimators of the concentration matrix for some special graphical models have been proposed in @xcite . for decomposable graphs , the mle exists with probability one if and only if @xmath49 is at least the size of the maximal clique of the given graph . however , in general , whether the mle exists , with probability one , for a given sample size @xmath49 and a given graph @xmath13 is a subtle matter ; see the recent paper by @xcite and the references therein . by contrast , the question whether for a given sample size _ any _ equivariant estimator exists , turns out to have a remarkably elegant answer for any graph @xmath13 . to state it , define the _ down set _ @xmath56 of an element @xmath57 $ ] to be the set of all @xmath58 $ ] with @xmath59 . [ thgeqexist ] let @xmath3,e)$ ] be an undirected graph . there exists a @xmath31-equivariant estimator @xmath60 if and only if @xmath61}|{\hspace*{-1pt}\downarrow\hspace{-3pt}}i|$ ] . to be precise , when @xmath49 is at least the bound in the theorem , a @xmath31-equivariant @xmath62 exists that is defined outside some measure - zero set ( in fact , an algebraic subvariety of positive codimension ) , while if @xmath49 is smaller than that bound , then not even any partially defined equivariant map @xmath62 exists . for our running example @xmath35 , we have @xmath63 and @xmath64 and @xmath65 , so theorem [ thgeqexist ] says that an equivariant estimator exists with probability one if and only if the sample size is at least @xmath66 , which in this case coincides with the condition for existence of the mle . theorem [ thgeqexist ] will be proved in section [ secequivariance ] , where we also establish upper bounds on the robustness of equivariant estimators , based on general theory from @xcite . another classical notion related to a group action on a statistical model is that of _ invariants _ on the sample space . in our case , these are maps @xmath67 defined on @xmath68 , possibly outside some measure - zero set , that are constant on @xmath31-orbits , that is , that satisfy @xmath69 for all @xmath55 . an invariant @xmath67 is called _ maximal _ if it distinguishes all @xmath31-orbits . in formulas this means that for @xmath49-samples @xmath70 , outside some set of measure zero , the equality @xmath71 implies that there exists a @xmath55 such that @xmath72 . any invariant map is then a function of @xmath67 . the relevance of maximal invariants in statistics lies in the fact that they facilitate inference for the maximum likelihood estimator in the case of transformation families [ see @xcite ] . in this case the maximal invariant is an ancillary statistics that one may chose to condition on . these ideas can be used also in the case of composite transformation families , where the inference for the index parameter @xmath73 is based on the marginal distribution of the maximal invariant statistics [ @xcite , section 5 ] . another important application of the maximal invariant is in the construction of _ invariant tests _ [ see @xcite ] . suppose , for instance , that we want to test the hypothesis that the distribution of the multivariate gaussian random vector @xmath19 lies in @xmath46 against the alternative that it does not , and suppose that for the @xmath49-sample @xmath74 the test would accept the hypothesis . then , since @xmath46 is stable under the action of any @xmath55 , it is natural to require that our test also accepts the hypothesis on observing @xmath75 . thus , the test itself would have to be @xmath31-invariant . our result on maximal invariants uses the equivalence relation @xmath76 on @xmath0 $ ] defined by @xmath77 if and only if both @xmath78 and @xmath59 , that is , if and only if @xmath79 . we write @xmath80 for the equivalence class of @xmath57 $ ] and @xmath0/\sim$ ] for the set of all equivalence classes . [ thmmmaxinv ] let @xmath3,e)$ ] be an undirected graph . suppose that @xmath81 . then the map @xmath82/\sim } { \mathbb{r}}^{n\times n}$ ] given by @xmath83^t \bigl(\mathbf{x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i ] \mathbf { x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i]^t\bigr)^{-1 } \mathbf{x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i ] \bigr)_{\bar{i}\in[m]/\sim},\ ] ] where @xmath84 \in{\mathbb{r}}^{| { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i|\times n}$ ] is the submatrix of @xmath52 given by all rows indexed by @xmath56 , is a maximal @xmath41-invariant . the lower bound on @xmath49 in the theorem ensures that the @xmath85-matrices @xmath84 \mathbf { x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i]^t$ ] are invertible for generic @xmath52 , and in particular for @xmath86 outside a set of measure zero . for the complete graph , theorem [ thmmmaxinv ] reduces to the known statement that @xmath87 is a maximal invariant , see example 6.2.3 in @xcite , while for our running example @xmath35 it says that the rank - one matrix @xmath88^t ( \mathbf{x}[1 ] \mathbf{x}[1]^t)^{-1 } \mathbf{x}[1]$ ] ( recording only the direction of the first row of @xmath52 ) and the rank - two matrices @xmath89^t ( \mathbf{x}[1,2 ] \mathbf { x}[1,2]^t)^{-1 } \mathbf{x}[1,2]$ ] and @xmath90^t ( \mathbf{x}[1,3 ] \mathbf{x}[1,3]^t)^{-1 } \mathbf{x}[1,3]$ ] together form a maximal invariant for @xmath41 . we stress that theorem [ thmmmaxinv ] gives a maximal invariant under the subgroup @xmath41 , rather than under all of @xmath31 . the proof of this theorem can be found in section [ secorbequiv ] . our final results concern the space @xmath92 of @xmath31-orbits in @xmath91 . when @xmath46 is a transformation family , this space consists of a single point and hence has dimension zero . conversely , it turns out that when the dimension of @xmath92 is zero , @xmath46 is a transformation family . by work of @xcite , it is known exactly for which graphs this happens . our result on @xmath92 is a combinatorial expression for its dimension . rather than capturing that expression in a formula , which we will do in section [ secorbits ] , we now describe it in terms of a combinatorial procedure . let @xmath3,e)$ ] be an undirected graph . color an edge @xmath93 red if @xmath77 , green if @xmath78 or @xmath59 but not both , and blue otherwise . next delete all green edges from @xmath13 , while retaining their vertices . then delete the blue edges sequentially , in each step not only deleting a blue edge but also its two vertices together with all further blue and red edges incident to those two vertices . continue this process until no blue edges are left . call the resulting graph @xmath94 ; it consists of red edges only . see figure [ figorbitspacedim ] for an example . one can show that , up to isomorphism , @xmath94 is independent of the order in which the blue edges with incident vertices were removed though in general it is larger than the graph obtained by deleting all blue edges , their vertices , and their incident edges at once . [ thmmorbitspacedim ] the dimension of @xmath92 equals the number of blue edges in the original graph @xmath13 minus the number of red edges in @xmath13 plus the number of remaining red edges in @xmath94 . in other words , that dimension equals the number of blue edges in @xmath13 minus the number of red edges deleted in the process going from @xmath13 to @xmath94 . this number is nonnegative : indeed , if in some step a blue edge @xmath95 is being deleted together with its vertices , then for each red edge @xmath96 being deleted along with @xmath39 there is also a blue edge @xmath97 being deleted , and for each red edge @xmath98 being deleted along with @xmath99 also a blue edge @xmath100 is deleted . this shows , in particular , that @xmath101 is zero if and only if @xmath13 has no blue edges , that is , if all edges run between vertices that are comparable in the pre - order . this is equivalent to the condition found in @xcite for @xmath46 to be a transformation family ; see theorem [ thmmmassam ] below . for our running example @xmath102 the model is a transformation family and similarly for complete graphs . for an example where @xmath92 has dimension @xmath103 , see figure [ figorbitspacedim ] . the proof of theorem [ thmmorbitspacedim ] can be found in section [ secorbits ] and in supplementary materials [ @xcite ] . has dimension @xmath103 . ] the remainder of the paper closely follows the structure of this introduction . first , in section [ secg ] we use structure theory of real algebraic groups to determine @xmath31 . in section [ secequivariance ] , we derive necessary and sufficient conditions for the existence , with probability one , of equivariant estimators of the concentration ( or covariance ) matrix , and we give an upper bound on the robustness of those estimators , measured by the finite sample breakdown point for generic samples . in section [ secorbequiv ] , we derive the maximal invariant of theorem [ thmmmaxinv ] . in section [ sectransitive ] , we discuss in some detail the case where @xmath31 acts transitively on @xmath91 providing general formula for an equivariant estimator , after which section [ secorbits ] is devoted to our combinatorial formula for the orbit space dimension in the general case . we conclude the paper with a short discussion . in the supplementary materials [ @xcite ] , we provide the proof of theorem [ thmmorbitspacedim ] . we also discuss further results on the combinatorial structure of the problem that link our work to @xcite . [ secg ] throughout this paper , we fix an undirected graph @xmath3,e)$ ] and define the group @xmath31 as in the : @xmath30 note that @xmath31 is , indeed , a subgroup of @xmath1 : first , if @xmath104 , then @xmath105 ; and second , if @xmath55 , then since @xmath91 linearly spans @xmath106 , the ( linear ) action of @xmath33 must map the linear space @xmath106 into itself . since @xmath33 is invertible , we then have @xmath107 ( which implies that @xmath108 holds instead of the apparently weaker defining inclusion ) . but then also @xmath109 . finally , the action by @xmath110 preserves positive definiteness , so that @xmath111 , as claimed . the general linear group @xmath1 has two natural topologies : the euclidean topology , and the weaker zariski topology in which closed sets are defined by polynomial equations in the matrix entries . the subgroup @xmath31 is closed in both topologies . indeed , by the above , its elements @xmath33 are characterized by the condition that @xmath112 for all @xmath113 , and this translates into quadratic equations in the entries of @xmath33 . as a zariski - closed subgroup of @xmath1 , the group @xmath31 is a real algebraic matrix group , and in particular a real lie group . for basic structure theory of algebraic groups , we refer to @xcite . in algebraic groups , the zariski - connected component containing the identity is always a normal subgroup , the quotient by which is finite . we first determine the identity component and then the quotient . observe that the group @xmath114 of all invertible diagonal matrices is contained entirely in @xmath31indeed , it just rescales the components of the random vector @xmath19 and therefore preserves the original conditional independence statements defining @xmath46 . the group @xmath115 has @xmath116 components in the euclidean topology , corresponding to the possible sign patterns of the diagonal entries , but it is connected in the zariski topology . for this reason , the zariski topology is slightly more convenient to work with , and in what follows our topological terminology refers to it . we will use that the connected component of @xmath31 containing the identity ( the _ identity component _ , for short ) is determined uniquely by its lie algebra @xmath117 . the following lemma helps us determine that lie algebra ; we use the standard notation @xmath118 for the matrix that has zeroes everywhere except for a one at position @xmath9 . [ lemifdiag ] let @xmath119 be a real algebraic matrix group containing the group @xmath115 . then the lie algebra of @xmath120 has a basis consisting of matrices @xmath118 with @xmath9 running through some subset @xmath121 of @xmath0 \times[m]$ ] . moreover , the set @xmath121 defines a pre - order on @xmath0 $ ] in the sense that @xmath122 lies in @xmath121 for all @xmath57 $ ] and that @xmath123 . conversely , the @xmath118 with @xmath9 running through any set @xmath124 \times[m]$ ] defining a pre - order on @xmath0 $ ] span the lie algebra of a unique closed connected subgroup of @xmath1 containing @xmath115 , namely , the group of all @xmath125 with @xmath126 unless @xmath127 . this lemma is well known , so we only sketch the key arguments . the commutative group @xmath115 acts by conjugation on the lie algebra of @xmath120 , which therefore must be a direct sum of simultaneous eigenspaces of the elements of @xmath115 in their conjugation action on the space of @xmath128-matrices . these simultaneous eigenspaces are the one - dimensional subspaces spanned by the @xmath118 , so the lie algebra of @xmath120 is spanned by some of these matrices . for this argument , see [ @xcite , section 8.17 ] . the inclusion @xmath129 implies that the @xmath130 are all in the lie algebra , and for @xmath131 in the lie algebra with @xmath132 , also the commutator @xmath133=e_{ik}$ ] lies in the lie algebra . the earliest relation to pre - orders that we could find is the paper @xcite . next , we determine which @xmath118 lie in @xmath117 . [ proph0gens ] for @xmath134 , $ ] the matrix @xmath118 lies in @xmath117 if and only if @xmath135 . as a consequence , the identity component of @xmath31 is the group @xmath136 from the . the element @xmath118 with @xmath137 lies in @xmath117 if and only if the one - parameter group @xmath138 lies in @xmath31 , that is , maps @xmath139 into itself . pick @xmath140 with nonzero entries on the diagonal and at all positions corresponding to edges of @xmath13 . we have @xmath141this takes into account the inverses and the transpose in the definition of the action . this action has the effect of subtracting @xmath142 times the @xmath39th row of @xmath28 from the @xmath99th row and subtracting @xmath142 times the @xmath39th column from the @xmath99th column . for suitable @xmath142 this will create zeroes at positions corresponding to nonedges of @xmath13 _ unless _ the positions of the nonzeroes in the @xmath39th row are among the positions of the nonzeroes in the @xmath99th row . this shows that @xmath143 is necessary for @xmath118 to lie in @xmath117 ; and repeating the argument for general @xmath28 shows that it is also sufficient . the second statement now follows from lemma [ lemifdiag ] . recall the running example @xmath14 from the . by proposition [ proph0gens ] , the lie algebra @xmath117 is spanned by @xmath144 together with @xmath145 and @xmath146 . the element @xmath145 lies in @xmath147 because @xmath148 . the inverse containment does not hold , so @xmath149 does not lie in @xmath147 . the group @xmath150 consists of invertible matrices of the form @xmath151,\ ] ] where the asterisk denotes an element which can be nonzero . it is useful in the remainder of the paper to have a thorough understanding of the pre - order @xmath152 . it can also be described in terms of the collection @xmath153 of _ maximal cliques _ in the graph @xmath13 , as follows : @xmath59 if and only if every @xmath154 containing @xmath99 also contains @xmath39 . recall that @xmath152 determines an equivalence relation @xmath76 on @xmath0 $ ] . it also determines a partial order on @xmath0/\sim$ ] , still denoted @xmath152 , defined by @xmath155 if @xmath78 . we denote the poset @xmath156/\sim,\preccurlyeq)$ ] by @xmath157 ; it was first introduced in @xcite but appeared also in other related contexts in @xcite . in figure [ figpg ] , we show three graphs and the _ hasse diagrams _ of the corresponding posets @xmath158 . we note in passing that not all posets arise as @xmath158 for some @xmath13 . two counterexamples are given in figure [ fignonposets ] . a more detailed study of the structure of @xmath159 is provided in the supplementary materials [ @xcite ] . . ] for any @xmath13 with collection @xmath153 of maximal cliques . ] imagine relabeling the vertices of @xmath13 by @xmath0 $ ] in such a way that the equivalence classes of @xmath76 are consecutive intervals and such that an inequality @xmath160 between equivalence classes implies that the interval corresponding to @xmath161 contains smaller integers than the interval corresponding to @xmath80 . then the matrices in @xmath41 are block lower triangular with square blocks along the diagonal corresponding to the equivalence classes . from this it is easy to see that @xmath41 is connected in the zariski topology , but not in the ordinary euclidean topology . its number of euclidean components is @xmath162/\sim|}$ ] , corresponding to sign patterns of the determinants of the diagonal blocks . the analogue of @xmath150 has been studied for other gaussian models . for lattice conditional independence models this group was named the group of generalized block - triangular matrices with lattice structure [ see @xcite , section 2.4 ] . the link between lattice conditional independence models and certain gaussian graphical models is discussed in @xcite and in the supplementary materials [ @xcite ] . now that we have determined the identity component @xmath41 of @xmath31 , we set out to describe the quotient @xmath163 , known as the component group . in the we observed that for our running example @xmath35 the permutation matrix @xmath164,\ ] ] lies in @xmath31 but not in @xmath41 . the key to generalizing this observation is the following . [ lemasperm ] every element @xmath165 can be written as @xmath166 , where @xmath167 and @xmath168 is a permutation matrix contained in @xmath31 . the subgroup @xmath169 is a maximal ( real , split ) torus in the real algebraic group @xmath41 . by a standard result in the theory of algebraic groups [ see , e.g. , @xcite , theorem 15.14 ] , maximal tori are conjugate under @xmath41 . hence , there exists a @xmath170 such that @xmath171 . then @xmath172 , that is , @xmath173 normalizes @xmath115 . but the normalizer of @xmath115 in @xmath1 consists of monomial matrices , that is , @xmath173 equals @xmath174 with @xmath175 and @xmath168 some permutation matrix . hence , @xmath176 . here the second factor is an element of @xmath41 , so that @xmath168 is a permutation matrix contained in @xmath31 . we can now prove theorem [ thmmg ] . proof of theorem [ thmmg ] by proposition [ lemasperm ] every element of @xmath31 can be written as @xmath177 with @xmath178 an element of @xmath41 and @xmath168 a permutation matrix belonging to @xmath31 , that is , preserving the zero pattern of matrices in @xmath139 . the only such permutation matrices are those coming from automorphisms of @xmath13 . this proves that @xmath179 , where we identify the automorphism group @xmath180 with the group of corresponding permutation matrices . this proves the theorem . as explained in the , the expression @xmath179 is not minimal in the sense that @xmath180 and @xmath41 may intersect . to get rid of that intersection , we define @xmath181 to be the graph with vertex set @xmath0/\sim$ ] and an edge between @xmath80 and @xmath161 if there is an edge between @xmath39 and @xmath99 in @xmath13 . define @xmath182/\sim\to{\mathbb{n}},\ \bar{i } \mapsto|\bar{i}|$ ] and view @xmath183 as a coloring of the vertices of @xmath181 by natural numbers . let @xmath184 denote the group of automorphisms of @xmath181 preserving the coloring . there is a lifting @xmath185 defined as follows : the element @xmath186 is mapped to the unique bijection @xmath187 \to[m]$ ] that maps each equivalence class @xmath80 to the equivalence class @xmath188 by sending the @xmath189th smallest element of @xmath80 ( in the natural linear order on @xmath0 $ ] ) to the @xmath189th smallest element of @xmath188 , for @xmath190 . on the right . ] [ thmmgsemidir ] the group @xmath31 equals @xmath191 , and the intersection @xmath192 is trivial , so @xmath31 is the semidirect product @xmath193 . by the proof of theorem [ thmmg ] , any @xmath55 can be written as @xmath194 with @xmath195 and @xmath196 . since @xmath76 is defined entirely in terms of the graph @xmath13 , the graph automorphism @xmath168 satisfies @xmath197 . this implies that @xmath168 determines an automorphism @xmath186 mapping @xmath80 to @xmath198 . now @xmath168 equals @xmath199 where @xmath200 maps each equivalence class @xmath80 into itself . but then @xmath201 lies in @xmath41 and hence @xmath33 equals @xmath202 times an element @xmath203 of @xmath41 . this proves the first statement . as for the second statement , observe that a permutation matrix can have the zero pattern prescribed by @xmath41 _ only _ if the permutation maps each equivalence class into itself . the only element of @xmath204 with this property is the identity matrix . as an example , we consider a special small graph bull graph_which is a graph on five vertices depicted in figure [ figbadcase ] . the continuous part of @xmath31 is given by the poset @xmath159 depicted on the right . there is only one nontrivial automorphism of @xmath13 . it permutes @xmath205 with @xmath206 and @xmath103 with @xmath66 . hence , the group @xmath207 consists of matrices of the following two types : @xmath208 \quad\mbox{and}\quad \left [ \matrix { 0 & 1 & 0 & 0 & 0 \vspace*{2pt}\cr 1 & 0 & 0 & 0 & 0 \vspace*{2pt}\cr 0 & 0 & 1 & 0 & 0 \vspace*{2pt}\cr 0 & 0 & 0 & 0 & 1 \vspace*{2pt}\cr 0 & 0 & 0 & 1 & 0 } \right]\cdot\left [ \matrix { * & 0 & 0 & 0 & 0 \vspace*{2pt}\cr 0 & * & 0 & 0 & 0 \vspace*{2pt}\cr * & * & * & 0 & 0 \vspace*{2pt}\cr * & 0 & 0 & * & 0 \vspace*{2pt}\cr 0 & * & 0 & 0 & * } \right]\ ] ] to see theorems [ thmmg ] and [ thmmgsemidir ] in some further examples , see section [ secexamples ] . to the coloured graph @xmath209 we can associate a gaussian graphical model @xmath210 with _ multivariate nodes _ , where node @xmath211 is associated to a gaussian vector of dimension @xmath212 . this model coincides with @xmath46 . this also shows , conversely , that our framework extends to general gaussian graphical models with multivariate nodes . suppose that in the inference of the unknown concentration matrix @xmath213 the observed @xmath49-sample @xmath214 leads to the estimate @xmath215 . then it is reasonable to require that the sample @xmath75 leads to the estimate @xmath216 . such a map @xmath217 , possibly defined only outside some ( typically @xmath31-stable ) measure - zero set and satisfying @xmath218 for all @xmath55 there , is called a ( @xmath31-)_equivariant estimator_. in this section we determine a sharp lower bound on @xmath49 for an equivariant estimator to exist , and then , building on theory from @xcite , we determine a bound on the robustness of such estimators . -equivariant map exists as soon as @xmath219 . however , the mle exists only when @xmath220 . ] the mle , when it exists , is automatically @xmath31-equivariant , since the likelihood function is @xmath31-invariant . a _ necessary _ condition for the mle to exist with probability @xmath103 is that the sample size @xmath49 be at least the largest clique size @xmath221 . a _ sufficient _ condition is that @xmath49 be at least the maximal clique size @xmath222 in a _ decomposable cover _ of @xmath13 , that is , a graph @xmath223,e^*)$ ] with @xmath224 that does not have induced @xmath189-cycles for @xmath225 . the exact minimal value of @xmath49 for which mle exists is not known explicitly in general , but interesting classes of graphs were analyzed in @xcite and @xcite . our theorem [ thgeqexist ] states that equivariant estimators of the concentration matrix ( or , equivalently by taking inverses , of the covariance matrix ) exist if and only if @xmath226 } than the necessary condition @xmath227 for the existence of mle . indeed , any down set @xmath56 is in fact a clique , because @xmath228 implies that @xmath229 , that is , @xmath99 and @xmath189 are either equal or connected by an edge . the inequality @xmath230 } |{\hspace*{-1pt}\downarrow\hspace{-3pt}}i| \leq \max_{c \in\mathcal{c } } |c|$ ] can be strict . for example , in the graph of figure [ figbibull ] the biggest maximal clique has cardinality @xmath231 , while @xmath232}|{\hspace*{-1pt}\downarrow\hspace{-3pt}}i|=2 $ ] . in consequence , our result does not shed new light on the existence of mle ; however , it does provide necessary and sufficient conditions in the search for other equivariant estimators . we now prepare the proof of theorem [ thgeqexist ] . in our arguing , we borrow some terminology from algebraic geometry : we say that some property holds for _ generic _ @xmath49-tuples @xmath233 if it holds for @xmath52 outside the zero set of some nonzero polynomial . note that if a property holds for generic @xmath49-tuples , then it holds with probability one for the random sample @xmath47 drawn from any nondegenerate probability distribution with continuous density function on @xmath234 . [ thdelta0 ] the minimal number @xmath49 for which the stabilizer in @xmath150 of a generic @xmath49-sample @xmath235 consists entirely of determinant-@xmath236 matrices equals @xmath237 } |{\hspace*{-1pt}\downarrow\hspace{-3pt}}i|$ ] . for that value of @xmath49 the stabilizer of a generic @xmath49-sample is , in fact , the trivial group @xmath238 . the condition that @xmath239 fixes one vector @xmath240 translates into @xmath241 linear conditions on the entries of @xmath33 , namely : @xmath242 the @xmath39th condition concerns only the entries in the @xmath39th row of @xmath33 . we therefore concentrate on that single row of @xmath33 , and regard the entries @xmath243 as variables to be solved from the linear equations above as @xmath244 ranges through the given @xmath49-tuple @xmath52 of vectors . since the given @xmath49-tuple is generic , those equations are linearly independent as long as @xmath49 is at most the cardinality of @xmath56 . hence , they determine the @xmath39th row uniquely as soon as @xmath49 is at least that number . hence , as soon as @xmath49 is at least the maximal cardinality of the sets @xmath56 over all @xmath39 the stabilizer of a generic sample @xmath52 will be trivial . what remains to be checked , is that for smaller @xmath49 the stabilizer of a generic @xmath235 does _ not _ consist entirely of determinant-@xmath236 matrices . this is most easily seen by considering the lie algebra of that stabilizer , which is the set of matrices @xmath245 in the lie algebra of @xmath41 satisfying the linear conditions @xmath246 . let @xmath39 be a row index for which @xmath56 has more than @xmath49 elements . then the linear conditions on @xmath245 do not fix the @xmath39th row of @xmath245 uniquely . moreover , by genericity , they do not fix the diagonal entry @xmath247 uniquely , either . as a consequence , they do not determine the trace of @xmath245 uniquely . this shows that the lie algebra of the stabilizer is not contained in the lie algebra of trace - zero matrices . but then the stabilizer is not contained in the lie group of determinant - one matrices ( whose lie algebra consists of the trace - zero matrices ) . proof of theorem [ thgeqexist ] , necessity of @xmath248 assume that there exists a @xmath31-equivariant estimator @xmath249 , possibly defined outside some measure - zero set . in particular , the @xmath150-equivariance of @xmath62 implies that the @xmath150-stabilizer of a generic sample @xmath52 is contained in the @xmath150-stabilizer of @xmath215 : @xmath250 now since @xmath251 , the stabilizer on the right - hand side is a generalized orthogonal group , and hence in particular compact in the eculidean topology . hence , the stabilizer on the left - hand side must be compact , as well . however , by ( the proof of ) theorem [ thdelta0 ] , that stabilizer is the intersection of @xmath1 with an affine subspace of @xmath252 . such a set is not compact in the euclidean topology unless it consists of a single matrix , and this happens only when @xmath253}|{\hspace*{-1pt}\downarrow\hspace{-3pt}}i|$ ] . to prove that @xmath254 is also _ sufficient _ for the existence of a @xmath31-equivariant estimator , we introduce the following construction . fix a natural number @xmath254 and construct a function @xmath255 \to[n]$ ] by induction , as follows : if @xmath256 has been defined on all elements of @xmath257 , then define @xmath256 on elements of @xmath80 to be the increasing bijection from @xmath258 ( with the natural linear order coming from @xmath0 $ ] ) to the @xmath259 smallest elements of the set @xmath260 \setminus f ( { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i \setminus\bar{i})$ ] . this automatically guarantees that @xmath256 is injective on any down set @xmath261 and that @xmath262 for all @xmath263 . now let @xmath264 be affine space of all matrices @xmath52 with the property that first , the matrix @xmath265 $ ] obtained by taking the rows labeled by @xmath80 and the columns labeled by @xmath266 is an identity matrix for each @xmath267/\sim$ ] , and second , the matrices @xmath268 $ ] are zero for all @xmath269 . in our running example @xmath35 , if the sample size @xmath49 is at least @xmath66 , then @xmath256 maps @xmath103 to @xmath103 and @xmath270 both to @xmath66 . the affine space @xmath271 then consists of all matrices of the form @xmath272.\ ] ] [ lemintol ] for generic @xmath235 , there exists a unique @xmath273 such that @xmath274 . the geometric content of this lemma is that @xmath271 is a _ slice _ transverse to ( most of ) the orbits of @xmath41 on @xmath68 . in our running example @xmath35 , one goes from a generic sample to a sample in @xmath271 by first multiplying the first row by @xmath275 so as to create a one at position @xmath276 ; then subtracting a multiple of the ( new ) first row from the second to create a zero at position @xmath277 and multiplying the second row by a constant to create a one at position @xmath278 ; and then similarly ( and independently ) for the third row . all of these operations are realized by elements of @xmath41 . the following proof in the general case is a straightforward generalization of this . proof of lemma [ lemintol ] for the existence of such a @xmath33 , proceed by induction . assume that the submatrix @xmath279 $ ] already has the required shape , and decompose @xmath280 $ ] into blocks as follows : @xmath281 & \mathbf{x}\bigl [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i \setminus\bar{i},f(\bar{i } ) \bigr]\vspace*{2pt } \cr \mathbf{x}\bigl[\bar{i},f ( { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i \setminus\bar{i})\bigr ] & \mathbf{x}\bigl[\bar{i},f(\bar{i})\bigr ] } { \right}].\ ] ] then take the block matrix @xmath273 which is the identity outside the blocks labeled by @xmath282 , and which in those blocks looks like @xmath283 & g[\bar{i},\bar{i } ] } { \right}].\ ] ] now straightforward linear algebra shows that , under the condition that both and @xmath284 $ ] are full rank , there are unique choices for the as yet unspecified components of @xmath33 such that @xmath285=0 $ ] and @xmath286=i$ ] . this shows the existence of @xmath33 such that @xmath287 . uniqueness can be proved by a similar induction . proof of theorem [ thgeqexist ] , sufficiency of @xmath248 now we show that @xmath253}|{\hspace*{-1pt}\downarrow\hspace{-3pt}}i|$ ] is also a sufficient condition for the existence of an equivariant map @xmath249 , defined for generic samples @xmath52 . indeed , by construction , the space @xmath271 is stable under @xmath204 . fix any @xmath204-equivariant map @xmath288 . such maps exist and can be found as follows : take @xmath289 _ any _ map , and then define @xmath290 an average over the finite group @xmath184 . we claim that @xmath62 extends to a unique @xmath31-equivariant map @xmath291 defined almost everywhere . indeed , this extension is defined as follows : given a generic sample @xmath52 , find the unique @xmath196 such that @xmath292 , and set @xmath293 . checking that the map @xmath62 thus defined ( almost ) everywhere is both @xmath184-equivariant and @xmath41-equivariant is straightforward . this proves the existence part of theorem [ thgeqexist ] . we stress that , apart from giving necessary and sufficient conditions for the existence of a @xmath31-equivariant estimator , the proof of theorem [ thgeqexist ] actually yields the general structure of _ any _ such estimator . of course , the usefulness ( bias , robustness , etc . ) of an equivariant estimator thus constructed depends on the ( free ) choice of @xmath294 , that is , on the restriction of @xmath62 to the slice @xmath271 . we do not know at present good conditions on @xmath294 that ensure usefulness of @xmath62 . note that the maps @xmath217 constructed in the proof of theorem [ thgeqexist ] are merely @xmath31-equivariant , and not necessarily invariant under permutation of the sample points . it is easy to see , though , that the lower bound @xmath295 also implies the existence of @xmath31-equivariant estimators that _ are _ invariant under permutations of the sample points . indeed , simply replace @xmath62 by its group average @xmath296 . an important notion for the robustness of parameter estimators is that of breakdown points [ @xcite ] . in a simple univariate situation , if the estimator is given by the sample mean , then a ( large ) change made to one of the observations leads to an arbitrarily large change in the value of the estimator . on the other hand , if the estimator is the sample median , then changing one observation in a sample of size larger than two can not lead to arbitrarily large changes in the estimator . this feature makes the median more robust to outliers in the sample . the ( _ finite sample _ ) _ breakdown point _ of an estimator @xmath62 at an @xmath49-sample @xmath297 is the minimal number of components of @xmath52 that need to be altered to force arbitrarily large changes in the value of the estimator ; this quantity is usually normalized by the sample size @xmath49 . for example , the sample mean above has breakdown point @xmath298 while the sample median has breakdown point roughly @xmath299 ( in fact , both independently of @xmath52 ) . so when it comes to robustness , the estimator with the highest breakdown point is preferred . in the multivariate gaussian setting , when estimating the concentration matrix ( or the covariance matrix ) , the change in the estimator value is often measured by means of the _ pseudo - metric _ @xmath300 on @xmath301 [ see , e.g. , @xcite ] @xmath302 for graphical models , robustness issues have been rarely looked at so far , although it has been known for some time that the classical estimators and model selection procedures are vulnerable to contaminated data [ @xcite ] . first , approaches toward robust covariance estimators for undirected gaussian graphical models can be found in @xcite . these papers suggest to replace the sample covariance matrix by the reweighted minimum covariance determinant ( mcd ) estimator . the paper @xcite proposes an m - type estimator instead . both in @xcite and in @xcite the assumption of normality is discarded , and replaced by the @xmath142-distribution or the general elliptical distribution , respectively , to model heavy tails . our modest contribution to robustness issues is an upper bound on the finite sample breakdown point for @xmath31-equivariant estimators of the concentration matrix for the graphical model @xmath46 . to this end , we specialize one of the key ideas from @xcite ( @xcite ) to our setting . suppose we have an @xmath49-sample @xmath303 and an equivariant estimator @xmath217 of the concentration matrix . assume that there exists an element @xmath55 with @xmath304 that fixes ( at least ) @xmath189 of the @xmath49 sample points @xmath305 of the sample @xmath86 . define @xmath306 and let @xmath307 since @xmath308 , for each natural number @xmath309 both @xmath310 and @xmath311 contain at least @xmath312 points of the original sample @xmath52 . by the triangle inequality , we have @xmath313 and on the other hand @xmath314 which is unbounded as @xmath315 . hence , changing not more than @xmath306 of the sample points in @xmath52 can already lead to arbitrarily large changes in the estimated concentration matrix , so that the finite sample breakdown point of @xmath62 at @xmath52 is at most @xmath316 . we now state and prove our upper bound on the robustness of equivariant estimators at generic samples . [ proprobustness ] assume that @xmath254 . then for any @xmath31-equivariant estimator @xmath217 the finite sample breakdown point at a generic sample @xmath52 is at most @xmath317 , there exist matrices @xmath273 with determinant @xmath318 that fix the first @xmath319 sample points . now the proposition follows from the discussion preceding it . writing @xmath320 , note that @xmath321 with equality if and only if @xmath13 is the complete graph , and that @xmath322 , with equality if and only if for each edge @xmath93 the vertex @xmath39 has neighbors that are not connected to @xmath99 ( and vice versa ) . examples of such graphs are @xmath241-cycles with @xmath323 . trees with @xmath324 vertices are examples of graphs with @xmath325 . note that for graphs with small @xmath326 the upper bound in proposition [ proprobustness ] is close to @xmath299 , even for relatively small sample sizes @xmath49 . on the other hand , the mle , as pointed out for example in @xcite , is typically the least robust with respect to potential outliers in the sample space . although we do not know whether the upper bound in the proposition is attained for any sensible estimator , our results do suggests the quest for more robust estimators , especially for graphs with small @xmath326 . in this section , we discuss a @xmath41-invariant map @xmath67 on the space @xmath68 of @xmath49-samples , defined almost everywhere , and prove that it is maximal in the sense that for two samples @xmath327 in the domain of definition of @xmath67 the equality @xmath71 implies that @xmath327 are in the same @xmath41-orbit . recall from the that @xmath67 is defined as @xmath328^t \bigl ( \mathbf{x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i ] \mathbf { x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i]^t\bigr)^{-1 } \mathbf{x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i ] \bigr)_{\bar{i}\in[m]/\sim},\ ] ] where we assume from now on that @xmath49 is at least @xmath329 , and where @xmath67 is defined on @xmath49-samples where @xmath84 $ ] has full rank for all @xmath39 . before we proceed , we recall the following known lemma . let @xmath330 be natural numbers , and consider the action of @xmath331 on @xmath332 . let @xmath333 be the open subset of the latter space consisting of matrices of full rank @xmath189 . then the map @xmath334 mapping @xmath52 to @xmath335 is a maximal invariant for the action of @xmath331 on @xmath333 . first , to see that @xmath336 is @xmath331-invariant , compute @xmath337 second , to see that @xmath336 is maximal , note that the row space of @xmath338 is also the row space of @xmath339 hence , if @xmath340 for a second @xmath341 , then @xmath310 has the same row space as @xmath86 . but this means that there exists a @xmath342 with @xmath343 . the proof of the lemma shows that @xmath339 determines the row space of @xmath52 ( and is determined by that ! ) . now we can prove theorem [ thmmmaxinv ] , which states that @xmath67 is a maximal @xmath41-invariant . this generalizes example 6.2.3 in @xcite , which deals with the case of complete graphs . proof of theorem [ thmmmaxinv ] the @xmath41-invariance of each of the components of @xmath67 follows from the observation that @xmath344=g [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i ] \mathbf { x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i]$ ] , together with the computation in the proof of the preceding lemma . for maximality , assume that @xmath71 . this means that the row space of @xmath84 $ ] equals that of @xmath345 $ ] , for all @xmath39 . if , by induction , we have replaced @xmath52 by an element in its orbit and achieved that @xmath346={\mathbf{y } } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i \setminus\bar{i}]$ ] , then it follows that @xmath347=h_1 \mathbf { x}[\bar{i } ] + h_2 \mathbf{x } [ { { \hspace*{-1pt}\downarrow\hspace{-3pt}}}i \setminus\bar{i}]$ ] for a suitable invertible @xmath348-matrix @xmath349 and a suitable full - rank @xmath350-matrix @xmath351 . these matrices @xmath352 can be assembled into a block matrix @xmath196 ( as in the proof of lemma [ lemintol ] ) such that @xmath353 coincides with @xmath52 outside the @xmath80-labeled rows and with @xmath310 in the @xmath80-labeled rows . doing this for all equivalence classes @xmath80 from the bottom to the top of @xmath157 , we move @xmath52 to @xmath310 by an element of @xmath41 . since every invariant test depends on @xmath52 only through the value of the maximal invariant [ @xcite , section 6.2 ] , theorem [ thmmmaxinv ] paves the way for @xmath41-invariant tests , for example , for testing the hypothesis that the distribution of @xmath19 lies in @xmath46 against the null - hypothesis that it does not . a more general question involves testing two alternative ( typically nested ) graphical models corresponding to graphs @xmath354 on @xmath0 $ ] . for this , it is natural to develop tests that are invariant with respect to matrices stabilizing _ both _ models . the identity component of the group of such matrices consists of all @xmath33 with @xmath126 unless @xmath59 in _ both _ pre - orders coming from @xmath354 . the same construction as above , now applied to the intersection of the pre - orders , gives the maximal invariant for this group . a simple example of a @xmath41-invariant test is the deviance test [ see , e.g. , @xcite , section 5.2.2 ] . when @xmath31 acts transitively on @xmath21 then @xmath46 forms an exponential transformation family [ see @xcite ] , which gives very efficient tools for dealing with the ancillary statistics in the hypothesis testing and inference . in particular the @xmath355-formula of @xcite , which gives an approximation for the density of the maximum likelihood estimator given the ancillary statistics is exact and the ancillary statistics is given by the maximal invariant @xmath356 . the following result tells us when the graphical gaussian model @xmath46 is an exponential transformation family under the group @xmath31 ( cf . theorem [ thmmorbitspacedim ] ) . [ thmmmassam ] let @xmath3,e)$ ] be an undirected graph . then @xmath31 acts transitively on @xmath46 if and only if one of the following equivalent conditions holds : * for any two neighbors @xmath357 $ ] we have either @xmath358 or @xmath359 ; * @xmath13 is decomposable and does not contain a @xmath205-chain @xmath360 as an induced subgraph ; * the hasse diagram of @xmath159 is a tree with a unique minimum . as we show in the supplementary materials [ @xcite ] , the transitive case is precisely the case when @xmath46 corresponds to a lattice conditional independence model . we also prove there the following lemma . [ lemequconds ] if @xmath13 satisfies the conditions of theorem [ thmmmassam ] , then @xmath361 is equal to the size of the biggest maximal clique of @xmath13 . in particular a @xmath31-equivariant estimator exists with probability one if and only is the mle estimator exists with probability one . in the transitive case construction of a @xmath31-equivariant estimator is particularly straightforward [ @xcite , chapter 6 , example 6.2 ] . this generalizes the case of a star - shape graph analyzed in @xcite . let @xmath362 be the mle of the covariance matrix and define @xmath363 . because @xmath364 is a sufficient statistic , without loss we can assume that every estimator based on the full sample satisfies @xmath365 . since @xmath366 , there exists @xmath367 such that @xmath368 . the construction of @xmath369 follows by the fact that in the transitive case there exists a well defined map @xmath370 , where @xmath371 is the stabiliser of the identity matrix . this map is the inverse of the canonical map from @xmath372 to @xmath21 . then @xmath369 is just a composition of @xmath373 followed by @xmath374 . by @xmath31-equivariance , @xmath375 where @xmath376 . we have just shown the following result . let @xmath13 be a decomposable graph without induced @xmath205-chains . define @xmath363 as above . then every @xmath31-equivariant estimator of the concentration matrix is of the form @xmath377 where @xmath378 is a constant matrix and @xmath379 is such that @xmath380 . since the function @xmath369 is uniquely identified the only way to obtain different equivariant estimators is by varying the constant matrix @xmath381 . this can be done with different optimality criteria in mind . an interesting problem is to find @xmath381 such that @xmath382 is an unbiased estimator of the concentration matrix . another motivation is that the mle for lattice conditional independence models ( and hence for @xmath46 in the transitive case by the theorem in the supplementary materials [ @xcite ] ) is not admissible [ see @xcite ] . a relevant question is to analyse equivariant estimators minimizing risk related to certain loss functions . this analysis has been already done for star - shaped models by @xcite . given an undirected graph @xmath3,e)$ ] , we have determined the group @xmath383 of all invertible linear maps @xmath384 stabilizing the cone @xmath91 . theorem [ thmmmassam ] characterizes when @xmath46 is a transformation family , that is , when @xmath31 has a single orbit on @xmath91 . for general @xmath13 , the orbit space @xmath92like many quotients of manifolds by group actions can conceivably be very complicated . in this section , we compute its first natural invariant , namely , its dimension . in the zero - dimensional case , we recover the class from theorem [ thmmmassam ] . basic lie group theory tells us that @xmath101 equals @xmath385 where @xmath386 is the stabilizer of a generic concentration matrix in @xmath91 . in this expression , the first term equals @xmath387 and the second term equals @xmath388/\sim } |\bar{i}| \cdot determine the generic stabilizer dimension . note that for the dimension it does not matter whether we consider the stabilizer in @xmath31 or in @xmath41 . the following theorem makes use of the colored quotient graph @xmath389 from section [ secg ] . [ propdimstab ] the dimension of the stabilizer @xmath390 in @xmath41 of a generic matrix in @xmath91 equals @xmath391/\sim } { n_{\bar{i}}\choose2}$ ] , where @xmath392 is defined by @xmath393 where the sum ranges over all neighbors @xmath161 of @xmath80 in @xmath181 that are not comparable to @xmath80 in the partial order @xmath152 . in words : starting from @xmath181 , one deletes all edges between vertices that are comparable in the partial order @xmath152 , and one subtracts from @xmath259 the sum of the @xmath394 for all neighboring @xmath161 in the new graph . if the result is positive , then this is @xmath392 ; otherwise , @xmath392 is zero . the expression above suggests that the identity component of @xmath390 is a product of special orthogonal groups of spaces of dimensions @xmath395 , which is indeed what the proof of this proposition , given in the supplementary materials [ @xcite ] , will show . we now use the proposition to explain the combinatorial procedure in the . proof of theorem [ thmmorbitspacedim ] by proposition [ propdimstab ] , we need to compute @xmath396/\sim } |\bar{i}||{\hspace*{-1pt}\downarrow\hspace{-3pt}}i| + \sum _ { \bar{i } \in[m]/\sim } \pmatrix{n_{\bar{i } } \cr 2}.\ ] ] the term @xmath241 cancels against the diagonal entries in @xmath41 in the second term . recall that in theorem [ thmmorbitspacedim ] we colored an edge @xmath95 in @xmath13 blue , green or red according to whether zero , one , or two of the statements @xmath78 and @xmath59 hold . the term @xmath397 counts blue plus green plus red . what remains of the second term after cancelling the diagonal entries against @xmath241 counts green edges once and red edges twice . thus , the first two terms count blue edges minus red edges . finally , the last term counts the number of red edges that survive when blue edges are deleted one by one . we conclude with few examples of the use of proposition [ propdimstab ] . let @xmath181 be the bull graph in figure [ figbadcase ] , with each vertex representing an equivalence class in @xmath0/\sim$ ] with cardinality @xmath398 for @xmath399 . in this case the only pair of connected but not comparable vertices is @xmath400 . with the convention that @xmath401 if @xmath402 , proposition [ propdimstab ] shows that the dimension of the stabilizer of a generic matrix in @xmath91 is @xmath403 let @xmath404 be a tree , where each vertex @xmath405 represents an equivalence class with cardinality @xmath406 . in this case , the dimension of the stabilizer of a generic matrix in @xmath21 is @xmath407 where the first sum is over all the inner vertices of @xmath404 and the second sum is over all the leaves ( vertices of valency @xmath103 ) of @xmath404 . in particular , if for some @xmath183 we have that @xmath408 for all @xmath409 then this formula degenerates to @xmath410 , where @xmath309 is the number of leaves . let @xmath411 denote the symmetric group on @xmath0 $ ] , @xmath412 the dihedral group of graph isomorphisms of an @xmath241-cycle . also recall that @xmath413 denotes the group of all diagonal invertible @xmath414 matrices . in table [ tabsmall ] , we provide the full description of @xmath31 for all undirected graphs on @xmath415 vertices . @lccc@ & & & + & & @xmath416 & @xmath417 + & & @xmath418&@xmath419 + & & @xmath420 & @xmath417 + & & @xmath421 $ ] & @xmath419 + & & @xmath422 & @xmath423 + & & @xmath424 & @xmath425 + & & @xmath426 & @xmath417 + & & @xmath427 $ ] & @xmath419 + & & @xmath428 & @xmath429 + & & @xmath430 $ ] & @xmath417 + & & @xmath431 & @xmath423 + & & @xmath432 $ ] & @xmath419 + & & @xmath433 & @xmath419 + @lccc@ & & & + & & @xmath434 $ ] & @xmath419 + & & @xmath435&@xmath419 + & & @xmath428 & @xmath436 + in this paper , we presented the complete description of the maximal subgroup of @xmath437 that stabilizes the gaussian graphical model @xmath46 for any given graph @xmath13 . the main motivation for this study was to put gaussian graphical models into the framework of ( composite ) transformation families . group invariance is a classical topic in multivariate statistics and there are many ways that statistical inference can be improved when the group action is better understood . while we have constructed the maximal invariant under this group on sample space , we have not yet used this invariant to develop explicit tests , for example , for model selection ; and while we have given theoretical bounds on when equivariant estimators for the concentration matrix exist , and how robust they can be , we have not yet constructed such new estimators . we regard our work as a step toward achieving these goals for general graphs , laying down the theoretical framework . on the other hand , in the case where @xmath31 acts transitively on the model , we already have a much better understanding . for instance , it seems feasible to extend the work of @xcite from star - shaped models to general models in the transitive case . once these transitive models are completely understood , it seems natural to move on to those where the orbit space of @xmath31 on the model is one - dimensional . here we expect beautiful mathematics and statistics to go hand in hand : combinatorics for characterizing which graphs lead to such models , geometry for a better understanding of the one - dimensional orbit space , and statistical inference tailored to the geometry of that space . s. kuhnt thanks the deutsche forschungsgemeinschaft ( sfb 823 , project b1 ) for funding .

this paper reviews advances in stein - type shrinkage estimation for spherically symmetric distributions . some emphasis is placed on developing intuition as to why shrinkage should work in location problems whether the underlying population is normal or not . considerable attention is devoted to generalizing the `` stein lemma '' which underlies much of the theoretical development of improved minimax estimation for spherically symmetric distributions . a main focus is on distributional robustness results in cases where a residual vector is available to estimate an unknown scale parameter , and , in particular , in finding estimators which are simultaneously generalized bayes and minimax over large classes of spherically symmetric distributions . some attention is also given to the problem of estimating a location vector restricted to lie in a polyhedral cone . .

1203.4724

we are happy to help celebrate stein s stunning , deep and significant contribution to the statistical literature . in 1956 , charles stein ( @xcite ) proved a result that astonished many and was the catalyst for an enormous and rich literature of substantial importance in statistical theory and practice . stein showed that when estimating , under squared error loss , the unknown mean vector @xmath0 of a @xmath1-dimensional random vector @xmath2 having a normal distribution with identity covariance matrix , estimators of the form @xmath3 dominate the usual estimator @xmath0 , @xmath2 , for @xmath4 sufficiently small and @xmath5 sufficiently large when @xmath6 . james and stein ( @xcite ) sharpened the result and gave an explicit class of dominating estimators , @xmath7 for @xmath8 , and also showed that the choice of @xmath9 ( the james stein estimator ) is uniformly best . for future reference recall that `` the usual estimator , '' @xmath2 , is a minimax estimator for the normal model , and more generally for any distribution with finite covariance matrix.=1 stein ( @xcite ) , considering general estimators of the form @xmath10 , gave an expression for the risk of these estimators based on a key lemma , which has come to be known as stein s lemma . numerous results on shrinkage estimation in the general spherically symmetric case followed based on some generalization of stein s lemma to handle the cross product term @xmath11 $ ] in the expression for the risk of the estimator . a substantial number of papers for the multivariate normal and nonnormal distributions have been written over the decades following stein s monumental results . for an earlier expository development of stein estimation for nonnormal location models see brandwein and strawderman ( @xcite ) . this paper covers the development of stein estimation for spherically symmetric distributions since brandwein and strawderman ( @xcite ) . it is not encyclopedic , but touches on only some of the significant results for the nonnormal case . given an observation , @xmath2 , on a @xmath1-dimensional spherically symmetric multivariate distribution with unknown mean , @xmath0 and whose density is @xmath12 ( for @xmath13 ) , we will consider the problem of estimating @xmath0 subject to the squared error loss function , that is , @xmath14 is a measurable ( vector - valued ) function , and the loss given by @xmath15 where @xmath16 and @xmath17.the risk function of @xmath18 is defined as @xmath19 unless otherwise specified , we will be using the loss defined by ( [ eq11 ] ) . other loss functions such as the loss @xmath20 will be occasionally used , especially when there is also an unknown scale parameter , and minimaxity , as opposed to domination , is the main object of study . we will have relatively little to say about the important case of confidence set loss , or of loss estimation . in section [ sec2 ] we provide some additional intuition as to why the stein estimator of the mean vector @xmath0 makes sense as an approximation to an optimal linear estimator and as an empirical bayes estimator in a general location problem . the discussion indicates that normality need play no role in the intuitive development of stein - type shrinkage estimators . section [ sec3 ] is devoted to finding improved estimators of @xmath0 for spherically symmetric distributions with a known scale parameter using results of brandwein and strawderman ( @xcite ) and berger ( @xcite ) to bound the risk of the improved general estimator @xmath21 . section [ sec4 ] considers estimating the mean vector for a general spherically symmetric distribution in the presence of an unknown scale parameter , and , more particularly , when a residual vector is available to estimate the scale parameter . it extends some of the results from section [ sec3 ] to this case as well as presenting new improved estimators for this problem . the results in this section indicate a remarkable robustness property of stein - type estimators in this setting , namely , that certain of the improved estimators dominate @xmath2 uniformly for all spherically symmetric distributions simultaneously ( subject to risk finiteness ) . in section [ sec5 ] we consider the restricted parameter space problem , particularly the case where @xmath0 is restricted to a polyhedral cane , or more generally a smooth cone . the material in this section is adapted from fourdrinier , strawderman and wells ( @xcite ) . in section [ sec6 ] we consider some of the advancements in bayes estimation of location vectors for both the known and unknown scale cases . we present an intriguing result of maruyama maruyama ( @xcite)which is related to the ( distributional ) robustness of stein estimators in the unknown scale case treated in section [ sec4 ] . section [ sec7 ] contains some concluding remarks . we begin by adding some intuition as to why stein estimation is both reasonable and compelling , and refer the reader to brandwein and strawderman ( @xcite ) for some earlier developments . the reader is also referred to stigler ( @xcite ) and to meng ( @xcite ) . the following is a very simple intuitive development for optimal linear estimation of the mean vector in @xmath22 that leads to the stein estimator . suppose @xmath23 = \theta$ ] , @xmath24 ( @xmath25 known ) , and consider the linear estimator of the form @xmath26 . what is the optimal value of @xmath4 ? the risk is given by @xmath27 and the derivative , with respect to @xmath4 , is @xmath28 hence , the optimal @xmath4 is @xmath29 and the optimal `` estimator '' is @xmath30 , which is , of course , not an estimator because it depends on @xmath0 . however , @xmath31 = p \sigma^ 2 + \|\theta\|^2 $ ] , so @xmath32 is a reasonable estimator of @xmath33 . hence , an approximation to the optimal linear `` estimator '' is @xmath34 which is the james stein estimator except that @xmath1 replaces @xmath35 . note that as @xmath1 gets larger , @xmath36 is likely to improve as an estimator of @xmath37 and , hence , we may expect that the dimension , @xmath1 , plays a role . strawderman ( @xcite ) considered the following general location model . suppose @xmath38 , where @xmath23 = \theta$ ] , @xmath39 ( @xmath40 known ) but that @xmath41 is otherwise unspecified . also assume that the prior distribution for @xmath0 is given by @xmath42 , the @xmath43 fold convolution of @xmath41 with itself . hence , the prior distribution of @xmath0 can be represented as the distribution of a sum of @xmath43 i.i.d . variables @xmath44 , where each @xmath45 is distributed as @xmath46 . also , the distribution of @xmath47 has the same distribution and is independent of the other @xmath45 s . the bayes estimator can therefore be thought of as @xmath48 = e[\theta|x - \theta + \theta ] \\ & = & e\biggl[\sum_{i = 1}^n u_i \big| \sum_{i = 0}^n u_i \biggr]\end{aligned}\ ] ] and , hence , @xmath49 } \\ & = & \frac{n}{n + 1}e\biggl[\sum_{i = 0}^n u_i \big|\sum_{i = 0}^n u_i\biggr ] \\ & = & \frac{n}{n + 1 } e[x|x ] = \frac{n}{n + 1}x\end{aligned}\ ] ] or , equivalently , @xmath50 = ( 1 - 1/\{n+1\})x$ ] . assuming that @xmath43 is unknown , we may estimate it from the marginal distribution of @xmath2 , which has the same distribution as @xmath51 . in particular , @xmath52 & = & e\biggl [ \biggl\|\sum_{i = 0}^n { u_i } \biggr\| ^2 \biggr ] \\ & = & \sum_{i = 0}^n e [ \| u_i \|^2 ] = ( n + 1)p\sigma^2,\end{aligned}\ ] ] since @xmath53 = 0 $ ] and @xmath54 , @xmath55=p\sigma^2 $ ] . therefore , @xmath56 can be estimated by @xmath57 . substituting this estimator of @xmath56 in the expression for the bayes estimator , we have an empirical bayes estimator @xmath58 which is again the james stein estimator , save for the substitution of @xmath1 for @xmath35 . note that in both of the above developments , the only assumptions were that @xmath59 and@xmath39 . the stein - type estimator thus appears intuitively , at least , to be a reasonable estimator in a general location problem . let @xmath60 , the loss be @xmath61 so the risk is @xmath62 $ ] . suppose an estimator has the general form @xmath21 . then @xmath63 \\ & = & e_{\theta } [ \| x + \sigma^2 g ( x ) - \theta\|^2 ] \\ & = & e_\theta [ \|x - \theta\|^2 ] + \sigma^4 e_\theta[\| g(x ) \|^2]\\ & & { } + 2\sigma^2 e_{\theta } [ ( x-\theta)'g(x)].\end{aligned}\ ] ] in the normal case , stein s lemma , given loosely as follows , is used to evaluate the last term . [ lem31 ] if @xmath64,then @xmath65 = \sigma^2 e_{\theta } [ \nabla'g(x)]$ ] [ where @xmath66 denotes the gradient of @xmath67 , provided , say , that @xmath68 is continuously differentiable and that all expected values exist . the proof is particularly easy in one dimension , and is a simple integration by parts . in higher dimensions the proof may just add the one - dimensional components or may be a bit more sophisticated and cover more general functions , @xmath68 . in the most general version known to us , the proof uses stokes theorem and requires @xmath69 to be weakly differentiable . using the stein lemma , we immediately have the following result . [ prop31 ] if @xmath70 , then @xmath71 + \sigma^4 e_{\theta}[\|g(x)\|^2 + 2 \nabla'g(x)]\end{aligned}\ ] ] and , hence , provided the expectations are finite , a sufficient condition for @xmath14 to dominate @xmath2 is @xmath72 a.e . ( with strict inequality on a set of positive measures ) . the key to most of the literature on shrinkage estimation in the general spherically symmetric case is to find some generalization of ( or substitution for ) stein s lemma to evaluate ( or bound ) the cross product term @xmath73 $ ] . we indicate two useful techniques below . brandwein and strawderman ( @xcite ) extended the results of stein ( @xcite ) to spherically symmetric distributions for estimators of the form @xmath74 . the following two preliminary lemmas are necessary to prove the result in theorem [ thmm31 ] . [ lem32 ] let @xmath2 have a distribution that is spherically symmetric about @xmath75 . then @xmath76 \\ & & \quad = p^{-1}r^2\mathrm{ave}_{b(r,\theta ) } \nabla'g(x),\end{aligned}\ ] ] provided @xmath77 is weakly differentiable . notation for this lemma : @xmath78and @xmath79 are , respectively , the ( surface of the ) sphere and ( solid ) ball , of radius @xmath80 centered at @xmath0 . note also that @xmath81 is the unit outward normal vector at @xmath2 on @xmath78 . also @xmath82 is the area measure on @xmath78 , while @xmath83 and @xmath84 denote area and volume , respectively . since the conditional distribution of @xmath85 given @xmath86 is uniform on the sphere of radius @xmath80 , it follows that @xmath87 \\ & & \quad= \operatorname{ave}_{s(r,\theta ) } \{(x - \theta)'g(x)\ } \\ & & \quad= \frac{r}{{a(s(r,\theta ) ) } } \oint_{s(r,\theta ) } \frac{(x - \theta)'g(x)}{r } \,d\sigma(x ) \\ & & \quad = \frac{r}{a(s(r,\theta ) ) } \int_{b(r,\theta ) } \nabla'g(x ) \,dx\\ & & \hspace*{86pt}\qquad { } \biggl(\mbox{since } \frac{v(b(r,\theta))}{a(s(r,\theta ) ) } = r / p\biggr)\\ & & \quad = \frac{r^2 } { pv(b(r,\theta ) ) } \int_{b(r,\theta ) } { \nabla'g(x ) } \,dx\\ & & \hspace*{119pt}\qquad { } ( \mbox{by stokes ' theorem } ) \\ & & \quad = p^{-1}r^2 \operatorname{ave}_{b(r,\theta ) } \nabla'g(x).\end{aligned}\ ] ] the following result is basic to the study of superharmonic functions and is well known ( see , e.g. , du plessis , @xcite , page 54 ) . [ lem33 ] let @xmath88 be superharmonic on @xmath89 , [ i.e. , @xmath90 , then @xmath91 . consider , now , an estimator of the general form @xmath92 , where @xmath4 is a scalar , and @xmath93 maps . [ thmm31 ] let @xmath2 have a distribution that is spherically symmetric about @xmath0 . assume the following : 1 . @xmath94 , [ as1thmm31 ] 2 . @xmath95 is superharmonic , @xmath96 $ ] is nonincreasing in @xmath80 for each @xmath0 , where @xmath97 has a uniform distribution on @xmath98 , [ as2thmm31 ] 3 . @xmath99\}$].[as3thmm31 ] then @xmath92 is minimax with respect to quadratic loss , provided @xmath69 is weakly differentiable and all expectations are finite . @xmath100 \bigr ] \\ & & \quad \le e\bigl [ e_\theta [ - 2a^2 h(x)\\ & & \hspace*{28pt}\qquad { } + 2a(x - \theta)'g(x ) & & \quad = e\bigl [ e_\theta [ - 2a^2 h(x ) | \|x - \theta\|^2 = r^2 ] \\ & & \quad\qquad { } + 2ae[\{r^2/p\ } \operatorname{ave}_{b(r,\theta ) } \nabla'g(x ) | r^2 ] \bigr ] \\ & & \quad \le e\bigl [ e_\theta [ - 2a^2 h(x ) | \|x - \theta\|^2 = r^2 ] \\ & & \quad\hspace*{10pt}\qquad { } + 2ae_\theta [ \{r^2/p\}e_\theta h(w ) | r^2 ] \bigr ] \\ & & \quad \le e\bigl [ e_\theta [ - 2a^2 h(w ) | r^2 ] \\ & & \quad\qquad{}+ 2ae_\theta [ \{r^2/p\}e_\theta h(w ) | r^2 ] \bigr ] \\ & & \hspace*{142pt}\qquad{}(\mbox{by lemma \ref{lem33 } } ) \\ & & \quad = 2ae\bigl [ e_\theta [ r^2 h(w ) | r^2 ] ( -a / r^2 + 1/p ) \bigr ] \\ & & \quad = 2ae [ e_\theta [ r^2 h(w ) | r^2 ] ] e[-a / r^2 + 1/p ] \\ & & \quad \le0\end{aligned}\ ] ] by the covariance inequality since @xmath101 $ ] is nonincreasing and @xmath102 is increasing and since @xmath103 . [ exam31 ] james stein estimators @xmath104 $ ] : in this case both @xmath105 and @xmath106 are equal to @xmath107 . conditions 1 and 2 of theorem [ thmm31 ] are satisfied for @xmath108 , provided @xmath109 since @xmath110 is superharmonic if @xmath109 , and since @xmath111 = e_{\theta / r } [ 1/\allowbreak\|x\|^2]$ ] is increasing by anderson s theorem . hence , by condition 3 , for any spherically symmetric distribution , the james stein estimator @xmath112 is minimax for @xmath113\}$ ] and @xmath114 . the domination over @xmath2 is strict for @xmath115\}$ ] , and also for @xmath116\}$ ] , provided the distribution is not normal . baranchik ( @xcite ) , for the normal case , considered estimators of the form @xmath117 under certain conditions on @xmath118 . under the assumption that @xmath118 is monotone nondecreasing , bounded between @xmath119 and @xmath120 , and concave , theorem [ thmm31 ] applies to these estimators as well , and establishes minimaxity for @xmath121\}$ ] and for @xmath114 . we note in passing that the results in this subsection hold for an arbitrary spherically symmetric distribution with or without a density . the calculations rely only on the distribution of @xmath2 conditional on @xmath122 , and , of course , finiteness of @xmath123 $ ] and @xmath124 $ ] . [ subsec32 ] berger ( @xcite ) gave a useful expression for the risk of a james stein estimator which is easily generalized to the case of a general estimator , provided the spherically symmetric distribution has a density @xmath125 . some form of this generalization ( and extensions to unknown scale case and the elliptically symmetric case ) has been used by several authors , including fourdrinier , strawderman and wells ( @xcite ) , fourdrinier , kortbi and strawderman ( @xcite ) , fourdrinier and strawderman ( @xcite ) , maruyama ( @xcite ) and kubokawa and srivastava ( @xcite ) , among others . [ lem34 ] suppose @xmath126 , and let @xmath127 and @xmath128 . then @xmath129\\ & & \qquad{}+ e_\theta [ \| { g(x ) } \|^2 + 2q(\| x-\theta\|^2 ) \nabla'g(x)].\end{aligned}\ ] ] the lemma follows immediately with the following identity for the cross product term : @xmath130\\ & & \quad= \int_{r^p } ( x - \theta)'g(x)f(\|x - \theta\|^2)\,dx \\ & & \quad= \int_{r^p } g(x)'\nabla f(\|x - \theta\|^2 ) \,dx\\ & & \quad= \int_{r^p } \nabla'g(x)f(\|x - \theta\|^2 ) \,dx \vadjust{\goodbreak}\\ & & \hspace*{96pt}\qquad(\mbox{by green 's theorem } ) \\ & & \quad= e [ q(\|x - \theta\|^2)\nabla'g(x)].\end{aligned}\ ] ] berger ( @xcite ) , maruyama ( @xcite ) and fourdrinier , kortbi and strawderman ( @xcite ) used the above result for distributions for which @xmath131 is bounded below by a positive constant . in this case , the next result follows immediately from lemma [ lem34 ] . [ thmm32 ] @xmath132suppose @xmath133 , and that @xmath134 . then the estimator @xmath135 dominates @xmath2 provided @xmath136 for all @xmath137 . as noted by berger ( @xcite ) , if @xmath41 is a scale mixture of normals , then @xmath131 is bounded below . to see this , note that if @xmath138and @xmath139 , then @xmath140 . similarly , @xmath141 hence , @xmath142 \ge e_0 [ v ] = \frac{\int_0^\infty v^{1 - p/2 } g(v)\,dv } { \int_0^\infty v^ { - p/2 } g(v)\,dv } \\ & = & \frac{e[v^{1-p/2}]}{e[v^{-p/2 } ] } = c > 0,\end{aligned}\ ] ] where @xmath143 denotes expectation with respect to the density proportional to @xmath144 . the inequality follows since the family has monotone likelihood ratio in @xmath145 . hence , for the james stein class @xmath7 , this result gives dominance over @xmath2 for @xmath146}{e[v^ { - p/2 } ] } \le0\ ] ] or @xmath147}{e[v^ { - p/2 } ] } .\ ] ] this bound on the shrinkage constant , @xmath4 , compares poorly with that obtained by strawderman ( @xcite ) , @xmath148 $ ] , which may be obtained by using stein s lemma conditional on @xmath149 and the fact that @xmath150 $ ] is monotone nondecreasing in @xmath149 . note that , again by monotone likelihood ratio properties ( or the covariance inequality),@xmath151)^{-1 } > e[v^{1 - p/2 } ] /e[v^ { - p/2 } ] $ ] . it is therefore somewhat surprising that maruyama ( @xcite ) and fourdrinier , kortbi and strawderman ( @xcite ) were able to use theorem [ thmm32 ] , applied to baranchik - type estimators , to obtain generalized and proper bayes minimax estimators . without going into details , the advantage of the cruder bound is that it requires only that @xmath152 be monotone , while strawderman s result for mixtures of normal distributions also requires that @xmath153 be monotone decreasing . other applications of lemma [ lem34 ] give refinedbounds on the shrinkage constant in the james stein or baranchik estimator depending on monotonicity properties of @xmath131 . typically , additional conditions are required on the function @xmath152 as well . see , for example , brandwein , ralescu and strawderman ( @xcite ) ( the calculations in that paper are somewhat different than those in this section , the basic idea is quite similar ) . applications of the risk expression in lemma [ lem34 ] are complicated relative to those in the normal case using stein s lemma , in that the mean vector , @xmath0 , remains to complicate matters through the function @xmath154 . it is both surprising and interesting that matters become essentially simpler ( in a certain sense ) when the scale parameter is unknown , but a residual vector is available . we investigate this phenomenon in the next section . [ sec4 ] in this section we study the model @xmath155 , where @xmath156 , and @xmath157 . the classical example of this model is , of course , the normal model @xmath158 . however , a variety of other models have proven useful . perhaps the most important alternatives to the normal model in practice and in theory are the generalized distributions @xmath159 or , more generally , scale mixture of normals of the form @xmath160 these models preserve the spherical symmetryabout the mean vector and , hence , the covariance matrix is a multiple of the identity . thus , the coordinates are uncorrelated , but they are not independent except for the case of the normal model . we look ( primarily ) at estimators of the form @xmath161 . the main result may be interpreted as follows : if , when @xmath162 ( @xmath40 known ) , the estimator @xmath163 dominates @xmath2 , then , under the model @xmath164 , the estimator @xmath165 dominates @xmath2 . that is , substituting the estimator @xmath166 for @xmath40 preserves domination uniformly for all parameters @xmath167 and ( somewhat astonishingly ) simultaneously for all distributions , @xmath41 . note that , interestingly , @xmath166 is the minimum risk equivariant estimator of @xmath40 in the normal case under the usual invariant loss . this wonderful result is due to cellier and fourdrinier ( @xcite ) . we refer the reader to their paper for the original proof based on stokes theorem applied to the distribution of @xmath2 conditional on @xmath168 . one interesting aspect of that proof is that even if the original distribution has no density , the conditional distribution of @xmath2 does have a density for all @xmath169 . we will approach the above result from two different directions . the first approach is essentially an extension of lemma [ lem34 ] . as in that case , the resulting expression for the risk still involves both the data and @xmath0 inside the expectation , but the function @xmath170 is a common factor . this allows the treatment of the remaining terms as if they are an unbiased estimate of the risk difference . the second approach is due to fourdrinier , strawderman and wells ( @xcite ) , and is attractive because it is essentially statistical in nature , depending on completeness and sufficiency . it may be argued also that this approach is somewhat more general in that it may be useful even when the function @xmath77 is not necessarily weakly differentiable . in this case an unbiased estimator of the risk difference is obtained which agrees with that in cellier and fourdrinier ( @xcite ) . this is in contrast to the above method whereby the expression for the risk difference still has a factor @xmath171 inside the expectation . @xmath172technically , our use of the term `` unknown scale '' is somewhat misleading in that the scale parameter may , in fact , be known . we typically think of @xmath41 as being a known density , which implies that the scale is known as well . it may have been preferable to write the density as @xmath173 , emphasizing the unknown scale parameter . this is more in keeping with the usual canonical form of the general linear model with spherically symmetric errors . what is of fundamental importance is the presence of the residual vector , @xmath174 , in allowing uniform domination over the estimator @xmath2 simultaneously for the entire class of spherically symmetric distributions . since the suppression of the scale parameter makes notation a bit simpler , we will , for the most part , use the above notation in this section . additionally , we continue to use the un - normalized loss , @xmath175 , and state results in terms of dominance over @xmath2 instead of minimaxity , since the minimax risk is infinite . in order to speak meaningfully of minimaxity in the unknown scale case , we should use a normalized version of the loss , such as @xmath176 . [ lem41 ] @xmath132suppose @xmath177 , where @xmath178 , @xmath179 . then , provided @xmath180 is weakly differentiable in each coordinate : 1 . @xmath181 = e_\theta[\|u\|^2 \nabla'_x g(x,\break \|u\|^2 ) q(\|x - \theta\|^2 + \|u\|^2 ) ] $ ] . [ 1lem41 ] 2 . @xmath182=e_\theta[h(x,\|u\| ^2)\cdot q(\|x - \theta\|^2 + \|u\|^2 ) ] $ ] , [ 2lem41 ] where @xmath183 and @xmath184 the proof of part 1 is essentially the same as the proof of lemma [ lem34 ] , holding @xmath174 fixed throughout . the same is true of part 2 , where the roles of @xmath2 and @xmath174 are reversed and one notes that @xmath185 which is given by ( [ eq41 ] ) , and , hence , @xmath186\\ & & \quad= e_\theta [ ( \|u\|^2 u)'u \|g(x,\|u\|^2)\|^2 ] \\ & & \quad = e_\theta [ \nabla'_u \{(\|u\|^2 u)\|g(x,\|u\|^2 ) \|^2 \ } \\ & & \hspace*{44pt}\qquad{}\cdot q(\|x - \theta\|^2 + \|u\|^2 ) ] \\ & & \quad= e_\theta [ h(x,\|u\|^2 ) q(\|x - \theta\|^2 + \|u\|^2 ) ] .\end{aligned}\ ] ] one version of the main result for estimators of the form @xmath187 is the following theorem . [ thmm41 ] suppose ( x , u ) is as in lemma [ lem41 ] . then : 1 . the risk of an estimator @xmath188 is given by @xmath189\\ & & \qquad{}+ e_\theta \biggl[\frac{{\|u\|^2 } } { { k + 2}}\{\|g(x)\|^2 + 2\nabla'g(x)\ } \\ & & \hspace*{52pt}\qquad{}\cdot q(\|x - \theta\|^2 + \|u\|^2 ) \biggr],\end{aligned}\ ] ] 2 . @xmath190 dominates @xmath2 provided @xmath191 . note that @xmath192 & & \quad= e_\theta[\|x - \theta\|^2 ] \\[-1pt ] & & \qquad{}+ e_\theta\biggl [ \frac{{\|u\|^4 } } { { ( k + 2)^2 } } \|g(x)\|^2 \\[-1pt ] & & \hspace*{30pt}\qquad{}+ 2\frac{{\|u\|^2 } } { { k + 2}}(x - \theta)'g(x ) \biggr ] \\[-1pt ] & & \quad= e_\theta [ \|x - \theta\|^2 ] \\[-1pt ] & & \qquad{}+ e_\theta \biggl [ \{\|g(x)\|^2 + 2\nabla'g(x)\ } \\[-1pt ] & & \hspace*{28pt}\qquad{}\cdot \frac{\|u\|^2q(\|x - \theta\|^2 + \|u\|^2)}{k+2 } \biggr]\end{aligned}\ ] ] by successive application of parts 1 and 2 of lemma [ lem41 ] . baranchik - type estimators : suppose the estimator is given by @xmath193 , where @xmath152 is nondecreasing , and @xmath194 , then for @xmath195 the estimator dominates @xmath2 simultaneously for all spherically symmetric distributions for which the risk of @xmath2 is finite . this follows since , if @xmath196,then @xmath197 james stein estimators : if@xmath198 , the baranchik estimator is a james stein estimator , and , since @xmath199 , the risk is given by @xmath200 + \frac{a^2 - 2a(p - 2)}{k+2}\\[1pt ] & & \hspace*{64pt}\quad{}\cdot e\biggl [ \frac{\|u\|^2}{\|x\|^2 } q(\|x - \theta\|^2 + \|u\|^2 ) \biggr].\end{aligned}\ ] ] just as in the normal case , @xmath201 is the uniformly best choice to minimize the risk . but here it is the uniformly best choice for every distribution . hence , the estimator @xmath202 is uniformly best , simultaneously for all spherically symmetric distributions among the class of james stein estimators ! a more refined version of theorem [ thmm41 ] which uses the full power of lemma [ lem41 ] is proved in the same way . we give it for completeness and since it is useful in the study of risks of bayes estimators . [ thmm42 ] suppose @xmath203 is as in lemma [ lem41 ] . then , under suitable smoothness conditions on @xmath69 : 1 . the risk of an estimator @xmath204 is given by @xmath205 & & \quad = e_\theta [ \|x - \theta\|^2 ] \\[1pt ] & & \qquad{}+ e_\theta [ \{(k+2)^{-1 } \|u\|^2 \|g(x,\|u\|^2 ) \|^2 \\[1pt ] & & \hspace*{32pt}\qquad{}+ 2\nabla'_x g(x,\|u\|^2 ) \\[1pt ] & & \hspace*{32pt}\qquad { } + 2(k+2)^{-2}\|u\|^4(\partial/\partial\|u\|^2)\\[1pt ] & & \hspace*{91pt}\qquad{}\cdot\| g(x,\|u\|^2 ) \|^2 \ } \\[1pt ] & & \hspace*{68pt}\qquad\cdot { } q(\|x-\theta\|^2 + \|u\|^2)],\end{aligned}\ ] ] 2 . @xmath206 dominates @xmath2 provided @xmath207 [ cor41 ] suppose @xmath208 . then @xmath209 dominates @xmath2 provided : 1 . @xmath210 and 2 . @xmath118 is nondecreasing . the result follows from theorem [ thmm42 ] by a straightforward calculation . we largely follow fourdrinier , strawderman and wells ( @xcite ) in this subsection . the nature of the conclusions for estimators is essentially as in theorem [ thmm41 ] , but the result is closer in spirit to the result of cellier and fourdrinier ( @xcite ) in that we obtain an unbiased estimator of risk difference ( from @xmath2 ) instead of the expression in theorem [ thmm41 ] where the function @xmath211 , which depends on @xmath0 , intervenes . the following lemma is the key to this development . [ lem42 ] let @xmath212 , where @xmath156 and @xmath157 . suppo - se @xmath69 and @xmath213 are such that when @xmath214,@xmath65 = e_\theta [ h(x)]$ ] . then , for @xmath203 as above , @xmath215\\ & & \quad= \{1/(k+2)\}e_\theta [ \|u\|^4 h(x)],\end{aligned}\ ] ] provided the expectations exist . typically , of course , @xmath88 is the divergence of @xmath77 , and , in all cases known to us , this remains essentially true . we choose this form of expressing the lemma because in certain instances of restricted parameter spaces the lemma applies even though the function @xmath69 may not be weakly differentiable , but the equality still holds for @xmath216 and @xmath217 , where @xmath218 is the indicator function of a set @xmath219 . proof of lemma [ lem42 ] suppose first , that the distribution of @xmath203 is @xmath220 and that @xmath0 is considered known . then by the independence of @xmath2 and @xmath174 we have by assumption that @xmath221\\ & & \quad = e_\theta [ ( 1/k)\| u \|^2 ( x - \theta)'g(x ) ] \\ & & \quad= e_\theta [ \{k(k+2)\}^{-1 } \| u \|^4 h(x ) ] .\end{aligned}\ ] ] hence , the claimed result of the theorem is true for the normal case . now use the fact that in the normal case ( for @xmath0 known ) , @xmath222 is a complete sufficient statistic . so it must be that @xmath223\\ & & \quad= e_\theta \biggl [ \frac{\| u \|^4 h(x ) } { k + 2 } \big| \| x - \theta\|^2 + \| u \|^2 \biggr]\end{aligned}\ ] ] for all @xmath224 except on a set of measure @xmath119 , since each function of @xmath225 has the same expected value . actually , it can be shown that these conditional expectations are continuous in @xmath80 and , hence , they agree for all @xmath80 ( see fourdrinier , strawderman and wells , @xcite ) . but the distribution of @xmath203 conditional on @xmath226 is uniform on the sphere centered at @xmath227 of radius @xmath80 , which is the same as the conditional distribution of @xmath203 conditional on @xmath226 for any spherically symmetric distribution . hence , the equality which holds for the normal distribution holds for all distributions @xmath41 . lemma [ lem42 ] immediately gives the following unbiased estimator of risk difference and a condition for dominating @xmath2 for estimators of the form @xmath228 . suppose @xmath229 and @xmath88 are as in lemma [ lem42 ] . then , for the estimator @xmath230 : 1 . the risk difference is given by @xmath231 \\ & & \quad= e_\theta \biggl [ \frac{\|u\|^4 } { ( k + 2)^2 } \{\|g(x)\|^2 + 2\nabla'g(x)\ } \biggr],\end{aligned}\ ] ] 2 . @xmath14 beats @xmath2 provided @xmath232 , with strict inequality on a set of positive measure , and provided all expectations are finite . we consider a simple version of the general restricted parameter space problem which illustrates what types of results can be obtained . suppose @xmath233 is distributed as in theorem [ thmm41 ] but it is known that @xmath234 , @xmath235 , that is , @xmath236 the first orthant . what follows can be generalized to the case where @xmath0 is restricted to a polyhedral cone , and more generally a smooth cone . the material in this section is adapted from fourdrinier , strawderman and wells ( @xcite ) . in the normal case , the mle of @xmath0 subject to the restriction that @xmath237 is @xmath238 , where the @xmath239th component is @xmath240 if @xmath241 and @xmath119 otherwise . here , as in the case of the more general restriction to a convex cone , the mle is the projection of @xmath2 onto the restricted cone . chang ( @xcite ) considered domination of the mle of @xmath0 when @xmath2 has a @xmath242 distribution and @xmath236 via certain stein - type shrinkage estimators . sengupta and sen ( @xcite ) extended chang s results to stein - type shrinkage estimators of the form @xmath243 , where @xmath244 is nondecreasing , and @xmath245 , and where @xmath246 is the ( random ) number of positive components of @xmath2 . hence , shrinkage occurs only when @xmath246 , the number of positive components of @xmath2 , is at least @xmath247 and the amount of shrinkage is governed by the sum of squares of the positive components . a similar result holds if @xmath0 is restricted to a general polyhedral cone where @xmath238 is replaced by the projection of @xmath2 onto the cone and @xmath246 is defined to be the dimension of the face onto which @xmath2 is projected . we choose the simple polyhedral cone @xmath237 because it will be reasonably clear that some version of the stein lemma [ lem31 ] applies in the normal case . we first indicate a convenient , but complicated looking , alternate representation of an estimator of the above form in this case . denote the @xmath248 orthants of @xmath22 , by @xmath249 , and let @xmath250 be @xmath251 . then we may rewrite ( a slightly more general version of ) the above estimator as @xmath252 where @xmath253 is the linear projection of @xmath2 onto @xmath254 , where @xmath254 is the @xmath246-dimensional face of @xmath255 onto which @xmath256 is projected . note that if @xmath257 , @xmath258 , the estimator is just the mle . [ lem51 ] suppose @xmath259 , and leteach @xmath260 be smooth and bounded . then : 1 . for each @xmath256 , @xmath261 is weakly differentiable in @xmath137 . [ 1lem51 ] 2 . further , @xmath262 \\ & & \quad = e_\theta \biggl[\biggl\ { \frac{(s - 2)r_i ( \|p_i ( x)\|^2 ) } { \|p_i ( x)\|^2 } \\ & & \hspace*{38pt}\qquad{}+ 2r'_i ( \|p_i ( x)\|^2 ) \biggr\ } i_{o_i } ( x)\biggr],\end{aligned}\ ] ] provided expectations exist . [ 2lem51 ] 3 . @xmath263 as given above dominates themle @xmath238 , provided @xmath264 is nondecreasing and bounded between 0 and @xmath265 . [ 3lem51 ] weak differentiability in part 1 follows since the function is smooth away from the boundary of @xmath256 and is continuous on the boundary except at the origin . part 2 follows from stein s lemma [ lem31 ] and the fact that ( essentially ) @xmath266 , since @xmath267 of the coordinates are @xmath119 . part 3 follows by stein s lemma [ lem31 ] as in proposition [ prop31 ] applied to each orthant . we omit the details . the reader is referred to sengupta and sen ( @xcite ) or fourdrinier , strawderman and wells ( @xcite ) for details in the more general case of a polyhedral cone . next , essentially applying lemma [ lem42 ] to each orthant and using lemma [ lem51 ] we have the following generalization to the case of a general spherically symmetric distribution . let @xmath268 where @xmath156 and @xmath157 and suppose that @xmath237 . then @xmath269 dominates the @xmath238 , provided @xmath264 is nondecreasing and bounded between 0 and @xmath265 . there have been advancements in bayes estimation of location vectors in several directions in the past 15 years . perhaps the most important advancements have come in the computational area , particularly markov chain monte carlo ( mcmc ) methods . we do not cover these developments in this review . admissibility and inadmissibility of ( generalized ) bayes estimators in the normal case with known scale parameter was considered in berger and strawderman ( @xcite ) and in berger , strawderman andtang ( @xcite ) where brown s ( @xcite ) condition for admissibility ( and inadmissibility ) was applied for a variety of hierarchical bayes models . maruyama and takemura ( @xcite ) also give admissibility results for the general spherically symmetric case . at least for spherically symmetric priors , the conditions are , essentially , that priors with tails no greater than@xmath270 give admissible procedures . fourdrinier , strawderman and wells ( @xcite ) , using stein s ( @xcite ) results ( especially proposition [ prop31 ] above , and its corollaries ) , give classes of minimax bayes ( and generalized bayes ) estimators which include scaled multivariate-@xmath145 priors under certain conditions . berger and robert ( @xcite ) give classes of priors leading to minimax estimators . kubokawa and strawderman ( @xcite ) give classes of priors in the setup of berger and strawderman ( @xcite ) that lead to admissible minimax estimators . maruyama ( @xcite ) and fourdrinier , kortbi and strawderman ( @xcite ) , in the scale mixture of normal case , find bayes and generalized bayes minimax estimators , generalizing results of strawderman ( @xcite ) . as mentioned in section [ sec3 ] , these results use either berger s ( @xcite ) ( a version of which is given in theorem [ thmm32 ] ) or strawderman s ( @xcite ) result for mixtures of normal distributions . fourdrinier and strawderman ( @xcite ) proved minimaxity of generalized bayes estimators corresponding to certain harmonic priors for classes of spherically symmetric sampling distributions which are not necessarily mixtures of normals . the results in this paper are not based directly on the discussion of section [ sec3 ] but are somewhat more closely related in spirit to the approach of stein ( @xcite ) . we give below an intriguing result of maruyama ( @xcite ) for the unknown scale case ( see also maruyama and strawderman , @xcite ) , which is related to the ( distributional ) robustness of stein estimators in the unknown scale case treated in section [ sec4 ] . first , we give a lemma which will aid in the development of the main result . [ lem61 ] suppose @xmath271 , the ( location - scale invariant ) loss is given by @xmath272 and the prior distribution on @xmath273 is of the form @xmath274 . then provided all integrals exist , the generalizedbayes estimator does not depend on @xmath41 . @xmath275 & & \quad= e[\theta\eta|x , u]/e[\eta|x , u ] \\[-2pt ] & & \quad = \biggl[\int_{r^p } \int_0^\infty \theta\eta^{(p + k)/2 + b+1}\\[-2pt ] & & \hspace*{38pt}\qquad{}\cdot f(\eta\{\|x - \theta\|^2 + \|u\|^2 \})\rho(\theta)\,d\eta \,d\theta\biggr]\\[-2pt ] & & \qquad{}\cdot \biggl[\int_{r^p } \int_0^\infty \eta^{(p + k)/2 + b+1 } \\[-2pt ] & & \hspace*{48pt}\qquad{}\cdot f(\eta\{\|x - \theta\|^2\\[-2pt ] & & \hspace*{82pt}\qquad{}+ \|u\|^2 \})\rho(\theta)\,d\eta \,d\theta\biggr]^{-1}.\vspace*{-2pt}\end{aligned}\ ] ] making the change of variables @xmath276 , we have @xmath277 & & \quad= \biggl[\int_{r^p } \theta(\|x - \theta\|^2 + \|u\|^2 ) ^ { - ( p + k)/2 + b + 2}\\[-2pt ] & & \hspace*{9pt}\quad\qquad{}\cdot\rho(\theta)\,d\theta\int_0^\infty w^{(p + k)/2 + b + 1 } f(w)\,dw \biggr]\\[-2pt ] & & \qquad{}\cdot\biggl[\int_{r^p } ( \|x - \theta\|^2 + \|u\|^2 ) ^ { - ( p + k)/2 + b + 2}\\[-2pt ] & & \hspace*{16pt}\quad\qquad{}\cdot\rho(\theta)\,d\theta\int_0^\infty w^{(p + k)/2 + b + 1 } f(w)\,dw \biggr]^{-1 } \\[-2pt ] & & \quad = \frac{\int_{r^p } \theta(\|x - \theta\|^2 + \|u\|^2 ) ^ { - ( p + k)/2 + b + 2 } \rho(\theta)\,d\theta } { \int_{r^p } ( \|x - \theta\|^2 + \|u\|^2 ) ^ { - ( p + k)/2 + b + 2 } \rho(\theta)\,d\theta}.\vspace*{-3pt}\end{aligned}\ ] ] hence , for ( generalized ) priors of the above form , the bayes estimator is independent of the sampling distribution provided the bayes estimator exists;thus , they may be calculated for the most convenient density , which is typically the normal . our next lemma calculates the generalized bayes estimator for a normal sampling density and for a class of priors for which @xmath278 is a scale mixture of normals . [ lem62 ] suppose the distribution of @xmath233 is normal with variance @xmath279 . suppose also that the conditional distribution of @xmath0 given @xmath280 and @xmath281 is normal with mean 0 and covariance @xmath282 , and the density of @xmath283 is proportional to @xmath284 , where @xmath285 . 1 . then the bayes estimator is given by @xmath286 , where @xmath287 and @xmath288 is given by @xmath289 & & \hspace*{26pt}{}\cdot(1 + w\lambda)^ { - k/2 - a - b/2 - 2 } \,d\lambda\biggr ] \nonumber \\[-10pt ] \\[-10pt ] \nonumber & & { } \cdot\biggl[\int_0 ^ 1 \lambda^{b/2 - 1 } ( 1 - \lambda)^{p/2 - b/2 - 1 } \\[-2pt ] & & \hspace*{26pt}{}\cdot ( 1 + w\lambda)^ { - k/2 - a - b/2 - 2 } \,d \lambda\biggr]^{-1}.\nonumber\end{aligned}\ ] ] this is well defined for @xmath290 , and @xmath291 . [ 1lem62 ] 2 . furthermore , this estimator is generalized bayes corresponding to the generalized prior proportional to @xmath292 , for any spherically symmetric density @xmath41 for which @xmath293 . [ 2lem62 ] part 1 . in the normal case , @xmath294}{e[\eta|x , u ] } \\ & = & x - \frac{\nabla_x m(x , u)}{2(\partial/\partial\|u\|^2 ) m(x , u)},\vadjust{\goodbreak}\end{aligned}\ ] ] where the marginal @xmath295 is proportional to @xmath296 & & \hspace*{32pt}\qquad { } \cdot\exp ( - \eta\{\|x - \theta\|^2 + \|u\|^2 \}/2 ) \\[-2pt ] & & \hspace*{32pt}\qquad{}\cdot \exp\biggl ( - \frac{\eta\lambda\|\theta\|^2}{2(1 - \lambda)}\biggr ) \,d\theta\ , d\eta\ , d\lambda\\[-2pt ] & & \quad = k ' \int_0 ^ 1 \int_0^\infty \eta^{b/2 + k/2 + a } \lambda^{b/2 - 1 } ( 1 - \lambda)^{p/2 - b/2 - 1}\\[-2pt ] & & \hspace*{70pt}{}\cdot\exp ( - \eta\{\lambda\|x\|^2 + \|u\|^2 \}/2 ) \,d\eta\ , d\lambda\\[-2pt ] & & \quad= k \int_0 ^ 1 ( \lambda\|x\|^2 + \|u\|^2 ) ^{- b/2-k/2-a-1 } \lambda^{b/2 - 1 } \\[-2pt ] & & \hspace*{46pt}{}\cdot(1 - \lambda)^{p/2 - b/2 - 1}\ , d\lambda.\end{aligned}\ ] ] hence , we may express the bayes estimator as @xmath297 , where @xmath298 & & \hspace*{34pt}{}\cdot\lambda^{b/2- 1 } ( 1 - \lambda)^{p/2-b/2 - 1 } \,d\lambda\biggr]\\[-3pt ] & & { } \cdot\biggl[- 2(d / d\|u\|^2)\\[-3pt ] & & \quad{}\cdot\int_0 ^ 1 ( \lambda\|x\|^2 + \|u\|^2)^{-b/2-k/2-a-1}\\[-3pt ] & & \hspace*{36pt}{}\cdot\lambda^{b/2 - 1 } ( 1-\lambda)^{p/2-b/2 - 1 } \,d\lambda\biggr]^{-1 } \\[-3pt ] & = & -x\biggl[\int_0 ^ 1 ( \lambda\|x\|^2 + \|u\|^2)^{-b/2-k/2-a-2 } \\[-3pt ] & & \hspace*{48pt}{}\cdot\lambda^{b/2 } ( 1-\lambda)^{p/2-b/2 - 1}\ , d\lambda\biggr]\\[-3pt ] & & { } \cdot \biggl[\int_0 ^ 1 ( \lambda\|x\|^2 + \|u\|^2 ) ^{-b/2-k/2-a-2 } \\[-3pt ] & & \hspace*{18pt}\quad{}\cdot\lambda^{b/2 - 1}(1 - \lambda)^{p/2-b/2 - 1 } \,d\lambda\biggr]^{-1 } \\[-3pt ] & = & -x\biggl[\int_0 ^ 1(\lambda w + 1)^{-b/2-k/2-a-2}\\[-3pt ] & & \hspace*{33pt}{}\cdot\lambda^{b/2 } ( 1-\lambda)^{p/2-b/2 - 1}\,d\lambda\biggr]\\[-3pt ] & & { } \cdot\biggl[\int_0 ^ 1 ( \lambda w + 1)^{-b/2-k/2-a-2}\\[-3pt ] & & \hspace*{26pt}{}\cdot\lambda^{b/2 - 1 } ( 1-\lambda)^{p/2-b/2 - 1 } \,d\lambda\biggr]^{-1 } \\[-3pt ] & = & - \frac{x}{w}r(w).\end{aligned}\ ] ] part a straightforward calculation shows that the unconditional density of @xmath273 is proportional to @xmath299 . hence , part 2 follows from lemma [ lem61 ] . the following lemma gives properties of @xmath288 . [ lem63 ] suppose @xmath300 and that @xmath301 . then , ( 1 ) @xmath288 is nondecreasing , and ( 2 ) @xmath302 . by a change of variables , letting @xmath303 in ( [ eq61 ] ) , then @xmath304\\ & & { } \cdot\biggl[\int_0^w { ( v + 1 ) } ^ { - b/2 - k/2 - a - 2}\\ & & \hspace*{28pt}{}\cdot v^{b/2 - 1 } ( 1 - v /w)^{p/2 - b/2 - 1 } \,dv\biggr]^{-1}.\end{aligned}\ ] ] so , we may rewrite @xmath288 as @xmath305 $ ] , where @xmath306 has density proportional to @xmath307}(v)$ ] . this density has increasing monotone likelihood ratio in @xmath308 as long as @xmath309 . hence , part 1 follows . the conditions of the lemma allow interchange of limit and integration in both numerator and denominator of @xmath288 as @xmath310 . hence , @xmath311\\ & = & \frac{\operatorname{beta}(b/2 + 1,k/2 + a + 1)}{\operatorname{beta}(b/2,k/2 + a + 2 ) } \\ & = & \frac{b/2}{k/2 + a + 1}.\end{aligned}\ ] ] combining lemmas [ lem61][lem63 ] with corollary [ cor41 ] gives as the main result a class of estimators which are generalized bayes and minimax simultaneously for the entire class of spherically symmetric sampling distributions ( subject to integrability conditions ) . suppose that the distributionof @xmath203 and the loss function are as in lemma [ lem61 ] , and that the prior distribution is as in lemmas [ lem62 ] and [ lem63 ] with a satisfying @xmath312 , and with @xmath300 . then the corresponding generalized bayes estimator is minimax for all densities @xmath41 such that the @xmath313th moment of the distribution of @xmath203 is finite , that is , @xmath314 . we note that the above finiteness condition,@xmath315 , is equivalent to the finiteness condition , @xmath316 in lemma [ lem62 ] . this paper has reviewed some of the developments in shrinkage estimation of mean vectors for spherically symmetric distributions , mainly since the review paper of brandwein and strawderman ( @xcite ) . other papers in this volume review other aspects of the enormous literature generated by or associated with stein s stunning inadmissibility result of 1956 . most of the developments we have covered are , or can be viewed as , outgrowths of stein s papers of 1973 and 1981 , and , in particular , of stein s lemma which gives ( an incredibly useful ) alternative expression for the cross product term in the quadratic risk function . among the topics which we have not covered is the closely related literature for elliptically symmetric distributions ( see , e.g. , kubokawa and srivastava , @xcite , and fourdrinier , strawderman and wells , @xcite , and the references therein ) . we also have not included a discussion of hartigan s ( @xcite ) beautiful result that the ( generalized or proper ) bayes estimator of a normal mean vector with respect to the uniform prior on any convex set in @xmath22 dominates @xmath2 for squared error loss . nor have we discussed the very useful and pretty development of the kubokawa ( @xcite ) ierd method for finding improved estimators , and , in particular , for dominating james stein estimators ( see also marchand and strawderman , @xcite , for some discussion of these last two topics ) . we nonetheless hope we have provided some intuition for , and given a flavor of the developments and rich literature in the area of improved estimators for spherically symmetric distributions . the impact of stein s beautiful 1956 result and his innovative development of the techniques in the 1973 and 1981 papers have inspired many researchers , fueled an enormous literature on the subject , led to a deeper understanding of theoretical and practical aspects of `` sharing strength '' across related studies , and greatly enriched the field of statistics . even some of the early ( and later ) heated discussions of the theoretical and practical aspects of `` sharing strength '' across unrelated studies have had an ultimately positive impact on the development of hierarchical models and computational tools for their analysis . we are very pleased to have been asked to contribute to this volume commemorating fifty years of development of one of the most profound results in the statistical literature in the last half of the 20th century.=-1

we propose a stochastic method for solving schwinger - dyson equations in large- quantum field theories . expectation values of single - trace operators are sampled by stationary probability distributions of the so - called nonlinear random processes . the set of all histories of such processes corresponds to the set of all planar diagrams in the perturbative expansions of the expectation values of singlet operators . we illustrate the method on the examples of the matrix - valued scalar field theory and the weingarten model of random planar surfaces on the lattice . for theories with compact field variables , such as sigma - models or non - abelian lattice gauge theories , the method does not converge in the physically most interesting weak - coupling limit . in this case one can absorb the divergences into a self - consistent redefinition of expansion parameters . stochastic solution of the self - consistency conditions can be implemented as a `` memory '' of the random process , so that some parameters of the process are estimated from its previous history . we illustrate this idea on the example of two - dimensional sigma - model . extension to non - abelian lattice gauge theories is discussed .

1009.4033

modern lattice qcd simulations are mostly based on direct evaluation of the path integral of the theory . such approach , while being very general and efficient for many applications , suffers from a number of problems , most notable of which are the sign problem at finite chemical potential , critical slowing down at small quark masses and large finite - volume effects as well as small signal - to - noise ratio in the analysis of excited states . these problems are inherent to standard monte - carlo simulations and can not be efficiently solved by simply increasing the computation power , since the required computing time quickly increases ( in the worst cases , exponentially ) with the required precision . such situation makes it tempting to devise alternative simulation algorithms for non - abelian lattice gauge theories . one of the efficient alternative numerical methods is the so - called diagrammatic monte - carlo , a method based on stochastic summation of all the terms in the strong- or weak - coupling expansion of the observable of interest @xcite . such a method in some cases allows one to reduce or avoid completely the sign problem in the original path integral , and does not suffer from finite - volume effects . furthermore , one can construct algorithms which yield particular correlation functions in terms of probability distributions of some random variables , which greatly facilitates the analysis of excited states @xcite . this is the idea of the `` worm '' algorithm by prokofev and svistunov @xcite , in which the probability distribution of the positions @xmath2 , @xmath3 of `` head '' and the `` tail '' of the worm yields the two - point green function @xmath4 . diagrammatic monte - carlo and the `` worm '' algorithm have been successfully applied to a number of statistical models with discrete symmetry groups such as the ising model , the xy model and unitary fermi gas and showed practically no critical slowing down near quantum phase transitions . however , application of such methods to lattice field theories with continuous field variables ( such as two - dimensional @xmath1 and @xmath5 sigma - models , abelian gauge theories and the @xmath6 theory ) resulted so far in quite complicated and model - dependent algorithms @xcite . a generalization of such algorithms to @xmath7 sigma - models or to non - abelian gauge theories is still not found . these algorithms are in essence based on the strong - coupling expansion , and while their applicability is not limited by the strong - coupling regime , one can expect that algorithms based on the weak - coupling expansion might show better performance near the continuum limit . typically , the weak - coupling expansion in such lattice theories is either quite complicated or non - convergent . up to now , divergent behavior of the weak - coupling perturbative expansions strongly limits the applicability of diagrammatic monte - carlo to field theories with continuous field variables . in a recent paper @xcite a method was proposed to construct convergent series which approximate the non - analytic path integrals with desired precision . this method , however , is difficult to generalize to physically interesting field theories such as non - abelian lattice gauge theories . another way to obtain convergent series while preserving important physical properties of the theory is to sum over diagrams with certain topology only . this corresponds to the large-@xmath0 limit in quantum field theories and matrix models , that is , the limit of infinite dimensionality of an internal symmetry group , such as @xmath1 or @xmath7 . for such theories , each feynman diagram acquires a factor @xmath8 , where @xmath9 is the euler character of this diagram @xcite . in the limit @xmath10 , the contribution of planar diagrams with @xmath11 dominates , and the sum over all planar diagrams typically has a finite convergence radius @xcite . in this paper we describe a stochastic method for summing over all planar diagrams in large-@xmath0 quantum field theories . the method is based on stochastic solution of schwinger - dyson equations , so that the correlators of field variables are obtained as stationary probability distributions of certain random variables . in this way we implement the idea of importance sampling , so that numerically small observables correspond to unlikely events . these probability distributions are sampled by the so - called nonlinear random processes . in contrast to conventional markov chains , stationary probability distributions of such random processes satisfy nonlinear equations , and hence they can be called `` nonlinear random processes '' or `` nonlinear markov chains '' in the terminology of @xcite . factorization of single - trace operators in the large-@xmath0 limit of quantum field theories corresponds to the phenomena of `` chaos propagation '' in random processes @xcite . while in the diagrammatic monte - carlo and in the `` worm '' algorithm the diagrams are stored in computer memory as a whole and are updated in such a way that the detailed balance condition is satisfied at each step , the method described in this paper works only with external lines . in contrast to the standard metropolis algorithm , one should not know explicitly the weight of each diagram , and the transition probabilities do not satisfy any detailed balance condition . unlike the quite popular `` numerical functional methods '' in continuum gauge theories ( see @xcite for a review ) , the proposed method does not require any truncation of the hierarchy of schwinger - dyson equations , and work only with singlet operators w.r.t . the internal symmetry group . another distinct feature is that the computational complexity of the method does not depend on @xmath0 , while the standard monte - carlo , the functional methods and the `` worm '' algorithm all require infinite computational resources in the limit @xmath10 . this feature might be advantageous for numerical checks of the predictions of the holographic models which are dual to large-@xmath0 quantum field theories @xcite . in section [ sec : sdeq_general ] we analyze the general structure of schwinger - dyson equations in large-@xmath0 quantum field theories on the example of a scalar matrix - valued field theory . when large-@xmath0 factorization is taken into account , schwinger - dyson equations become nonlinear equations with infinitely many unknowns . in section [ sec : recursive_process ] we describe nonlinear random processes of recursive type @xcite which can be used to stochastically solve such equations . in section [ sec : sds_stochastic_solution ] we apply such random processes to solve schwinger - dyson equations in several large-@xmath0 theories . in subsection [ subsec : phi4_general ] we consider the scalar matrix - valued field theory , for which the perturbative expansion yields the conventional feynman diagrams in momentum space . in subsection [ subsec : matrix_model ] this solution is compared with the exact solution of the simplest quantum field theory in zero dimensions , that is , the hermitian matrix model @xcite . the convergence of such solution and the strength of the sign problem is discussed . in subsection [ subsec : weingarten ] we consider the weingarten model @xcite and demonstrate how the proposed method can be used to simulate random surfaces on the hypercubic lattice . in this case , our method reproduces an ensemble of open , rather than closed , random surfaces , with critical behavior which is quite different from those of the closed planar random surfaces . since the structure of schwinger - dyson equations in the weingarten model is similar to the loop equations in large-@xmath0 non - abelian lattice gauge theories @xcite , studying this model might be helpful for further extensions of the present approach to non - abelian gauge theories . while the method described in section [ sec : recursive_process ] works well for non - compact field variables , for field theories with compact field variables , such as nonlinear sigma - models or non - abelian lattice gauge theories , a straightforward stochastic interpretation of schwinger - dyson equations is only possible in the strong coupling limit . in the weak - coupling limit one expects the field correlators to contain both the perturbative part in the coupling constant @xmath12 as well as nonperturbative corrections of the form @xmath13 with some constant @xmath14 . moreover , perturbative expansion in powers of @xmath12 typically results in asymptotic series , and nonperturbative corrections appear as a result of resummation of such series @xcite . in section [ sec : rps_with_mem ] we show how such nonperturbative corrections can be taken into account by a further relaxation of the markov property of the random process . the basic idea is to absorb the divergent part of the series into a self - consistent redefinition of the expansion parameter . these redefined parameters play the role of nonperturbative `` condensates '' @xcite . it turns out that the redefined expansion parameters can be estimated with increasing precision from the previous history of the random process which solves the schwinger - dyson equations , thus leading to the emergence of the `` memory '' of the random process . the approach of the redefined parameters to their self - consistent values is reminiscent somehow of the renormalization - group flow @xcite . such dependence on the previous history makes the random process essentially non - markovian , so that the stationary probability distribution also satisfies some nonlinear equation . we illustrate this idea on the example of @xmath1 sigma - model in two dimensions , which is equivalent to a bosonic random walk with a self - consistent mass . random process which simulates this model has the `` memory '' but no `` recursive '' structure . presumably , in order to sum up both perturbative and non - perturbative corrections which arise in non - abelian lattice gauge theories or @xmath15 sigma - models , one should devise the `` recursive '' nonlinear random process ( which would sum up perturbative corrections ) with memory ( which would generate nonperturbative quantities in a renormalization - group - like way ) . finally , in the concluding section we summarize the present work and discuss its extension to non - abelian lattice gauge theories in the limit of large @xmath0 . in order to analyze the general structure of schwinger - dyson equations for large-@xmath0 quantum field theories , let us first consider the theory of a hermitian @xmath16 matrix - valued field @xmath17 with the following lagrangian : @xmath18 } = n { { \rm tr } \ , } \phi { \left ( x \right ) } \ , { \left ( m^2 - \delta \right ) } \ , \phi { \left ( x \right ) } + \frac{n \lambda}{4 } { { \rm tr } \ , } \phi^{4 } { \left ( x \right ) } .\end{aligned}\ ] ] this theory is most convenient to illustrate the method described in this paper , since its perturbative expansion leads to conventional feynman diagrams in the momentum space . since this theory should be somehow regularized , let us assume from the very beginning that the action ( [ phi4_field_action ] ) is defined on the euclidean hypercubic @xmath19-dimensional lattice with total volume @xmath20 in lattice units . thus , the coordinates @xmath2 take integer values and @xmath21 is the lattice laplacian ( for definiteness , with periodic boundary conditions ) . schwinger - dyson equations for a theory with the action ( [ phi4_field_action ] ) read @xcite : @xmath22 where the single - trace correlators are @xmath23 , @xmath24 is the laplacian acting on @xmath25 and we have already taken into account the factorization property in the limit @xmath10 @xcite : @xmath26 these equations hold for any argument of the correlators , but the resulting system is redundant , and it is sufficient to consider only those schwinger - dyson equations which were obtained by the variation of the fields at @xmath25 . it is convenient now to go to the momentum representation , introducing the green functions in momentum space @xmath27 . in order to keep all expressions as symmetric as possible , we do not separate the factor @xmath28 in @xmath29 explicitly . this condition will be automatically satisfied by the nonlinear random process which we describe in subsection [ subsec : phi4_general ] . the equations ( [ phi4_sds_original_n2 ] ) , ( [ phi4_sds_original ] ) in the momentum representation are : @xmath30 where @xmath31 is the free scalar propagator on the hypercubic lattice . all momenta are assumed to lie in the first brillouin zone @xmath32 and are added modulo @xmath33 . the structure of these equations is schematically illustrated on fig . [ fig : phi4_sds_general ] , where dashed blobs denote the green functions @xmath29 and empty blobs denote @xmath34 . ) . dashed blobs denote the green functions @xmath29 and empty blobs denote the free propagator @xmath34.,title="fig:",width=226 ] + thus we have obtained an infinite system of quadratic functional equations for the set of functions @xmath29 with @xmath35 . such structure is common for large-@xmath0 quantum field theories : schwinger - dyson equations are quadratic equations for infinite set of unknown variables . in the case of scalar matrix field theory considered here , the unknown variables are the functions of the sequences of momenta @xmath36 for any even @xmath37 . in the case of lattice gauge theories or string theories schwinger - dyson equations are most naturally formulated in terms of the wilson loops , which are the functions defined on the discrete space of closed loops on the lattice @xcite . in this case , the equations are also quadratic w.r.t . the wilson loops . for @xmath1 sigma - model , schwinger - dyson equations are also quadratic equations which involve only the two - point function ( see section [ sec : rps_with_mem ] ) . typically , systems of equations with infinitely many unknowns can be efficiently solved by stochastic methods . it is advantageous to estimate the value of each unknown variable as a probability of observing some state of a random process . in this case the unknowns with numerically small values correspond to unlikely events , and the set of infinitely many unknown variables is automatically truncated to a set of unknowns with sufficiently large values . such methods are well - known mainly in the context of kinetic equations @xcite . recently they were also discussed in the context of probabilistic programming @xcite . in the next section we describe a discrete - time , discrete - space method of such type which is in our opinion most suitable for solving the schwinger - dyson equations in the large-@xmath0 limit . we consider nonlinear equations of the following form : @xmath38 where @xmath2 , @xmath3 , @xmath39 , @xmath40 are the elements of some space @xmath41 and @xmath42 with @xmath43 denotes summation or integration over all the elements of this space . we also assume that the functions @xmath44 , @xmath45 and @xmath46 satisfy the inequalities @xmath47 for any @xmath39 , @xmath40 . we would like to find a stochastic process for which @xmath48 is proportional to the probability of the occurrence of the element @xmath2 in some configuration space . obviously , ordinary markov process with configuration space @xmath41 can not solve such a problem , since stationary distributions of markov processes obey linear equations . in order to solve the nonlinear equation ( [ random_process_eq ] ) , one can , for example , extend somehow the configuration space . extensions of markov processes with stationary probability distributions which obey nonlinear equations have been considered recently in @xcite . in this section we concentrate on random processes similar to recursive markov chains of @xcite . the basic idea is that at any time one can leave the current chain and start a new one , then returning back to the old chain at some time . the initial state of a newly created chain depends on the states of older chains . thus one has not a single markov chain , but rather an infinite stack of chains . the random process which we describe below will be similar to these recursive markov chains , but instead of referring to `` recursion '' we will explicitly introduce the underlying stack structure . here we first consider the equations ( [ random_process_eq ] ) with the coefficients @xmath44 , @xmath45 and @xmath46 being all positive , and in appendix [ sec : arbitrary_coefficients ] we generalize to coefficients with arbitrary signs or complex phases . consider an extended configuration space which consists of ordered sequences @xmath49 for arbitrary @xmath50 , with @xmath51 . it is illustrative to interpret such configuration space as a stack of elements of the space @xmath41 , so that @xmath52 is at the top of the stack . the desired random process can be specified by the following prescriptions . at each discrete time step do one of the following : create : : : with probability @xmath44 create new element @xmath43 and push it to the stack . evolve : : : with probability @xmath53 pop the element @xmath3 from the stack and push the element @xmath2 to the stack . join : : : with probability @xmath46 consecutively pop two elements @xmath39 , @xmath40 from the stack and push a single element @xmath2 to the stack . restart : : : with probability @xmath54 , where @xmath39 , @xmath40 are the two topmost elements in the stack , empty the stack and push a single element @xmath43 into it , with probability distribution proportional to @xmath44 . the last action is also the procedure used to initialize the random process . the `` evolve '' action is just the evolution of a single markov chain at the top of the stack , with transition probabilities proportional to @xmath53 . the condition ( [ probability_ineq ] ) and the positivity requirement ensures that @xmath44 , @xmath53 and @xmath46 can be interpreted as probabilities . consider now an equation for the stationary probability distribution of such a markov chain . it has a general form @xmath55 , where @xmath56 is a stationary probability of the occurrence of a state @xmath57 and @xmath58 is the transition probability between the states @xmath59 and @xmath57 . let @xmath60 be the stationary probability to find the elements @xmath61 in the stack . this probability distribution function is obviously normalized to unity : @xmath62 the equation for the stationary probability distribution in our case reads : @xmath63 where @xmath64 and @xmath65 . by a direct substitution one can check that there is a factorized solution for @xmath60 : @xmath66 where @xmath48 obeys exactly the equation ( [ random_process_eq ] ) and @xmath67 obeys the following inhomogeneous linear equation : @xmath68 thus , for any equation of the form ( [ random_process_eq ] ) with positive coefficients which satisfy ( [ probability_ineq ] ) , there is a random process whose stationary distribution encodes the solution of this equation as in ( [ random_process_factorized_pd ] ) . the factorization of the stationary probability distribution of random processes with such an infinite configuration space is known as the `` propagation of chaos '' in random processes and was discovered for classical kinetic equations by mckean , vlasov and kac @xcite . comparing the equation ( [ random_process_eq ] ) with schwinger - dyson equations ( [ phi4_sds_original_n2 ] ) , ( [ phi4_sds_original ] ) , ( [ phi4_sds_momentum_n2 ] ) and ( [ phi4_sds_momentum ] ) , we conclude that this property corresponds to the factorization of single - trace operators in large-@xmath0 quantum field theories . it is interesting that time reversal of the random process described above leads to the so - called branching random process @xcite , which has quite different properties . this is due to the fact that for such random processes there is no detailed balance condition , and hence no time reversal symmetry . we do not consider here a subtle mathematical question of the existence of solutions to equation ( [ random_process_eq ] ) , since in our case it is ensured by the physical applications of this equation . finally , let us describe a practical procedure for finding @xmath48 by simulating the random process described above . by standard statistical methods , one should sample the probability distribution @xmath69 of the topmost element in the stack ( provided there is more than one element in it , otherwise we estimate @xmath67 rather than @xmath48 , see ( [ random_process_factorized_pd ] ) ) . from ( [ random_process_factorized_pd ] ) , we get @xmath70 , with @xmath71 . it should be stressed that @xmath48 is not normalized to unity , but rather satisfies the inequality @xmath72 . the value of the normalization constant @xmath73 can be also easily found numerically , since the probability to find @xmath74 elements in the stack decreases as @xmath75 for @xmath76 . after presenting the general method in section [ sec : recursive_process ] , we are ready to describe a stochastic numerical solution of schwinger - dyson equations ( [ phi4_sds_momentum_n2 ] ) , ( [ phi4_sds_momentum ] ) . for simplicity , let us assume that the coupling constant @xmath77 in ( [ phi4_field_action ] ) is negative . this allows us to apply directly the results of section [ sec : recursive_process ] , where all the coefficients in ( [ random_process_eq ] ) are assumed to be positive . in the case of positive @xmath77 , additional sign variables for each sequence of momenta can be easily introduced following appendix [ sec : arbitrary_coefficients ] . this will be done in the next subsection for the hermitian matrix model . note that while at finite @xmath0 the theory with negative coupling constant is not defined and the correlators are non - analytic in @xmath77 @xcite , in the leading order in @xmath0 perturbative series converge even when the coupling is negative , but not exceeding some critical value @xcite . correspondingly , in the planar approximation the correlators are analytic in @xmath77 . the space @xmath41 in ( [ random_process_eq ] ) should be the space of ordered sequences ( of any size ) of momenta @xmath36 , correspondingly , the extended configuration space is a stack which contains such sequences . it is convenient also to introduce two normalization constants @xmath78 and @xmath14 , so that the functions @xmath79 which will be estimated stochastically are defined as @xmath80 where @xmath20 is again the total volume of space . the constant @xmath14 can be thought of as the renormalization constant for the one - particle wave functions , and @xmath78 - as the overall wavefunction normalization . in terms of the functions @xmath81 the schwinger - dyson equations ( [ phi4_sds_momentum_n2 ] ) and ( [ phi4_sds_momentum ] ) read : @xmath82 comparing the schwinger - dyson equations ( [ phi4_sds_momentum_n2_rescaled ] ) , ( [ phi4_sds_momentum_rescaled ] ) with the general equation ( [ random_process_eq ] ) , we arrive at the random process which stochastically solves these equations . this random process is specified by the following probabilistic choice of actions at each discrete time step : create : : : with probability @xmath83 push a new sequence of momenta @xmath84 to the stack . add : : : with probability @xmath85 modify the topmost sequence of momenta @xmath36 in the stack by adding a pair of momenta @xmath84 either as @xmath86 or @xmath87 . create vertex : : : with probability @xmath88 replace the topmost sequence @xmath89 in the stack by @xmath90 . this action can only be performed if the topmost sequence contains more than two elements . join : : : with probability @xmath91 pop the two sequences @xmath36 , @xmath92 from the stack ( provided there are more than two elements in it ) and join them into a single sequence as @xmath93 . push the result to the stack . restart : : : otherwise restart with a stack containing a sequence @xmath94 , @xmath95 being distributed with the probability proportional to @xmath34 since the momenta are always added to the stack in pairs which sum up to zero , for all sequences in the stack the total sum of all momenta in the sequence is always zero . the @xmath96 factors in ( [ phi4_sds_momentum_n2_rescaled ] ) , ( [ phi4_sds_momentum_rescaled ] ) ensure that the probability distributions of the newly created momenta can be normalized to unity . let us check whether the inequalities ( [ probability_ineq ] ) are satisfied for such a process , that is , whether the total probability of all possible actions does not exceed unity . for the free propagator @xmath34 one has the inequalities @xmath97 and @xmath98 . the total probability of all possible actions can be then estimated as @xmath99 . clearly , for sufficiently small @xmath100 this estimate can be always made smaller than unity by increasing @xmath14 and @xmath78 . in subsections [ subsec : matrix_model ] and [ subsec : weingarten ] we will analyze such bounds on coupling constants in more details for hermitian matrix model and for the weingarten model . since the constructed process involves no permutations , one can trace the history of each momenta in the stack - from creation to joining into a vertex or a `` restart operation '' . by drawing all the momenta in stack as points on the vertical lines of some two - dimensional grid and connecting the corresponding points along the horizontal lines , all planar diagrams of the theory ( with an arbitrary number of external lines ) can be obtained . note also that the number of vertices in planar diagrams drawn by this random process can not exceed the number of time steps from the previous `` restart '' action . thus in order to maximize the mean order of diagrams which are summed up in some fixed number of time steps , it is advantageous to maximally reduce the rate of `` restart '' events , that is , to saturate the inequalities ( [ probability_ineq ] ) . one could also try to devise a random process which would solve the schwinger - dyson equations ( [ phi4_sds_original_n2 ] ) , ( [ phi4_sds_original ] ) directly in physical space - time , rather than in the momentum space . the configuration space of such a process would be the stack of sequences of points @xmath49 . as compared to the algorithm in the momentum space , there would be an additional choice of moving the last point @xmath52 in the topmost sequence to adjacent lattice sites , with the probability proportional to the hopping parameter @xmath101 . this would correspond to drawing the worldlines of virtual and real particles by bosonic random walks . interestingly , such worldlines can be mapped onto the string worldsheets in simplicial string theory @xcite . as well , the creation of a new interaction vertex would only be possible if three such random walks would intersect in one point . however , this is an unlikely event , with probability going to zero in the continuum limit . thus , solving the schwinger - dyson equations directly in the coordinate representation would lead to a less efficient numerical algorithm . note that for the theory ( [ phi4_field_action ] ) at finite @xmath0 the schwinger - dyson equations are linear equations , which are , however , defined on much larger functional space : the set of unknown functions includes also expectation values of multi - trace operators , such as @xmath102 . one can try to solve these linear equations by interpreting them as the equations for the stationary probability distribution of a markov process . the configuration space of such a process should be a space of sequences of the form @xmath103 , thus encoding the expectation values of all multi - trace operators . however , such a straightforward procedure leads to non - normalizable transition probabilities , indicating that the series which one tries to sum up are divergent . only when the terms subleading in @xmath104 are omitted from the schwinger - dyson equations , they can be interpreted as stochastic equations . at the same time , we obtain the markov process on the extended configuration space described in section [ sec : recursive_process ] , which we interpret as the stack of sequences . the property of the `` propagation of chaos '' @xcite ensures large-@xmath0 factorization of single - trace operators ( see equation ( [ random_process_factorized_pd ] ) ) . we are thus led to the random process of recursive type @xcite . to check the considerations of the previous subsection , let us consider the theory ( [ phi4_field_action ] ) in zero dimensions , that is , the hermitian matrix model with the following partition function : @xmath105 the green functions now depend only on one integer @xmath74 : @xmath106 . the schwinger - dyson equations ( [ phi4_sds_original_n2 ] ) , ( [ phi4_sds_original ] ) also take a very simple form : @xmath107 here we will assume that the coupling constant @xmath77 can be both positive and negative , in order to illustrate the method described in appendix [ sec : arbitrary_coefficients ] . let us again define the `` renormalized '' green functions @xmath108 as @xmath109 . in the case of arbitrary sign of @xmath77 , the configuration space of the random process should be the stack which contains integer positive numbers and additional sign variables . following appendix [ sec : arbitrary_coefficients ] , we introduce the variables @xmath110 and @xmath111 which are proportional to the probabilities to find the elements @xmath112 or @xmath113 at the top of the stack ( provided the stack contains more than one element ) . then @xmath114 . we thus arrive at the following random process for stochastic evaluation of @xmath115 . at each discrete time step one performs at random one of the following actions : * with probability @xmath116 add new element @xmath117 to the stack . * with probability @xmath118 increase the topmost element in the stack by @xmath119 and do not change its sign . * with probability @xmath120 decrease the topmost element in the stack by @xmath119 ( if it is greater than one ) and multiply its sign by the sign of @xmath77 . * with probability @xmath121 pop the two elements @xmath122 and @xmath123 from the stack ( provided there are more than two elements ) and push the element @xmath124 to the stack . * otherwise empty the stack and push into it a single element @xmath117 . note that for positive @xmath77 elements with the minus sign are not generated , so that @xmath125 and the random process automatically reduces to the one described in section [ sec : recursive_process ] . the inequalities ( [ probability_ineq ] ) read now : @xmath126 as discussed in subsection [ subsec : phi4_general ] , in order to increase the efficiency of the algorithm it is advantageous to saturate this inequality . it is easy to see that at the same time we saturate the upper bound on the absolute value of the coupling constant @xmath77 . maximizing this upper bound with respect to @xmath78 and @xmath14 , we see that @xmath100 can not exceed the value @xmath127 , where @xmath128 is the convergence radius of the planar perturbative expansion which can be found from the exact solution of the matrix model ( [ matrix_model_def ] ) @xcite . thus the described random process covers only some finite subrange of coupling constants for which the model ( [ matrix_model_def ] ) is defined . it is easy to understand the origin of this limitation : in fact the random process described above simulates an ensemble of diagrams with an arbitrary number of external legs , with the weight of each diagram being proportional to @xmath129 , where @xmath130 is the number of vertices . the number of open diagrams with a given number of vertices is obviously larger than the number of closed diagrams , hence the sums over open diagrams have smaller convergence radius . + for @xmath131 , there is a continuous set of @xmath78 , @xmath14 which saturate the inequality ( [ brezin_process_inequalities ] ) . one can , for example , fix @xmath14 and express @xmath78 as a function of @xmath77 : @xmath132 we call the solution with the minus sign in front of the square root `` branch 1 '' and the other solution `` branch 2 '' . + on fig . [ fig : brezin_moments ] we plot the green functions @xmath133 evaluated using the described random process as functions of @xmath77 up to @xmath134 . these results were obtained after @xmath135 discrete time steps at fixed @xmath136 and with @xmath78 given by the `` branch 1 '' of ( [ nn_vs_lambda ] ) . the error bars are smaller than the symbols on the plot . solid lines are the exact results for @xmath133 in the planar approximation , obtained using the saddle point method @xcite . + autocorrelation time and mean stack size for the described random process are plotted on figs . [ fig : autocorrelation_time ] and [ fig : mean_stack_size ] , respectively , as the functions of the coupling constant @xmath77 . the observable used to define the autocorrelation time was the sum of all numbers in the stack . first , we note that `` branch 1 '' is more advantageous for simulations , since with larger mean stack size one can gain more statistics . however , in this case the autocorrelation time is also larger . interestingly , for this branch both the autocorrelation time and the mean stack size have maximum near @xmath137 rather than near the `` critical point '' of the random process @xmath138 . for `` branch 2 '' , these quantities increase slowly towards @xmath138 . + in order to characterize the strength of the sign problem , we consider the quantity @xmath139 @xmath140 if the random process generates only elements with the plus sign and @xmath141 if the numbers of pluses and minuses exactly cancel . in practice , it is advantageous to have as large @xmath142 as possible , so that the difference @xmath143 can be estimated with maximal precision . @xmath142 are plotted on fig . [ fig : sign_problem_severity ] as the function of @xmath77 for @xmath144 . for @xmath145 , @xmath146 decreases with @xmath77 and @xmath74 . the sign cancelation is thus moderate for @xmath147 ( @xmath148 ) and becomes more and more important for higher - order correlators - @xmath149 at @xmath150 is close to zero . it is interesting that @xmath142 are almost equal for two different choices of @xmath78 in ( [ nn_vs_lambda ] ) . thus there are no indications of severe critical slowing down in the whole range of possible coupling constants @xmath151 . the sign problem is also moderate for low - order correlators , but becomes more severe for higher - order correlators . it could be extremely interesting to extend the applicability of the described random process up to @xmath152 while preserving these attractive features of the algorithm . weingarten model @xcite is a lattice field theory which in the large-@xmath0 limit reproduces the sum over all closed surfaces with genus one on the hypercubic lattice . the action for each surface is proportional to its area , thus the model can be considered as a lattice regularization of bosonic strings with nambu - goto action . although this model does not have a nontrivial continuum limit for any space dimensionality @xcite , the structure of the functional integral and of the schwinger - dyson equations in this model are similar to those in large-@xmath0 non - abelian lattice gauge theory , and the analysis of this model might be helpful for the extension of the approach described here to non - abelian gauge theories . in order to derive the schwinger - dyson equations , it is convenient to consider the reduced weingarten model @xcite , which in the large - n limit is equivalent to the original model , similarly to the eguchi - kawai model for non - abelian lattice gauge theory . it can be shown that in contrast to reduced lattice gauge theories , for the reduced weingarten model additional twisting is not necessary @xcite . ) in the weingarten model ( [ weingarten_model_def ] ) . these equations should hold for any link ( marked by a thick line ) which belongs to the loop.,title="fig:",width=226 ] + the reduced model is defined by an integral over complex @xmath153 matrices @xmath154 with @xmath155 : @xmath156 if one treats the second term in the exponent in ( [ weingarten_model_def ] ) as a perturbation and expands @xmath157 in powers of @xmath158 , the resulting sum over planar diagrams is equivalent to the sum over all possible closed surfaces of genus one on the lattice with the weight @xmath159 , where @xmath160 is the area of each surface . a basic observable in this model is the sum over all planar surfaces which are bounded by some closed loop @xmath161 . the loop @xmath161 can be uniquely specified by a sequence @xmath162 , where @xmath163 s take the values @xmath164 . in order to reconstruct the loop @xmath161 from the sequence , one should start from an arbitrary point on a hypercubical lattice and move along one link in the direction @xmath165 , forward if @xmath165 is positive and backward if @xmath165 is negative . from this new position one should similarly move in the direction @xmath166 , and so on . from the diagrammatic expansion one can see that such a sum over surfaces is given by the following correlator : @xmath167 where one takes the conjugate variable @xmath168 if @xmath169 is negative . this observable is similar to wilson loop in lattice gauge theory , but , unlike the wilson loop , it does not have a `` zigzag symmetry '' @xcite : passing a link forward and immediately backward changes the value of the wilson loop . in the large-@xmath0 limit the single - loop observables factorize , which allows us to obtain a closed set of schwinger - dyson equations for @xmath170 @xcite : @xmath171 these equations should hold for any lattice link @xmath172 belonging to the loop @xmath161 , but the resulting system of equations is redundant , and it is sufficient to consider only one link @xmath165 on the loop . the equations ( [ weingarten_loop_equations ] ) are schematically illustrated on fig . [ fig : weingarten_loop_equations ] , where the link @xmath165 is marked by a thick line . we see that the equations ( [ weingarten_loop_equations ] ) again take the form similar to ( [ random_process_eq ] ) . let us now define the `` renormalized '' observable @xmath173 by rescaling @xmath170 by the factors @xmath78 and @xmath174 as @xmath175 . one can interpret the factor @xmath176 as the mass attached to the boundaries of random surfaces , somewhat like the bare quark mass in qcd . the equations ( [ weingarten_loop_equations ] ) then take the following form : @xmath177 let us now devise a random process of the type described in section [ sec : recursive_process ] , which solves stochastically these equations . the configuration space is now a stack which contains closed loops , that is , sequences of indices @xmath178 . the desired random process is defined by the following possible actions at each discrete time step : create a new loop : : : with probability @xmath179 create a new elementary loop @xmath180 , where @xmath178 is random ( either positive or negative ) . join loops : : : with probability @xmath181 pop the two loops @xmath182 , @xmath183 from the stack and form a new loop @xmath161 by joining the loops @xmath184 , @xmath185 with a link in the random direction @xmath163 ( either positive or negative ) : @xmath186 . this action can only be performed if there are more than two loops in the stack . flatten loop : : : if the three links in the end of the sequence on the top of the stack form a boundary of the plaquette , that is , if the topmost loop has the form @xmath187 for some @xmath163 and @xmath188 , replace these three links by a single link in the direction @xmath188 with probability @xmath189 : @xmath190 . append to loop : : : with probability @xmath191 append a pair @xmath192 , where @xmath163 is random ( either positive or negative ) , to the topmost sequence @xmath162 in the stack as @xmath193 or @xmath194 . the probabilities of these two choices are equal . restart : : : otherwise start with a stack containing an elementary random loop @xmath180 , where @xmath178 is chosen randomly . again assuming that the sum of the probabilities of all possible actions is equal to one and the probability of `` restart '' events is minimized , we obtain an equation relating @xmath158 , @xmath78 and @xmath174 : @xmath195 maximization with respect to @xmath78 yields the relation between @xmath174 and @xmath158 : @xmath196 we call the solution with the minus sign in front of the square root `` branch 1 '' and the other solution `` branch 2 '' . + the value of @xmath158 in ( [ weingarten_q_vs_beta ] ) can not exceed the critical value @xmath197 . as we have already seen on the example of the matrix model , this critical value does not necessarily coincide with the true critical point @xmath198 at which the sum over planar surfaces diverges . indeed , @xmath199 does not exceed the lower bound @xmath200 obtained in @xcite , and is significantly lower than the critical values obtained numerically in @xcite . in fact , for @xmath201 all the observables are still dominated by the lowest - order perturbative contributions . expectation values of the observables @xmath202 ( @xmath203 loop ) and @xmath204 ( @xmath205 loop ) , which were obtained after @xmath206 iterations of the random process described above ( with @xmath174 given by `` branch 1 '' of ( [ weingarten_q_vs_beta ] ) ) , are plotted on fig . [ fig : wgtn_w1x1_w1x0 ] as the functions of the coupling constant @xmath158 . solid line corresponds to the first two terms in the perturbative expansion : @xmath207 , @xmath208 . within statistical errors , one sees only the lowest - order perturbative contributions . it should be stressed that the proposed random process implements stochastic summation of diagrams of _ all _ orders , but due to the smallness of @xmath209 , a very large computational time is required to see the contributions of higher - order terms . + note that when the coupling constant @xmath158 tends to zero ( that is , the `` bare string tension '' of the random surfaces tends to infinity ) and @xmath174 lies between the two solutions of ( [ weingarten_q_vs_beta ] ) , the above random process describes just the growth of `` branched polymers '' , whose branches are bosonic random walks and hence correspond to particles rather than `` strings '' . these branches consist of loops in which every lattice link is passed twice and which hence sweep out zero area . taking the limit @xmath210 in ( [ weingarten_q_vs_beta ] ) , we find that the minimal value of @xmath174 is @xmath211 . in order to understand this critical value we first note that in the limit @xmath210 the observables @xmath170 are all equal to one if the links @xmath212 form a loop which sweeps out zero area and zero otherwise . the probabilities to encounter such loops in the described random process is hence proportional to @xmath213 . simple examples of such loops , which can be also thought of as the random tree - like graphs on the lattice , are shown on fig . [ fig : branched_polymers_illustrated ] . now imagine adding to some loop @xmath95 links stemming from some lattice site . since the loop includes each link twice , the probability decreases by @xmath214 . the number of possible configurations of @xmath95 links is @xmath215 since each of @xmath95 links can point along any of @xmath19 directions both forward and backward . an additional factor of @xmath216 appears since the zero - area loops pass twice through each point and the new links can be inserted between the links pointing either forward or backward ( for example , compare the configurations on the left and on the right of fig . [ fig : branched_polymers_illustrated ] ) . finally , one can add any number @xmath217 of branches to any point belonging to the branched polymer . at the criticality , adding any number of random links to some configuration should not change its overall weight . therefore , the change of the weight due to the added links times the number of ways to add them should be equal to unity . we are thus led to the following equation for @xmath218 : @xmath219 or @xmath220 . thus , in the limit @xmath210 we indeed reproduce branched polymers with the correct critical behavior . at nonzero @xmath158 , deviation from trivial branched polymer configurations can be characterized by the rate of the `` flatten loop '' events . indeed , since the probability of @xmath74 such events is proportional to @xmath221 , such sequence of events corresponds to a random surface ( which is in general open ) consisting of @xmath74 lattice plaquettes , plus some number of random trees . one can therefore think of the described random process as of the process of drawing random loops which sweep out random planar surfaces . the average rate of `` flattening '' events is plotted on fig . [ fig : wgtn_mfr ] on the right as a function of the coupling constant @xmath158 for different dimensions @xmath19 and for different choices of @xmath174 in ( [ weingarten_q_vs_beta ] ) . one can see that in the whole range of coupling constants @xmath222 the rate of flattening events is numerically very small . on the other hand , the number of links in the loops , as well as the number of loops stored in the stack , are quite large . mean stack size and mean length of the topmost loop in the stack are plotted on fig . [ fig : wgtn_mss_mll ] as a function of the coupling constant @xmath158 for different dimensions @xmath19 and for different choices of @xmath174 in ( [ weingarten_q_vs_beta ] ) . one can conclude therefore that the `` branched polymers '' actually dominate in the properties of the described random process . critical behavior of these random trees is universal for any dimension @xmath19 , that is why such observables as the mean stack size or the mean loop length , which are mainly sensitive to the length of loops rather than to the area of random surfaces , practically do not depend on space dimensionality . while the closed planar surfaces in the vicinity of the true critical point @xmath198 of the weingarten model are also dominated by `` branched polymers '' @xcite , in our ensemble of open random surfaces this dominance can be thought of as the manifestation of the tachyonic instability of open , rather than closed , strings . the fact that the critical coupling @xmath199 in our case is smaller than the true critical point @xmath198 can be explained by the fact that the number of open surfaces with a given area is obviously larger than the number of closed surfaces with the same area . the true critical coupling constant @xmath198 can be quite easily found by a very simple re - weighting procedure , which will be described in details in a separate publication . in section [ sec : sds_stochastic_solution ] we have described a stochastic method for the solution of the schwinger - dyson equations for theories with noncompact variables . this method works only at small coupling constants and implements stochastic summation of perturbative series . for theories with compact variables , such as nonlinear @xmath223-models or lattice gauge theories , the structure of the schwinger - dyson equations is such that the described method can be straightforwardly applied only in the strong coupling regime , where one can stochastically sum all terms in the strong - coupling expansion . however , the continuum limit of such theories typically corresponds to the weak - coupling limit . an additional complication is that for physically interesting theories the observables can not be expressed as convergent power series in the small coupling constant @xmath12 , but rather contain non - analytic part which is typically of the form @xmath13 with some constant @xmath14 @xcite . in this section we point out one possible way to deal with this problem . the basic idea is to absorb non - perturbative corrections into some self - consistent redefinition of the expansion parameter @xcite . recently , a similar resummation method was also considered in @xcite . solving the self - consistency condition leads to the concept of a nonlinear random process with memory @xcite , in which all previous history of the process is used to estimate the value of the self - consistent expansion parameter . let us illustrate this idea on the simplest example of @xmath1 sigma model in the limit of large @xmath0 . the model is defined by the following path integral over unit @xmath0-component vectors @xmath224 living on the sites of the @xmath19-dimensional hypercubic lattice : @xmath225 where summation goes over all neighboring lattice sites . despite its simplicity , this model in @xmath226 dimensions is asymptotically free and has a mass gap which depends nonperturbatively on the coupling constant @xmath77 . schwinger - dyson equations in this theory can be written in terms of the two - point function @xmath227 , @xmath228 as : @xmath229 clearly , these equations have the structure similar to ( [ random_process_eq ] ) , but the inequalities ( [ probability_ineq ] ) are satisfied only for sufficiently large @xmath77 , that is , in the strong - coupling regime . therefore , the continuum limit at @xmath230 can not be reached by the method described in section [ sec : sds_stochastic_solution ] . let us , however , rewrite the equation ( [ sigma_model_sd ] ) as @xmath231 and introduce the `` hopping parameter '' @xmath232 now the equation ( [ sigma_model_sd ] ) in the form ( [ sigma_model_sd_as_rw ] ) looks like the equation for the free massive scalar propagator on the lattice with the mass @xmath233 in lattice units . note that in the weak - coupling limit @xmath230 @xmath234 , @xmath235 and we approach the continuum limit . let us now solve the equation ( [ sigma_model_sd_as_rw ] ) stochastically , assuming that @xmath236 is proportional to the stationary probability distribution @xmath48 of some random process : @xmath237 , @xmath238 . from ( [ sigma_model_sd_as_rw ] ) we get @xmath239 . the equation ( [ sigma_model_sd_as_rw ] ) now looks as @xmath240 combining this equation with the definition ( [ sigma_model_kappa_def ] ) , it is easy to show that @xmath241 obeys the following self - consistency condition : @xmath242 the equation ( [ sigma_model_rw_eq ] ) has the form ( [ random_process_eq ] ) without the nonlinear term and thus can be interpreted as the equation for the stationary probability distribution of the position of an ordinary bosonic random walk , defined by the following possible actions at each discrete time step : move : : : with probability @xmath243 move along the random unit lattice vector @xmath244 . restart : : : with probability @xmath245 start again at the origin @xmath246 . this ensures that @xmath247 and hence @xmath241 never exceeds its critical value @xmath248 . therefore @xmath236 and @xmath48 can be expanded in powers of @xmath241 . thus we have defined a new expansion parameter @xmath241 , which should obey the self - consistency equation ( [ sigma_model_kappa_return_prob ] ) , and obtained a well - defined convergent expansion , namely , the sum over all paths on the lattice with the weight @xmath249 , where @xmath250 is the length of the path . note that the quantity @xmath251 in fact plays the role similar to the gluon condensate ( which is expressed in terms of the mean plaquette in lattice theory ) in non - abelian gauge theory : one can absorb all the divergences into the self - consistent definition of condensates @xcite . the final step in the construction of the nonlinear random process which solves the equations ( [ sigma_model_sd ] ) is the solution of the self - consistency equation ( [ sigma_model_kappa_return_prob ] ) . one possible solution is to use the iterations @xmath252 here @xmath253 is the return probability of a bosonic random walk with hopping parameter @xmath241 . in practice , one should simulate the bosonic random walk at fixed @xmath254 for some number @xmath255 of discrete time steps , and then estimate @xmath256 as @xmath257 , where @xmath258 is the number of discrete time steps spent at @xmath246 . from ( [ sigma_model_iterations ] ) one then gets @xmath259 , and the process is repeated until the value of @xmath241 stabilizes with sufficient numerical precision . we call such algorithm `` algorithm a '' . one can also consider an ultimate case , for which the return probability is updated and estimated as @xmath260 every time the point @xmath246 is reached . now @xmath261 is the time from the start of the random process and @xmath258 is the number of time steps spent at @xmath246 . such algorithm will be called `` algorithm b '' . mathematically , such random processes are not markov processes , since the transition probabilities at each next step depend ( via @xmath262 ) on the behavior of the process at all previous time steps . stationary probability distributions of such processes obey nonlinear equations ( such as ( [ sigma_model_sd ] ) ) @xcite , and and hence they are also called nonlinear random processes . as an interesting side remark , let us discuss such a theory at finite temperature , which is described by a bosonic random walk on the cylinder . clearly , an ordinary bosonic random walk does not feel this compactification of space , and its stationary probability distribution is just a periodic linear combination of the corresponding distribution in infinite space . such behavior can not lead to any nonlinear finite - temperature effects such as phase transitions . on the other hand , if the parameters of the random walk depend on the return probability , as in ( [ sigma_model_kappa_return_prob ] ) , there is a nonlinear feedback mechanism since in the compactified space the returns are more likely . thus finite temperature indeed affects the local behavior of the random walker with memory and might lead to interesting critical phenomena . + in order to illustrate such a stochastic solution of the equations ( [ sigma_model_sd ] ) , we consider the case @xmath226 . in two dimensions the model ( [ sigma_model_pf ] ) is asymptotically free , and one can introduce the lattice spacing by fixing the value of mass in physical units ( we set @xmath263 ) : @xmath264 . the process of convergence of the lattice spacing to its exact value is illustrated on fig . [ fig : sigma_model_convergence ] for both the algorithms `` a '' and `` b '' . for algorithm `` a '' we have used @xmath265 . the algorithm `` a '' converges much faster than the algorithm `` b '' . the values of lattice spacing obtained using both algorithms are compared with the exact solution on fig . [ fig : spacing_vs_lambda ] . in agreement with asymptotic freedom , lattice spacing quickly decreases with @xmath77 . again , algorithm `` a '' yields more precise results in the same number of time steps . in this paper we have presented numerical strategies for the stochastic summation and re - summation of perturbative expansions in large-@xmath0 quantum field theories . our basic approach was to interpret the schwinger - dyson equations as the equations for the stationary probability distribution of some random process . since schwinger - dyson equations in such theories are nonlinear equations , we had to use so - called nonlinear random processes , rather than ordinary markov processes whose stationary probability distributions always obey linear equations . it is interesting to note that since the configuration spaces of random processes described in this paper are discrete , their numerical implementation require floating - point operations only for the random choice of actions . thus such algorithms can be potentially much faster than the standard monte - carlo simulations based on floating - point arithmetic , and can be advantageous for machines based on gpus . our final goal is to extend the presented approach to non - abelian lattice gauge theories . however , in this case direct stochastic interpretation of schwinger - dyson equations is only possible at strong coupling , while the continuum limit of such theories corresponds to the weak - coupling limit . in section [ sec : rps_with_mem ] , we have discussed a way to access the weak - coupling limit , which , however , was implemented numerically only for @xmath1 sigma - model at large @xmath0 . the basic idea is to absorb the divergences into a self - consistent redefinition of the expansion parameter and solve the self - consistency conditions using random processes with memory . in some sense , @xmath1 sigma - model can be thought of as the bosonic random walk in its own condensate , and the approach to the self - consistent value of mass gap ( see fig . [ fig : sigma_model_convergence ] ) - as a renormalization - group flow . for non - abelian gauge theories the redefined expansion parameters can emerge as the lagrange multipliers for the `` zigzag symmetry '' of the qcd string and should also satisfy some self - consistency conditions @xcite . zigzag symmetry means that when one adds a line which is passed forward and backward to the boundary of the fluctuating string , the amplitudes should not change . in lattice gauge theory , this condition is equivalent to the unitarity of the link variables @xmath266 , which is similar to the condition @xmath267 in @xmath1 sigma - model . these redefined parameters can be also related to the gluon condensate @xcite . by analogy with the sigma - model , one can think that non - abelian gauge theories are similar to strings moving in some self - consistent condensates . such a picture is also close to the idea of holographic ads / cft duality for non - abelian gauge theories , where the dual string lives in some self - consistent gravitational background , and the parameters of this background can be related to gluon condensates in gauge theory @xcite . in fact , the requirement that the metric of the holographic background approaches that of the ads space - time ensures the zigzag symmetry of the strings which end on the ads boundary @xcite . in view of these qualitative considerations , our hope is that the loop equations in non - abelian gauge theories can be solved stochastically by a random process similar to the one which was devised for the weingarten model of random surfaces ( see subsection [ subsec : weingarten ] ) , but with some self - consistent choice of parameters , which might be implemented as the `` memory '' in the random process . among other possible applications of the presented method one can think of the solution of schwinger - dyson equations in continuum gauge theories , combined with the renormalization group methods @xcite , numerical analysis of quantum gravity models described by various matrix models , and numerical solution of hydrodynamical equations @xcite . it should be noted here that several attempts at the stochastic solution of the loop equations in large-@xmath0 gauge theories have been already described in the literature quite a long time ago @xcite . these algorithms were , in essence , based on the so - called branching random processes , so that the wilson loop @xmath268 is proportional to the probability of transition from the initial loop configuration @xmath161 to the empty configuration with no loops . in particular , in contrast to the algorithm described in subsection [ subsec : weingarten ] , where one of the basic steps is to join loops , in the algorithms described in @xcite the basic step was to split a self - intersecting loop into two loops . as a result , these algorithms did not implement the importance sampling and were not able to produce any sensible results for the four - dimensional gauge theory . generally , branching random processes similar to those considered in @xcite can be obtained from the `` recursive '' nonlinear random process described in this paper by time reversal . however , since such processes do not satisfy any detailed balance condition , they are not invariant under this operation , and lead to very different numerical algorithms . i am grateful to drs . m. i. polikarpov , yu . m. makeenko , a. s. gorsky , n. v. prokofev and i. ya . arefeva for interesting and stimulating discussions . i d like also to thank drs . f. bruckmann and a. schaefer for their kind hospitality at the university of regensburg , where a part of this work was written . this work was partly supported by grants rfbr 09 - 02 - 00338-a , rfbr 08 - 02 - 00661-a , a grant for the leading scientific schools nsh-6260.2010.2 , by the federal special - purpose programme `` personnel '' of the russian ministry of science and education , and by personal grants from the `` dynasty '' foundation and from the fair - russia research center ( frrc ) . the random process described in section [ sec : recursive_process ] was devised under the assumption that the coefficients @xmath44 , @xmath269 and @xmath46 are real and positive for any @xmath2 , @xmath39 and @xmath40 . in this appendix we show how the solution of the equation ( [ random_process_eq ] ) with arbitrary signs or complex phases on the r.h.s . can be reduced to the solution of another equation of the form ( [ random_process_eq ] ) with all positive coefficients , provided the inequalities ( [ probability_ineq ] ) are satisfied . we begin by discussing the case of real but non - positive coefficients in details , and finally sketch the extension to complex - valued coefficients . to this end , let us represent the sign - alternating coefficients in ( [ random_process_eq ] ) as : @xmath270 where @xmath271 , @xmath272 and @xmath273 are all positive and also obey the following inequality : @xmath274 for any @xmath39 , @xmath40 . obviously , these inequalities can be satisfied if the inequalities ( [ probability_ineq ] ) are satisfied . let us now introduce two functions @xmath275 and @xmath276 , which satisfy the following equations : it is now easy to check that the difference @xmath278 satisfies the equation ( [ random_process_eq ] ) . on the other hand , the equations ( [ sign_alt_repr_eq ] ) again have the form of ( [ random_process_eq ] ) with all positive coefficients , but with the configuration space @xmath279 being the direct product @xmath280 . in other words , each variable @xmath2 now in addition carries the `` sign '' , @xmath281 or @xmath282 , which can be written as @xmath283 or @xmath284 . basing on the results presented in section [ sec : recursive_process ] , one can devise the random process which solves these equations : create : : : with probability @xmath271 create new element @xmath285 and push it to the stack . evolve : : : with probability @xmath286 pop the element @xmath287 from the stack and push the element @xmath288 to the stack , with probability @xmath289 do the same but flip the sign of @xmath3 . join : : : with probability @xmath290 consecutively pop two elements @xmath291 , @xmath292 from the stack and push a single element @xmath293 to the stack . that is , two pluses or two minuses associated with @xmath3 s give @xmath283 , but one plus and one minus give @xmath284 . with probability @xmath294 do the same , but flip the resulting sign , that is , push the element @xmath295 to the stack . restart : : : otherwise , empty the stack and push a single element @xmath285 into it , with probability proportional to @xmath271 . as in section [ sec : recursive_process ] , @xmath275 and @xmath276 are proportional to probabilities to find the elements @xmath283 or @xmath284 on the top of the stack , provided there is more than one element in it . the extension of this construction to complex - valued coefficients is quite straightforward . we represent the coefficients in ( [ random_process_eq ] ) as @xmath296 , where @xmath297 is real and positive , and similarly for the other coefficients . the configuration space @xmath279 becomes the direct product @xmath298 , where @xmath299 is the unit circle in the complex plane . the stack now contains the pairs @xmath300 , with @xmath301 } $ ] being the complex phase . the function @xmath48 is estimated as @xmath302 , where @xmath303 is the probability to find the element @xmath300 at the top of the stack . in the random process , new elements are created with probability distribution proportional to @xmath297 , and in the `` evolve '' and the `` join '' actions the phases of the elements and the coefficients in ( [ random_process_eq ] ) are added modulo @xmath33 , similarly to signs . note that this solution does not in general have the property of importance sampling . indeed , one can have some @xmath2 for which @xmath48 is numerically very small , but the random process can spend an almost equal large amount of time in the states @xmath283 and @xmath284 , so that numerically large @xmath275 and @xmath276 nearly cancel . whether this occurs or not depends on the particular system of equations and on the particular unknown variables , but potentially this feature can make numerical simulations less efficient .

we introduce a new oriented evolving graph model inspired by biological networks . a node is added at each time step and is connected to the rest of the graph by random oriented edges emerging from older nodes . this leads to a statistical asymmetry between incoming and outgoing edges . we show that the model exhibits a percolation transition and discuss its universality . below the threshold , the distribution of component sizes decreases algebraically with a continuously varying exponent depending on the average connectivity . we prove that the transition is of infinite order by deriving the exact asymptotic formula for the size of the giant component close to the threshold . we also present a thorough analysis of aging properties . we compute local - in - time profiles for the components of finite size and for the giant component , showing in particular that the giant component is always dense among the oldest nodes but invades only an exponentially small fraction of the young nodes close to the threshold . michel bauer and denis bernard service de physique thorique de saclay ce saclay , 91191 gif sur yvette , france

cond-mat0203232

evolving random graphs have recently attracted attention , see e.g. refs @xcite and references therein . this interest is mainly motivated by concrete problems related to the structure of communication or biological networks . experimental data are now available in many contexts @xcite . in these examples , the asymmetry and the evolving nature of the networks are likely to be important ingredients for deciphering their statistical properties . it is however far from obvious to find solvable cases that would possibly account for some relevant features of , say , the regulating network of a genome . although biology has strongly influenced our interest in evolving networks , the model we solve is not based on realistic biological facts but it nevertheless incorporates asymmetry and chronological order . understanding such simple evolving graphs may help understanding biological networks , at least by comparison and opposition . we were initially motivated by the study of the yeast genetic regulatory network presented in ref.@xcite . the authors studied in and out degree distributions and discovered a strong asymmetry : a single gene may participate to the regulation of many other genes the law for out - degrees seems to be large , but each genes is only regulated by a few other genes the law for in - degrees seems to have finite moments . this is why we consider oriented evolving random graphs in the sequel . a biological interpretation for the asymmetry is that the few promoter - repressor sites for each gene bind only to specific proteins , but that along the genome many promoter - repressor sites are homologous . however , this does not predict the precise laws . an understanding of the same features from a purely probabilistic viewpoint would be desirable as well . the recent experimental studies dealt with global statistical properties of evolving graphs , i.e. when the evolving network is observed at some fixed time with the ages of different vertices and edges not taken into account . there are simple experimental reasons for that : to keep track of the ages would in many cases dramatically reduce the statistics , and in other cases this information is even not available . our second motivation is a better understanding of the local - in - time statistical properties of evolving networks . this helps dating or assigning likely ages to different structures of the networks . as we shall later see , the global analysis , which is like a time average , gives a distorted view of the real structure of the networks . we shall present a detailed analysis of local - in - time features in our model . the model we study is the natural evolving cousin of the famous erds - renyi random graphs @xcite . starting from a single vertex at time @xmath0 , a new vertex is created at each time step so that at time @xmath1 , the size of the system , i.e. the number of vertices , is @xmath1 , and new oriented edges are created with specified probabilistic rules . a tunable parameter @xmath2 ranging from @xmath3 to @xmath4 describes asymptotically the average number of incoming edges on a vertex . precise definitions are given in the next section . our main results are the following : from very simple rules , we see an asymmetry emerging . the global in and out degree distributions are different . we also compute the local profiles of in and out degree distributions , and comment on the differences . we make a detailed global analysis for the structure and sizes of the connected components . we use generating function methods to write down a differential equation that implies recursion relations for the distribution of component sizes , see eqs.([cdiff],[crecur ] ) . a salient global feature of the model is a percolation phase transition at a critical value of the average connectivity . below this value , no single component contains a finite fraction of the sites in the thermodynamic limit , i.e. in the large @xmath1 limit . however , a slightly unusual situation occurs in that below the transition the system contains components whose sizes scale like a power of the total size of the graph , see eq.([eq : grosclu ] ) . correspondingly , the probability distribution for component sizes has an algebraic queue , see eq.([asympk ] ) , and its number of finite moments jumps at specific values of the average connectivity . above the transition , this probability distribution becomes defective , but its decrease is exponential , see eq.([pklarge ] ) . the transition is continuous . close to the threshold , the fraction of sites in the giant component the percolation cluster has an essential singularity , see eq.([eq : pof ] ) . we argue that this result is universal , with the meaning used in the study of critical phenomena . the essential singularity at the percolation threshold had already been observed numerically by @xcite in a different model which we show to be in the same universality class as ours for the percolation transition , and computed analytically for another class of models in @xcite . we then turn to the study of local - in - time profiles of connected components . guided by a direct enumeration based on tree combinatorics , we show that they satisfy recursion relations , and we give the first few profiles ( isolated vertices , pairs , triples ) explicitly . the profile of the giant component is given by a differential equation , from which we extract the singularity in the far past and the critical singularity in the present see eqs([eq : rho_0],[eq : rho_1 ] ) . in particular the giant component invades all the time slices of the graph above the transition . one strange feature of profiles , which would deserve a good explanation , is that in several instances the formal parameter involved in generating functions for global quantities is simply traded for the relative age to obtain interesting local - in - time observables , see eqs.([eq : young],[ddiff ] ) . we have compared our analytical results with numerical simulation whenever possible . while polishing this paper , we became aware of @xcite , whose goals overlap partly with ours . when they can be compared , the results agree we construct evolving random graphs with the following rules : + ( i ) we consider a triangular array of independent random variables @xmath5 , @xmath6 , where @xmath5 takes value @xmath0 with probability @xmath7 $ ] and value @xmath3 with probability @xmath8 . + ( ii ) we start from the graph made of single vertex at initial time @xmath9 . at time @xmath1 , @xmath10 , a vertex with label @xmath1 is added together with the _ directed _ edges @xmath11 $ ] for which @xmath12 . we shall often take the viewpoint that the ( biased ) coin tossings defining @xmath13 are done at time @xmath1 . we shall assume that @xmath14 at large time @xmath1 , with @xmath2 a parameter which we shall identify as half the average connectivity . this choice ensures the convergence of various distributions to stationary measures , most of them being independent of the precise values of the early probabilities . by construction all edges arriving at a given vertex are simultaneously created at the instant of creation of this vertex . as a consequence , these graphs are not only oriented but chronologically oriented this is unrealistic from the biological viewpoint . in this section we give the incoming and outgoing edge distributions . + let @xmath15 be the number of incoming edges at the vertex @xmath16 , and @xmath17 be the number of outgoing edges at this vertex at time @xmath1 . + let @xmath18 be the number of vertices with @xmath19 incoming edges , and @xmath20 be the number of vertices with @xmath19 outgoing edges at time @xmath1 . we may look either at the edge distributions at a given vertex , or we may look at the edge distributions defined by gathering averaged histograms over whole graphs . the former are specified by their generating functions , @xmath21 where @xmath22 denotes expectation value . it may depend on the specified vertex labeled by @xmath16 . the latter is defined by the generating functions , @xmath23 we remark that this global histogram distribution is the average of the local - in - time quantity @xmath24 . since at time @xmath1 the total number of vertices is @xmath1 , @xmath25 is properly normalized , @xmath26 , to define an averaged probability distribution function , independent of the vertices , for the incoming or outgoing edge variables @xmath27 : @xmath28 _ incoming vertices . _ the number of incoming edges @xmath29 at vertex @xmath30 asymptotically possesses a poisson distribution since @xmath31\simeq_{j\to\infty}\exp({\alpha}(z-1))\ ] ] the convergence of this distribution justifies our choice of asymptotic probabilities @xmath32 . only the vertices whose ages @xmath16 scale with the age of the graph , i.e. with @xmath33 fixed , @xmath34 , give non trivial contributions at large time to the averaged histogram ( [ histo ] ) and @xmath35 this expression may also be retrieved by looking at the evolution equation of @xmath36 . indeed , consider adding the new vertex at time @xmath1 . since the edges are oriented from older to younger vertices , . ] we have @xmath37 from the second definition in eq.([histo ] ) . this is equivalent to @xmath38 as @xmath39 at large time , the stationary limit is given by eq.([v- ] ) . this yields a poissonian distribution with probabilities @xmath40 _ outgoing vertices . _ at a given vertex @xmath30 , with @xmath33 fixed , the number of outgoing edges @xmath41 at vertex @xmath16 also have a poisson distribution at large time , @xmath42 \simeq_{t\to\infty } \exp(-{\alpha}\log\sigma(z-1))\ ] ] but with a parameter @xmath43 depending on the age of the vertex . approximating at large time the sum over @xmath16 in eq.([histo ] ) by an integral over @xmath44 gives the histogram distribution : @xmath45 as for incoming vertices , this formula follows from the evolution equation for @xmath46 . indeed , since the numbers of outgoing edges @xmath47 from vertex @xmath16 at time @xmath1 and @xmath48 differ by @xmath49 we have @xmath50 . from definition ( [ histo ] ) this gives @xmath51 where the first term is the contribution of the newly added vertex at time @xmath1 . the stationary limit is given by eq.([v+ ] ) . this is a geometric distribution , slightly larger than the poisson distribution , with probabilities @xmath52 _ mixed distribution_. let @xmath53 be the number of vertices with @xmath54 outgoing and @xmath55 incoming edges at time @xmath1 . as in eq.([histo ] ) , the generating function for the mixed histogram distribution is defined by @xmath56 by construction the outgoing and incoming edges variables @xmath57 are statistically independent for @xmath16 fixed , so that the last expectation values factorize . as above we may derive an evolution equation by evaluating the contribution of the newly added vertex at time @xmath1 this yields : @xmath58 its stationary limit is factorized : @xmath59 outgoing and incoming edges are statistically independent at large time . in this section , we present the main relations governing the probability distributions of connected components of the graphs . two vertices belong to the same connected component if they can be joined by a path made of edges , without any reference to orientation . this definition ensures that the property of being in the same connected component is an equivalence relation , but does not ensure that two points in a connected component can be joined by an oriented path . to partly avoid repetitions , the term _ cluster _ is used as a synonymous for _ connected component _ in the sequel . intuitively , the fact that the network is fragmented can be understood as follows : when a vertex @xmath60 is created , it has a finite probability to be isolated , and the probability that none of the vertices @xmath61 connects to vertex @xmath60 is @xmath62 which scales as @xmath63 . this quantity remains finite as long as @xmath64 does . this argument shows that there are isolated vertices in the system . a small extension of the argument shows that there are also finite components and that young vertices are more likely to be in small components than old ones . this will be made more rigourous in the study of profiles , see section [ sec : chropro ] . let @xmath65 be the number of connected components with @xmath19 vertices at time @xmath1 and let @xmath66 be the generating function , @xmath67 by definition , @xmath68 is the number of components and @xmath69 the total number of vertices , @xmath70 at any finite time . let us write an evolution equation for @xmath66 . at time @xmath71 , we add the vertex with label @xmath71 which may then be connected to @xmath72 connected components of size @xmath19 . this creates a new component of size @xmath73 , but also removes @xmath72 components of size @xmath19 . thus , at time @xmath71 we have : @xmath74 with @xmath75 the kronecker symbol . alternatively , @xmath76 as is apparent from this formulation , the transition probability from a given @xmath66 to a given @xmath77 can be given in closed form . to be precise , the admissible @xmath66 s ( describing the accessible distributions of components at time @xmath1 ) are polynomials with integral non - negative coefficients , whose derivative at @xmath78 have value @xmath1 . now suppose @xmath66 and @xmath77 are admissible . if the difference @xmath79 can not be written as @xmath80 for some set of nonnegative integers @xmath72 , the transition is forbidden . if it can , then the @xmath72 s are uniquely defined and the transition probability is @xmath81 the meaning of this equation is simple . at time @xmath71 , the new vertex is added , and for each of the former @xmath1 points a ( biased ) coin is tossed to decide the value of the edge variables @xmath82 . the tossings are independent with the same law , so the probability that the new point does not attach to a given component of size @xmath19 is @xmath83 , and distinct components are independent . hence for each @xmath19 one makes @xmath65 independent bernoulli trials with failure probability @xmath83 , and the transition from @xmath84 requires exactly @xmath72 successes . this shows that the graph evolution is a ( time inhomogeneous ) markov process on the space of components distributions , a fact that we shall use for the purpose of numerical simulations . this explicit representation of the transition probability could be used to average equation ( [ ntbis ] ) . alternatively , one can represent the number @xmath72 of components of size @xmath19 which are connected to the new vertex in terms of the edge variables @xmath5 as @xmath85=1}^{n_k(t)}[1 - \prod_{j\in[k]}(1-\ell_{j , t+1})]\ ] ] where @xmath86 $ ] runs over connected components of size @xmath19 . since the edge variables @xmath82 are statistically independent of the earlier edge variables , @xmath87 with @xmath88 , and therefore also independent of the @xmath65 s , we have for any @xmath89 , @xmath90^{n_k(t ) } \rangle } \label{ndist}\end{aligned}\ ] ] in particular , @xmath91 we can now take the average value of eq.([ntbis ] ) to get @xmath92^{n_k(t ) } \rangle } \nonumber \\ & = & { \langle n_{t}(zq_{t+1 } ) \rangle } + z q_{t+1}^t { \langle \prod_{k\geq 1}[1+(q_{t+1}^{-k}-1)z^k]^{n_k(t ) } \rangle } \label{nmoyen}\end{aligned}\ ] ] assume , but this will be justified later using the asymptotic behavior of the @xmath93 s , that at large time and for fixed component size @xmath94 is self - averaging , meaning that @xmath95 with @xmath96 equals to its averaged value with probability one and with the remaining other @xmath97 terms random . this in particular implies that for any finite size @xmath19 the number of connected components of size @xmath19 scales thermodynamically with the graph size . if both sides of eq.([nmoyen ] ) are expended in powers of @xmath98 , a given degree involves only components of bounded ( @xmath1-independent ) size . so , order by order in @xmath98 , self - averaging applies and we conclude that taking @xmath99 , the following is an accurate approximation for the last term in eq.([nmoyen ] ) at large times : @xmath100^{n_k(t ) } \simeq_{t\to \infty } \exp(-{\alpha}+{\alpha}\sum_{k\geq 1}kz^kc_k).\ ] ] the averaged evolution equation ( [ nmoyen ] ) then gives a deterministic differential equation @xmath101 for the generating function @xmath102 the function @xmath103 has a direct probabilistic interpretation . indeed , @xmath104 is the fraction of points in clusters of size @xmath19 , or equivalently the probability that a randomly chosen vertex belongs to a connected component of size @xmath19 . by construction @xmath105 is a taylor series with positive coefficients . the series @xmath106 is convergent because it counts the fraction of points in finite clusters which is @xmath107 . as a consequence the radius of convergence of @xmath105 is at least @xmath0 : if we denote by @xmath108 the radius of convergence @xmath105 , we know that @xmath109 for @xmath110 . we show in appendix a that analogous methods can be used to describe the connected components of the erds - renyi random graph model . most of our subsequent analysis will be based on ( [ cdiff ] ) . it recursively determines the @xmath96 s . the first few are : @xmath111 more generally , the combinations @xmath112 are rational functions in @xmath2 . they are more efficiently computed using a second order differential equation which follows from derivation of the logarithm of eq.([cdiff ] ) : @xmath113 this leads to @xmath114 a general feature of the recursion relations is that the rational function @xmath112 has no poles except possibly at the points @xmath115 . eq.([nmoyen ] ) is not a closed formula for the probability distributions of the number of connected components . but as shown in appendix b , one may improve slightly the above argument to obtain a closed system ( see eq.[mast ] ) which can be used to prove systematically that the variables @xmath96 are self - averaging , a task we perform for @xmath116 and @xmath117 in appendix b. the proof becomes more and more tedious for large @xmath19 . but we are confident that the self averaging property is general because of the close relationship with the erds - renyi model presented in appendix a. up to now , our analysis of connected components has always concentrated on finite components : the thermodynamic limit @xmath118 has been taken while keeping @xmath19 , the component size , arbitrary but fixed . we have argued that in this regime , the number of components of size @xmath19 is proportional to @xmath1 . this has lead to a satisfactory description of @xmath96 . our arguments did not use any hypothesis on whether or not only finite components play a role in the thermodynamic limit . however , as we observed already , the sum @xmath119 measures exactly the fraction of the sites that are either isolated or in components of size @xmath120 or @xmath121 or @xmath122 . to rephrase it more vaguely , @xmath119 counts the fraction of vertices that belong to finite connected components . if only clusters of finite size contribute to the thermodynamic limit , then @xmath123 . else , by standard physical arguments , a single giant component of size @xmath124 accounts for the deficit . the giant component is also called the percolating cluster . its relative size , which we denote by @xmath125 , is @xmath126 and @xmath127 . left : the fraction of sites occupied by the giant component . right : the variation of @xmath128 close to the transition . _ _ ] and @xmath127 . left : the fraction of sites occupied by the giant component . right : the variation of @xmath128 close to the transition . _ _ ] to discriminate between the two situations , @xmath125 may be computed numerically by evaluating a large number of coefficients @xmath104 using the recursion relation ( [ crecur ] ) . as we shall see later , the convergence of the series is slow for the whole relevant range of @xmath2 . the result of such a partial summation @xmath129 is plotted in fig.[fig : perconum ] for @xmath130 . it reveals a phase transition at a value @xmath131 between @xmath132 and @xmath133 , going from a regime where finite components contain all vertices to a regime where they do not . below @xmath132 and above @xmath133 , the plots corresponding to different values of @xmath134 are hard to distinguish , but in the transition region , large values of @xmath19 make substantial contributions to @xmath135 . the transition is also manifest on an analogous study of @xmath136 . we shall show later that @xmath137 is the exact threshold in this model and that @xmath136 is discontinuous at the transition . the growth of @xmath125 just above the threshold seems to start with many vanishing derivatives , in strong contrast with what happens in the erds - renyi random graph , for which the growth of the giant component is linear close to the transition . this can be related to the following observations : as we have recalled in appendix a , in the erds - renyi model the components of size @xmath19 occupy a fraction @xmath138 . as a function of @xmath2 , this fraction has a single maximum at @xmath139 . these values accumulate at @xmath140 , the well - known transition point for the standard random graph . then for @xmath141 , the fraction of sites occupied by components of size @xmath19 decreases with a finite slope for all @xmath19 s , and so does the sum , so that the growth of the giant component is linear close to the transition . in the model studied in this paper , the behavior of @xmath142 as a function of @xmath2 for generic @xmath19 is not so easy to get at . however , a simple numerical analysis leads to the following picture : @xmath143 is a decreasing function of @xmath2 , but for @xmath144 , @xmath104 has a single maximum , at say @xmath145 . this sequence starts with @xmath146 , @xmath147 @xmath148 , is maximum for @xmath149 with @xmath150 and then decreases very slowly ( @xmath151 , @xmath152 , @xmath153 ) , apparently getting closer and closer to @xmath137 . at the transition , most @xmath142 s are still increasing , and close enough to the transition a finite but large number of them is still increasing . so subtle compensation mechanisms can take place , leading possibly to the vanishing of ( infinitely ) many derivatives of @xmath125 at @xmath154 . this is confirmed in the following subsections , which are also devoted to a more precise description of the distributions of finite and infinite clusters . to summarize : @xmath155 we turn to the examination of the consequences of eq.([cdiff ] ) , the equation that determines the generating function for the number of clusters of given finite size . we know that @xmath108 , the radius of convergence of @xmath105 , is at least @xmath0 . to analyze the behavior of @xmath105 around @xmath78 , we define @xmath157 from eq.([cdiff ] ) @xmath158 satisfies @xmath159 below the transition , there is no percolating cluster so that @xmath160 or alternatively , @xmath161 this makes clear that the normalized positive numbers @xmath104 are the probabilities for a vertex to be in a connected component of size @xmath19 . for @xmath162 this translates into the boundary condition @xmath163 we first look for a formal solution @xmath164 of eq.([feq ] ) in the form of a taylor series in @xmath165 : @xmath166 the taylor coefficients are the moments of the measure @xmath104 : @xmath167 . for example @xmath168 is the average proportion of vertices per component . as eq.([cd2 ] ) the differential equation eq.([cdiff ] ) may be turned into a second order differential equation for @xmath169 , @xmath170 eq.([cdiff ] ) with @xmath171 gives @xmath172 . eq.([ydiff ] ) then allows us to recursively compute the @xmath173 s : @xmath174 with @xmath175 the square root singularity in the expression of @xmath168 indicates that the initial boundary condition ( [ mesp ] ) becomes pathological at @xmath154 and thus signals the percolation transition . the recursion relation for the higher taylor coefficients shows that @xmath176 possesses a pole at @xmath177 . it actually changes sign , from positive to negative , across the pole . as a function of @xmath178 , the @xmath176 s have a simple pole at @xmath179 , @xmath180 where the numerator does not vanish at @xmath181 . since @xmath182 at least for @xmath183 , this simple pole codes for the asymptotic behavior of @xmath104 at large @xmath19 . indeed , recall that if @xmath184 as @xmath185 then @xmath186 diverges as @xmath187 from below with a simple pole at this value . thus assuming that we may extend the simple pole ( [ simplet ] ) to non - integers value of @xmath178 we learn that @xmath188 with @xmath189 this means in particular that the probability distribution @xmath104 ( which describes the proportion of the system occupied by clusters of size @xmath19 ) is large and only the moments @xmath176 for @xmath183 , exist . the fact that the taylor coefficients @xmath176 , computed from the recursion relation ( [ mu < ] ) , can not consistently be interpreted as moments of a probability distribution for @xmath190 indicates that the expansion ( [ trunc ] ) is only up to @xmath191 terms . indeed , the differential equation ( [ feq ] ) is compatible with an expansion of @xmath162 for @xmath192 , @xmath193 , of the form : @xmath194 with @xmath195 and @xmath196 . as a function of the complex variable @xmath98 , @xmath197 has thus a branch cut starting at @xmath78 : @xmath198 for @xmath179 an integer , this formula should becomes @xmath199 . the cut implies the asymptotic behavior ( [ asympk ] ) for @xmath200 . at the transition but from below , i.e. @xmath201 , there are logarithmic correction to the scaling behavior ( [ asympk ] ) , and the branch cut equation is @xmath202 . this ensures that the first moment is finite , and its value , computed from ( [ mu < ] ) , is @xmath203 the scaling law ( [ asympk ] ) may be linked to the typical growth rate of large clusters in the system . for concreteness , consider the component of vertex @xmath205 for any given @xmath205 . for very large @xmath1 , the number of arrows emerging from vertex @xmath205 grows like @xmath206 . then we infer that the size of the genealogical tree of @xmath205 will grow like @xmath207 ( under the hypothesis that the genealogical tree is indeed tree - like , a reasonable assumption for small @xmath2 ) . this counting of descendants gives a crude lower bound for the size of the connected component of @xmath205 . hence we expect that the system contains components whose sizes grow like a power of @xmath1 . to estimate this power , we argue as follows . consider a given large cluster of size @xmath208 at time @xmath1 : \i ) @xmath209 is @xmath3 with probability @xmath210 and @xmath211 with probability @xmath212 times the probability that vertex @xmath71 connects to @xmath213 clusters of size @xmath214 apart from the large cluster ; \ii ) removing the given large cluster does not change the thermodynamical properties of the graph , so the probability that vertex @xmath71 connects to @xmath213 clusters of size @xmath214 apart from the large cluster is simply the probability that vertex @xmath71 connects to @xmath213 clusters of size @xmath214 . hence for large @xmath1 , from eq.([nbis ] ) , @xmath215 ; \iii ) suppose we add @xmath216 new vertices with @xmath217 but @xmath218 . between time @xmath1 and time @xmath219 , many new clusters have be connected to the given large components so @xmath220 , but this has not changed the thermodynamical properties of the graph . hence we can average the equation in i ) to get a deterministic equation @xmath221 this leads to @xmath222 we find that the growth rate of large clusters is universal . as expected , their growth exponent is larger than @xmath2 , the genealogical tree growth , because it takes not only descendants into account but the whole component . the difference is maximum at the transition , where @xmath223 . the fact that the same exponent , @xmath224 , governs the asymptotic behavior of @xmath104 at large @xmath19 and the size of large clusters for large @xmath1 can be understood directly as follows . first consider one realization of the random graph for a given @xmath1 and suppose that there is a single component of maximal size , say @xmath225 . observe that @xmath226 is by definition the number of points in components of size larger than @xmath227 . this number is strictly larger than @xmath227 if @xmath228 , but vanishes if @xmath229 . so @xmath225 is characterized by the identity @xmath230 from this we infer by taking the average that for large @xmath1 the relation @xmath231 gives a sensible characterization for @xmath232 , the order of magnitude of the size of large components in the graph . we write @xmath233 and use that for @xmath234 , @xmath123 to write @xmath235 . for large @xmath232 , the asymptotics of the first sum is @xmath236 . the second sum is made of finite size corrections . if we assume that these that these are not too large , we conclude that @xmath237 , i.e. that @xmath238 . in fact , experience from finite size scaling suggests that the two sums give contributions of the same order of magnitude . the idea is that @xmath239 is the main contribution to @xmath240 not only when @xmath239 is of order @xmath1 , but even when simply @xmath241 so that self averaging remains valid . this means that @xmath242 and @xmath240 become of the same order of magnitude only when @xmath19 is so large that @xmath243 . so again we see that @xmath232 is characterized by @xmath244 , and we conjecture that @xmath245 is a scaling variable . above the transition there is an giant component . let @xmath125 be its relative size . by definition , @xmath247 or alternatively @xmath248 this makes clear that the @xmath104 s and @xmath125 define the probability distribution of vertices among the clusters of different sizes , with @xmath125 the probability for a vertex to be in the percolating cluster . the size of the giant component @xmath249 increases linearly with time : @xmath250 for @xmath1 large . the slope may be evaluated as follows . imagine adding a new vertex at time @xmath71 . it is connected to the percolating cluster with probability @xmath251 so that @xmath252 where , as in eq.([nt ] ) , @xmath213 are the numbers of components of size @xmath214 connected to the new vertex . this can be rearranged as @xmath253 . the quantity @xmath254 is self averaging , so using @xmath255 and @xmath256 at large time we infer @xmath257 which by use of eq.([nbis ] ) leads to @xmath258 as shown below the above equation is exact , but it does not determine @xmath125 as this requires knowing the moment @xmath168 . above the transition , @xmath162 still satisfies the differential equation ( [ feq ] ) but with a different boundary condition : @xmath259 with @xmath125 vanishing at the transition i.e. as @xmath260 . this modifies the behavior of its moments . so let us expand @xmath162 in taylor series around the origin : @xmath261 with @xmath262 . eq.([feq ] ) implies eq.([pinf1 ] ) which is thus exact . the second order differential equation ( [ ydiff ] ) does not fix @xmath125 but determines recursively the @xmath173 s which depend parametrically on @xmath125 . it remains to decipher what the behavior of the size of the giant component is , at least close to the transition . this will follow from an analysis of the behavior of @xmath162 close to its singularities . as we are going to show , @xmath162 possesses square root branch cut at a point @xmath263 but exponentially close to the origin . this means that above the transition the singularity of the function @xmath197 is at @xmath264 , with @xmath265 depending on @xmath2 , a behavior which has to be compared with the cut at @xmath78 below the transition , eq.([cut1 ] ) . more precisely , we show that @xmath266 with @xmath267 defined by @xmath268 to prove eq.([eq : sing ] ) we shall look at the behavior of @xmath158 in the neighborhood of three different points : at the origin @xmath269 where @xmath162 takes the boundary value ( [ fpinfty ] ) , at the branch point @xmath263 with @xmath270 , @xmath271 and , around the point @xmath272 specified by the condition @xmath273 . let us first show @xmath274 and @xmath275 exist . the function we are interested in is defined by @xmath276 , and satisfies the differential equation eq.([feq ] ) . so @xmath277 and @xmath158 decreases from @xmath278 at @xmath279 to @xmath3 at a point @xmath274 which is non - negative since @xmath280 . @xmath158 also satisfies the obvious inequalities @xmath281 for @xmath192 and @xmath282 . so there is a single point @xmath283 , with @xmath284 , at which @xmath285 and @xmath286 has the sign of @xmath287 . both @xmath274 and @xmath275 vanish at the transition , @xmath288 . it turns out that most of these properties could be proved using only the differential equation . more precisely , take any solution @xmath289 of ( [ feq ] ) with an initial condition at @xmath290 such that @xmath291 . the function @xmath289 can be extended on the right as a positive function on a maximal interval @xmath292 . then @xmath289 is strictly decreasing for @xmath293 , @xmath294 , @xmath295 , and @xmath296 if @xmath297 . this discussion emphasizes the intrinsic role played by @xmath298 in this problem . before proving eq.([eq : sing ] ) let us get intuition from a simpler , but more universal , version of the differential equation ( [ feq ] ) . if we look at a region where @xmath158 is small and we know that such regions exist generically we can estimate @xmath299 by @xmath300 with @xmath301 approximating @xmath158 . this leads to the simplified , but still non - linear , differential equation @xmath302 it turns out that this simpler equation can be solved in closed form . before doing that , let us further assume that @xmath165 is close to @xmath274 so that @xmath301 can be neglected compared to @xmath165 . then we get @xmath303 , with solution @xmath304 leading to the announced square root singularity . but this does not give informations on the location of the branch point . consider now the following functional of an arbitrary function @xmath301 of @xmath165 @xmath305 to fix conventions , we specify the function @xmath306 by demanding that it is continuous and takes value in @xmath307-\pi/2,\pi/2[$ ] . the total derivative of this functional with respect to @xmath165 is @xmath308 it vanishes if @xmath301 is a solution of eq.([eq : univ ] ) , so we have indeed solved in closed form equation ( [ feq ] ) for small @xmath158 . as this is the domain we are interested in , it is tempting to argue that @xmath158 and @xmath301 should exhibit the same singular behaviour . this turns out to be true , but there are some subtleties because the limit @xmath260 is singular : the size of the domain for which @xmath301 is a uniform approximation to @xmath158 shrinks to @xmath3 . our strategy is to use the invariant ( [ eq : int ] ) for the approximate equation ( [ eq : univ ] ) to derive exact inequalities for @xmath158 . let us observe that the functional ( [ eq : int ] ) is singular at points where @xmath309 where it has a jump of amplitude @xmath310 . with this in mind , we define the functional @xmath311 in this definition , @xmath312 is a priori independent of @xmath2 , the value for which @xmath162 is considered . the functional @xmath313 is a smooth function of @xmath165 on @xmath307-\infty,{\tau}_c[$ ] . using the differential equation for @xmath158 , its total derivative with respect to @xmath165 is found to be @xmath314 if @xmath315 , the right - hand side is always negative , so @xmath316 is decreasing , and comparing its values at @xmath317 , at @xmath275 , at @xmath3 and at @xmath274 we find @xmath318 we know that @xmath319 , so taking a fixed @xmath320 , we can take the limit @xmath260 . at point @xmath165 , @xmath158 is analytic in @xmath321 , so @xmath322 has a finite limit , and we get exponentially small upper - bounds for @xmath323 and @xmath274 . for instance , @xmath324 is bounded above when @xmath260 . for @xmath325 , @xmath326 changes sign at a point @xmath327 solution of @xmath328 . this point is unique as @xmath158 is decreasing and goes from @xmath278 at @xmath329 to @xmath3 at @xmath274 . let us choose @xmath330 such that @xmath331 and such choices exist . then @xmath326 goes from negative to positive if @xmath165 increases so that @xmath316 increases for @xmath165 varying from @xmath327 to @xmath274 . this leads to lower bounds @xmath332 take @xmath333 , in such a way that @xmath334 vanishes exactly at @xmath275 . then @xmath335 , and @xmath336 when @xmath260 . comparing the lower bounds with the upper bounds obtained above , we see that @xmath337 to get a lower bound for @xmath338 we take @xmath339 such that @xmath334 vanishes at @xmath340 . then @xmath341 and @xmath342 so we have a crude bound @xmath343 . taking for instance @xmath344 , so that @xmath345 and @xmath346 as @xmath347 . reporting in @xmath348 yields @xmath349 . thus , from ( [ eq : borninf ] ) , @xmath350 is bounded below when @xmath260 . the above upper and lower bounds then imply @xmath351 we restate the physically most important result : up to an algebraic prefactor , the size of the percolation cluster close to percolation is @xmath352 in fact , we have obtained a better estimate . we expect that @xmath353 as a finite limit for @xmath347 for a certain @xmath354 , or equivalently @xmath355 we have proved that if @xmath354 exists , @xmath356 . the asymptotic behavior of the probabilities @xmath104 is not the same below and above the transition as @xmath105 does not have the same analytical properties on the two sides of the transition . since above the transition the branch point is located at @xmath357 , the @xmath104 s now decrease exponentially . more precisely , @xmath197 possesses a square root branch point at @xmath265 , @xmath358 so that @xmath359 behave as @xmath360 to be compared with eq.([asympk ] ) . we give now a universality argument suggesting that @xmath361 and this is confirmed by solving numerically the differential equation ( [ feq ] ) . our invocation of universality rests on eq.([pklarge ] ) , in which @xmath274 , or equivalently ( in the vicinity of the critical point ) @xmath362 , controls the exponential decrease of the @xmath104 s and plays the role of a mass gap . this mass gap is exponentially small close to the transition , and by analogy we may argue that increasing @xmath267 is a marginally relevant perturbation of the percolating critical point . introducing the wilson - callan - symanzik beta function @xmath363 , we expect a relation @xmath364 comparison with our formula gives @xmath365 . it is known from field theory that the coefficient @xmath366 , which dictates the exponent of the algebraic prefactor , is universal . such universal features are controlled by the continuum limit . in our framework , the continuum region is reached at small @xmath165 , and small @xmath158 , and we expect that the continuum limit is governed by ( [ eq : univ ] ) . note that both @xmath274 and @xmath338 are exponentially small close to the transition , and @xmath162 remains small for @xmath165 between these points . this a posteriori justifies looking at the approximate equation ( [ eq : univ ] ) . consequently , as far as universal quantities are concerned , the inequalities in ( [ eq : bornsup ] ) can be replaced by equalities . this leads to @xmath361 , or @xmath367 i.e. @xmath368 as announced . as universality could suggest , eq([eq : univ ] ) turns out to describe the continuum limit for a larger class of evolving networks than just the specific one we are studying . this is the case for the model studied in @xcite . we refer to the original paper for the definitions . it suffices to say that eq.([feq ] ) is replaced by @xmath369 in this equation , @xmath370 is a generating function , the coefficient of @xmath371 giving the fraction of points in components of finite @xmath19 , so @xmath372 is the fraction of sites occupied by the giant component . the parameter @xmath373 is the average number of edges created at each time step , so this is is precisely the equivalent of our @xmath2 . to study @xmath374 close to @xmath269 , the approximation is @xmath375 , and we retrieve ( [ eq : univ ] ) , with @xmath2 replaced by @xmath376 . the percolation threshold is @xmath377 ( as expected , percolation thresholds are not universal ) , and the size of the infinite component behaves like @xmath378 . the prefactor @xmath379 compares quite well with the numerical value @xmath380 obtained in the original paper . this fact was already noted in @xcite . we have accumulated evidence that the percolation phase transition is very similar to the kosterlitz - thouless phase transition in the xy model . for instance , the probabilities @xmath104 follow the scaling laws ( [ asympk ] ) below the transition with scaling exponents varying continuously with the parameter @xmath2 while they decrease exponentially above the transition . the discussion of universality in the previous section suggests to look for scaling functions describing the neighborhood of the critical point . for instance , from eq.([ydiff ] ) it follows that the moments @xmath173 , @xmath381 , diverge as @xmath382 with @xmath383 in particular , @xmath168 is discontinuous at the transition , jumping from @xmath384 to @xmath385 . these scaling relations imply that the function @xmath386 has a finite limit , denoted by @xmath387 , as @xmath260 for any fixed @xmath388 , @xmath389 the differential equation ( [ feq ] ) at @xmath390 then reduces to @xmath391 . with the boundary condition @xmath392 , this is integrated as : @xmath393 the left - hand side is very reminiscent of the limit @xmath394 of eq.([eq : int ] ) . as a consequence @xmath387 has a square root branch point at @xmath395 at which @xmath396 vanishes and @xmath397 for @xmath398 large actually , a more precise computation based on the lagrange formula presented in appendix c yields the exact value of the scaling coefficients @xmath399 : @xmath400 one can make a direct counting of the average number of connected components in the random graph at time @xmath1 which are copies of a given finite labeled graph . this will allow us to retrieve the results of section [ sec : clusters ] . but this also leads to a detailed local - in - time description of the connected components which illustrates the consequences of the chronological memory of our model . let @xmath401 be a labeled graph with vertices @xmath402 . we let @xmath403 be the number of edges connecting vertex @xmath16 to a vertex with smaller label , so that @xmath404 is the number of edges of @xmath401 . we look for the average number of increasing maps @xmath405 from @xmath406 $ ] to @xmath407 $ ] such that the vertices @xmath408 span a connected component of the random graph isomorphic to @xmath401 . this number is the average number of connected components isomorphic to @xmath401 in the random graph . by the rules of construction of the random graph , the probability that the vertices @xmath409 with @xmath410 span a connected component of the random graph isomorphic to @xmath401 is @xmath411 with the convention @xmath412 . the average we look for is obtained by summing this expression over the @xmath413 s . for large @xmath1 , using the asymptotic behavior @xmath414 for large @xmath16 , the sum can be formally reinterpreted as a riemann sum , leading to a contribution @xmath415 in deriving this formula , we have not treated carefully the contribution of small values of the @xmath413 s . this is reflected in the fact that the integral can be divergent if some @xmath416 s are too large . however , the prefactor @xmath417 is the sign that in this case the sum over the @xmath413 s is nevertheless negligible in the large @xmath1 limit , as a more careful treatment would show . the most salient feature of this formula is that only connected graphs with @xmath418 give a contribution proportional to @xmath1 , i.e. contribute to @xmath96 . since @xmath19 is the number of vertices and @xmath398 the number of links , this relation characterizes trees as follows from the euler formula : only trees contribute to the thermodynamic limit . a given tree on @xmath19 vertices with incoming degrees @xmath416 , @xmath419 gives a contribution @xmath420 to @xmath96 . observe that for a tree , @xmath421 for @xmath419 so that all integrals are well - defined and finite for real non - negative @xmath2 . it is amusing to note also that the contribution of a single tree of size @xmath19 can contain poles in @xmath2 at values that are not in the list @xmath115 . these poles have to cancel between different trees in the sum over trees of size @xmath19 , because we know that they are absent in @xmath96 . but we have no simple explanation for this cancellation . this explicit formula makes it easy to show that @xmath105 is ( complex ) analytic in @xmath2 in a neighborhood of @xmath3070,+\infty[$ ] for every @xmath98 such that @xmath422 . indeed , we know that for @xmath4230,+\infty[$ ] , and @xmath422 the series @xmath424 is absolutely convergent . but @xmath142 is a sum of non - negative contributions , each tree giving a contribution of the form ( [ eq : tree ] ) . now suppose that @xmath2 is complex with positive real part . for each tree contribution and for fixed @xmath425 , the modulus of @xmath426 is maximal when @xmath427 and then the expression is real and positive . this is because the statement is true for every factor . taking the sum over trees we infer that @xmath428 . so the series @xmath424 is absolutely convergent if @xmath429 . this equation defines , for fixed @xmath422 , a neighborhood of @xmath4230,+\infty[$ ] in which we have an absolutely convergent sum of analytic functions analytic of @xmath2 . hence the sum is analytic in @xmath2 as claimed . to resum more explicitly the contribution of all trees of a given size , we need the generating function for labeled trees with given incoming degrees . suppose more generally that we give a weight @xmath430 for each edge leaving vertex @xmath431 ( i.e. connecting @xmath431 to a @xmath432 ) and a weight @xmath433 for each edge entering vertex @xmath431 ( i.e. connecting @xmath431 to a @xmath434 ) . the generating function @xmath435 for weighted trees on @xmath178 vertices factorizes nicely as @xmath436 this generalization of the famous caley tree formula labeled trees on @xmath178 vertices . ] implies it immediately . it seems to be little known , although it is implicit in the mathematical literature @xcite . gilles schaeffer @xcite provided us with a clean proof using a refined version of one of the standard proofs of the caley tree formula , putting trees on @xmath19 vertices in one to one correspondence with applications from @xmath406 $ ] to @xmath406 $ ] fixing @xmath0 and @xmath19 , see e.g. @xcite . this formula can be specialized to @xmath437 and @xmath438 for @xmath419 to give @xmath439 if one integrates only over a subset of the @xmath44 s , one gets marginal distributions . for instance , for @xmath440 , we get that in the thermodynamic limit the fraction of sites with age close to @xmath441 that are isolated is @xmath442 . for @xmath443 , if we integrate over @xmath444 , we get that the fraction of sites with age close to @xmath441 that are the older vertex of a tree on two vertices is @xmath445 while if we integrate over @xmath446 , we get that the fraction of sites with age close to @xmath441 that are the younger vertex of a tree on two vertices is @xmath447 . the sum , @xmath448 , gives the probability that a site with age close to @xmath441 belongs to a tree on @xmath120 vertices . our explicit representation in terms of trees shows that in this model , and at least for questions concerning connected components , the thermodynamic limit applies not only to the full system , but also to slices of fixed relative age @xmath449 . in the next section we shall study @xmath44-dependent profiles . for small @xmath19 s , we have done all the integrals and checked the agreement with the value of @xmath96 obtained by the recursion relation . but a general proof valid for all @xmath19 s is lacking . to illustrate the evolving nature of our model , we now determine the local - in - time distribution of the cluster sizes . this means determining for any given age interval what is the proportion of vertices of these ages which belong to clusters of given size . define @xmath450 to be the probability that vertex @xmath205 belongs to a component of size @xmath19 at time @xmath1 . guided by the previous tree representation , we infer that in the thermodynamical limit @xmath451}\ , p_k(t , t')\simeq t\rho_k(\sigma)d\sigma$ ] , with @xmath452 a deterministic function . by construction , @xmath453 , the total fraction of points that belong to components of size @xmath19 . the reasoning leading to ( [ cdiff ] ) can be generalized : one writes down a recursion relation for @xmath454 and then takes the average , a step justified by the explicit tree representation . the event that vertex @xmath205 belongs to a component of size @xmath19 at time @xmath71 is the exclusive union of several events . \i ) vertex @xmath205 belonged to a component of size @xmath19 at time @xmath1 and this component is not linked to the new vertex @xmath71 . this has probability @xmath455 . \ii ) vertex @xmath205 belonged to a component of size @xmath456 at time @xmath1 , this component is linked to the new vertex @xmath71 , and together with the other components linked to @xmath71 ( say @xmath457 components of size @xmath398 ) , it builds a component of size @xmath458 . this has probability @xmath459 to perform the explicit sum over @xmath227 and the @xmath457 s , we introduce again generating functions and set @xmath460 . this leads to @xmath461 in the large @xmath1 limit this complicated formula simplifies if we use again the hypothesis of self - averaging and asymptotic independence . defining @xmath462 this leads to @xmath463 together with the sum rule relating @xmath464 to @xmath465 , this fixes completely the profiles @xmath452 . a relation between @xmath466 and @xmath105 is obtained by integrating ( [ profit ] ) for @xmath44 between @xmath3 and @xmath0 . using the defining equation for @xmath105 , eq.([cdiff ] ) , this leads to @xmath467 . thus , we can summarize our knowledge on component time profiles with the four relations : @xmath468 expansion in powers of @xmath98 leads to recursion relations for the @xmath452 s . they can be shown to be alternating polynomials of degree @xmath19 in @xmath469 with vanishing constant coefficient . the first few polynomials are @xmath470 again , we have checked that for small @xmath19 s the values of @xmath471 as computed from iterated tree integrals and from the generating function coincide , but we have no general proof . as an application , let us look at the profile of vertices of relative age close to @xmath44 which are the youngest in components of size @xmath19 . these are vertices that created , when they appeared , a component of size @xmath19 this happens with probability @xmath472 which was then left untouched for the rest of the evolution this happens with probability @xmath473 . so the distribution of vertices that are the youngest in their connected ( finite ) component is @xmath474 in this expression , the formal parameter @xmath98 has been replaced by @xmath475 . the giant component , when it exists , also has a well - defined profile to which we turn now . we now determine the profile of the giant component . so let @xmath476 be the fraction of vertices whose ages are between @xmath477 and @xmath478 which belong to the giant component . by definition @xmath479 . we know from the tree representation that for finite clusters the thermodynamic limit applies not only on the full graph , but also on time slices . the giant component is the complement of finite components , so we expect that the density @xmath480 is self - averaging as is the size of the percolating cluster . to derive an equation fixing this density we look for the probability for a site of age @xmath481 not to be in the percolating cluster . on the one hand , by definition of the density this probability is @xmath482 ) = 1-\rho_{\infty}(\sigma)\ ] ] on the other hand , this probability may be evaluated by demanding the vertex @xmath16 not to be connected to the older and younger vertices of the percolating cluster : @xmath482)= \prod_{k < j , k\in[k_\infty]}(1-\frac{{\alpha}}{j } ) \prod_{k > j , k\in[k_\infty]}(1-\frac{{\alpha}}{k})\ ] ] at large time , the first above product converges to @xmath483 and the second to @xmath484 . so we get the relation : @xmath485 as @xmath486 is positive , @xmath487 is an increasing function of @xmath44 . so @xmath480 is decreasing and has a right limit at @xmath3 . unless @xmath488 ( i.e. @xmath489 ) , the integral has a logarithmic divergence , and henceforth @xmath490 . more precisely , @xmath491 this means that the early vertices belong to the giant component with probability @xmath0 . on the other hand , by definition @xmath492 so taking @xmath493 in ( [ profil ] ) we get that @xmath494 this means that the late vertices always belong to the giant component with a non vanishing probability although this probability is exponentially small close to the threshold . hence the giant component invades all time slices above the threshold , and the term percolation transition is appropriate even with this unusual interpretation . our results are illustrated on fig.[fig : profinfin ] . random graphs of size @xmath495 . from top to bottom , the values of @xmath2 are @xmath0 , @xmath496 and @xmath497.__,scaledwidth=70.0% ] to conclude this discussion , let us observe that @xmath486 can be expressed in terms of the profiles of finite components that we studied before : @xmath498 with @xmath499 defined in eq.([defrhoz ] ) . we end up with yet another equation constraining @xmath499 , @xmath500 if we expand @xmath501 in powers of @xmath502 as @xmath503 and define @xmath504 , we can check that @xmath505 in particular , by differentiation this implies that @xmath506 the resemblance with equations ( [ cdiff ] ) and ( [ crecur ] ) for the function @xmath105 is rather striking , but we have not been able to use this more deeply . one difference is that ( [ cdiff ] ) determines @xmath116 from scratch , whereas ( [ ddiff ] ) does it in a two step process . if @xmath507 is the solution of this equation satisfying @xmath508 , then @xmath509 , and then @xmath510 has to satisfy @xmath511 another difference is that the sequence @xmath512 alternates in sign . again , the formal parameter @xmath98 receives a simple physical interpretation @xmath513 where @xmath44 labels the relative date of birth of vertices . in this study , we have described detailed global and local - in - time features of evolving random graphs with uniform attachment rules . concerning global properties , we have shown that the model has a percolation phase transition at @xmath154 . below the transition , the system contains clusters whose sizes scale like @xmath514 . above the transition , a single component , the giant component , grows steadily with time . we have shown that , close to the threshold , the fraction of sites in the giant component has an essential singularity and behaves as @xmath515 . the behaviors below and above the transitions are strongly reminiscent of the two dimensional @xmath516 model , so that our model can be interpreted as some algorithmic equivalent of it , but it seems unlikely that a direct connexion exists . this analogy and further scaling properties we present call for an alternative renormalization group approach to the transition . by describing local - in - time profiles , we have shown that they offer a more accurate vision of the specificities of evolving graphs . it would be desirable to generalize this approach by answering the following question : assume that a procedure to assign ages to the vertices of some evolving graphs has been given , what informations on the microscopic evolution rules of these graphs can be decoded from the knowledge of local - in - time statistics ? aknowledgements : we thank gilles schaeffer for his clarifying help in tree generating functions and sergei dorogovtsev for his comments and interest . we show how our arguments of section [ sec : clusters ] can be modified to describe the erds - renyi random graph model . in this model , one starts with @xmath178 points , and any two points are connected by an edge with probability @xmath517 ( so in this model all the points are equivalent ) . then a limit @xmath518 is taken . this famous model describes a static graph , but it can also be rephrased as an evolving graph in the following way : set @xmath519 and suppose that points are added one by one , from @xmath0 to @xmath1 , each new point connecting to any previous one with probability @xmath520 . from this point of view , it can be seen that looking only at the first @xmath205 vertices ( @xmath521 ) amount to look at an erds - renyi random graph with a modified connectivity parameter @xmath522 . to get a recursion relation , we start from an erds - renyi random graph of size @xmath1 with connectivity parameter @xmath2 . we add vertex @xmath71 and connect any older vertex to it with probability @xmath523 , so that the effective connectivity parameter for the graph on @xmath71 vertices is @xmath524 . then the derivation proceeds as before , with the little proviso that in eq.([nmoyen ] ) , on the left - hand side @xmath525 has an effective connectivity parameter @xmath524 instead of @xmath2 . so when we take the thermodynamic limit , nothing changes on the right - hand side of ( [ nmoyen ] ) , but an additional term @xmath526 contributes to the left - hand side . if @xmath527 denotes the analog of @xmath465 but for the erds - renyi random graph , then @xmath528 this little modification in the equation has drastic consequences . the new equation has a single solution regular at @xmath529 , namely @xmath530 which is the well - known result . we conclude that our self - averaging hypothesis is valid for the erds - renyi model . this makes it more plausible that it works for our original model as well , a fact also confirmed by numerical simulations . our goal is to prove the following formula and to investigate a few of its consequences . we claim that @xmath531^{n_m(t ) } \rangle } \label{mast}\end{aligned}\ ] ] where the contour integral is around the origin . it is a fokker - planck equation for the markov process formed by the @xmath65 s , which could be used to prove systematically that the variables @xmath96 are self - averaging , a task we perform for @xmath116 and @xmath117 at the end of this appendix . we start from eq.([nt ] ) which may be rewritten as @xmath532 to compute the last term we insert the tautological identity @xmath533 in the r.h.s . to get @xmath534 to compute the r.h.s . expectation value we use a contour integral representation of the kronecker symbol @xmath535 this yields @xmath536 the r.h.s . can now be computed using eq.([ndist ] ) and gives eq.([mast ] ) . this computation has a rather simple combinatorial reinterpretation . we set @xmath537 and observe that going from time @xmath1 to @xmath71 , we add vertex @xmath71 and edges from the rest of the graph to @xmath71 . suppose that the component of vertex @xmath71 has size say @xmath19 . this component was build by `` eating '' some components of the graph at time @xmath1 . a component of size @xmath16 is swallowed with probability @xmath538 and survives with probability @xmath539 . on the other hand , if one expands @xmath540 in powers of @xmath541 , the term of degree @xmath542 enumerates all the possibilities to build the component of vertex @xmath71 with the correct probability . defining @xmath543 , summation over @xmath19 gives @xmath544 as obtained previously . the fokker - planck eq.([mast ] ) may also be formulated as a difference equation , similar to a discrete schrodinger equation . let us for instance specify it for @xmath545 with @xmath546 , @xmath547 , a set of complex numbers and let @xmath548^{n_k(t ) } \rangle } \nonumber\end{aligned}\ ] ] this parametrization is similar to that used in matrix theory where @xmath549 may be thought of as the trace of the @xmath550 power of matrix whose eigenvalues are the @xmath546 s . the contour integral in eq.([mast ] ) can then be explicitly evaluated by deforming the integration contour to pick the simple pole contributions located at the points @xmath551 . this gives : @xmath552 eq.([mast ] ) or ( [ diffshro ] ) may be used to prove that the numbers @xmath96 are self - averaging . let us choose for example two parameters @xmath553 and @xmath554 . then , only the clusters of size @xmath0 give a non trivial contribution to @xmath555 at large time so that @xmath556 at large time , the difference equation ( [ diffshro ] ) then reduces to a differential equation for @xmath557 , @xmath558 it implies that @xmath559 is linear in @xmath98 at large time which means that @xmath116 is self - averaging and stationary at large time . more precisely , integrating the above equation gives : @xmath560 in agreement with eq.([lesc ] ) . similarly , to prove that @xmath117 possesses a finite self - averaging limit as @xmath561 we choose three parameters @xmath553 , @xmath562 so that only clusters of size two survive in @xmath555 at large time , @xmath563 eq.([diffshro ] ) then gives a first order differential equation for @xmath564 which implies that @xmath565 in agreement with eq.([lesc ] ) . although we do not have a global argument this proof may clearly be extended to recursively prove self - averageness of any @xmath96 by choosing the parameters @xmath566 with @xmath567 . here we present the proof of eq.([gscal ] ) . let us first recall lagrange formula . consider a variable @xmath568 defined by the implicit relation @xmath569 for some given analytic function @xmath301 . the solution of this equation is supposed to be unique so that @xmath568 is function of @xmath570 . given another analytic function @xmath571 we look for the taylor series in @xmath570 of @xmath572 . this composed function may be presented as a contour integral : @xmath573 expanding the integrated rational function in taylor series in @xmath570 gives lagrange formula : @xmath574 let @xmath575 . eq.([eqgc ] ) translates into @xmath576 or @xmath577 for @xmath578 and @xmath579 . we now apply lagrange formula with @xmath580 and @xmath581 . this gives @xmath582 with @xmath583 and @xmath578 , this proves eq.([gscal ] ) .

the transition states and dividing surfaces used to find rate constants for bimolecular reactions are shown to undergo morse bifurcations , in which they change diffeomorphism class , and to exist for a large range of energies , not just immediately above the critical energy for first connection between reactants and products . specifically , we consider capture between two molecules and the associated transition states for the case of non - zero angular momentum and general attitudes . the capture between an atom and a diatom , and then a general molecule are presented , providing concrete examples of morse bifurcations of transition states and dividing surfaces . morse bifurcations of transition states in bimolecular reactions + r s mackay and d c strub + mathematics institute and centre for complexity science , university of warwick , coventry cv4 7al , u.k . + bimolecular reactions , collision theories , transition state theory , dividing surfaces , morse bifurcations + published in * nonlinearity * , http://dx.doi.org/10.1088/0951-7715/28/12/4303[28(12):4303-4329 ] +

1501.00266

one way of finding rates of reaction is to consider rates of transport in a low dimensional hamiltonian system representing the specific reaction . some of the first examples studied using transition state theory consisted of bimolecular reaction in gaseous phase , i.e. @xmath4 for two ( polyatomic ) molecules @xmath5 and @xmath6 . provided the born - oppenheimer approximation holds , we can pass from the quantum mechanical system to a classical one , namely the hamiltonian system for the motion of the nuclei interacting via a potential given by the ( ground state ) energy of the electrons as a function of the internuclear coordinates . then , if this extremely high dimensional hamiltonian system is , at any instant , the product of `` reacting '' two molecule sub - systems that are independent of each other , we may consider the evolution of an ensemble of individual reactions in this low dimensional hamiltonian sub - system . for this , we require the gas to be sufficiently dilute . see keck @xcite for a nice review of transition state theory and discussion of the assumptions involved . finally , since the low dimensional system is hamiltonian and energy is conserved , we can restrict our attention to the energy levels , and consider the flow of ergode , as a function of the energy . finding ( microcanonical ) reaction rates translates to finding the rate of transport of ergode between regions representing reactants and products . transition state theory provides upper bounds on rates of transport via the flux of ergode through a _ dividing surface _ that separates the regions of interest . the dividing surface approach , which we reviewed in @xcite , can be traced back to the works of marcelin ( * ? ? ? * chapter 2 ) and wigner @xcite . in order to obtain a useful upper bound , the dividing surface is chosen to have locally minimum flux in a given direction . for this to be the case , the dividing surface must be the union of surfaces of unidirectional stationary flux with no local recrossings , which in turn is the case for surfaces spanning a closed , invariant , orientable , codimension-2 submanifold of the energy level , known as a _ transition state_. the simplest transition states and dividing surfaces are found in the basic transport scenario of `` flux over a saddle '' @xcite . these are transition states diffeomorphic to @xmath7 for energies just above the index-1 critical point of the hamiltonian for @xmath8 degree of freedom systems , with dividing surfaces diffeomorphic to @xmath9 . for two degree of freedom systems , these transition states are hyperbolic periodic orbits . as the energy is increased , they may lose normal hyperbolicity and bifurcate . the bifurcations of periodic orbits are well known , see e.g. ( * ? ? ? * section 8.6 ) , and those of periodic orbit transition states have been studied for a long time , see e.g. @xcite . instead , for higher degrees of freedom than two the transition states for the basic scenario are normally hyperbolic @xmath10 spheres ( within the energy levels ) and very little is known about their bifurcations . what had been overlooked , and was explained in @xcite , is that the transition states may lose normal hyperbolicity by becoming singular , i.e. the manifold structure fails , at a higher critical energy value but then regain their normal hyperbolicity for values above the critical energy . these are _ morse bifurcations _ * appendix b ) , leading to a change in diffeomorphism class of the transition states and dividing surfaces spanning them . they occur when the union of transition states over a range of energies forms a smooth normally hyperbolic submanifold of state space , which we call a _ transition manifold _ , and there is a critical value of the hamiltonian function restricted to the transition manifold . by definition , the transition states are the level sets of this restricted hamiltonian and undergo a bifurcation . similarly , the dividing surfaces , which span the transition manifold , undergo a morse bifurcation . actually , they occur because the energy levels undergo a morse bifurcation themselves so the dividing surfaces , and therefore the transition states , must also undergo a change of diffeomorphism class in order to still separate these in two . morse theory tells us that we can continue the transition states and dividing surfaces , with respect to the energy , through critical values , as well as giving the diffeomorphism class of the submanifolds . this is useful , for example , when the transition states above some critical energy are disjoint , which we shall see in the atom - frozen diatom system , as we then know exactly what objects to look for . recently , there has been a lot of interest in general transport problems ( as opposed to the basic one ) , non - minimum energy paths @xcite , and roaming reactions @xcite . these transport scenarios may emerge from the basic one via a morse bifurcation , in which case by considering these bifurcations , we obtain transition states and dividing surfaces that can be used to find the rates of transport . we shall consider transport between the region representing two distant molecules ( reactants ) and the region in which the molecules are close . the latter does not however generally constitute the products . this is the _ capture _ transport problem associated with the necessary first step of the molecules getting close enough to react . the _ capture rate _ ( sometimes also called _ collision rate _ ) provides an upper bound on the reaction rate , as we do not expect all captured trajectories to proceed to the products region @xcite . note that there might actually be multiple product regions , but for a two - body capture process there is only one final region of interest . capture rates are crucial for many physical processes , and have a long history dating back at least to 1905 with langevin s early contribution @xcite , see review by chesnavich and bowers @xcite . two assumptions are usually found in the literature . firstly , the reacting hamiltonian systems are assumed to have euclidean symmetry , that is to be invariant under translations and rotations . this is the case for gas phase reactions with no background ( electro - magnetic ) field . the hamiltonian system can then be reduced to a family of systems , in centre of mass frame , parametrised by the angular momentum . the hamiltonian function then contains both coriolis and centrifugal terms . secondly , the energy is taken to be below those at which the two molecules dissociate and centrifugal and coriolis forces to be sufficiently weak such that the molecules are well defined and in the small vibrations regime . these assumptions allow to distinguish between intermolecular degrees of freedom ( distance and relative attitudes of the molecules ) and intramolecular ones . we too shall consider systems that satisfy these assumptions . we want to find the rate of capture , which we shall assume can be thought of as transport between regions on either side of a non - degenerate maximum @xmath11 of the effective potential ( centrifugal terms plus potential ) with respect to the intermolecular distance @xmath12 . in the literature , this maximum is generally a centrifugal maximum obtained by balancing the repulsive centrifugal terms with the attractive long distance potential energy . alternatively , @xmath11 could be a non - degenerate chemical maximum of the bimolecular potential and therefore of the effective potential for small angular momentum . provided @xmath11 is sufficiently large , such that capture occurs in a region where the potential is only weakly dependent upon the attitudes of the molecules , and sufficiently non - degenerate , we shall see that fixing the intermolecular distance degree of freedom to the maximum value gives a normally hyperbolic transition manifold in state space , which can be spanned by a dividing manifold . the restriction of these manifolds to the energy levels gives dividing surfaces and transition states , which we shall refer to as _ capture _ transition states . the literature often refers to them as _ orbiting _ or _ loose _ transition states @xcite . the central field model , in which the attitudes of the colliding pair are ignored , is attributed to langevin @xcite . in this very early work , one already finds capture periodic orbit transition states . langevin considers the capture process using scattering theory . for an introduction to scattering theory see e.g. ( * ? ? ? * section 3.10 ) , whereas for a comparison with the dividing surface approach see @xcite . non - central fields were considered later , also usually from a scattering theory perspective , starting with the works of pechukas @xcite and chesnavich , su and bowers @xcite . as the energy is increased above the critical energy for first connection between reactants and `` products '' , the intermolecular attitude and angular momentum degrees of freedom will generally be involved in interesting sequences of morse bifurcations causing changes in the diffeomorphism class of the energy levels , as the `` cone of acceptance '' ( i.e. orientations for which capture is possible ) opens up , and therefore changes of the capture transition states and dividing surfaces . instead , the intramolecular degrees of freedom are assumed to consist of small vibrations , thus not playing a role in the morse bifurcations . using morse theory , one can know which the correct transition states and dividing surfaces are , as we shall outline for two examples , and then study rates of transport for larger energies than previously thought possible . here , we shall focus on the morse bifurcations , the reduction and the details of bimolecular capture problems . some analysis of `` reaction dynamics '' of rotating molecules has been done recently in @xcite , but we are interested in the interaction of two rotating molecules . whether captured pairs then go on to react can be thought of as a further transport problem with its own ( _ reaction _ ) transition states and dividing surfaces , possibly ( but not necessarily ) associated to other maxima @xmath13 of the effective potential . these are often referred to as _ tight _ transition states . the capture and reaction transition states are therefore in series . the simplest case will be when these are distant and the level sets of separate transition manifolds . however , even when `` separate '' their stable and unstable manifolds , which act as transport barriers , will intersect , determining the `` reaction channels '' . some trajectories joining reactants and products might roam in the region between the ( capture and reaction ) dividing surfaces , that is follow trajectories with a non - monotonic intermolecular distance in time , before finally crossing the reaction dividing surface . this is to be expected due to coupling between degrees of freedom , and was recently given as an explanation of what has been called _ roaming reactions _ @xcite . reaction rates have an equally long history as capture rates , and bimolecular reactions have played the role of test problems since the early days of transition state theory ( as noted in @xcite ) . the transport problems associated with reaction tend to be harder because the chemical potentials are at best not simple and often degenerate . similarly to how the first capture models were simplified by making the fields central , reaction rates were first , and largely still , considered for collinear and planar systems with zero angular momentum . note that reaction transition states will also generally undergo morse bifurcations . hamiltonian systems with symmetry , such as the molecular @xmath14 body systems with euclidean symmetry considered here , have conserved quantities by noether s theorem ( see e.g. ( * ? ? ? * appendix 5 ) , ( * ? ? ? * section 2.7 ) ) , and can be reduced via symplectic reduction . the euclidean symmetry group of our @xmath14 body systems is a semi - direct product @xmath15 and can be reduced in stages ( * ? ? ? * chapter 4 ) . the translational symmetry and associated linear momentum are easily reduced , whereas the rotational symmetry , which does not act freely on the whole of state space , and the associated angular momentum require singular reduction , see e.g. @xcite , ( * ? ? ? * section 8) . the reduced state space is therefore a stratified symplectic manifold , however the hamiltonian flow leaves the connected components of the strata invariant so the reduced dynamics can be studied on each stratum individually . the stratified symplectic manifold for our reduced molecular @xmath14 body systems is composed of three strata . both singular strata have zero angular momentum , whereas on the principal stratum the angular momentum is arbitrary . this is discussed in @xcite . actually , reactions with initial positions and momenta confined to a plane remain in this plane for all successive times , and reduced planar systems have no angular momentum degree of freedom and no coordinate singularities at collinear configurations . in section [ planaratomdiatom ] , we therefore consider planar capture between an atom and a diatom and find our first examples of morse bifurcations . this example and that of two interacting diatoms were considered in less detail in @xcite . a spatial example , of an atom interacting with a molecule , is considered in section [ spatialbimol ] . we choose to use the coordinates that one obtains via the gauge theoretic approach to cotangent bundle reduction as outlined by littlejohn and reinsch @xcite , and reviewed in appendix [ bigauge ] , where we discuss the symplectic form and consider the transformation to serret - andoyer coordinates on the angular momentum sphere , which is not done in @xcite . this approach gives coordinates , which are furthermore physically meaningful since coordinates and momenta are not mixed , however it involves considering the rotational @xmath16 action on configuration space as opposed to the lifted action in state space , and therefore introduces coordinate singularities . these are avoided by considering non - collinear molecules . ways to deal with them are discussed in the conclusions in section [ conclude ] . the simplest non - trivial capture example is planar reactions between an atom and a diatom . the planar reduced three - body hamiltonian system that represents an atom interacting with a diatom with no background field is the family of mechanical systems @xmath17 , parametrised by the angular momentum @xmath18 , with @xmath19 where @xmath20 , @xmath21 , @xmath12 is the intermolecular distance , @xmath22 the attitude and @xmath23 the length of the diatom , see figure [ jacobifig ] . @xmath24 is the effective potential with the centrifugal term . the parameters are the reduced masses @xmath8 and @xmath25 , and the magnitude of the angular momentum @xmath26 . we shall restrict our attention to energies below that at which the diatom dissociates , so we have a two - body capture problem . specifically , we shall consider the capture of an atom by a strongly bonded diatom that is distant and rotating slowly , assuming that the potential of the pair has a non - degenerate maximum @xmath11 with respect to @xmath12 for large @xmath12 , and that the centrifugal and coriolis forces are not too large . thus , for energies below that at which the diatom dissociates , the reduced coordinates and their momenta split into intermolecular @xmath27 and intramolecular @xmath23 degrees of freedom . our assumptions imply that the diatom is in the small vibrations regime , i.e. @xmath28 is an elliptic degree of freedom , and rotating slowly , so in the neighbourhood of @xmath11 we will find a codimension-2 normally hyperbolic submanifolds @xmath29 and span it by a dividing manifold @xmath30 . this will be the capture dividing manifold that trajectories must cross in order for capture to occur . the assumptions will be introduced by scaling the coordinates , after which the dynamics and the existence of the transition manifold @xmath29 will be clear . then , we will show that the level sets of @xmath29 and @xmath30 , the transition states @xmath31 and dividing surfaces @xmath32 , generically undergo morse bifurcations involving the @xmath33 degree of freedom , as the energy is varied . for simplicity , the masses of the atom and the diatom are set to @xmath34 . a better approach would be to non - dimensionalise the variables . note that normalised or mass - weighted jacobi coordinates do not remove the mass dependence . the intermolecular terms of the potential will be repulsive at short ranges , possibly have a number of maxima ( and therefore minima ) with respect to the intermolecular distance @xmath12 in the mid ranges , and be attractive at long ranges with inverse @xmath35-power of @xmath12 terms @xcite , see figure [ atomdiatomxpot ] . the potential is then summed to the repulsive centrifugal term to give the effective potential . in the long ( physical ) range , provided the attractive potential falls off faster than the centrifugal potential as @xmath36 , i.e. @xmath37 , the effective potential has a centrifugal maximum @xmath38 . in the short ( chemical ) range of the potential , the chemical maxima of @xmath39 with respect to @xmath12 imply chemical maxima of @xmath24 for @xmath26 small with respect to the slope of @xmath39 at the maxima . in either case , as @xmath26 increases the maxima will `` collide '' with the minima and disappear , see figure [ atomdiatomxpot ] . we shall think of capture as transport between the regions on either side of the largest maximum @xmath40 , assumed large with respect to the length of the diatom @xmath23 . we introduce the capture scale by setting @xmath41 and @xmath42 with @xmath43 , and taking @xmath11 to be of order 1 . essentially , @xmath44 is the ratio between the size of the diatom and its distance to the atom , however in practice we assume that for distant pairs , the potential energy is weakly dependent on the attitude of the diatom , since the pair are distant , and we can choose @xmath44 such that @xmath45 this is the case for potentials that are inverse power functions of the intermolecular distance , which can be expanded using legendre polynomials . the hamiltonian function expands into @xmath46 where we have dropped the tildes . we note that @xmath22 does not appear until order @xmath47 , so @xmath48 with constant @xmath49 . furthermore , the system separates into slow and fast degrees of freedom , i.e. at order @xmath50 we find the fast oscillations of the diatom plus a `` centrifugal '' term , at order @xmath51 there is the intermolecular ( capture ) dynamics , and then there are the higher order terms . up to order @xmath51 , the @xmath12 and @xmath23 degrees of freedom are uncoupled , and @xmath52 . the diatom will have an equilibrium configuration , which corresponds to a non - degenerate minimum @xmath53 of the joint atom - diatom potential with respect to the intramolecular distance @xmath23 . we assume this minimum to be highly non - degenerate , i.e. that the diatom is strongly bonded , and that the centrifugal and coriolis forces on the diatom are not too strong , so the diatom will vibrate about its equilibrium without significant distortion . we therefore , linearise @xmath23 about @xmath54 by setting @xmath55 and @xmath56 with @xmath57 . the constant @xmath58 is chosen such that @xmath59 where @xmath60 , @xmath61 is order one , and @xmath62 . recall that we are assuming @xmath63 to be large . this scaling ensures that the leading order terms of the potential with respect to the coordinates are of the same order as their conjugate momenta in the kinetic energy . the hamiltonian function , again dropping the tildes , becomes @xmath64 we shall further simplify our hamiltonian by setting @xmath65 , i.e. considering @xmath66 . general @xmath67 is considered in the disconnecting example of @xcite . the scaled system consists of @xmath68 and @xmath69 from which we get the equations of motion up to order @xmath70 , namely @xmath71 by assumption , the intermolecular distance degree of freedom is hyperbolic , and the intramolecular distance is in the small vibrations regime , i.e. elliptic . these dynamics are uncoupled to this order . as the diatom rotates , the attitude degree of freedom will display both kinds of motion . provided the @xmath72 degree of freedom is more strongly hyperbolic that the @xmath33 one , that is the maximum @xmath11 is sufficiently non - degenerate , the submanifold @xmath73 is almost invariant , or more precisely invariant to order @xmath70 , and normally hyperbolic . by normal hyperbolicity theory , there is a true normally hyperbolic submanifold @xmath29 nearby . given @xmath74 , we could find a better approximation to @xmath29 as explained in @xcite . however , for the purpose of studying the morse bifurcations of the transition states , @xmath74 is a sufficiently good approximation . the normally hyperbolic submanifold @xmath29 is a transition manifold , as it can be spanned by a dividing manifold . the approximate transition manifold @xmath74 is spanned by @xmath75 the transition states @xmath31 and dividing surfaces @xmath32 are then approximately the level sets of the hamiltonian restricted to the approximate transition and dividing manifolds . as the energy varies , we expect these to bifurcate . the transition states may lose normal hyperbolicity . for atom - diatom reactions , @xmath31 are 3-dimensional manifolds , so it is not well understood how they lose normal hyperbolicity . freezing the diatom , the system only has two degrees of freedom and @xmath31 is a periodic orbit . in @xcite , we saw that these disappear in a centre - saddle bifurcation . however , before the loss of normal hyperbolicity , the capture transition states will undergo changes of diffeomorphism class via morse bifurcations . if we write @xmath76 we find that the hamiltonian function restricted to the transition manifold @xmath29 is independent of @xmath77 and @xmath78 up to order @xmath47 , namely @xmath79 where @xmath80 is the non - canonical momentum , and we dropped constant terms . for @xmath81 below that at which the diatom dissociates , the intramolecular degree of freedom contributes only positive - definite terms to the restricted hamiltonian function and is not involved in any morse bifurcations , which can be studied by considering the simpler function obtained by freezing the diatom , i.e. minimising @xmath82 over @xmath28 by setting @xmath83 and @xmath84 giving @xmath85 different reactions , i.e. choices of atom and diatom , will have different potentials and different sequences of morse bifurcations . an example frozen restricted hamiltonian @xmath86 is depicted in figure [ atomdiatom ] . considering this @xmath87 , we find that @xmath88)$ ] bifurcates from @xmath89 to @xmath90 , and so the level sets @xmath91 from @xmath92 to @xmath93 , where @xmath94 denotes a ball and @xmath95 a sphere . passing to the full system , we find that the transition manifold @xmath96 bifurcates from @xmath97 to @xmath98 , the transition state @xmath31 from @xmath99 to @xmath100 , and the dividing surface @xmath32 from @xmath101 to @xmath102 . thus , the energy levels are separated by the dividing surfaces for all energies , even above the critical one of the morse bifurcation . other examples of @xmath103 are considered in @xcite . ( 11,4.5 ) [ b][t]@xmath104 [ t][b]@xmath81 ( 2,2.1 ) as a function of energy @xmath81 , for the capture transition state of the atom - diatom system , with potential as in figure [ atomdiatom ] , undergoing a disconnecting morse bifurcation at @xmath105 . left : two degree of freedom system with frozen diatom . right : three degree of freedom system.,title="fig:",scaledwidth=45.0% ] ( 8.7,2.1 ) as a function of energy @xmath81 , for the capture transition state of the atom - diatom system , with potential as in figure [ atomdiatom ] , undergoing a disconnecting morse bifurcation at @xmath105 . left : two degree of freedom system with frozen diatom . right : three degree of freedom system.,title="fig:",scaledwidth=45.0% ] given a dividing surface @xmath32 , we can compute the flux of ergode through it , denoted @xmath106 , and use this to find the rate constant . the flux varies @xmath107 smoothly through the morse bifurcations , for @xmath8 degrees of freedom , but not @xmath108 @xcite . graphs of @xmath104 as a function of @xmath81 for the capture dividing surface of the atom - diatom system with the example potential presented in figure [ atomdiatom ] can be seen in figure [ fluxfig ] . for the periodic orbit transition state of the two degree of freedom system with the frozen diatom , we see the log - like infinite slope singularity at the index-1 morse bifurcation . this is not present in the graph of the flux through the dividing surface of the three - degree of freedom system . we now consider the capture of an atom by a polyatomic molecule . the molecule shall be assumed to have a non - degenerate non - collinear equilibrium , of its @xmath109 atoms , about which it is vibrating fast , and the system to have energy below that at which the molecule dissociates , so we can use the charts provided by the bundle approach to cotangent bundle reduction , see appendix [ bigauge ] . the capture scenario will be the same as the atom - diatom one , modulo differences due to its spatial nature , such as an additional angular momentum degree of freedom . we shall again assume that the effective potential has a non - degenerate maximum @xmath110 with respect to the intermolecular distance @xmath12 when the pair is distant , that the attitude terms are not dominant in this range , and that @xmath6 is a normal molecule with a very non - degenerate equilibrium , i.e. that it is strongly bounded . in order to have @xmath6 in the ( elliptic ) small vibrations regime , and the attitude and angular momentum degrees of freedom dominated by the hyperbolic ( in the neighbourhood of @xmath11 ) intermolecular one , we ask that the molecule is rotating slowly with respect to the atom , that it is 3d with distinct principal moments of inertia and that the energy is bounded such that , roughly speaking , the angular momentum @xmath111 is almost perpendicular to the line of centres . these extra constraints due to the spatial nature shall be discussed in detail after we scale the intermolecular and intramolecular distance . finally after also scaling the angular momentum coordinates and the attitude momenta , we shall find a normally hyperbolic transition manifold @xmath29 and dividing manifold @xmath30 about @xmath11 and discuss the possible morse bifurcations of the transition states @xmath31 and dividing surfaces @xmath32 . these will now also involve the angular momentum degree of freedom , degeneracies when the moment of inertia tensor has equal moments , and require an understanding of the diffeomorphism class of the reduced space , which we shall discuss . capture between two non - collinear molecules can be studied following the same steps as those below , and will be presented in a future publication . we denote the reduced molecular @xmath112 body hamiltonian system by @xmath113 . the state space @xmath114 is the subset of the principal stratum with non - collinear configurations . it is a smooth @xmath115-manifold diffeomorphic to a ( generally non - trivial ) @xmath116 fibre - bundle with base space @xmath117 , where @xmath118 is the non - collinear subset of the translation reduced configuration space @xmath119 , and fibre the angular momentum sphere @xmath120 ( * ? ? ? * section 2.3 ) . in canonical coordinates @xmath121 , we have @xmath122 where @xmath123 are the internal coordinates , which we shall choose shortly , @xmath26 the magnitude of the angular momentum and @xmath124 canonical serret - andoyer coordinates on the angular momentum sphere , such that e.g. the angular momentum is given by @xmath125 more than one chart is necessary , due to the coordinate singularities both for @xmath126 , with @xmath127 bodies , as well as on the angular momentum sphere . @xmath24 is the effective potential with the centrifugal terms , @xmath128 is the reduced metric , @xmath129 is the moment of inertia tensor , and @xmath5 is the gauge potential , present in the coriolis terms . these are introduced in apprendix [ bigauge ] . we are uninterested in scaling effects due to the mass , so we set @xmath130 . this affects @xmath128 and @xmath129 . the rotating frame for the reduction is chosen such that the jacobi vector along the line of centres @xmath131 is parallel to the @xmath132 axis , and the remaining @xmath133 symmetry about the @xmath132-axis is used to orient the equilibrium configuration of the molecule @xmath6 such that its moment of inertia tensor @xmath134 with @xmath135 . this determines the reduction of the @xmath133 symmetry about @xmath132 and the definition of the attitude of the molecule . the most natural choice of reduced coordinates @xmath123 is the distance between atom and molecule @xmath12 and two angles @xmath136 for the attitude of the molecule , which are intermolecular coordinates , as well as @xmath137 coordinates @xmath23 for the intramolecular degrees of freedom of @xmath6 , so @xmath138 , unless @xmath6 has further symmetries of its own that can be reduced . the intramolecular coordinates @xmath23 shall be chosen in order to simplify the hamiltonian along the lines of the eckart @xcite and sayvetz @xcite conventions for normal and anomalous molecules in the small vibration regime . essentially , we shall consider an eckart convention for a normal molecule in the small vibrations regime interacting with an atom , for which the intermolecular coordinates are similar to the large amplitude coordinates of anomalous molecules considered by sayvetz ( see apprendix [ spatialdetials ] ) . in the eckart convention the rotations and vibrations are decoupled to leading order since the intramolecular coordinates @xmath23 are chosen to be riemann normal coordinates for which the gauge potential @xmath139 vanishes at the equilibrium configuration . this is discussed from a geometric perspective by littlejohn and mitchell @xcite . first , we scale the distances and their momenta . we are interested in large @xmath12 about the ( capture ) maximum @xmath11 , so we scale @xmath140 and @xmath141 . then , by passing to the intermolecular time , we scale the hamiltonian such that @xmath12 is of order one . also , we consider the molecule in the small vibrations regime about its non - degenerate minimum , and shift the intramolecular coordinates to have @xmath142 at equilibrium , and then scale @xmath143 and @xmath144 . as for the planar case , we assume that the potential scales to @xmath145 and then choose @xmath58 such that @xmath146 that is , we are assuming that the molecule is strongly bonded , @xmath147 . note that we have chosen normal mode intramolecular coordinates for which @xmath148 . the reduced kinetic and centrifugal energies contain both intermolecular and intramolecular terms and are scaled in appendix [ spatialdetials ] . the hamiltonian function becomes @xmath149 where @xmath150 and @xmath151 for some @xmath152 , see appendix [ spatialdetials ] . we are using angular momenta @xmath111 and the non - canonical momenta @xmath153 where @xmath154 , as place - holders . ( 12,4.5 ) ( 2.5,2.2 ) ( 9.5,2.2 ) ( 3.3,1.12)@xmath155 ( 2.6,4.6)@xmath156 ( 0,2)@xmath157 ( 7.1,2)@xmath157 next , we consider the angular momentum degree of freedom @xmath158 . from the scaled hamiltonian , we note that its dynamics are coupled to the internal one and that it appears in terms of orders both @xmath50 and @xmath51 . in order to ensure that we have a normally hyperbolic capture transition manifold about @xmath11 , we need that the intermolecular distance degree of freedom @xmath159 is more hyperbolic than the angular momentum degree of freedom . we will therefore find the necessary scaling for this to be the case and apply it . considering the atom and the molecule as constituting a single `` body '' , if the molecule is in equilibrium with itself and with respect to the atom , we obtain a rigid body and the hamiltonian reduces to the centrifugal terms . rigid bodies follow closed curves on the angular momentum sphere with equilibrium points when @xmath111 is parallel to the eigenvectors of @xmath160 , called _ principal axes _ @xcite . typical rigid body dynamics for the case of distinct eigenvalues , or _ principal moments _ , is depicted in figure [ amcentrif1 ] . in general , the angular momentum degree of freedom does nt follow closed curves on @xmath120 , since it is coupled with the internal `` deformation '' dynamics . the equilibria of the system occur when @xmath123 is a critical point of the effective potential @xmath24 , @xmath161 , and again @xmath111 is parallel to the principal axes . we therefore consider the centrifugal energy @xmath162 which is not conserved , in more detail . first , we consider the moment of inertia tensor and its principal moments and axes . the eigenvectors @xmath163 of @xmath160 are the same as those of @xmath129 , whereas the eigenvalues @xmath164 are the reciprocals . @xmath129 has real eigenvalues , since it is a real positive - definite matrix , and if these are distinct then the eigenvectors are orthogonal . to order @xmath165 , the eigenvalues and eigenvectors of @xmath160 are @xmath166 where @xmath167 and @xmath168 . thus in order to have distinct principal moments and axes , we require either @xmath169 , or @xmath170 . for our 3d molecule @xmath6 with three distinct moments , as it rotates relative to the distant atom , we expect to find three pairs of points ( on the attitude sphere @xmath171 ) at which two of the moments of @xmath172 , and so @xmath173 , are non - distinct . then , we consider the actual centrifugal energy and its critical points . given a fixed configuration @xmath123 with distinct @xmath174 , we write @xmath175 i.e. consider serret - andoyer coordinates @xmath176 obtained by projecting onto the principal axes , cf . appendix [ bigauge ] . then @xmath177 so the critical points are @xmath178 , @xmath179 , @xmath180 , @xmath181 , and @xmath182 . the superscript denotes the principal axis to which @xmath111 is parallel . for non - distinct eigenvalues @xmath183 , choosing generalised eigenvectors for @xmath184 , gives @xmath185 so the critical points are the degenerate @xmath186 , and @xmath187 . this is depicted in figure [ amcentrif1 ] . for arbitrary @xmath188 , in order to ensure that the centrifugal terms are higher order than the intermolecular @xmath159 ones , we restrict our attention to energies just above the critical @xmath189 such that the angular momentum degree of freedom is confined to a small annulus @xmath190 that does nt contain @xmath191 . if all the energy of the system is in @xmath192 , then for the non - distinct case we have @xmath193 at @xmath194 , whereas for the distinct case @xmath195 furthermore , @xmath196 , so bounding @xmath197 , with @xmath198 small , gives @xmath199 . thus , we scale @xmath200 , giving @xmath201 next , we consider the rotational degree of freedom , with non - canonical rotational momenta @xmath202 and ensure that the @xmath159 dynamics dominates this one also . even though we have removed the coupling of vibrations and rotations up to order @xmath70 , @xmath203 are not conserved since the reduced metric @xmath128 is a function of @xmath22 . however , the rate of change of @xmath67 is a function of @xmath204 up to order @xmath47 , so we consider a molecule that is initially rotating slowly , which will take a long time to increases its rotational velocity . thus , we scale @xmath205 , so @xmath206 . finally @xmath207 and @xmath208 giving the equations of motion , up to order @xmath70 , as @xmath209 where @xmath210 . thus , about the maximum @xmath11 , the @xmath159 degree of freedom is more hyperbolic than both the attitude @xmath33 and the angular momentum @xmath211 ones . the submanifold @xmath212 is therefore invariant to order @xmath70 , and normally hyperbolic . taking @xmath74 as an approximation to the true normally hyperbolic submanifold @xmath29 nearby , and @xmath213 as an approximate dividing manifold , we can find the restricted hamiltonian functions @xmath214 modulo constant terms , and @xmath215 to leading orders . these give the transition states @xmath216 and dividing surfaces @xmath217 , respectively . as for the planar examples , it is simpler to study the morse bifurcations if we minimise the reduced hamiltonians over the positive - definite coordinates , namely @xmath218 , @xmath219 and @xmath220 . this can be simplified by using canonical angular momentum coordinates @xmath176 aligned with the principal axes , and setting @xmath221 in @xmath222 , to obtain @xmath223 we are then interested in the level - sets of @xmath224 and their bifurcations . the domain of @xmath224 is a subset of @xmath29 , which is codimension-2 in the reduced state space @xmath114 . the latter is some @xmath120 fibre bundle over @xmath117 , so also @xmath29 and @xmath30 could in general be non - trivial bundles . we shall restrict our attention to subsets for which the bundle is trivial . furthermore , we are considering energies below that at which the molecule dissociates , and up to just above the centrifugal energy for the angular momentum aligned with the @xmath225 principal axis . the critical points @xmath226 of the frozen , restricted effective potential @xmath224 are given by @xmath227 the first equation is satisfied trivially for @xmath228 at which the two principal moments are equal @xmath229 . we shall consider examples of @xmath224 that are morse functions , i.e. have non - degenerate critical points @xmath226 with @xmath230 , satisfying either @xmath231 or @xmath232 for @xmath233 , cf . * section iv.e ) . furthermore , the morse function @xmath224 has at least two non - degenerate minima at @xmath234 due to the symmetry of the centrifugal terms . the sequence of morse bifurcations of the level sets of @xmath224 , and therefore of @xmath31 and @xmath32 , depends on the relative size of the centrifugal and the reduced potential @xmath103 energies . this will determine the relation of the different critical energies . critical points with the same attitude @xmath22 but the angular momentum aligned with different principal axes have energies that differ by @xmath235 , whereas the difference in energy for different attitudes depends on the atom - molecule pair . the simplest case is when the first morse bifurcation encountered as the energy is increased from the minima involves the angular momentum angle , and the system goes from rotating about the @xmath236 axis to rotating more freely about @xmath237 as well . this bifurcation occurs at the critical energy for the @xmath238 critical points . in this case both the domain of @xmath224 and the subsets of @xmath29 and @xmath30 of interest are bundles over a contractible base space and so trivial ( * ? ? ? * theorem 11.6 ) . the frozen energy levels @xmath239 bifurcate from @xmath240 to @xmath241 , so the transition states @xmath31 go from @xmath242 to @xmath243 , and similarly the dividing surfaces . ( 12,11.5 ) ( 2.5,8.5 ) torus , where darker regions represent lower energies . top row : for no values @xmath244 with non - distinct principal moments . bottom row : for non - distinct principal moments @xmath245 and @xmath246 at @xmath244 depicted by vertical back lines . case in which the value of the function at @xmath247 is smaller than that at @xmath248 on the left , and vice - versa on the right.,title="fig:",scaledwidth=33.0% ] ( 9.5,8.5 ) torus , where darker regions represent lower energies . top row : for no values @xmath244 with non - distinct principal moments . bottom row : for non - distinct principal moments @xmath245 and @xmath246 at @xmath244 depicted by vertical back lines . case in which the value of the function at @xmath247 is smaller than that at @xmath248 on the left , and vice - versa on the right.,title="fig:",scaledwidth=33.0% ] ( 12.2,5.65)@xmath249 ( 6.6,11.2)@xmath250 ( -0.4,11.2)@xmath250 ( -0.4,6.9)@xmath251 ( -0.4,8.2)@xmath252 ( -0.7,9.5)@xmath253 ( -0.7,10.7)@xmath254 ( 5.2,5.65)@xmath249 ( 3.4,5.65)@xmath255 ( 0.9,5.65)@xmath256 ( 2.5,2.5 ) torus , where darker regions represent lower energies . top row : for no values @xmath244 with non - distinct principal moments . bottom row : for non - distinct principal moments @xmath245 and @xmath246 at @xmath244 depicted by vertical back lines . case in which the value of the function at @xmath247 is smaller than that at @xmath248 on the left , and vice - versa on the right.,title="fig:",scaledwidth=33.0% ] ( 9.5,2.5 ) torus , where darker regions represent lower energies . top row : for no values @xmath244 with non - distinct principal moments . bottom row : for non - distinct principal moments @xmath245 and @xmath246 at @xmath244 depicted by vertical back lines . case in which the value of the function at @xmath247 is smaller than that at @xmath248 on the left , and vice - versa on the right.,title="fig:",scaledwidth=33.0% ] ( 12.2,-0.3)@xmath249 ( 6.6,5.2)@xmath250 ( -0.4,5.2)@xmath250 ( 5.2,-0.3)@xmath249 ( 0.6,-0.3)@xmath244 ( 3.5,-0.3)@xmath257 as the energy is increased further , we will reach critical values at which also the attitude coordinates are involved in morse bifurcations . we will consider the case in which the energy does not change significantly as the molecule rotates in one direction , with respect to the atom , but does when it tries to rotate in the other direction and a potential @xmath103 on @xmath116 that has a minimum @xmath258 , a saddle @xmath259 and two maxima @xmath260 , restricting our attention to the annulus @xmath261 containing @xmath258 and @xmath259 . therefore , choosing the angles appropriately , only one is involved in morse bifurcations , whereas the other contributes positive definite terms . the subset of the transition manifold @xmath29 of interest is a bundle over @xmath262 which we claim is trivial . firstly , we note that it is equivalent to the product of a bundle over @xmath92 with @xmath263 via homotopy - type arguments ( * ? ? ? * theorem 11.4 ) , cf . bundles over contractible spaces being trivial . the characterisation of bundles over spheres with structure group @xmath264 depends on certain homotopy groups of @xmath264 ( * ? ? ? * theorem 18.5 ) . our fibres are diffeomorphic to @xmath116 , or subsets of it , and the diffeomorphism group of @xmath116 is the orthogonal group @xmath265 . however , @xmath29 is orientable so both elements of the product must be orientable . thus , given that the bundle over the circle is orientable , we restrict our attention to the orientation preserving diffeomorphisms @xmath16 and find that the bundle over the circle is a product , and therefore our original bundle is also trivial ( * ? ? ? * section 26 ) . note however that not all orientable surface bundles over the circle are product bundles , as we can construct non - trivial bundles with fibres diffeomorphic to @xmath266 , for example . @xmath224 can be minimised over the irrelevant attitude to obtain a function on the torus @xmath266 for @xmath267 , say . there are two possible scenarios for this case , the first is that @xmath266 does not contain points @xmath228 at which the @xmath245 , @xmath246 principal moments become equal . the order of the bifurcations then depends on the relative heights of the critical energies , and both cases are straightforward , see figure [ spatialcase1 ] . the other scenario is when @xmath266 does contain @xmath228 . we shall consider the case in which it contains only one pair of such points . contour plots for the restricted function on @xmath266 are given in figure [ spatialcase1 ] . if the centrifugal energy is smaller than the attitude potential , then the points @xmath244 at which the moments @xmath245 , @xmath246 are not distinct do not play a role in the morse bifurcations , which are the same as those for the case when @xmath266 does not contain @xmath228 , as we can see by comparing the left hand side of figures [ spatialcase1 ] . instead , when the molecular potential is smaller than the centrifugal one , depicted on the right in figure [ spatialcase1 ] , we see that the points @xmath244 do play a significant role in the bifurcations and the sub - level sets of the torus bifurcate as follows @xmath268 where @xmath269 and @xmath270 can be written as handlebodies using morse theorem b @xcite . therefore , the sub - level sets of the capture transition manifold @xmath96 have the following sequence of bifurcations @xmath271 and the transition states @xmath272 similarly for the dividing surfaces . finally , if we were to consider higher energies for this choice of @xmath103 , the other attitude would also become involved in morse bifurcations . here again the @xmath228 points would most likely lead to interesting sequences of morse bifurcations , however we would also have to deal with the non - trivial nature of the fibre bundle . after the morse bifurcations at the index-2 critical points @xmath260 , the base space would contain a 2-sphere , and many examples of non - trivial orientable bundles over these can be found . thus before we can consider the full sequence of morse bifurcations of the dividing surfaces and transition states and the transport for a larger range of energies , the bundle class of the reduced state space needs to be understood . the purpose of this article was to show that morse bifurcations must be considered when studying transport problems for larger ranges of energies , and more specifically to show the existence of morse bifurcations of capture transition states and dividing surfaces for bimolecular reactions . by considering the different sequences of morse bifurcations we were able to find interesting new transition states and dividing surfaces for general reactions with non - zero angular momentum , thus extending the dividing surface approach beyond the well known basic transport scenario . other choices of molecules than those considered here will have similar capture transport problems and therefore similar transition states and bifurcations . the flux of ergode through a dividing surface as a function of energy and the effect of the morse bifurcations was considered in @xcite and commented upon in section [ atomdiatom ] for planar atom - diatom reactions . seeing as the systems representing these examples have more than 2 degrees of freedom , apart from the unrealistic planar atom - frozen diatom case , the morse bifurcations do not have a significant effect on the flux , which varies @xmath107 smoothly through these . the actual use of capture rates as bounds on reaction rates is debatable , but largely depends on the reaction being considered . however , these were only chosen to provide relatively simple examples of morse bifurcations , and to show the importance of the attitude and angular momentum degrees of freedom in bimolecular reactions . we considered bimolecular reactions with euclidean symmetry and reduced them accordingly . even though ( symplectic ) reduction theory is a highly developed subject , we faced a number of difficulties when considering these examples . setting aside the fact that singular reduction was required , due to the nature of the rotational symmetry , and that singular cotangent bundle reduction is still not a complete theory , there is a large gap between the reduction theory literature and applications . even if one restricts one s attention to the principal non - singular stratum , it is not an easy task to find suitable charts . some of the literature avoids charts altogether , focusing instead on the global geometric properties of the reduced spaces , whereas the celestial mechanics literature considers charts for different regions of the reduced space . the most common approach in the molecular literature is to restrict one s attention to non - collinear configurations such that the gauge theoretic approach to cotangent bundle reduction provides a set of charts , as reviewed in appendix [ bigauge ] . however , here we face the opposite issue , namely the reduced space is an @xmath116 fibre bundle , due to the angular momentum degree of freedom , but the global nature of this bundle is generally not discussed in the literature . we feel that more work is needed , both on charts for the reduced spaces and on their global nature , and that this would improve our understanding of molecular reactions , and also other @xmath14-body systems . by considering normal molecules , with a fixed equilibrium configuration and energies below that at which either of the two molecules dissociates , collinearity becomes a decreasing concern with increasing size of the molecules , namely codimension-@xmath273 where @xmath274 is the number of atoms in the @xmath275th molecule ( @xmath276 ) , and the chemistry of the molecule is not taken into account . however , for smaller molecules , higher energies , or other transport problems we may need to consider collinear configurations . for non - zero angular momentum , collinear configurations are a subset of the principal reduced stratum . however due to collinear configurations having non - trivial configuration space isotropy , we can not find charts via the gauge theoretic approach to cotangent bundle reduction . the issue is therefore not one of reduction per se , but only of finding suitable coordinates . the transport problem and bifurcations of transition states will be the same as those considered in section [ spatialbimol ] . for more than seventy years , chemists have been using charts obtained by modifying gauge theoretic cotangent bundle reduction @xcite . the idea is to pass to a rotating frame in which the collinear ( equilibrium ) configuration is along a chosen axis , say the @xmath132-axis , but retain the remaining rotational symmetry ( about @xmath132 ) as an internal coordinate . then by choosing the eckart convention and the non - gauge invariant form of the kinetic energy , we find that the lagrangian is not a function of the angular velocity about the collinear axis @xmath277 , so we can obtain a hamiltonian that is not a function of the first angular momentum component @xmath278 . that is , @xmath278 is replaced by the canonical momentum conjugate to the `` internal '' rotation about the @xmath132-axis . these charts were first considered by sayvetz @xcite , though nowadays they are often attributed to watson @xcite . this procedure can be justified geometrically by applying the slice theorem ( see e.g. ( * ? ? ? * section 2.3.14 ) ) to configuration space in a neighbourhood of the collinear configurations , and then lifting the charts obtained to the cotangent bundle @xcite . actually , with this understanding , charts can be obtained that are not those of the eckart convention , i.e. other gauges and internal coordinates . this was used in examples by kozin et al . note that this is just the splitting of coordinates into internal coordinates and rotations and not an actual reduction , cf . appendix [ bigauge ] . with these charts , we can not simply pass to serret - andoyer coordinates to reduce the symmetry , seeing as the hamiltonian is not a function of @xmath278 . this is generally not addressed in literature . the transport problems associated with reaction are usually a lot more complicated than the capture ones considered here , so their transition states may undergo a number of different bifurcations . one simple example which should display much the same bifurcations as those seen here is isomerization reactions involving only one molecule . these will be the topic of a future publication . we have concentrated here on how the transition state and dividing surface vary with energy , but the exact dependence of the capture transition states on the angular momentum and the possible loss of normal hyperbolicity for large values should be considered in detail . due to the high degrees of freedom of these systems , this is not a straightforward task . it is also interesting to ask whether the dividing surface method can be extended to consider reactions with a varying external field or laser pulse , which need to be modelled as a non - autonomous hamiltonian system ; or reactions that are not in the ( dilute ) gas phase , for which the product kinetic approximation leading to a low dimensional hamiltonian system is not valid ; or even reactions out of equilibrium . reviews of the dividing surface approach applied to the basic transport scenario of flux over a saddle for non - autonomous systems and langevin systems can be found in @xcite and @xcite , respectively . however , more work is required to consider general transport scenarios and fully understand transport in these systems . we thank mark roberts , robert littlejohn and kevin mitchell for pointing us to @xcite and @xcite , respectively , and two anonymous referees for comments and suggestions leading to improvements to the article . strub was supported by an epsrc studentship . reduced charts for @xmath14-body hamiltonian systems in the non - collinear configurations region can be found by considering the euclidean action of @xmath279 on configuration space . it is because one considers the @xmath280 action on configuration space that the configurations must be non - collinear in order to avoid coordinate singularities when the action is not free . this is a different issue from the stratification of the reduced state space . note that generally the configuration and state space isotropy subgroups for a given lie group @xmath264 action are not the same , instead we have that @xmath281 for @xmath20 . the gauge theoretic approach to cotangent bundle reduction is nicely reviewed by littlejohn and reinsch @xcite . they however do not consider the final step required to reduce the rotational symmetry and fix the angular momentum . this is achieved by introducing serret - andoyer coordinates , as explained by deprit @xcite ( see also @xcite ) . these introduce inevitable coordinate singularities on the angular momentum sphere , which is probably why littlejohn and reinsch avoid them . we shall briefly review the gauge theoretic approach for general @xmath14-body systems , and introduce our notation . a specific choice of charts for @xmath14-body system representing bimolecular reactions is given in section [ spatialbimol ] . consider a translation - reduced , rotation invariant @xmath14-body systems restricted to the non - collinear subset ( i.e. the trivial configuration isotropy - type submanifold ) of configuration space @xmath282 and written in the lagrangian formalism @xmath283 where @xmath284 are some choice of jacobi vectors , and @xmath285 the reduced masses . the jacobi vectors are not normalised or mass - weighted . we believe that the mass parameters are best dealt with by non - dimensionalising the system . the potential @xmath39 is assumed to be invariant under the action of @xmath16 . pass from the inertial frame @xmath286 to a convenient rotating frame @xmath287 , which will depend on the problem at hand , and write @xmath288 for @xmath289 where @xmath290 is the rotation parametrised by the euler angles @xmath291 , and @xmath292 are the jacobi vectors in the rotating frame . the rotating jacobi vectors can be expressed in terms of @xmath293 _ internal _ coordinates @xmath123 for @xmath294 by specifying @xmath295 , which is called the _ gauge _ in the physics literature @xcite . we are effectively considering a fibre bundle @xmath296 , and @xmath123 are coordinates for the base space . then , @xmath297 is a section , and the euler angles @xmath291 are coordinates for the fibre , diffeomorphic to @xmath16 . in the new coordinates , the kinetic energy is @xmath298 where @xmath299 is the angular velocity , that is the vector corresponding to the skew - symmetric matrix @xmath300 , for which @xmath301 , for any 3-vector @xmath302 . we are therefore considering an _ anholonomic _ frame ( or _ vielbein _ ) @xmath303 for the tangent space at @xmath304 , with @xmath305 ( * ? ? ? * appendix c ) . the pseudo - metric @xmath306 satisfies @xmath307 this is the restriction of the euclidean metric on the ( translation - reduced ) configuration space @xmath308 to the section @xmath309 , and hence a `` pseudo - metric '' on the internal space @xmath310 . it is of no importance in gauge theoretic terms , but nonetheless features prominently in the molecular literature . the moment of inertia tensor @xmath311 is given by @xmath312 or @xmath313 where @xmath314 is the tensor , or outer , product for which @xmath315 , and the gauge potential @xmath316 associated with the coriolis effect , which is caused by the coupling term , is @xmath317 where @xmath318 and @xmath319 equivalently @xmath320 where @xmath321 are components of @xmath173 , and @xmath322 the levi - civita symbols is @xmath323 if @xmath324 is an even permutation of @xmath325 , @xmath326 if it is an odd permutation , and @xmath327 if any index is repeated . ] . the kinetic energy is gauge invariant , i.e. independent of the choice of internal coordinates , but the individual terms are not ( see ( * ? ? ? * section iv.a ) ) . it is therefore rewritten , in a gauge - invariant form , as @xmath328 where the metric @xmath329 is an actual ( riemannian ) metric on the internal space , obtained by projecting the metric on configuration space @xmath308 down to the internal space . it is therefore positive definite , but non - euclidian due to the nature of the space ( * ? ? ? * section iv.c ) . finally , pass to the hamiltonian formalism . the momenta are found ( via the fibre derivative of the lagrangian ) to be @xmath330 where @xmath111 is the angular momentum in the rotating frame , i.e. @xmath331 . the hamiltonian is then the legendre transform of the lagrangian , namely @xmath332 where the potential is a function of the internal coordinates only , due to the assumption of rotational invariance , and the euler angles are ignorable . the symplectic form is @xmath333 alternatively , the poisson bracket for two smooth functions @xmath334 , @xmath264 is @xmath335 littlejohn and reinsch derive this in ( * ? ? ? * section iv.d ) . the momenta @xmath336 are gauge dependent because of coriolis term @xmath337 . passing to gauge - independent non - canonical momenta . we shall use the same notation , hoping that it will not lead to any confusion , even though it gives @xmath338 . ] , @xmath339 , simplifies the hamiltonian to @xmath340 where @xmath24 is the effective potential combining the centrifugal term and the potential , and @xmath341 where we have introduced the _ coriolis tensor _ @xmath342 which is a curvature form on fibre bundle ( see ( * ? ? ? * section iii.g ) ) , and simplifies the equations of motion . effectively , this transformation moves the coriolis effect from the hamiltonian to the symplectic form , in the second and third terms . it is similar to using non - canonical coordinates for a charged particle in a magnetic field , such that the effect of the lorentz force comes from the symplectic form , see e.g. ( * ? ? ? * section 2.10 ) . the molecular literature usually does not pass to the gauge - invariant form of the kinetic energy , see discussion in ( * ? ? ? * section iv.f ) . by introducing the rotating frame , we have split the coordinates into internal coordinates @xmath123 , ( ignorable ) rotations @xmath343 and their momenta , but we have not actually reduced the system . since @xmath344 are non - canonical , the fact that @xmath343 are ignorable does nt lead to @xmath111 being constant . we can however pass from the non - canonical @xmath344 to canonical serret - andoyer coordinates @xmath345 which consist of the total angular momentum @xmath346 plus two projections of @xmath111 , which we are free to choose , and three angles . the choice of projection onto the @xmath132 and @xmath347-axis is shown in figure [ bimolando ] . we immediately note that @xmath348 which we need to transform the hamiltonian function , whereas the relations between the new angles @xmath349 and the non - canonical angular momentum coordinates is less straightforward and depends on the original choice of euler angles . these are of no use to us here , but can be found in @xcite and @xcite , where @xmath345 are shown to be canonical coordinates . the hamiltonian in these new coordinates is @xmath350 therefore , the system is reduced by eliminating the ignorable degree of freedom @xmath351 , fixing @xmath352 , which is the constant absolute value of the angular momentum , and eliminating @xmath353 . the remaining angular momentum coordinates @xmath354 are the canonical latitude and longitude on the angular momentum sphere @xmath120 , henceforth denoted @xmath124 , and there is a coordinate singularity at @xmath355 . the reduced hamiltonian function is @xmath356 and @xmath357 the choice of projection is equivalent to a choice of which axis to use as a longitude for @xmath358 . the transformation for other projections is equivalent . by considering e.g. minor and major principal axes , we get two charts that cover the whole of @xmath358 . this appendix contains the scaling of the moment of inertia tensor , the gauge potential and the reduced metric for the spatial atom - diatom molecule capture problem of section [ spatialbimol ] . the intermolecular distance degree of freedom is scaled as @xmath140 and @xmath141 , whereas the intramolecular coordinates are shifted such that @xmath142 at the equilibrium , and then scaled as @xmath143 and @xmath144 . thus , the scaled , rotating frame jacobi vectors are @xmath359 where @xmath360 are equilibrium configuration vectors , @xmath361 determines the orientation of @xmath6 and the @xmath362 constants @xmath363 determine the intramolecular coordinates @xmath23 and shall be chosen along the lines of the eckart @xcite and sayvetz @xcite conventions for normal and anomalous molecules in the small vibration regime . if we write the moment of inertia tensor as @xmath364 then @xmath365 where @xmath366 . thus , it scales to @xmath367 where @xmath368 , and we choose @xmath134 with @xmath135 , via @xmath369 and the rotation of @xmath6 about @xmath132 , i.e. @xmath370 . the inverse moment of inertia matrix exists , since @xmath129 is a real , symmetric and positive definite for @xmath6 non - collinear , and scales as @xmath371 for our choice of coordinates and gauge , the gauge potential @xmath372 has @xmath373 and we ask that @xmath374 for all @xmath375 , i.e. @xmath376 this is known as the eckart condition , and imposes @xmath377 conditions on @xmath378 . the gauge potential scales to @xmath379 where @xmath380 and @xmath381 for @xmath382 . finally , for the reduced metric @xmath383 , which is a real , symmetric , positive definite matrix , we consider the pseudo - metric @xmath384 and write @xmath385 where @xmath386 and @xmath387 we ask that @xmath388 for all @xmath389 . that is we choose williamson normal form coordinates for the intramolecular degrees of freedom . these are @xmath390 conditions on @xmath391 . furthermore , we claim that due to eckart condition @xmath392 for all @xmath389 , i.e. @xmath393 let us consider the case with @xmath394 . the euler angles @xmath22 can be chosen in a number of ways , and the rotation matrix written as @xmath395 , where @xmath396 is a rotation by @xmath397 about some axis @xmath398 . recall that the symmetry about @xmath132 has been reduced and @xmath399 . thus @xmath400 where @xmath401 is a rotation about @xmath402 by @xmath403 and simultaneously a contraction in the @xmath402 direction to zero . this can be seen by considering planar rotation matrices . then @xmath404 where @xmath405 and @xmath406 by the eckart condition . thus @xmath407 , and the same is true for @xmath408 . the gauge dependent term of @xmath383 instead scales to @xmath409 so @xmath410 and @xmath411 by inverting the matrix blockwise , and expanding inverse matrices in formal power series .

power spectrum estimation and evaluation of associated errors in the presence of incomplete sky coverage ; non - homogeneous , correlated instrumental noise ; and foreground emission is a problem of central importance for the extraction of cosmological information from the cosmic microwave background . we develop a monte carlo approach for the maximum likelihood estimatation of the power spectrum . the method is based on an identity for the bayesian posterior as a marginalization over unknowns , and maximization of the posterior involves the computation of expectation values as a sample average from maps of the cosmic microwave background and foregrounds given some current estimate of the power spectrum or cosmological model , and some assumed statistical characterization of the foregrounds . maps of the cmb and foregrounds are sampled by a linear transform of a gaussian white noise process , implemented numerically with conjugate gradient descent . for time series data with samples , and pixels on the sphere , the method has a computational expense $ ] , where is a prefactor determined by the convergence rate of conjugate gradient descent . preconditioners for conjugate gradient descent are given for scans close to great circle paths , and the method allows partial sky coverage for these cases by numerically marginalizing over the unobserved , or removed , region .

astro-ph0209560

power spectrum estimation and evaluation of the associated errors in the presence of incomplete sky coverage ; non - homogeneous , correlated instrumental noise ; and foreground emission is a problem of central importance for the extraction of cosmological information from the cosmic microwave background . from a bayesian point of view , power spectrum estimation involves the maximization of the posterior probability density , with error bars given by the set of cosmological parameters or power spectrum whose integrated posterior density achieves some specified level of confidence . a bayesian approach to cmb analysis for large data sets involving a direct evaluation of the likelihood is intractable due to the @xmath4 expense associated with computing the inverse of non - sparse matrices , or their determinants ( @xcite , @xcite ) . the goal of this paper is the development of alternative numerical methods , specifically monte carlo techniques , for the bayesian analysis of the cmb , including the complications of incomplete sky coverage , correlated noise and foregrounds . previous work has demonstrated that for a certain class of scanning strategies , the signal and inverse noise matrices are block diagonal . the block diagonal properties of these matrices give an exact @xmath5 bayesian method , and therefore tractable for data sets as large as will be returned from planck . the complications of this method are that it can not easily accommodate partial sky converge or precessing scan strategies . the method of @xcite computes the maximum of the likelihood through a newton raphson method . the numerical innovations of this method involve monte carlo simulations and the use of conjugate gradient descent , giving an overall expense @xmath5 . the method was proposed and numerically demonstrated in the context of uncorrelated noise , and a region of sky coverage of azimuthal symmetry , where a good preconditioner can be constructed . however , the algorithm is in fact more general , provided there is sufficient memory for storage of the needed matrices , and that conjugate gradient descent converges quickly enough ( i.e. there is a good preconditioner ) . as suggested in @xcite , we can use the ring set approach to supply preconditioners . an outstanding problem to be solved is a way of retaining the mathematical advantages of a ring set scan ( block diagonal inverse noise and signal matrices ) while accommodating partial sky coverage . the approach formulated in this paper handles the problem of partial sky coverage by embedding the data in an azimuthally symmetric region of sky , and using a monte carlo markov chain to numerically marginalize over the unobserved part . for scans close to ring sets , we therefore inherit good preconditioners , allowing an extension of both the ring set and conjugate gradient methods to scan strategies as planned for planck . for observations @xmath6 , where @xmath7 are the cmb signal , foregrounds , and noise respectively , our approach to power spectrum estimation is motivated by the identity ( derived in appendix [ appposterioridentity ] ) @xmath8 p(s , f|\gamma_{0},y)\ ] ] where @xmath9 is any parameterization of conclusions ( such as the power spectrum or cosmological parameters ) , and @xmath10 is any fixed guess . the bayesian posterior ratio on the left is given as an integral over the unknown quantities which are assumed to generate the observed data . maximization of the posterior involves computing the gradient of equation [ posterioridentity ] which will be shown to depend on the expectation value of the power spectrum with respect to the random field @xmath11 , @xmath12 = \int d(s , f ) \ c_{l}(s ) \ p(s , f |\gamma_{0},y)\ ] ] we maximize the posterior ratio in equation [ posterioridentity ] by the expectation maximization algorithm ( @xcite ) which proceeds by iteratively setting @xmath13 $ ] . the algorithm converges to the posterior maximum for a uniform prior , and gives an un - biased , consistent estimator ( see appendix [ appestimator ] ) . in this paper , we focus on computation of the expectation value of the power spectrum @xmath14 $ ] given the data and some guess @xmath15 under the assumption of perfect foreground separation ( although we comment on how the approach can be generalized to include foregrounds later in the paper , and leave its numerical demonstration for future work ) . we compute the expectation value of the power spectrum @xmath14 $ ] numerically with a monte carlo approach , where we sample maps of the cmb from the probability density @xmath16 . conditioning on some estimate of the foregrounds , the method exploits the fact that @xmath17 is a gaussian random field , and therefore completely characterized by the mean field map and covariance matrix of fluctuations about that map . maps are sampled from @xmath17 by first computing the mean field map with conjugate gradient descent , and then sampling fluctuations about the mean field map from a zero mean gaussian field with covariance matrix @xmath18 ( where @xmath19 is the inverse noise matrix , and @xmath20 is the inverse covariance matrix for the cmb ) . these fluctuation maps are sampled by a linear transformation , numerically computed with conjugate gradient descent , of a spatial white noise gaussian process thereby generating maps with all the same statistical properties as samples from @xmath17 . each step of conjugate gradient descent involves a multiplication by the matrix @xmath21 , which can be done very quickly by multiplication by @xmath19 in the basis in which it is diagonal , followed by a transform to the spherical harmonic basis where @xmath22 is diagonal . for spatially uncorrelated noise and circularly symmetric beams , we only need to transform from the pixel to the spherical harmonic domain , with an expense @xmath23 @xcite . in order to accommodate spatially correlated noise , we transform to the time domain , followed by a transform to the spherical harmonics , giving an expense @xmath24 $ ] where @xmath0 is the number of time samples , and @xmath3 depends on the convergence rate of conjugate gradient descent . including the full complications of asymmetric beams , we would need to compute a convolution on the sphere . using the convolution method of ( @xcite ) , the expense of our method is @xmath25 $ ] . the computational feasibility of this method is limited by finding a numerical implementation of conjugate gradient descent which converges quickly so that the prefactor @xmath3 above is small . the strategy here is to embed the data in a region covered by an exact ring set scan , following the intuition that good preconditioners can be constructed for scan strategies close to ring sets @xcite . embedding the data in a region on the sky with no observations ( or where they have been removed ) is accommodated by numerically marginalizing over the missing observations . moreover , the same techniques can be used to marginalize over the foregrounds , and provide monte carlo estimates of the confidence intervals for cosmological parameters . the paper is organized as follows . we first review complications with a direct computation of the likelihood , and provide an overview of our approach . we then discuss a technique we call transformed white noise sampling , which allows us to sample maps representing fluctuations about the mean field map for some guess of the power spectrum . we demonstrate the method with a flat sky @xmath26 test case , including incomplete sky coverage , with uncorrelated , non - homogeneous noise . we close with a discussion of further complications encountered in real cmb experiments , and how they can be accommodated in the framework presented here . we begin with a brief review of the likelihood and complications with its computational evaluation . the data returned from an experiment is a vector in the time domain @xmath27 , which is related to the cmb signal on the sky @xmath28 through some linear mapping and additive gaussian noise , @xmath29 \int dn '' \ b(n',n '' ) s(n '' ) \right ] + \eta(t)\ ] ] where @xmath30 is the beam of the instrument , and @xmath31 is gaussian noise , assumed to be stationary with a noise correlation matrix @xmath32 . denoting the linear mapping from time domain to the sky as @xmath33 \int dn '' \ b(n',n'')\ ] ] we will simply write @xmath34 . the likelihood for the power spectrum @xmath35 given the data is @xmath36 any linear transformation of the data vector will generate a gaussian form for the likelihood ( as reviewed in appendix [ applikelihoods ] ) . for example , we can transform the data to an estimate of the cmb map @xmath37 ( as discussed in detail in @xcite , @xcite ) . the covariance matrix of this map is @xmath38 & = & c + ( r^{t } n^{-1 } r)^{-1 } \end{aligned}\ ] ] which shows that the map @xmath39 can be thought of as signal @xmath22 with additive gaussian noise with covariance matrix @xmath40 . the likelihood can can equivalently be written in terms of the map @xmath41 according to @xmath42 & = & - { \hat s}(d ) [ ( r^{t } n^{-1 } r)^{-1 } + c ] ^{-1 } { \hat s}(d ) - \log \det [ ( r^{t } n^{-1 } r)^{-1 } + c ] \end{aligned}\ ] ] we can also write the likelihood in the original time domain , @xmath43\ ] ] directly evaluating the likelihood in either the spatial or time domain results in an @xmath4 computation , as it involves inversion , or computation of the determinant , of @xmath44 $ ] or @xmath45 $ ] respectively . the computational expense is due to the fact that we do not know the eigenbasis for either of these matrices , and computing this basis is generally an @xmath4 problem . the method of @xcite solves the likelihood in the spatial domain by evaluating the determinant with a monte carlo algorithm . the method involves conjugate gradient descent to solve a linear problem , and as such involves matrix multiplication , which carries an @xmath5 expense . the method was proposed and numerically demonstrated in the context of uncorrelated noise , and a region of sky coverage of azimuthal symmetry . however , the algorithm is in fact more general , provided there is sufficient memory for storage of the needed matrices , and that conjugate gradient descent converges quickly enough ( i.e. there is a good preconditioner ) . for a certain class of observing strategies , we can exactly compute the likelihood , and for perturbations about these cases , we can use the approximate case as a preconditioner ( as suggested in @xcite ) . the approach formulated in this paper provides a consistent way to do this , and involves a monte carlo markov chain approach to numerically marginalizing over the unobserved part of azimuthally symmetric regions of the sky . the method to be developed in this paper involves embedding the data in a region for which the signal and noise matrices have desirable properties . the likelihood for the data in the context of some model is given as an integral over the part of the embedding region which was not observed . this gives the identity for the bayesian posterior for the power spectrum or cosmological model ( denoted by @xmath9 ) , given the time domain data @xmath27 as the integral @xmath46 p(s^{(1)},s^{(2)},f|\gamma_{0},d)\ ] ] where we have explicitly written the cmb maps @xmath47 in terms of the part of the sky where we have data @xmath48 and the complementary region @xmath49 . for the case of full sky coverage and prior knowledge @xmath50 , the log posterior ratio is given as @xmath51\ ] ] where the power at a given multipole order is @xmath52 for other regions with azimuthal symmetry ( an annulus or polar cap ) , the posterior ratio would involve a similar form in terms of block diagonal matrices . the simple form of the log posterior ratio for full sky coverage is one of the motivations for treating the problem of partial sky coverage as missing data , and marginalizing over it ( detailed and justified in appendix [ apppartialcoverage ] ) . in order to estimate the power spectrum , we would like to find @xmath9 which maximizes the posterior given the noisy data . differentiating the posterior with respect to the parameters @xmath35 gives a gradient in the direction @xmath53}{c_{l}^{2}(\gamma_{0 } ) } - \frac{1}{c_{l}(\gamma_{0 } ) } \right]\ ] ] where @xmath54 $ ] was given previously in equation [ expectation ] . an improvement of our current estimate of the power spectrum @xmath10 can then given according to a newton - raphson iterative scheme ( @xcite and @xcite ) , where our current guess is updated using an approximation to the curvature of the likelihood @xmath55 $ ] , according to @xmath56 \left . \frac{\partial \log p(\gamma|d)}{\partial c_{l } } \right|_{\gamma_{n}}\ ] ] in appendix [ appestimator ] , it is shown that the curvature matrix is given in terms of the expectation value @xmath57 = \int ds \ c_{l}(s ) c_{l'}(s ) \ p(s|d , \gamma_{0})\ ] ] in practice , we might want to avoid computing the inverse of the curvature matrix , and simply use the diagonal elements . for this paper , we instead implemented the simpler ( although more slowly converging ) expectation maximization algorithm @xcite . this method essentially follows from jensen s inequality , giving the lower bound to the posterior @xmath58 \left ( \frac{1}{c_{l}(\gamma ) } - \frac{1}{c_{l}(\gamma_{0 } ) } \right ) + \log \frac{c_{l}(\gamma ) } { c_{l}(\gamma_{0 } ) } \right]\ ] ] for a uniform prior , the lower bound is maximized by @xmath59 $ ] . in appendix b , we prove that this estimator iteratively converges to the maximum of the posterior , and is a consistent and unbiased estimator ( for a uniform prior ) . in order to iteratively converge to the optimal , consistent estimator of the power spectrum , we need to compute , for any current guess of the power spectrum @xmath10 , the expectation value @xmath60 $ ] as defined in equation [ expectation ] . defining the mean field map @xmath61^{-1 } n^{-1 } y\ ] ] and associated power spectrum estimate @xmath62 the expectation value can be written by integrating over fluctuations about the mean field map @xmath63 as @xmath64 & = & \frac{1}{2l+1 } \sum_{m } \int d \xi \ \langle { \hat s } + \xi | lm \rangle \langle lm | { \hat s } + \xi \rangle \ \xi [ n^{-1 } + c^{-1}(\gamma_{n } ) ] \xi } } { \int d\xi ' \ e^{- \xi ' [ n^{-1 } + c^{-1}(\gamma_{n } ) ] \xi ' } } \nonumber \\\end{aligned}\ ] ] since @xmath65 = 0 $ ] , the expectation value is @xmath64 & = & c_{l}({\hat s } ) + \frac{1}{2l+1 } \sum_{m } \int d \xi \ \langle \xi | lm \rangle \langle lm | \xi \rangle \ \frac{e^{- \xi [ n^{-1 } + c^{-1}(\gamma_{n } ) ] \xi } } { \int d\xi ' \ e^{- \xi ' [ n^{-1 } + c^{-1}(\gamma_{n } ) ] \xi ' } } \end{aligned}\ ] ] we refer to these two terms as the mean field map power spectrum estimate ( known to be biased ) and the correction term . analytically , we know that the correction term is given by @xmath66^{-1 } matrix @xmath67^{-1}$ ] . our strategy is to compute the correction term with a monte carlo method described below . due to the computational intractability of computing the matrix inverse @xmath68 , our strategy is to compute the expectation value of the correction term from fluctuation maps @xmath69 sampled from the zero mean gaussian random field with covariance matrix @xmath68 . we could easily sample fluctuation maps @xmath69 if we could compute the eigenvectors and eigenvalues of the matrix @xmath70 , since in this basis the gaussian probability density for @xmath69 factors . however , computing the eigenvectors and eigenvalues is again an @xmath4 operation . because of these difficulties , we look for an alternative way to sample maps . defining @xmath71 , we can write the log density , up to the normalization constant , as @xmath72 the transformed gaussian process has the covariance matrix @xmath73 , making it easy to sample from . specifically , we can sample maps from this gaussian process by drawing two independent white noise maps @xmath74 , and setting @xmath75 . since both white noise maps are drawn independently from a zero mean gaussian process , the resulting covariance matrix is @xmath76 = n^{-1 } + c^{-1}$ ] ( as discussed in appendix [ appsampling ] ) . the maps with the correct statistical properties are @xmath77 , which can be solved numerically for a given map @xmath78 . a numerically stable implementation involves setting @xmath79 ( as also noted in @xcite ) , and using conjugate gradient descent to solve @xmath80 the resulting maps @xmath69 have the correct statistical properties , since @xmath81 = ( n^{-1 } + c^{-1 } ) ^{-1}$ ] ( see appendix [ appsampling ] ) , allowing us to compute the correction term to the power spectrum estimate of the mean field map as a sample average . in order to actually sample fluctuation maps by transforming a gaussian white noise process , we need to obtain the cholesky decomposition of both the signal and noise matrices . if we have observations with uncorrelated noise on the sky , then @xmath82 is known in the spatial domain . however , the scan strategy of the instrument will result in complicated correlations , so that computing @xmath82 is intractable . the noise is simple in the time domain , which suggests that instead of choosing white noise maps in the spatial domain , we instead draw from white noise gaussian processes in the time domain , where we know @xmath82 in the fourier basis , followed by a transformation to the sky , where we can operate with the signal matrix @xmath22 . for a realization of a white noise process in the time domain @xmath83 and a white noise map in the spatial domain @xmath84 , we can compute a fluctuation map according to @xmath85 \xi = c^{1/2 } \omega + c r^{t } n^{-1/2 } \tau\ ] ] where @xmath82 is known in the fourier basis associated with the time domain . in appendix [ appsampling ] , this procedure is justified with a proof that the covariance matrix of the fluctuation maps is @xmath81 = [ c^{-1}+ ( r^{t } n^{-1}_{t } r)^{-1}]^{-1}$ ] . the overall computational expense is fixed , for each iteration of the power spectrum estimate , by the expense of matrix multiplication and number of iterations needed to converge with conjugate gradient descent . in order to multiply by the matrix @xmath86 we need to : * transform to the time domain with the matrix @xmath87 . * compute a time domain fft . * multiply by @xmath19 - this is a diagonal matrix in the time domain fourier basis * compute a time domain inverse fft . * transform back to the spatial domain with @xmath88 . * compute a spherical harmonic transform . * multiply by @xmath22 - this is a diagonal matrix in the spherical harmonic domain if the embedding region is the full sky . for the case of circularly symmetric beams , the convolution with the beam is not needed when operating with the matrix @xmath87 or its transpose , giving an expense @xmath89 $ ] , where @xmath0 is the number of time samples , and @xmath3 is the prefactor related to the convergence rate of conjugate gradient descent . for cases where the beam is not circularly symmetric , the convolution with the beam would have to be computed , increasing the expense to @xmath90 $ ] . the simulations presented here involve the assumption of spatially uncorrelated , but non - homogeneous , noise , as shown by the upper left in figure @xmath91 . we also restrict the problem to power spectrum estimation from a small patch of sky , and neglect curvature ( and therefore work with discrete fourier basis instead of spherical harmonics ) . our goal with these numerical simulations is to demonstrate the approach in action . future work will involve numerical implementations on the sphere . a cmb power spectrum was generated using cmbfast @xcite , followed by the creation of a full sky map on the sphere using the * synfast * routine in the healpix package @xcite . a smaller patch of the sky was then selected , and projected on a rectangular grid . this map @xmath28 was taken to be the noise free map , as shown in the upper right of figure 1 . we then generated a noise map @xmath92 by selecting independently at each pixel a gaussian random number with variance scaled as shown according to the upper left of figure 1 . noise was added to the noise free map , and data then removed in a rectangular hole ( as shown in figure 1 ) . this was taken to be a simulated data set @xmath93 with partial coverage of the rectangular patch of sky . the inverse noise matrix was given in terms of the variance at the ith pixel @xmath94 as @xmath95 as an initial estimate of the power spectrum , we computed the the power spectrum of the noisy , incomplete data ( as computed in the two - dimensional fourier basis since we neglected curvature ) and subtracted the power spectrum of a single simulated noise map ( on the region of sky where we have data ) . we then iteratively adjusted the power spectrum with the expectation maximization as above until convergence . we found that preconditioning was in fact necessary to achieve a reasonable convergence rate . using the initial estimate of the power spectrum @xmath96 , and computing the diagonal elements of the inverse noise matrix in the fourier domain @xmath97 with monte carlo noise maps , gave the preconditioner @xmath98^{-1}\ ] ] conjugate gradient descent was then used to solve the linear equations for the mean field and fluctuation maps @xmath99 { \hat s } & = & m^{-1 } cn^{-1 } y \nonumber \\ m^{-1 } [ i + cn^{-1 } ] \xi & = & m^{-1 } \delta\end{aligned}\ ] ] where @xmath100 was computed from two independently chosen spatial white noise maps @xmath74 , and @xmath82 vanished in the unobserved part of the sky , and elsewhere given by @xmath101 . the result of iterating the algorithm to convergence is shown in figure 2 . uncertainties in the power spectrum estimate were computed by monte carlo , in which new cmb maps were generated , noise added , and the algorithm run again . the methods presented above can be generalized to handle other additional problems faced in the bayesian analysis of the cmb . we do not provide numerical examples , but briefly include comments on how to use transformed white noise sampling to estimate error bars and include foregrounds . the ability to sample maps of the cmb given some estimate of the power spectrum can be used to construct a markov chain monte carlo algorithm which converges to the bayesian posterior @xmath102 itself . previously , markov chain monte carlo techniques have been proposed for the extraction of marginal densities for cosmological parameters from approximate bayesian posterior densities for the power spectrum ( @xcite , @xcite , @xcite , @xcite ) we want to construct a transition matrix for the _ joint _ density of cmb maps and the power spectrum . this can be done by following the metropolis hastings algorithm , where we first assume detailed balance @xmath103 from the condition of detailed balance we see that @xmath104 which shows that the bayesian posterior is the marginalized equilibrium density , generated by repeatedly taking steps generated with the transition matrix @xmath105 . given any approximation to the joint density @xmath106 , repeated application of the transition matrix will reach the equilibrium density . we can construct the transition matrix as follows . we first assume the proposal density is independent of the past @xmath107 and given by @xmath108 s } { \tilde p}(\gamma_{2 } | y)\ ] ] where @xmath109 is any approximation to the bayesian posterior itself . we therefore have , as before , the conditional density @xmath110 s } } { \int ds ' \ e^{- ( d - rs ' ) n^{-1 } ( d - rs ' ) - s ' c^{-1}[\gamma_{2 } ] s ' } } \ ] ] and the marginal density for the power spectra given by our approximation @xmath109 . with this choice of proposal matrix , the acceptance matrix can be chosen to be @xmath111 = \min \left [ 1 , \frac{p(\gamma_{2 } , s_{2 } | y)}{p(\gamma_{1 } , s_{1 } | y ) } \frac{\rho(\gamma_{1},s_{1 } | y ) } { \rho(\gamma_{2 } , s_{2 } | y ) } \right]\ ] ] what makes this computationally tractable is that we know the ratios @xmath112 s_{2 } - \log \| c(\gamma_{2 } ) \| } q[\gamma_{2 } ] } { e^{- ( d - rs_{1})n^{-1 } ( d - rs_{1 } ) - s_{1 } c^{-1}[\gamma_{1 } ] s_{1 } - \log \| c(\gamma_{1 } ) \| } q[\gamma_{1 } ] } \ ] ] as well as @xmath113 s_{1 } } { \tilde p}(\gamma_{1 } | y ) } { e^{- ( d - rs_{2 } ) n^{-1 } ( d - rs_{2 } ) - s_{2 } c^{-1}[\gamma_{2 } ] s_{2 } } { \tilde p}(\gamma_{2 } | y)}\ ] ] which simplifies to the explicitly computable acceptance probability @xmath111 = \min \left [ 1 , \frac{e^{- \log \| c(\gamma_{2 } ) \| } q[\gamma_{2 } ] } { e^{- \log \| c(\gamma_{1 } ) \| } q[\gamma_{1 } ] } \frac{{\tilde p}(\gamma_{1 } | y ) } { { \tilde p}(\gamma_{2 } | y ) } \right]\ ] ] in summary , a single step of the markov chain involves : * choose a new guess for the power spectrum from @xmath109 . * sample a map from @xmath114 according to * * compute the mean field map @xmath115^{-1 } r^{t } n^{-1 } y$ ] . * * for two independently sampled white noise maps in the spatial and time domains @xmath116 , compute @xmath69 as the solution to @xmath117 . * * set @xmath118 . * accept the transition @xmath119 with probability @xmath120 $ ] . * continue for circularly symmetric beams , each step of the markov chain has expense @xmath89 $ ] , giving a total expense @xmath121 $ ] , where @xmath122 is the number of realizations of @xmath123 , and @xmath124 quantifies the `` efficiency '' of the markov chain ( suggestively denoted by @xmath125 since we intuitively expect it to vary inversely with the average acceptance probability ) . by the efficiency of the markov chain , we mean the number of proposed moves that must be made until one is accepted . this efficiency depends entirely on how good the approximation to the posterior @xmath109 actually is . we can always write the acceptance matrix as @xmath111 = \min \left [ 1 , \frac{p(s_{2 } | \gamma_{2 } , y)}{p(s_{1 } | \gamma_{1 } , y ) } \frac{p(\gamma_{2}| y)}{p(\gamma_{1 } | y ) } \frac{\rho(s_{1 } | \gamma_{1 } y ) } { \rho(s_{2 } | \gamma_{2 } y ) } \frac{{\tilde p } ( \gamma_{1 } | y ) } { { \tilde p}(\gamma_{2 } | y ) } \right]\ ] ] however , by construction , this is exactly @xmath111 = \min \left [ 1 , \frac{p(\gamma_{2}| y)}{p(\gamma_{1 } | y ) } \frac{{\tilde p } ( \gamma_{1 } | y ) } { { \tilde p}(\gamma_{2 } | y ) } \right]\ ] ] so that if our approximation was exact , the acceptance matrix is unity for all proposals , and we reduce to an expense @xmath126 $ ] . provided that we have reasonable approximations to the posterior , we can converge to the exact posterior with this method , and the use of transformed white noise sampling . we also briefly comment on some practical issues with this approach . in order to run the markov chain , we need to compute the mean field map _ for every proposed power spectrum_. while this is feasible , we can speed up the markov chain by sampling a grid of power spectra from our approximation @xmath127 , and pre - computing the mean field maps for each @xmath9 on our grid . then , for any proposed power spectrum , we can start with the initial condition for conjugate gradient descent with the closest grid point . we conclude with a brief discussion of the inclusion of foregrounds . the inclusion of the foregrounds involves sampling both cmb and foreground maps given some estimate of the cmb power spectrum and some prior for the foregrounds . for a gaussian prior for the foregrounds , the same approach to sampling can be followed . including foregrounds the data for the jth frequency channel is given by @xmath128 where @xmath129 is the mapping from the sky to the time domain for the jth frequency channel , @xmath130 is the response of the cmb at the jth frequency , and we sum over the foreground components @xmath131 . for a statistical characterization of the foregrounds @xmath132 and a guess of the cmb power spectrum @xmath9 , we need to sample from the density given by ( up to normalization ) @xmath133 we need to compute the expectation value @xmath134 = \int d(s , f ) \ c_{l}(s ) \ p(s , f | d,\gamma_{0 } , \beta)\ ] ] this expectation value can be numerically computed by sampling maps @xmath135 by the time average of a markov chain with equilibrium density @xmath136 . one legitimate way to construct such a markov chain is to alternately sample maps from the conditional densities @xmath137 and @xmath138 . we briefly comment on sampling maps from each of these conditional densities . given some estimate of the foregrounds , we need to sample maps from the conditional density @xmath139 sampling from @xmath137 proceeds in essentially the same way as described above , but generalized for multi - frequency data . specifically , the mean field map includes the subtracted response from the foreground estimate @xmath140^{-1 } a_{0 } r^{t } n^{-1 } ( d - af)\ ] ] and the fluctuation maps include a time domain white noise sample for each frequency channel @xmath141 note that the covariance matrix of the fluctuations is @xmath142 = ( c^{-1 } + r^{t } a_{0 } n^{-1 } a_{0 } r)^{-1}\ ] ] where the proof depends on the independence of the time domain white noise maps for every frequency channel . given some estimate of the cmb , we need to sample from the conditional density @xmath143 if the prior for the foregrounds @xmath144 is gaussian @xmath145 then we can also use transformed white noise sampling . first , we compute the mean field foreground map , @xmath146^{-1 } r^{t } a n^{-1 } ( d - a_{0 } s)\ ] ] and then sampling fluctuations @xmath147 according to the transform of white noise samples @xmath148 multiplying by the foreground signal matrix @xmath149 is either done in the pixel or spherical harmonic basis depending on the basis in which it is sparse . if we use a non - gaussian prior ( such as the maximum entropy method , or other priors ) , then we will need to employ more general sampling techniques to sample foreground maps , such as the metropolis algorithm . in this case the needed expectation value is to be computed as the _ time average _ from a markov chain with equilibrium density @xmath150 . the fundamental hurdle to numerically implementing an exact bayesian approach to cmb analysis , including complications of partial sky coverage , correlated noise , and foregrounds , is finding efficient ways to solve the linear problem @xmath151 for any vector @xmath78 . solving the linear equation has an expense @xmath89 $ ] for circularly symmetric beams , and the algorithm provides a tractable approach provided @xmath3 can be made small enough . the strategy we presented in this paper allows the data to be embedded in an azimuthally symmetric region of the sky covered by a wandelt ring set scan , with the intuition that , provided the true scan of the instrument is close enough to the exact scan , we inherit good preconditioners . we also commented on how the method of transformed white noise sampling can be used in monte carlo markov chain for the entire bayesian posterior . the feasibility of this approach depends on a good approximation to the posterior itself . previous work has demonstrated several computationally feasible , unbiased estimates of the power spectrum and associated error covariance matrix . any of these methods could therefore be used , in principle , to give an approximate posterior , so that a markov chain approach can be used as a final consistency check . future work will incorporate the foregrounds in the algorithm presented here , generalized for multifrequency data . maximization of the likelihood of the power spectrum given the data again leads to the computation of the expectation value @xmath152 $ ] , but now the marginalization includes the foregrounds as well . if the prior for the foregrounds is gaussian , then we can also use transformed white noise sampling to sample a new foreground map while conditioning on the cmb . if the prior used is non - gaussian , other sampling schemes can be used , including gibbs sampling or the metropolis algorithm . this research was carried out at the nasa jet propulsion lab , under nasa aisrp grant , and support from the long wavelength center . we also thank eric hivon and ben wandelt for interesting discussions during the course of this work . the data returned from a cmb experiment is a vector @xmath27 in the time domain , generated from scanning the cmb signal @xmath28 and foregrounds @xmath153 on the sky and adding independent gaussian noise . the bayesian posterior is given directly as an integral over unknown quantities by @xmath154 for the case of gaussian random fields , this integral can be done analytically ( as will be discussed in appendix [ applikelihoods ] ) , but evaluation of the resulting likelihood leads to computationally intractable matrix manipulations . for any estimate @xmath10 @xmath155 p(\gamma_{0},s , f|d)\ ] ] since we have @xmath156 this becomes @xmath157 p(s , f|\gamma_{0},d)\ ] ] where we have used @xmath158 . by the assumed independence of the noise , the joint density over which we are marginalizing is @xmath159 where @xmath50 is a prior for the parameters , @xmath160 is a prior for the foregrounds , and @xmath161 is completely determined by the noise properties of the instrument , the beam , and scan strategy . given some estimate of the noise free cmb signal , the density for a new guess of the power spectrum is independent of the data , @xmath162 , as shown by @xmath163 therefore , for _ any estimate _ @xmath10 , our identity now reads @xmath164 p(s , f|\gamma_{0},d)\ ] ] or for the likelihood ratio @xmath165 p(s , f|\gamma_{0},d)\ ] ] the likelihood of the data , generally in the time domain for some experiment , is @xmath166 we first consider the case of transforming the time ordered data to a map according to the transformation @xmath37 . the likelihood is the probability density evaluated at the data . therefore , we first need to find the probability density of _ maps _ given by transforming the density for time ordered data , according to @xmath167 = \int \delta d \ \delta[{\hat s } - ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } d ] \ e^{- ( d - rs)n^{-1 } ( d - rs ) } e^{-s c^{-1 } s}\ ] ] this is equivalently @xmath168 & \propto & \int \delta d \ \delta[{\hat s } - ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } d ] \ e^{- ( d - rs)n^{-1 } ( d - rs ) } e^{-s c^{-1 } s } \nonumber \\ & = & \int \delta d \ \int dk \ e^{- i k \cdot { \hat s } } e^{+i k \cdot ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } d } \int ds \ e^{- ( d - rs)n^{-1 } ( d - rs ) } e^{-s c^{-1 } s } \nonumber \\ & = & \int dk \ e^{- i k \cdot { \hat s } } \int ds \ e^{-s c^{-1 } s } \int \delta d \ e^{+i k \cdot ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } d } e^{- ( d - rs)n^{-1 } ( d - rs ) } \nonumber \\ & = & \int dk \ e^{- i k \cdot { \hat s } } \int ds \ e^{+ i k \cdot s } e^{-s c^{-1 } s } \int \delta d \ e^{+i k \cdot ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } ( d - rs ) } e^{- ( d - rs)n^{-1 } ( d - rs ) } \nonumber \\ & = & \int dk \ e^{- i k \cdot { \hat s } } e^{- k ( r^{t } n^{-1 } r)^{-1 } k } \int ds \ e^{+ i k \cdot s } e^{-s c^{-1 } s } \nonumber \\ & = & \int dk \ e^{- i k \cdot { \hat s } } e^{- k ( r^{t } n^{-1 } r ) ^{-1 } k } e^{- k c k } \nonumber \\ & = & e^{- { \hat s } [ ( r^{t } n^{-1 } r ) ^{-1 } + c ] ^{-1 } { \hat s } } \end{aligned}\ ] ] notice that there are many time series vectors which correspond to the same spatial map , since we can add any vector such that @xmath169 . this is possible since @xmath170 vanishing means that averages at the same point vanish . we can also write the likelihood in terms of the data in the original time domain . this can be done by writing the characteristic function @xmath171 which shows that @xmath172 the likelihood for the cmb given the theory for partial sky coverage can be written as the marginalization over the unobserved part of the sky . denoting the cmb @xmath173 as the cmb in the observed and unobserved regions of sky respectively , we have @xmath174 therefore , the posterior for partial sky coverage can be written @xmath175 p(f|\beta ) \nonumber \\ & = & q(\gamma ) \int d(s^{(1)},s^{(2)},f ) \ p(y|s^{(1)},f ) p(s^{(1)},s^{(2 ) } |\gamma ) p(f|\beta)\end{aligned}\ ] ] this gives the identity , explicitly written for arbitrary sky coverage , @xmath176 p(s^{(1)},s^{(2)},f|\gamma_{0},y)\ ] ] because @xmath177 is supported on the full sky , the log posterior ratio is @xmath178\ ] ] and the conditional density from which we are to sample @xmath179 is @xmath180 this is equivalent to setting the inverse noise matrix to zero for the part of the sky where there is no data . two limiting cases of our identity involve `` no data '' and `` no noise '' . in the no data limit , the inverse noise matrix vanishes everywhere , so that @xmath181 and the posterior is given by the prior , as shown by @xmath182 p(s | \gamma_{0 } ) \nonumber \\ & = & \int ds \ \left [ \frac{p(s|\gamma ) q(\gamma)}{p(s|\gamma_{0 } ) q(\gamma_{0 } ) } \right ] p(s | \gamma_{0 } ) \nonumber \\ & = & \frac { q(\gamma ) } { q(\gamma_{0 } ) } \int ds \ p(s|\gamma ) \nonumber \\ & = & \frac { q(\gamma ) } { q(\gamma_{0})}\end{aligned}\ ] ] in the noise free limit , the conditional density @xmath183 , independent of our choice of @xmath10 , so that we converge to the noise - free posterior ratio . recall the the algorithm used involves iterating @xmath184\ ] ] and therefore the fixed point satisfies @xmath185 $ ] . we can prove that this estimator maximizes the posterior , or equivalently the log posterior . a direct calculation shows that @xmath186 \frac{p(\gamma|s)}{p(\gamma_{0}|s ) } p(s|y,\gamma_{0})\ ] ] so that at the maximum @xmath187 p(s|y,\gamma_{0})\ ] ] at the maximum we therefore have @xmath188 which is identically the fixed point of the expectation maximization algorithm . therefore , the iteration converges to the maximum of the posterior ( for a uniform prior ) . after we have computed the maximum likelihood estimator of the power spectrum ( or maximum posterior estimate ) , we want to find a confidence interval . there are several ways this confidence interval might be approximated . one approach is to compute the inverse curvature matrix of the likelihood , and take the diagonal entries as an estimate of the error bars ( refs ) . we comment below on how this can be done with the same method used to compute @xmath189 $ ] . approximating the likelihood as a gaussian functional of the @xmath190 is equivalent to a second order taylor expansion of the log likelihood about the maximum . as before we have the identity for the likelihood @xmath191 it is more convenient to parameterize the likelihood ratio in terms of @xmath192 , so that when embedding the data on the full sky , we have @xmath193 c_{l}(s ) - \log \frac{\theta_{l}(\gamma)}{\theta_{l}(\gamma_{0 } ) } \right)\ ] ] denoting the curvature matrix of the likelihood @xmath194 \equiv - \frac{\partial^{2 } \log p(d|\gamma ) } { \partial c_{l } \partial c_{l ' } } \ ] ] we have the relation @xmath195 & = & \frac{1}{c^{2}_{l}(\gamma_{0 } ) } \frac{\partial^{2 } \log p(d|\gamma ) } { \partial \theta_{l } \partial \theta_{l ' } } \frac{1}{c^{2}_{l'}(\gamma_{0})}\end{aligned}\ ] ] the curvature of @xmath196 evaluated at the maximum , where @xmath197 $ ] is therefore @xmath198 - c_{l}(\gamma_{0 } ) c_{l'}(\gamma_{0 } ) \right]\end{aligned}\ ] ] where we have defined the expectation value @xmath57 = \int d\xi \ c_{l}({\hat s } + \xi ) c_{l'}({\hat s } + \xi ) \ \xi [ n^{-1 } + c^{-1}(\gamma_{0 } ) ] \xi } } { \det |n^{-1 } + c^{-1}(\gamma_{0 } ) |}\ ] ] this can , in principle , be computed with the conjugate gradient descent method of transforming samples from a white noise process . future work will study the accuracy and convergence properties of estimating the curvature matrix from transformed white noise sampling . the correctness of the algorithm for power spectrum estimation presented in this paper is established by proving that the covariance matrix of two linearly transformed vectors has some specific form . for simplicity of notation , we choose to denote the covariance matrix of two vectors @xmath199 as the expectation value of the outer product of the vectors @xmath200 \equiv e[x \otimes y]_{ij}$ ] . for two matrices @xmath201 and @xmath149 , an identity used repeatedly in computing covariance matrices is @xmath202 = a e[x \otimes y ] b^{t}\ ] ] which is shown simply by checking for each matrix element @xmath203_{mn } & = & \sum_{ij } e[a_{mi } x_{i } b_{nj } y_{j } ] \nonumber \\ & = & \sum_{ij } a_{mi } e[x_{i } y_{j } ] b^{t}_{jn}\end{aligned}\ ] ] one example of this identity is in computing the expectation value of maps computed from time ordered data . one form of making a map from time ordered data is given by @xmath204 as mentioned in section 2 and in @xcite . according to the identity above , the covariance matrix of this map is @xmath205 & = & ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } e[d \otimes d ] ( r^{t } n^{-1 } ) ^{t } [ ( r^{t } n^{-1 } r)^{-1}]^{t } \end{aligned}\ ] ] substituting @xmath34 , this is @xmath205 & = & ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } e[(rs + \eta ) \otimes ( rs + \eta ) ] n^{-1 } r ( r^{t } n^{-1 } r)^{-1 } \nonumber \\ & = & ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } \left ( r e[s \otimes s ] r^{t } + r e[s \otimes \eta ] \right . \nonumber \\ & & \left . + e[\eta \otimes s ] r^{t } + e[\eta \otimes \eta ] \right ) n^{-1 } r ( r^{t } n^{-1 } r)^{-1 } \nonumber \\ & = & ( r^{t } n^{-1 } r)^{-1 } r^{t } n^{-1 } \left ( r c r^{t } + n \right ) n^{-1 } r ( r^{t } n^{-1 } r)^{-1 } \nonumber \\ & = & c + ( r^{t } n^{-1 } r)^{-1}\end{aligned}\ ] ] where we have used the independence of the signal and noise , and also that both are zero mean processes . this result is also proven in appendix b below using the characteristic function of the transformed time ordered data . the estimator given by the expectation maximization algorithm ( also equivalent to the maximum likelihood estimator ) is given by @xmath206 = \int ds \ [ s \otimes s ] \ \frac{e^{- ( y - rs ) n^{-1 } ( y -rs ) - sc^{-1 } s } } { \int ds ' \ e^{- ( y - rs ' ) n^{-1 } ( y -rs ' ) - s ' c^{-1 } s ' } } \ ] ] for partial sky coverage , this is equivalent to embedding the data on the full sky and marginalizing over it @xmath206 = \int d(s , y^{(2 ) } ) \ [ s \otimes s ] \ \frac{e^{- ( y^{(2 ) } - s ) { \tilde n}^{-1 } ( y^{(2 ) } - s ) } e^{- ( y - rs ) n^{-1 } ( y -rs ) - sc^{-1 } s } } { \int d(s',y ' ) \ e^{- ( y ' - s ' ) { \tilde n}^{-1 } ( y ' - s ' ) } e^{- ( y - rs ' ) n^{-1 } ( y -rs ' ) - s ' c^{-1 } s ' } } \ ] ] where we have arbitrarily chosen the full sky inverse noise matrix @xmath207\ ] ] the expectation of the covariance matrix is then @xmath208 = \int dy \ \int d(s , y^{(2 ) } ) \ [ s \otimes s ] \ \frac{e^{- ( y^{(2 ) } - s ) { \tilde n}^{-1 } ( y^{(2 ) } - s ) } e^{- ( y - rs ) n^{-1 } ( y -rs ) - sc^{-1 } s } } { \int d(s',y ' ) \ e^{- ( y ' - s ' ) { \tilde n}^{-1 } ( y ' - s ' ) } e^{- ( y - rs ' ) n^{-1 } ( y -rs ' ) - s ' c^{-1 } s ' } } \ ] ] consistency of the estimator is then shown by proving that @xmath209 $ ] . using the augmented noise matrix ( which now has an inverse on the full sky ) , we can define the mean field map @xmath210^{-1 } n^{-1 } ( y , y^{(2)})$ ] , and write @xmath211 & = & ( n^{-1 } + c^{-1})^{-1 } n^{-1 } e[y \otimes y ] n^{-1 } ( n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & c(n+c)^{-1 } ( n+c ) ( n+c)^{-1 } c \nonumber \\ & = & c(n+c)^{-1 } c\end{aligned}\ ] ] the expectation value of the correction is @xmath212 , so that @xmath211 & = & c ( n+c)^{-1 } c + ( n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & c(n+c)^{-1 } c + n(n + c)^{-1 } c \nonumber \\ & = & c\end{aligned}\ ] ] therefore , the expectation maximization algorithm converges to the maximum likelihood estimator for a uniform prior , which is also a consistent estimator for arbitrary sky coverage . as derived in section ? above , the algorithm to sample maps from the gaussian process with covariance matrix @xmath68 is * draw @xmath213 from a white noise process . * compute @xmath214 . * compute @xmath215 with conjugate gradient descent . we can prove that for any @xmath19 , even one which does not have an inverse ( i.e. as in the case of partial coverage of the chosen embedding region ) , we have @xmath81 = ( n^{-1 } + c^{-1})^{-1}$ ] . from the above we have @xmath216\ ] ] which gives @xmath217 & = & ( n^{-1 } + c^{-1})^{-1 } e [ ( c^{-1/2 } \omega ' + n^{-1/2 } \omega ) \otimes ( c^{-1/2 } \omega ' + n^{-1/2 } \omega ) ] ( n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & ( n^{-1 } + c^{-1})^{-1 } \left ( c^{-1/2 } e[\omega ' \otimes \omega ' ] c^{-1/2 } + c^{-1/2 } e[\omega ' \otimes \omega ] n^{-1/2 } \right . \nonumber \\ & & \left . + n^{-1/2 } e[\omega \otimes \omega ' ] c^{-1/2 } + n^{-1/2 } e[\omega \otimes \omega ] n^{-1/2 } \right ) ( n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & ( n^{-1 } + c^{-1})^{-1 } \left ( c^{-1 } + n^{-1 } \right ) ( n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & ( n^{-1 } + c^{-1})^{-1 } \end{aligned}\ ] ] where by independence of the two white noise maps , @xmath218 = e[\omega ] \otimes e[\omega']$ ] which vanishes since the white noise process is zero mean . an important point to notice is that the matrices @xmath19 or @xmath82 can be singular in the sense that they do not have inverses on the full sky ( i.e. are generated by incomplete scanning of the sky ) . in fact @xmath219 vanishes in the null space of @xmath19 ( where we do not have data ) . for a realization of a white noise process in the time domain @xmath83 and a white noise map in the spatial domain @xmath84 , we compute a fluctuation map according to @xmath220\ ] ] where @xmath82 is known in the fourier basis associated with the time domain . we can prove that the maps @xmath69 have the correct covariance matrix , @xmath81 = ( r^{t } n^{-1 } r + c^{-1})^{-1}$ ] by the direct calculation @xmath217 & = & ( r^{t } n^{-1 } r + c^{-1})^{-1 } e [ ( c^{-1/2 } \omega + r^{t } n^{-1/2 } \tau ) \otimes ( c^{-1/2 } \omega + r^{t } n^{-1/2 } \tau ) ] ( r^{t } n^{-1 } r + c^{-1})^{-1 } \nonumber \\ & = & ( r^{t } n^{-1 } r + c^{-1})^{-1 } \left ( c^{-1/2 } e[\omega \otimes \omega ] c^{-1/2 } + + r^{t } n^{-1/2 } e[\tau \otimes \tau ] n^{-1/2 } r \right ) ( r^{t } n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & ( r^{t } n^{-1 } r + c^{-1})^{-1 } \left ( c^{-1 } + r^{t } n^{-1 } r \right ) ( r^{t } n^{-1 } + c^{-1})^{-1 } \nonumber \\ & = & ( r^{t } n^{-1 } r + c^{-1})^{-1 } \end{aligned}\ ] ] where again , the cross terms vanish since independence gives @xmath221 = e[\omega ] \otimes e[\tau]$ ] , which vanishes since both are zero mean processes . we can also find the likelihood for the data embedded in an azimuthally symmetric region covered by a ring set . as discussed in @xcite , we represent the signal on the ring set with coefficients @xmath222 , so that @xmath223 where both indices range from @xmath224 . therefore , the signal on the sky , parameterized on the ring set , is given by a two - d inverse fft . for any specific power spectrum , the corresponding signal matrix on the ring set covering the embedding region is block diagonal . we denote the ring set covariance matrix @xmath225 $ ] . given the noise free signal on the ring set @xmath226 , any observed data set is given by projecting into the sub - region where we have data , @xmath227 where @xmath228 vanishes on the regions of the ring set where we have no observations . as before we would first compute the mean field and fluctuation maps . in order to sample fluctuation maps , we would again compute , @xmath229 where @xmath83 is a time domain white noise process , @xmath84 is a spatial white noise process on the full sky , @xmath230 is the cholesky decomposition of the ring set covariance matrix . although it is possible to compute the cholesky decomposition of the ring set covariance matrix , we might instead sample @xmath231 on the full sky , and then project into the region covered by the ring set , giving maps @xmath232 . then the covariance matrix for these maps is , using the usual identity @xmath233 & = & r c^{1/2 } e[\omega \otimes \omega ] c^{1/2 } r^{t } \nonumber \\ & = & r c r^{t}\end{aligned}\ ] ] which is the correct covariance matrix on the ring set region . the point is that we do not have to compute the cholesky decomposition of the signal matrix on the ring set , but can instead transform white noise maps with the choelsky decomposition of the full sky signal covariance matrix ( diagonal in the spherical harmonic basis ) , and then project down to the ring set region . we can construct a preconditioner as follows . define the projection from the time domain to the ring set following a wandelt scan strategy as @xmath234 . the for the same time domain noise matrix @xmath19 , we can use a preconditioner given by @xmath234 and define @xmath235^{-1}\ ] ] it is shown in @xcite that @xmath234 is block diagonal on the ring set , so that @xmath236 can be computed in @xmath5 operations . we can then solve the linear equations for the mean field and fluctuation maps @xmath237 { \hat s } & = & m^{-1}t q^{t } n^{-1 } d \nonumber \\ m^{-1 } [ i + t q^{t } n^{-1 } q ] \xi & = & m^{-1 } \left ( t^{+1/2 } \omega + t q^{t } n^{-1/2 } \tau \right)\end{aligned}\ ] ] for scans which are close to the ring set scan the intuition is that conjugate gradient descent will converge quickly for the above equations . stompor , r. , et al . , astro - ph/0106451 . seljack , u. and zaldarriaga , m. , apj , 469 , 437 , 1996 . gorski , e.k . , hivon , e. , and wandelt , b. , in proceedings of the mpa / eso garching conference 1998 , eds . banday , a.j . , sheth , k. , and l. da costa . see http://www.eso.org kgorski / healpix/ borrill , j. , phys . d59 , 027302 , 1999 . bond , j.r . et al , comput.sci.eng . 1 , 21 , 1999 . dempster , a.p . , laird , n.m . , and rubin , d.b . , j. roy . stat b , vol . 39 , no . 1 , pp . 1 - 38 , 1977 . tegmark , m. , phys . d55 , 5895 - 5907 , 1997 . , spergel , d.n . , and hinshaw , g. , apj,510 , 551 , 1999 . wandelt , b. and hansen , f. , astro - ph/0106515 , 2001 . wandelt , b. and gorski , k. , phys . rev d , 63,123002 , 2001 . bond , j.r . , jaffe , a.h . , knox , l. , phys . d , 57 , 2117 , 1998 . christenson , r. , et al . grav . , 18,2677 , 2001 . knox , l. , et al . , apj , 563 , l95 , 2001 . lewis , a. , and bridle , s. , astro - ph/0205436 . runbino - martin , et al . , astro - ph/0205367 . * _ figure 1 _ - the top left image shows the variance of the noise in each pixel . the black region is an unobserved `` hole '' . the upper right plot is a noise free realization , and the lower left is a mean field map given the simulated noisy data . the bottom right shows a typical fluctuation map computed with conjugate gradient descent . * _ figure 2 _ - plot showing the power spectrum estimation after @xmath238 iterations starting from a flat initial guess . the green dots represent the results of separate runs on each of 10 simulated data sets , with each shade of green representing a different run . for each data set we have produced a new noise - free map ( drawn from the theoretical power spectrum given by the solid black line ) , and added inhomogenous noise and a hole as shown in figure 1 . the initial guess used was the expected ( flat ) noise spectrum , multiplied by a random number between 0 and 1 . at low @xmath239 the snr is high , and the spread in the dots is caused by noise and sample variance . at high @xmath239 , the snr is low and the spread in the dots is caused by variation in the initial guess , resulting in an upper bound as shown in the plot . also shown are sample power spectra for a single simulated data set ( light blue line ) , and noise - free map ( dark blue line ) .

we consider cosmology in the einstein - ther theory ( the generally covariant theory of gravitation coupled to a dynamical timelike lorentz - violating vector field ) with a linear - lagrangian . the spacetime splitting approach is used to derive covariant and gauge invariant perturbation equations which are valid for a general class of lagrangians . restricting attention to the parameter space of these theories which is consistent with local gravity experiments , we show that there are tracking behaviors for the field , both in the background cosmology and at linear perturbation level . the primordial power - spectrum of scalar perturbations in this model is shown to be the same that predicted by standard general relativity . however , the power - spectrum of tensor perturbation is different from that in general relativity , but has a smaller amplitude and so can not be detected at present . we also study the implications for late - time cosmology and find that the evolution of photon and neutrino anisotropic stresses can source the field perturbation during the radiation and matter dominated epochs , and as a result the cmb and matter power spectra are modified . however these effects are degenerate with respect to other cosmological parameters , such as neutrino masses and the bias parameter in the observed galaxy spectrum .

0709.4581

for more than two decades milgrom s modified newtonian dynamics ( mond ) mond has been able to explain galaxy rotation curves which are conventionally considered as an evidence of cold dark matter ( cdm ) on galactic scales . mond modifies newton s second law of motion to @xmath1 where @xmath2 and @xmath3 are the acceleration and newtonian gravitational potential , respectively ; @xmath4 is an effectively free function tending to unity in the limit @xmath5 , with @xmath6 being a new fundamental constant , which must have a numerical value of @xmath7 in order to match observations on a galactic scale . this theory looks like newton s when accelerations are large but is significantly different when accelerations are small . on galactic scales , @xmath8 , so the newtonian dynamics is modified , but in a way that can fit spiral galaxy rotation curves . subsequently , bekenstein @xcite built a relativistic theory which has mond as a non - relativistic , weak - field limit , thus making the study of cosmology possible . in addition to the conventional tensor gravitational field , bekenstein s theory involves a vector and a scalar field , and is therefore dubbed teves . interestingly , it has been argued that teves could also explain the large - scale structure formation of the universe without recurring to cdm @xcite , thanks to the presence of the vector field @xcite . recently , the authors of ref . @xcite showed that teves is equivalent to a vector - tensor theory of gravitation where the vector field has a non - fixed norm . they also showed that the correct mondian limit could be realized with a single vector field having non - canonical kinetic terms @xcite . these results indicate that a vector field in the gravity sector might be an interesting component of the universe ( indeed the model in @xcite and its generalized version we shall presented below in eq . ( [ eq : ouraction ] ) could be used to explain dark energy and dark matter in background cosmology ) and merits more detailed investigations . the idea of a vector field coupled to gravity has a long worldline ( see for example @xcite for a review and @xcite for further references ) , but in this work we will focus on the model described in @xcite , which is the most well - studied one , and investigate its cosmological implications . this particular theory is based on a dynamical vector field coupled to gravitation that picks up a preferred frame and preserves general covariance . this vector field is unit - norm , timelike , and violates local lorentz invariance . it is called the ther field ( or simply - field ) and we will refer to the associated einstein - ther theory as - theory defined by the - lagrangian . the - lagrangian considered in @xcite is a special case of our general model introduced in eq . ( [ eq : ouraction ] ) below for a unit - norm vector field that includes terms up to second order in derivatives , and it has been extensively studied in various contexts @xcite . in refs . @xcite , the background cosmology and primordial power spectra of perturbations from inflation of a slight different model were also considered . here , we investigate these for the model presented in ref . @xcite , and also study the evolution of linear perturbation to the - field during the radiation and matter - dominated epochs . as we will show below , if we restrict the parameter space of the underlying theory so as to satisfy the local experimental gravity constraints , @xcite , this perturbation becomes sourceless and decays during the epoch of inflation and late - matter domination . however , it is sourced by the evolutions of the photon and neutrino anisotropic stresses during the radiation era and early matter era , which have some imprints on the cosmological observables . our presentation is organized as follows . in sec . [ sect : equations ] , we briefly introduce the general - theory and derive perturbation equations for the background friedmann - like cosmology in the covariant and gauge invariant ( cgi ) formalism ( see @xcite for a derivation of the perturbation equations in conformal newtonian gauge ) . in sec . [ sect : cosmology ] we shall use these equations to discuss the perturbation dynamics for the cosmological models of ref . @xcite . first , we summarize the existing constraints on the model in sec . [ sect : parameterspace ] ; then , in sec . [ sect : perturbationevolution ] , we present the evolution equations for the perturbation variables and then use them to show how the primordial spectra of scalar and tensor perturbations in this theory are unmodified and modified , respectively , on comparing them with the predictions of general relativity ( gr ) . the late - time evolution of the - field perturbation and its effects on cosmological observables are also studied there . finally , our discussion and conclusions are presented in sec . [ sect : conclusion ] . throughout this work our convention is @xmath9u^{c } = r^{\ \ c}_{ab\ d}u^{d } , r_{ab } = r_{acb}^{\ \ \ c}$ ] , where @xmath10 are respectively the riemann tensor and ricci tensor ; the metric signature is @xmath11 and the universe is assumed to be spatially flat , filled with photons , baryons , cdm , 3 species of neutrinos and a cosmological constant . in this section we briefly introduce the general form of the -theory and derive the cgi perturbation equations which we will use to analyze the cosmological effects of the -field . the equations presented here are for general lagrangians . later , we shall focus on a specific class of such theory characterized by a linear lagrangian ( with @xmath12 , see below ) . the model we consider here is a slight generalization of that presented in @xcite . it is characterized by a general gravitational action of the form @xmath13\end{aligned}\ ] ] in which @xmath14 is the _ bare _ newtonian gravitation constant , @xmath15 is a lagrange multiplier ensuring that -field has a unit norm , and @xmath16 is the -lagrangian expressed by @xmath17 with @xmath18 and @xmath19 being an arbitrary analytic function of its arguments . as long as the norm of @xmath20 is fixed ( @xmath21 ) , the form of @xmath22 is the most general possibility we can have for our vector - field lagrangian . notice that eq . ( [ eq : kabcd ] ) here differs from that given in @xcite by the @xmath23 term and we shall refer to our model and that of @xcite as the gea ( generalized einstein ther ) model to distinguish from the one considered in @xcite ( for which we instead called ea ( einstein ther ) ) . the matter lagrangian is taken to be the same as in standard @xmath24 model . we treat the -field @xmath25 and inverse metric @xmath26 as the dynamical degrees of freedom and vary the action with respect to them to obtain the field equations . the former gives the -field equation of motion ( eom ) : @xmath27 where we have defined @xmath28 , @xmath29 and @xmath30 . the variation with respect to the metric leads to a modified einstein equation . one could retain the form of einstein equations in standard gr by treating the vector field as a new contribution ( denoted by @xmath31 ) to the total energy - momentum tensor in the universe , in addition to that of the conventional fluid matter which is denoted by @xmath32 . then , according to the definition @xmath33 in which @xmath34 , we have @xmath35\right\ } + \frac{1}{2}g_{ab}f + \left[\text{\ae}_{d}\nabla_{c}(fj^{cd } ) - c_{4}f(\text{\ae}^{d}\nabla_{d}\text{\ae}_{c})(\text{\ae}^{e}\nabla_{e}\text{\ae}^{c})\right]\text{\ae}_{a}\text{\ae}_{b}\nonumber\\ & & - f\left[c_{1}\nabla_{a}\text{\ae}^{c}\nabla_{b}\text{\ae}_{c } - c_{1}\nabla^{c}\text{\ae}_{a}\nabla_{c}\text{\ae}_{b } - c_{4}(\text{\ae}^{c}\nabla_{c}\text{\ae}_{a})(\text{\ae}^{d}\nabla_{d}\text{\ae}_{b})\right].\end{aligned}\ ] ] we also note that by varying the action with respect to the lagrangian multiplier @xmath36 we simply get the normalization relation of the -field , @xmath37 , as mentioned above . the cgi perturbation equations in general -theories are derived in this section using the method of @xmath0 decomposition @xcite ( see mg3 + 1 for applications of this method in modified - gravity models ) . first , we briefly review the main ingredients of @xmath0 decomposition and their application to standard general relativity @xcite for ease of later reference . the main idea of @xmath0 decomposition is to make spacetime splits of physical quantities with respect to the 4-velocity @xmath38 of an observer . the projection tensor @xmath39 is defined as @xmath40 and can be used to obtain covariant tensors perpendicular to @xmath41 . for example , the covariant spatial derivative @xmath42 of a tensor field @xmath43 is defined as @xmath44the energy - momentum tensor and covariant derivative of the 4-velocity are decomposed respectively as @xmath45 in the above , @xmath46 is the projected symmetric trace - free ( pstf ) anisotropic stress , @xmath47 the heat flux vector , @xmath48 the isotropic pressure , @xmath49 the pstf shear tensor , @xmath50}$ ] the vorticity , @xmath51 ( @xmath52 is the mean expansion scale factor ) the expansion scalar , and @xmath53 the acceleration ; the overdot denotes time derivative expressed as @xmath54 , brackets mean antisymmetrisation , and parentheses symmetrization . the 4-velocity normalization is chosen to be @xmath55 . the quantities @xmath56 are referred to as _ dynamical _ quantities and @xmath57 as _ kinematical _ quantities . note that the dynamical quantities can be obtained from the energy - momentum tensor @xmath58 through the relations @xmath59 decomposing the riemann tensor and making use the einstein equations , we obtain , after linearization , five constraint equations @xcite : @xmath60 \varepsilon _ { b)ec}^{\ \ \ \ d}u^{e } ; \\ \label{eq : constraintphi } \hat{\nabla}^{b}\mathcal{e}_{ab } & = & \frac{1}{2}\kappa \left[\hat{\nabla}^{b}\pi_{ab}+\frac{2}{3}\theta q_{a}+\frac{2}{3}\hat{\nabla}_{a}\rho \right];\\ \hat{\nabla}^{b}\mathcal{b}_{ab } & = & \frac{1}{2}\kappa \left[\hat{\nabla}_{c}q_{d}+(\rho + p)\varpi _ { cd}\right ] \varepsilon _ { ab}^{\ \ cd}u^{b},\end{aligned}\ ] ] and five propagation equations , @xmath61 } & = & 0 ; \\ \label{eq : propagationphi } \frac{1}{2}\kappa \left[\dot{\pi}_{ab } + \frac{1}{3}\theta\pi_{ab}\right ] - \frac{1}{2}\kappa \left[(\rho + p)\sigma_{ab}+\hat{\nabla}_{\langle a}q_{b\rangle}\right ] \nonumber \\ -\left [ \dot{\mathcal{e}}_{ab}+\theta \mathcal{e}_{ab}-\hat{\nabla}^{c}% \mathcal{b}_{d(a}\varepsilon _ { b)ec}^{\ \ \ \ d}u^{e}\right ] & = & 0;\ \ \ \ \\ _ { d(a}\varepsilon _ { b)ec}^{\ \ \ \ d}u^{e } \nonumber \\ + \frac{\kappa } { 2}\hat{\nabla}^{c}\mathcal{\pi } _ { d(a}\varepsilon _ { b)ec}^{\ \ \ \ d}u^{e } & = & 0.\end{aligned}\ ] ] here , @xmath62 is the covariant permutation tensor , @xmath63 and @xmath64 are respectively the electric and magnetic parts of the weyl tensor @xmath65 , defined by @xmath66 and @xmath67 . the angle bracket means taking the trace - free part of a quantity . besides the above equations , it is useful to express the projected ricci scalar @xmath68 into the hypersurfaces orthogonal to @xmath38 as @xmath69the spatial derivative of the projected ricci scalar , @xmath70 , is then given as @xmath71and its propagation equation by @xmath72 finally , there are the conservation equations for the energy - momentum tensor : @xmath73 as we are considering a spatially - flat universe , the spatial curvature must vanish on large scales and so @xmath74 . thus , from eq . ( [ eq : spatialriccicurvature ] ) , we obtain @xmath75 this is the friedmann equation in standard general relativity , and the other background equations ( the raychaudhuri equation and the energy - conservation equation ) can be obtained by taking the zero - order parts of eqs . ( [ eq : raychaudhrui ] , [ eq : energyconservation ] ) , yielding : @xmath76 in what follows , we will only consider scalar perturbation modes , for which the vorticity @xmath77 and magnetic part of weyl tensor @xmath78 are at most of second order @xcite , and so will be neglected in our first - order analysis . in the einstein - ther theories where we consider the -field as a new species of matter , the gravitational field equations eqs . ( eq : constraintvarpi - [ eq : backgroundenergyconservation ] ) listed above preserve their forms , but the dynamical quantities @xmath79 appearing there should be replaced by the effective total quantities of the same type . for simplicity , we shall _ always _ use variables _ without _ superscripts to denote these _ effective total _ quantities , while for those of a specified matter species we shall add corresponding superscripts ( _ e.g. _ , @xmath80 denotes the energy density of the -field @xmath81 ) . the vector -field , @xmath20 , requires further discussion . as we mentioned above , it has the normalization relation @xmath82 . in the background friedmann - robertson - walker ( frw ) universe the requirements of homogeneity and isotropy require that @xmath20 is just equal to @xmath83 , which is unambiguously chosen as the 4-velocity of the fundamental observers . but in a perturbed , almost - frw , universe this is no longer true and we can write @xmath84 where @xmath85 is another ( first - order ) vector field that vanishes in a frw universe : we call it the perturbation of the -field . furthermore , the relation @xmath86 implies that @xmath87 , _ i.e. _ , @xmath85 is a _ spatial _ vector field which is perpendicular to @xmath83 , up to first order in perturbation . this fact is used extensively in deriving the perturbation equations ( _ e.g. _ , @xmath88 _ etc_. ) . with these preliminaries at hand , and after some lengthy manipulations , the -field eom eq . ( [ eq : aeeom ] ) can be written as ( up to first order ) @xmath89 + c_{14 } \left[f\dot{a}_{a}+\dot{f}a_{a}+\frac{2}{3}f\theta a_{a}\right ] - \left[\frac{1}{3}(\alpha - c_{14})(\dot{f}\theta + f\dot{\theta } ) - \frac{2}{9}c_{14}f\theta^{2}\right]\text{\ae}_{a}\nonumber\\ + \frac{1}{3}\alpha\hat{\nabla}_{a}(f\theta ) + \frac{1}{3}\alpha f\hat{\nabla}_{a}\hat{\nabla}^{b}\text{\ae}_{b } + c_{13}f\hat{\nabla}^{b}\left(\sigma_{ab}+\hat{\nabla}_{\langle a}\text{\ae}_{b\rangle}\right ) & = & 0,\end{aligned}\ ] ] and from the definitions eq . ( [ eq : defdynamicalquantity ] ) the -field energy density , isotropic pressure , heat - flux vector and anisotropic stress can be identified from eq . ( [ eq : aeemt ] ) ( again up to first order ) as @xmath90,\\ \label{eq : aeq } \kappa q^{\text{\ae}}_{a } & = & -c_{14 } \left[f\dot{a}_{a}+\dot{f}a_{a}+\frac{2}{3}f\theta a_{a}\right ] -c_{14}\left[f\ddot{\text{\ae}}_{a}+\left(\dot{f } + f\theta\right)\dot{\text{\ae}}_{a}\right ] + \left[\frac{1}{3}(\alpha - c_{14})(\dot{f}\theta + f\dot{\theta } ) - \frac{2}{9}c_{14}f\theta^{2}\right]\text{\ae}_{a}\\ \label{eq : aeq2 } & = & \frac{1}{3}\alpha\hat{\nabla}_{a}(f\theta ) + \frac{1}{3}\alpha f\hat{\nabla}_{a}\hat{\nabla}^{b}\text{\ae}_{b } + c_{13}f\hat{\nabla}^{b}\left(\sigma_{ab}+\hat{\nabla}_{\langle a}\text{\ae}_{b\rangle}\right),\\ \label{eq : aepi } \kappa\pi^{\text{\ae}}_{ab } & = & -c_{13}(\dot{f}+f\theta)\left[\sigma_{ab } + \hat{\nabla}_{\langle a}\text{\ae}_{b\rangle}\right ] - c_{13}f\left[\dot{\sigma}_{ab } + ( \hat{\nabla}_{\langle a}\text{\ae}_{b\rangle})^{\cdot}\right],\end{aligned}\ ] ] where , in eq . ( [ eq : aeq2 ] ) , we have used eq . ( [ eq : aeeom ] ) . here , we have defined the new parameters @xmath91 , @xmath92 and @xmath93 ; @xmath94 is given by @xmath95 . including these -contributions to eqs . ( [ eq : constraintvarpi ] - [ eq : backgroundenergyconservation ] ) one obtains the modified gravitational field equations for the general -theory . it is also easy to check that the above results satisfy ( separately ) the conservation of the -field s energy - momentum tensor , eqs . ( [ eq : energyconservation ] , [ eq : heatfluxevolution ] ) . note that eqs . ( [ eq : aerho ] - [ eq : aepi ] ) are the general expressions of energy density , pressure , heat flux and anisotropic stress in the @xmath0 decomposition which include both zeroth order ( background ) and first order terms ; to calculate the actually density contrast _ etc . _ ( see eqs . ( [ eq : aerho_2 ] - [ eq : aepi_2 ] ) below ) one needs to take the covariant spatial derivatives of these equations @xcite . in above we presented the field equations for general -theories , but in what follows we shall only analyze the cosmology of a specific edition of the theory which is defined by choice of a linear lagrangian : @xmath96 ( note that the minus sign is because of our sign convention ) . this model is by far the most well known , in the sense that it has been investigated in the contexts of static weak - field limit @xcite ( see also @xcite for the weak - field limit of the - model considered in @xcite ) , background cosmology @xcite , the radiation and propagation of the -gravitational waves @xcite , compact stars @xcite , and black holes @xcite . some of these studies have imposed stringent constraints on the viable parameter - space of @xmath97s . in view of these restrictions we will confine our study to this constrained subset of possible theories . note that the perturbation dynamics of the -model has also been analyzed in aetherperturbation in the absence of the @xmath23 term . here , we shall include this term and perform a similar analysis but in slightly different manner and in more detail ; we will also discuss on some detailed features which lead to our conclusions being different . the late - time evolution of the -perturbation is also investigated . in particular , we shall find that within the locally - constrained parameter space the -model will leave slightly different signatures on the perturbation evolutions from those left by the standard @xmath24 paradigm in general relativity , and so cosmological data on cosmic microwave background ( cmb ) and matter power spectra might place some constraint on the parameter space . in this subsection we briefly summarize the constraints on the -model described in eq . ( [ eq : aelagrangian ] ) . it has been well known that in the weak - field , slow - motion , limit and in the background cosmology the -model displays tracking behavior . for the former environment , it can be shown that in the presence of the -field the observed _ effective _ gravitational constant @xmath98 is a rescaling of the bare one @xmath99 @xcite by ( note that here our @xmath97s have different signs from those in @xcite ) @xmath100 and in the latter environment the observed gravitational constant @xmath101 is also a rescaling of @xmath102 but with a different factor @xcite @xmath103 correspondingly , we define @xmath104 and @xmath105 to be used below . note that the rescaled @xmath106 is generally not equal to @xmath14 and as such the background cosmic expansion rate will be different from that in standard gr . however , we note that the numerical value of gravitational constant we find in the textbooks and use in the numerical calculations is _ not _ @xmath14 but rather the locally - measured @xmath98 . it is possible to obtain limits on non - local values of @xmath107 by considering the primordial nucleosynthesis of light elements ( see for example ref . if @xmath108 then we have @xmath109 , which indicates that the background cosmological dynamics will be exactly the same as _ assuming _ standard gr @xcite ; otherwise we can use constraints from primordial nucleosynthesis on the value of gravitational constant to show that @xmath110 @xcite . however , note that the particle - horizon size at the epoch of neutron - proton freeze - out ( @xmath111 ) , which is most sensitive to variations in the value of @xmath107 , is only @xmath112 and this causally linked region expands in size by about a factor of @xmath113 by the present - day to a size @xmath114 which is a sub - galactic scale but there has then been local gravitational collapse by a factor of @xmath115 . such collapse may however also affect the local value of the gravitational constant @xcite , but we will not consider this in the present work . there are also constraints from the observations of parameterized post - newtonian ( ppn ) parameters @xcite . it is shown in @xcite that for all the ppn parameters to coincide with those in gr ( otherwise the parameters may need to be fine - tuned ) one reduces the full four - parameter space of the model to a two - parameter subspace characterized by @xmath116 in addition , -theories contain five gravitational and -wave modes , which include the two usual spin-2 gravitational waves , and three additional modes : two spin-1 transverse -gravity waves and one spin-0 longitudinal -gravity wave . the speeds of the three additional modes are generally not equal to @xmath117 . it has been shown in ref . @xcite that if these speeds are less than 1 then the high - energy particles will produce @xmath118erenkov radiation when passing through vacuum , which imposes stringent constraints on the model . however , as suggested in @xcite , these constraints do not apply if these modes propagate superluminally . the requirement that the additional -gravity waves do not propagate subluminally further limits the parameter space to @xmath119 finally , when the above constraints eq . ( [ eq : parameterspace2 ] ) are satisfied , the positive energy requirement and the stability of additional wave modes @xcite also hold . in addition , we will have @xmath120 so that the big bang nucleosynthesis constraint does not apply . thus , we can see that , even after using all the current constraints , there is still a large parameter space remaining for the model . in what follows we shall ask whether linear - order cosmological observations such as the cmb and the form of the matter power spectrum could reduce this parameter space further , and as we will show , the answer is positive . however , the modifications are small and depend weakly on the model parameters , which mean that the data on cmb and matter power spectra can not give very stringent limits on the parameter space . in this subsection we consider the perturbation evolutions of our -model . for generality , we will derive the equations for arbitrary choices of parameters and only later confine ourselves to the parameter space described in eq . ( [ eq : parameterspace2 ] ) . besides , since the presence of the -field in general will modify the cosmology at all times , we will also investigate its effects during the inflationary era as in @xcite ; after that we will turn to its effects on late time cosmology . following @xcite , we shall make the following harmonic expansions of our perturbation variables @xmath121 in which @xmath122 is the eigenfunction of the comoving spatial laplacian @xmath123 satisfying @xmath124 and @xmath125 are given by @xmath126 . note that @xmath127 is dimensionless . in terms of the above harmonic expansion coefficients , eqs . ( eq : constraintq , [ eq : constraintphi ] , [ eq : propagationsigma ] , eq : propagationphi , [ eq : constrainteta ] , [ eq : propagationeta ] ) can be rewritten as @xcite @xmath128,\\ \label{eq : propagationsigma2 } k(\sigma ' + \mathcal{h}\sigma ) & = & k^{2}(\phi+a ) -\frac{1}{2}\kappa a^{2}\pi,\\ \label{eq : propagationphi2 } k^{2}(\phi'+\mathcal{h}\phi ) & = & \frac{1}{2}\kappa a^{2}\left[k(\rho+p)\sigma+kq-\pi ' -\mathcal{h}\pi\right],\ \ \ \ \\ \label{eq : constrainteta2 } k^{2}\eta & = & \kappa\mathcal{x}a^{2 } - 2k\mathcal{h}\mathcal{z},\\ \label{eq : propagationeta2 } k\eta ' & = & -\kappa qa^{2 } - 2k\mathcal{h}a\end{aligned}\ ] ] where @xmath129 and a prime denotes the derivative with respect to the conformal time @xmath130 ( @xmath131 ) . also , eq . ( [ eq : heatfluxevolution ] ) and the spatial derivative of eq . ( [ eq : energyconservation ] ) become @xmath132 recall that we shall always neglect the superscript @xmath133 for the total dynamical quantities and add appropriate superscripts for individual matter species . note that @xmath134 and that the @xmath94 appearing above is the bare ( not necessarily the measured ) one . furthermore , for convenience we define the frame - independent ( fi ) variables @xcite @xmath135 @xmath136 is fi according to eq . ( [ eq : aepi ] ) because we know that the anisotropic pressure tensor @xmath137 is frame invariant . hence , it follows from eq . ( [ eq : constraintphi ] ) , that @xmath138 is also fi up to first order in perturbation . in the zero - shear frame ( the newtonian gauge ) , we have simply @xmath139 and @xmath140 . before presenting the evolution equations , we first write the dynamical quantities of the -field in terms of the harmonic coefficients introduced above : @xmath141,\\ \label{eq : aep_2 } \kappa\mathcal{x}^{p , \text{\ae } } a^{2 } & = & - \frac{\alpha}{3}k\left[(\mathcal{z}+ \text{\ae})^{\prime } + 2\mathcal{h}(\mathcal{z}+\text{\ae } ) \right]\nonumber\\ & & - \alpha(\mathcal{h}^{\prime } -\mathcal{h}^{2})a , \\ \label{eq : aeq_2 } \kappa q^{\text{\ae } } a^{2 } & = & -\frac{1}{3}\alpha k^{2}(\mathcal{z}+\text{\ae } ) - \frac{2}{3}c_{13}k^{2}\tilde{\text{\ae}},\\ \label{eq : aepi_2 } \kappa\pi^{\text{\ae } } a^{2 } & = & c_{13}k(\tilde{\text{\ae}}^{\prime } + 2\mathcal{h}\tilde{\text{\ae } } ) .\end{aligned}\ ] ] in these expressions we have used both @xmath127 and @xmath142 because not all these quantities are fi . however it can be shown that the two quantities @xmath143 \nonumber \\ & & - \left(\frac{1}{3 } + \frac{\xi^{\prime } } { 3\mathcal{h}\xi}\right)c_{14 } \left(k\tilde{\text{\ae } } ^{\prime } + k\mathcal{h}\tilde{\text{\ae } } - k^{2}\phi\right ) \nonumber \\ & & - \frac{1}{3}\alpha k\tilde{\text{\ae } } ^{\prime } - \frac{1}{3}\alpha k\mathcal{h% } \tilde{\text{\ae } } + \frac{\xi^{\prime } } { 3\mathcal{h}\xi}\alpha k\mathcal{h}% \tilde{\text{\ae } } \end{aligned}\ ] ] and @xmath144 \nonumber \\ & = & \frac{1}{2}\alpha\kappa a^{2}\tilde{q } - \frac{1}{3}\alpha k^{2}\tilde{% \text{\ae } } - \frac{2}{3}c_{13}k^{2}\tilde{\text{\ae } } , \end{aligned}\ ] ] which will be used in the derivations , are fi , as they are expressed in terms of fi variables only . note that in the above we have defined @xmath145 for convenience , where @xmath146 is the newtonian gravitational potential , and @xmath147 we now investigate in detail eqs . ( [ eq : fiadiabatic ] , [ eq : fiaetherq ] ) . on large scales , where @xmath148 , the terms involving @xmath149 ( these include the @xmath150 term in eq . ( [ eq : fiadiabatic ] ) ) can be safely disregarded , and as a result we have @xmath151 where the superscript @xmath152 means the fluid matter . we see that on large scales these attributes of the -field track those for other matter species in the universe . as the combination @xmath153 determines the type of perturbations ( for example , the perturbation is adiabatic if the combination is equal to zero ) , this indicates that the -field will not alter the type of the scalar perturbation produced in the inflationary era . note that the above tracking behaviors are the same as that in the background cosmology , _ i.e. _ , rescaling the gravitational constant by a same factor . now we can proceed to derive the evolution equations for our -model . the propagation equation for the -field eq . ( [ eq : aeeom ] ) , in terms of the fi variables , becomes @xmath154 in which we have changed the variable to @xmath155 for simplicity . taking the time derivative of eq . ( [ eq : propagationphi2 ] ) , adding to it @xmath156 times the same equation , and using eq . ( [ eq : heatfluxevolution2 ] ) to eliminate the @xmath157 term , we obtain the following equation second - order differential equation for @xmath158 : @xmath159.\end{aligned}\ ] ] eqs . ( [ eq : propagationepsilon ] , [ eq : propagationphi ] ) are the evolution equations for the coupled @xmath160 system that we are looking for ( note that @xmath161 where @xmath162 is the fluid matter anisotropic stress and @xmath163 can also be expressed in terms of @xmath164 and its time derivatives by virtue of eq . ( [ eq : aepi_2 ] ) ) . they are not closed if @xmath165 and @xmath162 are unknown and to know these quantitie we would need to know the matter content of the universe . in the analysis above , we have mentioned that the presence of the -field does not affect the form of the scalar perturbation produced during inflation . but we also need to know whether other features of the inflationary power spectrum , such as the spectral index and the amplitude , are modified by the -field , as compared with the predictions in standard gr . here , we will investigated this issue ( see @xcite for a study in the absence of the @xmath23 term in the -lagrangian ) by considering a single - field model of inflation @xcite in the presence of the -field . during the inflationary epoch a scalar inflaton field @xmath166 slowly rolls along its potential and has an almost constant energy density which drives an almost exponential expansion of the universal scale factor . the comoving hubble length ( the horizon ) decreases with time so that in this process the quantum vacuum fluctuations of the inflaton field @xmath166 on the scales of interest to us leave the horizon ( their scales become larger than the horizon ) . the curvature perturbations they generate remain constant during their subsequent super - horizon evolution , until these scales eventually reenter the horizon long after the inflationary period has ended . during the radiation - dominated era when these modes stay outside the horizon , the metric perturbation @xmath146 becomes a constant , which drives the density perturbations of different matter species , and leads to the observed cmb and matter power spectra after horizon re - entry . there are no couplings between the scalar field and gravitational or -fields , so we can write down its dynamical quantities for the scalar field @xmath166 as @xmath167 making the following harmonic expansion for @xmath168 @xmath169 it is easy to get @xmath170 where @xmath171 . then parallel to eqs . ( [ eq : fiadiabatic ] , [ eq : fiaetherq ] ) we have , following the standard procedure , @xmath172\ \ \ \\end{aligned}\ ] ] and @xmath173 \nonumber \\ & = & \kappa k\varphi^{\prime } \tilde{\chi}\end{aligned}\ ] ] where we have defined the fi variable @xmath174 substituting eqs . ( [ eq : fiadiabatic ] , [ eq : fivarphiq ] ) into eq . ( eq : propagationphi ) , and using eq . ( [ eq : constraintphi2 ] ) to eliminate the term proportional to @xmath175 , we arrive at the following equation @xmath176\frac{k}{a}\epsilon ' + \frac{1}{1-\frac{1}{2}\alpha } ( c_{1}+c_{2}+c_{3})\frac{\xi'}{2\xi}\frac{k}{a}\epsilon - \frac{1}{k}c_{13}\left(\frac{\epsilon'''}{a } - \frac{\xi'}{\xi}\frac{\epsilon''}{a } - \xi\frac{\epsilon'}{a}\right).\end{aligned}\ ] ] eqs . ( [ eq : propagationepsilon ] , [ eq : propagationphi2 ] ) that form a closed set of evolution equations for the coupled -inflaton system ; they are a generalization of the results presented in aetherperturbation , and from them we can perform our analysis of the observational effects . let us look first at the large - scale evolution of @xmath158 . in this limit @xmath177 so the last term on the left - hand side , and the first two terms on the right hand side , of eq . ( [ eq : propagationphi2 ] ) can be dropped . notice that during the inflationary era @xmath178 and eq . ( eq : propagationepsilon ) become @xmath179 substituting this into eq . ( [ eq : propagationphi2 ] ) one could see that @xmath180 so that we finally have @xmath181 on super - horizon scales . note that one should not simply set @xmath182 in this limit as was done in ref . aetherperturbation , although that leads to the same result @xcite . this equation is still valid _ outside _ the inflationary epoch within the parameter space given by eq . ( [ eq : parameterspace2 ] ) . the solution to eq . ( [ eq : propagationphils ] ) is given by @xmath183 where @xmath184 is a constant . long after leaving the horizon , the potential @xmath185 and the scalar perturbation @xmath186 have time to evolve , and after inflation @xmath186 ceases to exist . however , as mentioned above , during the radiation era on super - horizon scales @xmath158 finally becomes a constant which is related to @xmath184 by @xmath187 once @xmath184 is known we could set the initial conditions of @xmath158 in the radiation era for the subsequent evolutions . the quantity @xmath184 could be fixed by matching to the inflaton perturbation @xmath186 at the time of horizon exit ( @xmath188 ) as follows : substituting eqs . ( eq : fiaetherq , [ eq : fivarphiq ] ) into eq . ( [ eq : propagationphi2 ] ) we get @xmath189 \nonumber \\ & & - c_{13}\left[\epsilon^{\prime \prime } + ( \mathcal{h}^{\prime } -\mathcal{h% } ^{2})\epsilon\right].\end{aligned}\ ] ] now we begin to confine ourselves within the parameter space eq . ( eq : parameterspace ) . in that case @xmath190 and the -field eom eq . ( eq : propagationepsilon ) becomes @xmath191 & = & ( c_{1}+c_{2}+c_{3})k^{2}\epsilon.\ \ \\end{aligned}\ ] ] so , from these two equations we find that @xmath192 on the other hand , from eq . ( [ eq : lsphi ] ) one can write @xmath193 obviously , matching these two expressions gives the value of @xmath184 @xmath194 where the three variables @xmath195 and @xmath186 are all evaluated at the horizon exit . as a result , the initial power spectrum for @xmath158 , @xmath196 where @xmath197 means the ensemble average , is given by @xmath198 in which we have used the relation @xmath199 @xcite ( as discussed in ref . @xcite , this relation is not affected by the -field up to first order because the -field is not coupled to the inflaton ) and the hubble expansion rate @xmath200 is also evaluated at the time of horizon exit . note that the @xmath146 power spectrum eq . ( [ eq : pps ] ) has exactly the same form as in standard gr . to compare their magnitudes , let us write @xmath201 in which we have used the fact that at the time of horizon exit @xmath202 should be the same in the two models . if we further assume the same inflation potential , then @xcite because the background expansion in our ther model is indistinguishable from that of gr , we have @xmath203 that is , the power spectra in these two models are exactly the same , in both the shape and the magnitude . it should be noticed that if the background evolution , @xmath204 , of the - model during the inflation is different from gr , then the evolutions of @xmath205 will also be different because of the scalar field equation of motion ; as a result the slow - roll parameters at the horizon crossing are generally different in these two models and so will be the spectral indices . what is the fate of the vector and tensor modes of the -field perturbation ? as shown in @xcite , the vector mode is also sourceless unless exotic matter such as cosmological defects exist . hence , it should decay similarly all the way up to the present and leave no traces in the cmb observables . as for the tensor modes , from eqs . ( [ eq : aerho ] - [ eq : aepi ] ) one can see that the contribution of the -field to the tensor modes lies only in @xmath206 and is independent of the model parameter @xmath23 . consequently , the power spectrum of the tensor perturbations given in @xcite @xmath207 is still valid here ( note that what appears in this formula is the true bare gravitational constant ) . following the argument above again , we can show that @xmath208 according to eq . ( [ eq : parameterspace2 ] ) , @xmath209 and @xmath210 , so the amplitude of the gravitational - wave spectrum predicted in our -model is smaller than that arising in standard gr . although this may provide a discrimination between the two models , the difficulty in observing the tensor spectrum will be a great hurdle for the use of this discriminator : if our universe turns out to be described by the -model , then this spectrum is even more difficult to observe than in gr . we have seen in the above that in our -model the primordial scalar - mode power spectrum is exactly the same as assuming standard gr and using the measured value @xmath211 in calculations . the next question is if possible deviations from gr will emerge in the subsequent evolutions of the perturbations . interestingly , we find that the answer is yes . the reason is that , although the -perturbation is sourceless during the inflationary era , in subsequent stages where the fluid anisotropic stresses are nonzero it becomes sourced . to see the consequences explicitly , we now write the -field eom within the parameter space eq . ( [ eq : parameterspace ] ) as @xmath212^{\prime } & = & \frac{\alpha+2c_{13}}{3}k^{3}\tilde{% \text{\ae } } . \nonumber\end{aligned}\ ] ] during the inflationary epoch @xmath213 and @xmath214 , thus the solution to the above equation is @xmath215 where we have defined @xmath216 which is positive according to eq . ( eq : parameterspace2 ) ; @xmath217 are integration constants . meanwhile , during this epoch the scale factor @xmath52 undergoes exponential growth , which means that @xmath136 decays exponentially and its initial value is washed out soon . versus the cosmic scale factor @xmath52 . we have shown it for 3 different scales @xmath218 respectively , as indicated beside the curves . the model parameters are @xmath219 , which satisfy the constraints eq . ( [ eq : parameterspace2 ] ) . ] in the radiation dominated epoch , @xmath220 and @xmath221 ( here subscripts @xmath222 denotes photon and neutrino respectively ) . now the homogeneous part of the -field eom becomes @xmath223 whose solution is @xmath224 \nonumber \\ & & + \frac{b_{2}}{\tau^{2}}\left[\sin(c_{s}k\tau)c_{s}k\tau + \cos(c_{s}k\tau)\right]\end{aligned}\ ] ] where @xmath225 are integration constants and a subscript @xmath226 means the general solution . we can see that in the limit @xmath227 @xmath228 decays as @xmath229 and when @xmath230 it decreases as @xmath231 . meanwhile there is a particular solution of @xmath136 which involves a weighted integration of @xmath232 over time . deep into the radiation - dominated epoch , the decaying general solutions of @xmath233 have become negligible and the growing particular solution is still tiny , so we can reasonably take the initial conditions of @xmath234 as @xmath235 in our numerical calculations . we have checked that , for the scales we are interested in , this choice of initial conditions is robust and the numerical results are not sensitive to ( not dramatically ) different initial conditions , and in fig . [ fig : aether ] we have depicted the time evolution of @xmath136 at different scales ( or different @xmath149 ) for the model @xmath236 . it can be seen there that , at early times @xmath233 remains close to zero ; later it grows as @xmath237 deviates significantly from 0 , and finally when @xmath238 becomes tiny in the matter - dominated era it undergoes a ( oscillatory ) decay again . now what kind of signatures will this behavior of @xmath136 imprint on the perturbation evolution ? in order to answer this question , we find it useful to rewrite the total density perturbation , pressure perturbation , heat flux and anisotropic stress ( with the contributions from the @xmath239-field included ) as follows : @xmath240 where the quantities with subscript @xmath241 @xmath242 are the tracking parts because they are exact rescalings of the corresponding quantities for the fluid matter [ we remind that @xmath243 , while those with subscript @xmath244 are the non - tracking parts expressed as @xmath245 only the measured value @xmath211 appears in the final expressions . it can be easily checked that the tracking and non - tracking parts satisfy the energy momentum conservation equation separately by utilizing the -field eom . in particular , the non - tracking part effectively represents a purely perturbed contribution to the total energy momentum tensor that has no background counterparts . we also note that @xmath246 . taking the time derivative of eq . ( [ eq : constrainteta2 ] ) , using eq . ( eq : propagationeta2 ) to eliminate the @xmath247 term and cancelling the @xmath248 appearing there with eq . ( eq : energyconservation2 ) , one arrives at the following evolution equation of @xmath249 : @xmath250 or equally @xmath251 that is , the non - tracking part does not contribute to the evolution of @xmath249 . in this subsection we work in the cdm frame where the 4-acceleration @xmath252 , and thus @xmath253 @xcite ( where @xmath254 is the density contrast ) which means the growth of matter perturbation is not affected by the non - tracking part . however , since @xmath255 , which is nonzero at early times , appears in eq . ( [ eq : constrainteta2 ] ) , the values of @xmath256 and @xmath249 in this equation are slightly modified that change the subsequent evolution of @xmath249 a little . these can be seen in fig . [ fig : pk ] , where we have displayed the matter power spectrum of the -model for different choices of parameters . obviously the spectrum depends weakly on the parameters ; furthermore , its shape is essentially the same as that in @xmath257 model so that the parameters will degenerate with the bias relating the matter power spectrum to the observed spectrum of galaxies . in addition to that , there is also the degeneracy with respect to the neutrino masses , since the latter affects its anisotropic stress and the time at which they become non - relativistic . taken together this indicates that the data for @xmath258 will not place very stringent constraints on the -model . and @xmath259 ( which is just @xmath257 ) respectively . the other parameters common for all curves are @xmath260 and @xmath261 where @xmath262 are the fractional energy density of baryons and cdm , and @xmath263 . the differences among these spectra are small although their parameters are quite different . ] from eqs . ( [ eq : constraintphi2 ] - [ eq : propagationphi2 ] ) one can see that @xmath264 are also influenced by the non - tracking part , and because these variable determine the cmb power spectrum , we could expect the latter to be modified by the presence of the -field as well . this is correct , as shown fig . [ fig : cmb ] , where we have plotted the cmb spectra for different choices of the model . obviously the cmb power spectrum is changed by the -field , though this effect is not very strong and so it might be difficult to place stringent limits on the parameter space of the theory . from top to bottom the curves correspond to @xmath265 and @xmath259 ( which is just @xmath257 ) respectively . the other parameters common for all curves are @xmath260 and @xmath261 where @xmath262 are the fractional energy density of baryons and cdm , and @xmath263 . ] there are two points that need to be noticed . firstly , in this work we are assuming that the recent cosmic acceleration is caused by a cosmological constant , the anisotropic stress of which is zero . if in contrast the dark energy has a nonzero anisotropic stress ( as in @xcite ) then the -field perturbation @xmath142 probably will not decay at late times , and this might lead to further modifications of the cmb power spectrum . secondly , when @xmath266 is not simply a linear function of @xmath267 , the -field will be sourced by the evolution of the gravitational potential as well , and its effects will be more intricate and interesting . such new features may appear in more complicated models like teves @xcite and -fields with non - canonical kinetic terms @xcite . in total , these results indicate that within the parameter space eq . ( [ eq : parameterspace ] ) the background cosmology of our -model is the same as in gr , while the cmb and matter power spectra differ slightly from the predications of gr . in principle , these features could be used to constrain the parameters @xmath268 and @xmath269 of the present model . this will generally involve a full search of the parameter space using for example the markov chain monte carlo method , which is beyond the scope of this work . however , as we discussed above , the constraints from the linear spectra may not be very stringent anyway . in this paper we have studied the cosmology of the einstein- ther theory . after presenting the general field equations for such theories in the cgi formalism , we focussed on a specific class of models described in eq . ( [ eq : aelagrangian ] ) , and confined ourselves to the parameter space of models described by eq . ( [ eq : parameterspace2 ] ) which pass the ppn and @xmath118erenkov constraints . this parameter space is known to have the same locally- and cosmologically - felt gravitational constants ( @xmath270 , which are different from the true bare @xmath94 ) and this tracking behavior indicates that we can consider its background expansion just by using the measured value @xmath271 and ignoring the presence of the -field ( since the only effect of the ther field is to track other matter species , the model in eq . ( [ eq : aelagrangian ] ) clearly can not explain dark energy . more general models , such as the @xmath266 proposed in @xcite and our sec . [ sect : equations ] , may serve this purpose ) . we find that it is a general feature that the tracking behavior not only occurs at the background level but also at the linear order in perturbation theory . for example , the quantity @xmath272 tracks that of the other matter species on super - horizon scales . this indicates that , whatever type of perturbation is generated during inflation , the presence of -field will not alter it . in particular , in the single - field inflation model we consider , no isocurvature perturbation is produced . this is an important characteristic , and it would be interesting to see whether similar behavior occurs for general , higher - order choices of @xmath266 . we generalized the analysis of primordial power spectra for the -theory @xcite to our model . for the parameter ranges which satisfy local gravity bounds , we find that the evolution of the large - scale gravitational potential , @xmath158 , is unmodified as compared with that in gr , and show that the primordial power spectrum of @xmath158 also has the same form as in gr [ c.f . ( [ eq : pps ] ) ] . if we assume that the bare gravitational constants in gr and in the -model are the same , then the magnitudes of the spectra are different in these two models . however , contrary to the discussion in @xcite , we argue that we do not know the true bare @xmath273 , but only know the measured value , @xmath274 . in both gr and our -model , @xmath275 is equal to the cosmological value @xmath276 while in the latter @xmath277 . as a result , we show that with the same inflationary potential the primordial power spectra of @xmath158 in these two models should have the same shape and the same magnitude . meanwhile , we also find that the power spectrum of tensor perturbations in our model is smaller in magnitude than is predicted in gr , and so currently we can not use it to distinguish between gr and the -model . for the late - time evolution of the perturbations , it is shown that , the -field perturbation is driven by the evolution of the anisotropic photon and neutrino stresses the @xmath278^{\prime}$ ] term in its propagation equation although it is sourceless during inflation . therefore , it decays away exponentially during the inflationary era , grows again in the radiation - dominated epoch when @xmath279^{\prime}$ ] is significant , and finally diminishes again , oscillating when @xmath279^{\prime}$ ] eventually becomes negligible . as a result , the cmb and matter power spectra are modified by the existence of the -field . we also remark that , depending on the nature of the dark energy , it is possible that the -field perturbation will have non - trivial dynamics driven by the @xmath280 term , where @xmath281 is the possible dark - energy anisotropic stress @xcite . we also note that there recently appeared a later paper @xcite which investigated in details the structure formation of the model proposed in @xcite . in their sec . v the authors considered a simple power - law model @xmath282 in which @xmath283 corresponds to our lagrangian eq . ( [ eq : aelagrangian ] ) ( except for the @xmath23 term ) . they obtained criteria for there to be a growing mode in the evolution of the ( scalar - mode ) - field perturbation . however their criteria can not be applied directly to the model we consider here . to see why , note that in eq . ( 32 ) of @xcite the quantities @xmath284 also contain @xmath285 ( their @xmath285 is equivalently our @xmath286 ) . however , by their argument the @xmath285 terms in @xmath287 are suppressed by a small quantity @xmath288 ( which in _ our _ notation is just @xmath289 ) and are ultimately neglected from the - field equation of motion , their eq . ( 43 ) . in our model , there is no argument that @xmath290 ( its magnitude actually _ is _ 1 ) and so we can no longer neglect the @xmath286 terms appearing in @xmath291 . when these are taken into account we obtain a different equation of motion , _ i.e. _ , the @xmath292s in eq . ( 43 ) of @xcite are different in our work , and consequently the criteria for the existence of growing modes are different as well . in addition , are the facts that we have a @xmath23 term in our lagrangian and use a different parameter space , eqs . ( [ eq : parameterspace ] , [ eq : parameterspace2 ] ) , which also contribute to the differences between our results and those in @xcite . as an aside , we stress that the key relation in our work , @xmath293 , is a consequence of eq . ( [ eq : parameterspace ] ) and has nothing to do with whether or not the - gravity waves propagate superluminally . in fact , we could do our calculation dropping the constraint that these waves propagate superluminally , and in this case the perturbation dynamics might place some constraints on the parameter space ; for example , in some portion of the parameter space the growth of @xmath286 may become unstable . in our work we choose the parameters leading to superluminal gravity waves simply to avoid constraints from @xmath294erenkov radiation . there are still some debates about this kind of choice , see however @xcite for a different and conservative point of view . there are also other general differences between the ea model we consider here and the gea model of @xcite . first , for the static and weak field limit , the @xmath23 term and the choice of parameter space eq . ( [ eq : parameterspace ] ) are crucial for the ea model to evade ppn tests @xcite ; in @xcite there is no @xmath23 term , but the nonlinearity in the ther field lagrangian guarantees that the local gravitational tests are not a problem because at high densities the modification is negligible . second , as we have seen above , our conclusion that the primordial power spectrum of density perturbation is unmodified compared with gr relies on the fact @xmath293 so that the background expansion during inflation in our model is the same as in gr ( again the @xmath23 term is crucial here ) . if the @xmath23 term is not included , the local and cosmological gravitational constants are different , meaning that the background evolution during inflation is different from gr ; as a result the background quantities at horizon crossing , which determine the observables in inflationary models through the slow - roll parameters , may be different from those in gr this means that the ea model will predict different shape of primordial density spectrum other than gr @xcite . for the gea model considered in @xcite , again the nonlinear @xmath266 makes the modifications to gr during early times small enough to be neglected , and as such the primordial power spectrum is also the same as in gr . note however that in the latter case @xcite the late time evolution of vector field perturbation is also modified in a complex way , making the corresponding cosmological behaviors different from ours . so , in conclusion , although the background expansion of our -model is exactly the same as that in gr , the cosmological data on cmb and matter power spectra can be used to constrain the -model . however , the constraint is not expected to be very stringent because the linear perturbation spectra depend weakly on the model parameters . other considerations of the behaviour in strong gravitational fields , such as those studies of the compact stars or black holes @xcite would enable better constraints on the present model to be obtained . alao , as general interests for cosmology , studies in more complicated - field lagrangian as those performed in @xcite need to be explored thoroughly . we thank ted jacobson , eugene lim , constantinos skordis , xiaoting wang , hongsheng zhao , tom zlosnik and the referee for helpful discussions . the numerical calculation in this work uses a modified version of the public camb code @xcite . bl is supported by the overseas research studentship , cambridge overseas trust and the damtp at university of cambridge . dfm acknowledges the humboldt foundation . arianto , f. p. zen , bobby , e. gunara , triyanta and supardi ( 2007 ) , arxiv:0709.3688 [ hep - th ] ; s. kanno and j. soda , phys . d * 74 * , 063505 ( 2006 ) ; a. tartaglia and n. radicella ( 2007 ) , arxiv:0708.0675 [ gr - qc ] ; m. libanov , v. rubakov , e. papantonopoulos , m. sami and s. tsujikawa , jcap * 0708 * , 010 ( 2007 ) ; p. g. ferreira , b. m. gripaios , r. saffari and t. g. zlosnik , phys . d * 75 * , 044014 ( 2007 ) . and , * * , ( ) ; and , in _ theoretial and observational cosmology , _ ed . m. lachize - rey ( springer , new york , 1998 ) ; and , * * , ( ) ; , phd dissertation , queens college and astrophysics group , cavendish laboratory , cambridge university , 2000 . and , * * , ( ) [ arxiv : astro - ph/0610486 ] ; and , * * , ( ) [ arxiv : gr - qc/0701111 ] ; , and , * * , ( ) [ arxiv : astro - ph/0610794 ] ; , and , * * , ( ) [ arxiv:0705.3795 [ gr - qc ] ] ; t. koivisto and d. f. mota , phys . lett . b * 644 * , 104 ( 2007 ) ; t. koivisto and d. f. mota , phys . rev . d * 75 * , 023518 ( 2007 ) ; , and , phys . d * 76 * , 104047 ( 2007 ) ( arxiv:0707.2664 [ gr - qc ] ) . we want to elaborate this point as it will also be used below . suppose we have a universe described by either standard gr or the -theory eq . ( [ eq : aelagrangian ] ) and we have a measured gravitational constant @xmath295 . if we assume gr , then we have @xmath296 , while otherwise we have @xmath297 , so that whenever we need @xmath298 we could relate it to the textbook - value @xmath295 by @xmath299 ( in the gr case ) , or @xmath300 ( in the -case ) . obviously , when @xmath301 , @xmath302 the background cosmologies are the same in gr and our -model . t. clifton , d. f. mota and j. d. barrow , mon . not . * 358 * ( 2005 ) 601 [ arxiv : gr - qc/0406001 ] . , _ theory and experiment in gravitational physics _ ( cambridge university press , cambridge , 1993 ) . t. koivisto and d. f. mota , phys . d * 73 * , 083502 ( 2006 ) [ arxiv : astro - ph/0512135 ] . t. koivisto and d. f. mota , arxiv:0707.0279 [ astro - ph ] . d. f. mota , j. r. kristiansen , t. koivisto and n. e. groeneboom , arxiv:0708.0830 [ astro - ph ] . with general choices of parameters @xmath303 , eq . ( eq : propagationphils ) is only valid if @xmath304 . thus , in the radiation - dominated era it can not apply even for superhorizon scales , and its solution eq . ( [ eq : lsphi ] ) should be modified as well .

this paper unifies problems and results related to ( embedding ) universal and homomorphism universal structures . on the one side we give a new combinatorial proof of the existence of universal objects for homomorphism defined classes of structures ( thus reproving a result of ) and on the other side this leads to the new proof of the existence of dual objects ( established by ) . our explicite approach has further applications to special structures such as variants of the rational urysohn space . we also solve a related extremal problem which shows the optimality ( of the used lifted arities ) of our construction . our method also relates to weakly indivisible homomorphism defined classes of structures .

0909.4939

it is an old mathematical idea to reduce a study of a particular class of objects to a certain single `` universal '' object . it is hoped that this object may then be used to study the given ( infinite ) set of individual problems in a more systematic and perhaps even efficient way . for example the universal object may have interesting additional properties ( such as symmetries and ultrahomogeneity ) which in turn can be used to classify the finite problems . a good example of this is the classification of ramsey classes of structures via the classification program of generic structures @xcite . in this paper we deal with universal objects from the homomorphism point of view . we define and study homomorphism - universal ( shortly hom - universal @xcite ) objects and show their relationship to embedding - universal ( shortly universal ) objects . given a class @xmath0 of countable structures , an object @xmath1 is called _ hom - universal _ ( or _ universal _ ) if for every object @xmath2 there exists a homomorphism ( or an embedding ) @xmath3 . the characterization of those classes @xmath0 which have an universal ( or , more recently , hom - universal ) object is a well known open problem which was studied intensively , see e.g. @xcite . the whole area was inspired by the negative results ( see @xcite ) : for example the class of graphs not containing @xmath4 ( = cycle of length @xmath5 ) fails to be universal for any @xmath6 . for universal structures the strongest results in the positive direction were obtained by cherlin , shelah and shi in @xcite . particularly , they proved the following ( as consequence of theorem 4 @xcite ) : [ 1thm ] for every finite set @xmath7 of finite connected graphs the class @xmath8 has an universal object . here @xmath9 denotes the class of all graphs ( and more generally relational structures ) @xmath10 for which there is no homomorphism @xmath11 for every @xmath12 . formally @xmath13 . ( in this paper the structures are denoted by bold letters . this applies also to graphs . but a cycle of length @xmath5 is still denoted by @xmath4 and complette graph on @xmath14 vertices by @xmath15 . ) similarly by @xmath16 we denote the class of all graphs ( and more generally relational structures ) @xmath10 for which there is no embedding @xmath11 for every @xmath12 . ( note that the @xcite proves a stronger result : the universal graph exists for any class with finite algebraic closure and , moreover , that the existence of @xmath17-categorical universal graph is equivalent to this condition . this result was extended to relational structures in @xcite . our motivation comes from study of graph homomorphisms and thus in this paper we are interested in this weaker version of their result as presented by theorem [ 1thm ] . ) the proof of theorem [ 1thm ] given in @xcite is based on techniques of model theory and it is possible to say that no explicit universal object is constructed `` in any very explicite way '' ( @xcite ) . in this paper we give such a proof . homomorphism universal structures lead to a very different area of constraint satisfaction problems . clearly every universal object is also hom - universal . this however does not hold conversely as shown by examples of classes all with of graphs bounded chromatic numbers . another example is provided by the class of all planar graphs : @xmath18 is hom - universal for the class of planar graphs by virtue of the @xmath19-color theorem while no universal graph exists @xcite . of special interest are classes @xmath9 which have finite hom - universal graph . such universal graphs are called _ duals _ and in this context the classes with finite hom - universal objects were characterized @xcite as follows ( for general finite relational structures ) @xmath9 has a finite hom - universal object ( i.e. dual ) if and only if all structures @xmath20 are ( relational ) trees . in this paper we give a new construction of duals ( corollary [ nasdual ] ) . it is the aim of this paper to investigate both universal and hom - universal objects in the context of a ( seemingly much more restricted ) class of generic objects . main result ( theorem [ univthm ] ) proves theorem [ 1thm ] by means of _ shadows ( reducts ) _ and _ lifts ( expansions ) _ of a generic ( i.e. ultrahomogeneous universal ) structure . these classical model theoretic notions will be , for the sake of completeness , briefly reviewed now ( see e.g. @xcite ) : a _ structure _ @xmath21 is a pair @xmath22 where @xmath23 ( i.e. @xmath24 is a @xmath25-ary relation on @xmath26 ) . the finite family @xmath27 is called the _ type _ @xmath28 . the type is usually fixed and understood from the context . ( note that we consider relational structures only , and no function symbols . ) if set @xmath26 is finite we call _ @xmath29 finite structure_. a _ homomorphism _ @xmath30 is a mapping @xmath31 satisfying for every @xmath32 , @xmath33 . if @xmath34 is 1 - 1 , then @xmath34 is called an _ embedding_. the class of all ( countable ) relational structures of type @xmath28 will be denoted by @xmath35 . the class @xmath36 , @xmath37 finite , is fixed throughout this paper . unless otherwise stated all structures @xmath38 belong to @xmath35 . now let @xmath39 be a type containing type @xmath28 . ( by this we mean @xmath40 and @xmath41 for @xmath33 . ) then every structure @xmath42 may be viewed as structure @xmath43 together with some additional relations @xmath44 for @xmath45 . to make this more explicite these additional relations will be denoted by @xmath46 . thus a structure @xmath42 will be written as @xmath47 and , by abuse of notation , briefly as : @xmath48 we call @xmath49 a _ lift _ of @xmath21 and @xmath21 is called the _ shadow _ ( or _ projection _ ) of @xmath49 . in this sense the class @xmath50 is the class of all lifts of @xmath35 . conversely , @xmath35 is the class of all shadows of @xmath50 . note that a lift is also in the model theoretic setting called an _ expansion _ and a shadow is often called a _ reduct_. ( our terminology is motivated by a computer science context , see @xcite . ) we will use letters @xmath51 for shadows ( in @xmath35 ) and letters @xmath52 for lifts ( in @xmath53 . for lift @xmath54 we denote by @xmath55 the relational structure @xmath21 i.e. its shadow . ( @xmath56 is called the _ forgetful functor_. ) similarly , for a class @xmath57 of lifted objects we denote by @xmath58 the class of all shadows of structures in @xmath57 . as it is well known , an universal object may be constructed by an iterated amalgamation of finite objects and this leads to a stronger notion of _ @xmath0-generic object _ : for a class @xmath0 we say that an object @xmath59 is @xmath0-generic if it is both @xmath0-(embedding ) universal and it is _ ultrahomogeneous_. the later notion means the following : every isomorphism @xmath60 between two finite substructures @xmath61 and @xmath62 of @xmath59 may be extended to an automorphism of @xmath59 . the notion of ultrahomogeneous structure is one of the key notions of modern model theory and it is the source of the well known classification programme @xcite . the ultrahomogeneous structures are characterized by the properties of finite substructures . denote by @xmath63 the class of all finite substructures of @xmath59 . by a following classical result @xcite structure @xmath59 is ultrahomogeneous iff @xmath64 is a countable class with the following three properties : 1 . ( hereditary property ) for every @xmath2 and an induced substructure @xmath61 of @xmath21 we have @xmath65 ; 2 . ( joint embedding property ) for every @xmath66 there exists @xmath67 such that both @xmath62 contains @xmath21 and @xmath61 as induced substructures ; 3 . ( amalgamation property ) for @xmath68 and @xmath60 embedding of @xmath62 into @xmath21 , @xmath69 embedding of @xmath62 into @xmath61 , there exists @xmath70 and embeddings @xmath71 and @xmath72 such that @xmath73 . see figure [ amalgamfig ] . + and @xmath61 over @xmath62 . ] any class @xmath0 satisfying the three conditions @xmath74 is called an _ amalgamation class_. in this case we denote by @xmath75 the ( up to isomorphism uniquely determined ) generic object @xmath59 . @xmath76 is called the _ limit _ of @xmath0 . in this paper we show that universal object for classes @xmath77 may be constructed as a shadow ( reduct ) of a generic object of an explicitly defined lifted class . the following is the principal result of this paper : [ univthm ] for every finite family @xmath78 of finite connected relational structures of type @xmath28 there exists a type @xmath79 containing @xmath28 and a finite family @xmath80 of finite structures of type @xmath79 such that : * @xmath81 is amalgamation class whose shadow is @xmath77 , * the shadow @xmath59 of the generic @xmath82 is universal structure for @xmath77 . among others this implies that @xmath83 is ultrahomogeneous and @xmath59 is @xmath17-categorical ( of course @xmath17-categoricity for universal objects for classes @xmath77 was proved in @xcite as well . ) note that , for lifted objects we have to consider forbidden - embedding classes @xmath81 ( and not just classes @xmath84 ) . this is necessary as we show in lemma [ stupne ] . we prove theorem [ univthm ] in three steps : first we prove that there is a class @xmath85 of lifted objects which is an amalgamation class ( theorem [ mainthm ] ) and then we define @xmath85 by a set of forbidden rooted substructures ( theorem [ reprezent ] ) and finally by the set of forbidden embeddings ( theorem [ reprezent2 ] ) . thus we can get universal objects for classes @xmath77 in a particularly efficient way as shadow of a limit of certain ( explicitly defined ) lifted class @xmath81 . theorem [ univthm ] and its combinatorial proof have several other consequences . for example , we can prove , that many classes @xmath77 are weakly indivisible . in the another direction we show that in the case that all @xmath86 are relational trees it suffices to consider only a class of monadic lifts ( i.e. all relations @xmath87 for @xmath45 are monadic ; this case corresponds to structures endowed with a coloring of its vertices ) . this is stated in section [ monadic ] . consequently , in this case the generic structure @xmath83 has not only the universal shadow @xmath59 but in fact this shadow @xmath59 has a finite core @xmath88 which is thus a finite hom - universal structure and thus a dual . this is stated as corollary [ nasdual ] . in this way we reprove the finite duality theorem @xcite . this is particularly pleasing : ( in this sense ) duals are generic objects and this indicates yet another context of hom - dualities . in section [ urysohnsection ] we include further corollaries of our method to metric spaces ( which can be treated analogously to forbidding cycles ) . in section [ bounded ] we solve a related extremal problem : we show that our method lead to the optimal arities of added lifts . this is based on a non - trivial ramsey - type result ( lemma [ sets ] ) . let us add a few more definitions . for a structure @xmath89 the _ gaifman graph _ ( in combinatorics often called _ 2-section _ ) is the graph @xmath90 with vertices @xmath26 and all those edges which are a subset of a tuple of a relation of @xmath21 : @xmath91 where @xmath92 iff @xmath93 and there exists tuple @xmath94 such that @xmath95 . a _ cut _ in @xmath21 is a subset @xmath96 of @xmath26 such that the gaifman graph @xmath90 is disconnected by removing set @xmath96 ( i.e. if @xmath96 is graph theoretic cut of @xmath90 ) . by a _ minimal cut _ we always mean an inclusion minimal cut . if @xmath96 is a set of vertices then @xmath97 will denote a tuple ( of length @xmath98 ) from all elements of @xmath96 . alternatively , @xmath97 is arbitrary linear ordering of @xmath96 . the following is our basic notion which resembles decomposition techniques standard in graph theory and thus we adopted similar terminology . for a relational structure @xmath21 and minimal cut @xmath100 in @xmath21 , a _ piece _ of relational structure @xmath29 is pair @xmath101 . here @xmath102 is structure induced on @xmath21 by union of @xmath100 and vertices of some connected component of @xmath103 . tuple @xmath104 consist of the vertices of cut @xmath100 in a ( fixed ) linear order . note that from the inclusion - minimality of the cut @xmath100 follows that all pieces of a connected structure are connected structures . it follows that for connected structure @xmath21 the pieces of @xmath21 determined by any minimal cut @xmath96 cover the whole @xmath21 . all pieces are considered as rooted structures : a piece @xmath105 is a structure @xmath102 rooted at @xmath104 . accordingly , we say that pieces @xmath106 and @xmath107 are _ isomorphic _ if there is function @xmath108 that is isomorphism of structures @xmath109 and @xmath110 and @xmath60 restricted to @xmath111 is the monotone bijection between @xmath111 and @xmath112 ( we denote this @xmath113 ) . [ petersoni ] observe that for relational trees , pieces are equivalent to rooted branches . fig [ petersoni ] shows all pieces of the petersen graph . let @xmath106 be a piece of structure @xmath29 and @xmath107 a piece of @xmath109 . if @xmath114 , then @xmath115 is also a piece of @xmath29 . [ kousky ] denote by @xmath116 connected component of @xmath117 that produces @xmath118 ( i.e. @xmath119 ) . denote by @xmath120 component of @xmath121 that produces @xmath115 ( i.e. @xmath122 ) . as @xmath114 one can check that then @xmath120 is contained in @xmath116 and every vertex of @xmath21 connected by tuple to any vertex of @xmath120 is contained in @xmath109 . thus @xmath120 is also connected component of @xmath21 created after removing vertices of @xmath123 . let @xmath78 be a fixed countable set of finite relational structures of ( finite ) type @xmath28 . for construction of universal structure of @xmath77 we use special lifts , called @xmath78-lifts . let @xmath124 be all pieces of all relational structures @xmath86 . notice that there are only finitely many pieces . this enumeration of pieces is fixed throughout this section . relational structure @xmath125 is called _ @xmath78-lift _ of relational structure @xmath29 when the arities of relations @xmath126 , correspond to @xmath127 . for relational structure @xmath29 we define _ canonical lift @xmath128 _ by putting @xmath129 iff there is homomorphism @xmath60 from @xmath130 to @xmath26 such that @xmath131 denote by @xmath85 the class of all induced substructures of lift @xmath132 , @xmath133 . for @xmath134 we denote by @xmath135 a structure @xmath136 such that structure @xmath49 is induced on @xmath137 by @xmath132 . @xmath135 is called a _ witness _ of the fact that @xmath49 belongs to @xmath85 . [ mainthm ] let @xmath78 be a finite set of finite connected relational structures . the class @xmath85 is an amalgamation class . consequently , there is a generic structure @xmath59 in @xmath85 and its shadow @xmath138 is a universal structure for class @xmath77 . [ mainth ] by definition the class @xmath85 is hereditary , isomorphism closed , has a joint embedding property . @xmath85 is countable , because there are only countably many structures in @xmath77 ( because type @xmath28 is finite ) and thus also countably many lifts . to show that @xmath85 is amalgamation class it remains to verify that @xmath85 has the amalgamation property . consider @xmath139 . assume that structure @xmath140 is substructure induced by both @xmath49 and @xmath141 on @xmath142 and without loss of generality assume that @xmath143 . put @xmath144 @xmath145 @xmath146 because @xmath85 is closed under isomorphism , we can still assume that @xmath21 and @xmath61 are vertex disjoint with exception of vertices of @xmath62 . let @xmath88 be free amalgam of @xmath21 and @xmath61 over vertices of @xmath62 : vertices of @xmath88 are @xmath147 and there is @xmath148 iff @xmath149 or @xmath150 . we claim that the structure @xmath151 is ( not necessarily free ) amalgam of @xmath152 over @xmath140 and thus also amalgam of @xmath153 over @xmath140 . first we show that the substructure induced by @xmath154 on @xmath26 is @xmath132 and that the substructure induced by @xmath154 on @xmath155 is @xmath156 . in the other words no new tuples to @xmath152 ( and thus also to @xmath49 and to @xmath141 ) were introduced . assume the contrary that there is a new tuple @xmath157 and among all tuples and possible choices of @xmath14 choose one with the minimal number of vertices of the corresponding piece @xmath158 . such @xmath158 is fixed throughout this proof . by symmetry we can assume that @xmath159 . explicitely , we assume that there is homomorphism @xmath60 from @xmath160 to @xmath88 such that @xmath161 . the set @xmath162 of vertices of @xmath160 mapped to @xmath132 is nonempty , because it contains all vertices of @xmath163 . @xmath164 is nonempty because there is no homomorphism @xmath69 from @xmath160 to @xmath21 such that @xmath165 ( otherwise we would have @xmath166 . @xmath167 is also non empty as there are no edges from vertices @xmath168 to vertices @xmath169 in @xmath170 and because pieces are connected . moreover the vertices of @xmath167 form a cut of @xmath160 . denote by @xmath171 all connected components of substructure induced on @xmath172 by @xmath173 . for each component @xmath174 , @xmath175 there is vertex cut @xmath176 of @xmath173 constructed by all vertices of @xmath162 connected to @xmath177 . this cut is always contained in @xmath167 . because @xmath158 is a piece of some @xmath178 and because @xmath179 are pieces of @xmath173 , by lemma [ kousky ] , they are also pieces of @xmath180 . we denote by @xmath181 all pieces isomorphic to pieces @xmath182 via isomorphisms @xmath183 . now we use the minimality of the piece @xmath158 . all the pieces @xmath184 have smaller size than @xmath158 ( as @xmath167 is a cut of @xmath185 . thus we have that tuple @xmath186 of @xmath187 is as well a tuple of @xmath132 . thus there exists homomorphism @xmath188 from @xmath174 to @xmath170 such that @xmath189 for every @xmath190 . in this situation we define @xmath191 as follows : 1 . @xmath192 when @xmath193 for some @xmath190 . @xmath194 . it is easy to see that @xmath195 is homomorphism from @xmath173 to @xmath132 . this is a contradiction . it remains to verify that @xmath196 . we proceed analogously . assume that @xmath60 is homomorphism of a @xmath86 to @xmath88 . because @xmath197 , @xmath60 must use vertices of @xmath62 and @xmath167 forms cut of @xmath198 . denote by @xmath199 a minimal cut contained in @xmath167 . @xmath200 must contain tuples corresponding to all pieces of @xmath198 having @xmath199 as roots in @xmath140 . this is a contradiction with @xmath201 . in theorem [ mainthm ] we found an amalgamation class @xmath202 of lifted objects such that the shadow of the limit of @xmath85 is universal object of @xmath77 . in this section we further refine this result by giving an explicit description of the amalgamation class @xmath85 in terms of forbidden substructures . we prove that @xmath85 is equivalent to a class @xmath81 for an explicitly defined class of lifts @xmath80 ( derived from the class @xmath78 ) . first we show a more explicit construction of a witness . [ enumerace ] for a piece @xmath203 such that @xmath204 is @xmath205-tuple and for an @xmath205-tuple @xmath206 of vertices of @xmath21 we denote by @xmath207 relational structure created as disjoint union of @xmath21 and @xmath130 identifying vertices of @xmath208 along @xmath206 ( i.e. @xmath209 is free amalgamation over @xmath210 ) . we put @xmath211 where @xmath212 . finally we put @xmath213 . we will call @xmath214 the _ universal witness _ of @xmath215 . now we develop alternative and more explicite description of the class @xmath85 ( introduced in section 2 ) . we preserve all the notation introduced there . [ uwlem ] lift @xmath215 belongs to @xmath85 iff @xmath216 and @xmath215 is induced on @xmath137 by @xmath217 ( in the other words @xmath216 is a witness of @xmath49 ) . assume that @xmath134 . @xmath144 @xmath218 it follows from the construction that there exists a homomorphism @xmath219 which is identity on @xmath49 . if there was homomorphism @xmath69 from some @xmath198 to @xmath214 then by composing with @xmath60 , there exists also a homomorphism from @xmath198 to @xmath135 . this is not possible , since @xmath21 is witness . let us assume now that @xmath49 is not induced by @xmath156 on @xmath137 . from the construction of @xmath156 we have trivially that for each @xmath220 , there is also @xmath221 . assume that there is some @xmath221 consisting only from vertices of @xmath49 such that @xmath222 . let @xmath69 be the homomorphism @xmath223 such that @xmath224 . again by composing with @xmath60 we obtain homomorphism @xmath225 , a contradiction with @xmath222 . thus @xmath49 is induced by @xmath226 on @xmath156 . in the reverse direction if @xmath214 was witness , then @xmath134 . the conditions listed in the lemma are precisely the conditions on @xmath214 being witness . [ roothomo ] for rooted structure @xmath227 we define @xmath228-rooted homomorphism @xmath229 as a homomorphism @xmath230 such that @xmath231 iff @xmath232 . for relational structure @xmath21 and @xmath49 sublift of @xmath132 , we say that @xmath215 is _ @xmath21-covering _ iff there is homomorphism @xmath233 . similarly for piece @xmath203 and @xmath49 a sublift of @xmath234 such that @xmath235 , we say that @xmath215 is _ @xmath203-covering _ iff @xmath137 contains all roots of @xmath208 and there is homomorphism @xmath236 such that @xmath34 is identity on @xmath208 . our first characterization of classes @xmath85 is in the form of rooted homomorphism and coverings . [ reprezent ] for fixed @xmath78 , class @xmath85 ( defined above before theorem [ mainthm ] ) satisfies : @xmath134 if and only if 1 . there is no homomorphism @xmath237 , for @xmath141 being @xmath198-covering for some @xmath86 ; 2 . for every @xmath238 and every @xmath203-covering @xmath140 there is no @xmath228-rooted homomorphism @xmath239 . ( up to isomorphisms and permutations of roots ) , inclusion minimal @xmath240-covering sublifts , and inclusion minimal @xmath118-covering and @xmath115-covering sublifts . roots are denoted by dotted line . ] conditions @xmath241 and @xmath242 holds for every @xmath134 . fix @xmath134 . assume that @xmath241 does not hold for some @xmath141 that is @xmath198-covering for @xmath243 . since there is homomorphism @xmath244 and a homomorphism @xmath245 we also have homomorphism @xmath246 , a contradiction with lemma [ uwlem ] . @xmath242 is a rooted analogy of the same proof . take lift @xmath49 such that @xmath247 . by lemma [ uwlem ] we have one of the following cases : * @xmath49 is not induced by @xmath248 on @xmath49 . + in this case we have some homomorphism of @xmath249 such that @xmath250 . assume that @xmath228 is chosen such that number of vertices of @xmath130 is minimal . + denote by @xmath141 maximal ( non - induced ) sublift of @xmath234 such that @xmath34 is also homomorphism from @xmath141 to @xmath49 . we need to show that @xmath141 is @xmath203-covering to get contradiction with @xmath242 . + denote by @xmath251 components of @xmath252 . now denote by @xmath253 pieces corresponding to these components and by @xmath254 homomorphisms @xmath255 mapping non - roots of @xmath256 to the corresponding component and roots to vertices of @xmath141 . + because @xmath257 is homomorphism @xmath258 we have from minimality of counterexample @xmath259 and thus also @xmath260 . this holds for every @xmath261 and thus we have also @xmath262 such that it is identity on @xmath263 . this prove that @xmath141 is @xmath203-covering . * there is a homomorphism @xmath34 of some @xmath243 to @xmath214 . + assume that @xmath264 is empty . in this case there is @xmath228 such that @xmath265 is contained in vertices of copy of @xmath203 in @xmath266 . in this case lift @xmath49 is covering , because it contains tuple in @xmath267 . a contradiction . + denote by @xmath141 maximal ( non - induced ) sublift of @xmath268 so @xmath34 is also a homomorphism @xmath245 . because there is non - empty intersection of @xmath265 and @xmath137 , @xmath263 is nonempty . we can show that @xmath141 is covering by same argument as in @xmath37 getting contradiction with @xmath241 too . observe that properties @xmath241 and @xmath242 directly translate to the family @xmath80 with the desired property that the shadow of @xmath81 is @xmath77 . this leads to the desired explicit characterization of the class @xmath85 . [ reprezent2 ] let @xmath80 be a class of @xmath78-lifts satisfying the following : 1 . @xmath269 for every lift @xmath49 such that there is @xmath198-covering lift @xmath141 for some @xmath86 together with a surjective homomorphism @xmath245 , 2 . @xmath269 for every structure @xmath49 such that there is @xmath270 and @xmath203-covering rooted structure @xmath271 together with a surjective @xmath228-rooted homomorphism @xmath245 , 3 . @xmath80 contains no other structures . then we have : 1 . @xmath80 is a finite family ; 2 . @xmath272 and thus @xmath81 is amalgamation class . the shadows @xmath59 of the generic @xmath82 is universal structure for @xmath77 . @xmath80 is necessarily finite , because number of vertices of lifts @xmath273 is bounded by number of vertices of structures @xmath274 . from construction above it follows that @xmath81 is precisely class of structures satisfying conditions @xmath241 and @xmath242 . we used notion of rooted homomorphisms ( and thus classes @xmath81 ) to define our lifted classes . it is easy to see that classes @xmath84 are not powerful enough to extend the expressive power . [ stupne ] assume that there is a class @xmath78 and a lifted class @xmath80 such that @xmath84 contains a generic object whose shadow is universal graph of class @xmath77 . then the class @xmath77 itself contains a generic object . observe that all classes @xmath84 are monotonous . that is for any @xmath275 a relational structures @xmath141 created from @xmath49 by removing some of its edges also belongs to @xmath84 . in particular @xmath84 is closed for constructing shadows and thus @xmath77 may be thought as subclass of @xmath84 ( modulo the signature of relational structures ) . now take any @xmath276 and their lifts @xmath277 such that they contain no new edges . these lifts are in class @xmath84 . now consider @xmath278 amalgam of @xmath49 and @xmath141 over @xmath140 and its shadow @xmath88 . @xmath88 is amalgam of @xmath21 and @xmath61 over @xmath62 . the expressive power of lifts can be limited in several ways . for example , it is natural to restrict arities of the newly added relations . it follows from the above proof that the arities of new relations in our lifted amalgamation class @xmath85 depend on the size of maximal inclusion minimal cut of the gaifman graph of the forbidden structure . in this section we completely characterize the minimal arity of ultrahomogeneous lifts of classes @xmath77 . this involves a non - trivial ramsey type statement stated bellow as lemma [ sets ] . as a warm up , we first show that generic universal graph for the class @xmath279 can not be constructed by finite monadic lifts : consider , for contradiction , a monadic lift @xmath83 which is both a generic relational structure and whose shadow @xmath59 is universal for the class @xmath279 . since all extended relations are monadic , we can view them as finite coloring of vertices . for @xmath280 we will denote by @xmath281 the color of @xmath282 or , equivalently , set of all extended relations @xmath283 such that @xmath284 . since graphs in @xmath279 have unbounded chromatic number , we know that chromatic number of @xmath59 is infinite . consider decomposition of @xmath59 implied by @xmath285 . since the range of @xmath285 is finite , one of the graphs in this decomposition has infinite chromatic number . denote this subgraph by @xmath286 . in fact is suffices that @xmath286 is not bipartite . thus @xmath286 contains an odd circle . the shortest odd cycle has length @xmath287 and thus @xmath286 contains induced path of length 3 formed by vertices @xmath288 . additionally there is a vertex @xmath282 of degree at least 2 . because the graph is triangle free , the vertices @xmath289 and @xmath290 connected to @xmath282 are not connected by an edge . from genericity of @xmath59 we know that the partial isomorphism mapping @xmath291 and @xmath292 can be extended to an automorphism @xmath60 of @xmath83 . the vertex @xmath293 is connected to @xmath294 and @xmath295 and thus together with @xmath296 contains either triangle or @xmath297-cycle . it follows that generic @xmath83 can not be defined by monadic lifts of @xmath77 . in this section we prove that there is nothing special here about arity 2 and about the pentagon . one can determine the minimal arity of the ultrahomogeneous lifts for general classes @xmath77 . towards this end we shall need a ramsey type statement which we formulate after introducing the following : let @xmath298 be finite set with partition @xmath299 . for @xmath300 we denote by @xmath301 the index @xmath228 such that @xmath302 . similarly , for a tuple @xmath303 of elements of @xmath298 we denote by @xmath304 the tuple @xmath305 . we make use of the following : [ sets ] for every @xmath306 , @xmath307 and @xmath308 integers there is relational structure @xmath309 , with vertices @xmath310 ( sets @xmath311 are mutually disjoint ) and single relation @xmath312 of arity @xmath313 with the following properties : 1 . every @xmath314 , @xmath315 , @xmath316 satisfies @xmath317 2 . for every @xmath318 , @xmath319 , @xmath320 and @xmath321 has at most @xmath322 common vertices . 3 . for every coloring of tuples on @xmath298 of size @xmath322 using @xmath308 colors there is tuple @xmath323 such that every two tuples @xmath324 consisting of vertices of @xmath320 such that @xmath325 have same color . this statement follows from results obtained by neetil and rdl @xcite . although not stated explicitely , this is `` partite version '' of the main result of @xcite . it can be also obtained directly by means of amalgamation method , see @xcite . in this paper this result plays auxiliary role only and we omit the proof . given a relational structure @xmath309 with relation @xmath312 of arity @xmath313 and rooted relational structure @xmath326 of type @xmath28 with @xmath327 , we denote by @xmath328 the following relational structure @xmath61 of type @xmath28 : the vertices of @xmath155 are equivalence classes of equivalence @xmath329 on @xmath330 generated by the following pairs : @xmath331 @xmath332 @xmath333 denote by @xmath334 $ ] the equivalence class of @xmath329 containing @xmath335 . we put @xmath336 iff @xmath337,[\vec{u},v_2],\ldots [ \vec{u},v_t])$ ] for some @xmath338 and @xmath339 . this construction is commonly used in the graph homomorphism context as _ indicator construction _ , see @xcite . it essentially means replacing every tuple of @xmath312 by disjoint copy of @xmath21 with roots @xmath104 identified with vertices of the tuple . for given vertex @xmath282 of @xmath328 such that @xmath340 $ ] ( or @xmath341 $ ] ) we will call vertex @xmath342 ( or @xmath343 respectively ) _ the vertex corresponding to @xmath282 in @xmath286_. note that this gives correspondence between vertices of @xmath286 and @xmath328 restricted to vertices @xmath334 $ ] and @xmath344 $ ] . [ hconstruct ] finite family finite relational structures is called _ minimal _ iff there all structures in @xmath78 are cores and there is no homomorphism in between two structures in @xmath78 . the following is the main result of this section . [ aritythm ] denote by @xmath78 minimal family of finite connected relational structures . there is lift of class @xmath77 that contains new relations of arity at most @xmath322 that is amalgamation class iff all minimal cuts of @xmath86 consist of at most @xmath322 vertices . construction of lifted class @xmath85 in proof of theorem @xmath345 adds relations of arities corresponding to the sizes of minimal cuts of @xmath86 so one direction of theorem [ aritythm ] follows directly from the proof of theorem @xmath345 . in the opposite direction fix class @xmath78 , @xmath346 and a relational structure @xmath86 containing minimal cut @xmath347 of size @xmath348 . assume , for contradiction , that there exists lift @xmath0 of class @xmath77 such that @xmath0 is an amalgamation class and contains new relations of arities at most @xmath322 . denote by @xmath308 the number of different relational structures on @xmath322 vertices appearing in @xmath0 . for the brevity , assume that @xmath349 has only two connected components . denote by @xmath106 and @xmath107 the pieces generated by @xmath96 such that @xmath350 . ( for 3 and more pieces we can proceed analogously . ) now we construct relational structure @xmath351 as follows : @xmath352 and put @xmath353 @xmath354 with no other tuples . in other words , @xmath351 is disjoint union of @xmath109 and @xmath110 . we will consider @xmath351 rooted by tuple @xmath355 take relational structure @xmath286 from lemma [ sets ] and put @xmath356 . this construction for the two pieces of 5-cycle is shown at figure [ hconstruct ] . for vertex @xmath300 denote by @xmath357 the vertex of @xmath88 corresponding to @xmath282 ( if it exists ) or arbitrary vertex of @xmath88 otherwise . denote by @xmath34 mapping @xmath358\mapsto a$ ] for @xmath359 and @xmath360\mapsto a$ ] for @xmath361 . it is easy to check that @xmath34 is homomorphism @xmath362 . additionally for @xmath363 put @xmath364 observe that for vertex @xmath363 such that @xmath365 , @xmath366 may contain multiple tuples , while for all other vertices @xmath366 contains precisely one tuple . assume , to the contrary , that there is homomorphism @xmath367 . by composition we have that @xmath368 is homomorphism @xmath369 . because @xmath198 is core , we also know that @xmath368 is automorphism of @xmath198 . it follows that @xmath60 is injective @xmath370 . for @xmath371 denote by @xmath366 the set @xmath372 where @xmath373 is uniquely defined by @xmath374 . it follows that for @xmath375 , @xmath366 consists of single tuple . for tuple @xmath376 , there is tuple @xmath377 iff sets @xmath378 have nonempty intersection ( i.e. all belongs to single copy of some piece @xmath130 ) and thus also sets @xmath379 have nonempty intersection . as relational systems @xmath380 are connected , it follows that all @xmath381 are equivalent singleton sets . denote by @xmath382 tuple such that @xmath383 for @xmath384 and by @xmath385 tuple such that @xmath386 for @xmath387 . because copies of pieces in @xmath88 corresponding to single tuple @xmath388 are not connected , we have @xmath389 . finally because every vertex in @xmath96 is connected to @xmath390 by some tuple ( from minimality of cut @xmath96 ) , we have @xmath391 for every @xmath392 and analogously @xmath393 for every @xmath392 . it follows that sets @xmath382 and @xmath385 overlaps on whole @xmath96 . this @xmath382 and @xmath206 overlaps on @xmath322 or more vertices that is a contradiction with construction of relational system @xmath286 . it follows that there is no homomorphism @xmath394 . there is also no homomorphism @xmath395 for any @xmath396 because composing such homomorphism with @xmath34 would lead to homomorphism @xmath397 that does not exist . it follows that @xmath196 . take generic lift @xmath398 . every embedding @xmath399 ( @xmath138 is shadow of @xmath59 ) imply @xmath308 coloring of @xmath322-tuples from elements of @xmath400 ( colors are defined by the additional relations of @xmath59 ) and thus also @xmath308 coloring of @xmath322-tuples of @xmath286 . subsequently , using lemma [ sets ] , there is tuple @xmath401 , such that @xmath402 and the relations added by lift @xmath0 are equivalent on @xmath403 and @xmath404 . thus @xmath59 induce on both sets @xmath405 and @xmath406 same lift @xmath49 . ( @xmath49 is lift of relational structure induced by @xmath198 on @xmath96 . ) subsequently there is partial isomorphism of @xmath59 mapping @xmath407 . from genericity of the @xmath59 this partial isomorphism extends to automorphism @xmath408 of @xmath59 . from the construction of relational system @xmath88 this mapping @xmath408 sends a root of image of piece @xmath118 to corresponding roots of image of piece @xmath115 and thus shadow of @xmath59 contains copy of @xmath86 , a contradiction . by theorem [ aritythm ] it follows that only minimal classes of finite relational structures @xmath78 such that class @xmath77 has monadic lift that forms an amalgamation class are precisely classes @xmath78 such that all minimal vertex cuts of their gaifman graph have size 1 . examples forming an amalgamation class include graphs with all its blocks being complete graphs . consider even more restricted classes @xmath78 of structures consisting from ( relational ) trees only ( relational trees are defined bellow ) . in this case we can claim a much stronger result : there exists a finite universal object @xmath88 which is a retract of an universal structure @xmath59 . a constraint satisfaction problem ( csp ) is the following decision problem : instance : a finite structure @xmath21 question : does there exists a homomorphism @xmath409 ? we denote by @xmath410 the class of all finite structures @xmath21 with @xmath409 . it is easy to see that class @xmath410 coincides with a particular instance of lifts and shadow . a _ finite duality _ ( for structures of given type ) is any equation @xmath411 where @xmath78 is s finite set @xcite . @xmath88 is called _ dual of @xmath78_. we also write @xmath412 for dual of @xmath78 ( it is easy to see that @xmath412 is up to homomorphism equivalence uniquely determined ) . the pair @xmath413 is called _ dual pair_. in a sense duality is a simple constraint satisfaction problem : the existence of homomorphism into @xmath88 ( i.e. a @xmath88-coloring ) is equivalently characterized by a finite set of forbidden substructures . dualities play a role not only in the complexity problems but also in logic , model theory , partial orders and categories . particularly it follows from @xcite and @xcite that dualities coincide with those first order definable classes which are homomorphism closed . finite dualities for monadic lifts include all classes @xmath414 . we formulate this as follows : [ prop21 ] for a class @xmath0 of structures the following two statements are equivalent : 1 . @xmath415 for finite @xmath351 . 2 . there exists a class @xmath57 of monadic lifts such that 1 . shadow of @xmath57 is @xmath0 . 2 . @xmath416 where @xmath80 is a finite set of monadic covering lifts of edges ( i.e. every @xmath417 contains at most one non - unary tuple . ) while every vertex belongs to a unary lifted tuple . obviously imply 2 . in the opposite direction construct @xmath351 as follows : let @xmath418 be lift with vertex for every consistent combination of new relations @xmath419 and relations @xmath420 empty . now construct lift @xmath351 on same vertex set as @xmath418 with @xmath421 . put tuple @xmath422 iff the structure induced by @xmath206 is on @xmath418 with @xmath206 added to @xmath423 is in @xmath84 . in the language of dualities this amounts to saying that the classes @xmath410 are just classes described by shadow dualities of the simplest kind : forbidden lifts are just vertex colored edges . a ( relation ) tree can be defined as follows : the _ incidence graph _ @xmath424 of relational structure @xmath21 is the bipartite graph with parts @xmath26 and @xmath425 and edges @xmath426 $ ] such that @xmath427 . ( here we write @xmath428 when there exists an index @xmath14 such that @xmath429 ; @xmath430 is a multigraph . ) @xmath21 is called a _ tree _ when @xmath424 is a graph tree ( see e.g. @xcite ) . the definition of relational trees by the incidence graph @xmath424 allows us to use graph terminology for relation trees . finite dualities are characterized ( see @xcite ) for every type @xmath28 and for every set @xmath78 of relational trees there exists a dual @xmath28-structure @xmath412 . up to homomorphism equivalence there are no other dual pairs . [ dualpairs ] given @xmath78 various constructions of structure duals are known @xcite . it follows from this section that we have a yet another approach to this problem : [ nasdual ] let @xmath78 be a set of finite tree - structures , then there exists finite set of lifted structures @xmath80 with the following properties : * @xmath81 is amalgamation class ( and thus there is universal @xmath431 ) , * all lifts in @xmath81 are monadic , * @xmath432 is universal for @xmath0 , * @xmath83 has a finite retract @xmath433 and consequently @xmath434 is a dual of @xmath78 . observe that the inclusion minimal cuts of a relational tree are all having arity 1 . thus for fixed family @xmath78 of finite relational trees our theorem [ mainthm ] establishes the existence monadic lift that give an generic structure @xmath83 whose shadow is ( homomorphism ) universal for @xmath77 . this structure @xmath83 is countable . to get a dual , we find finite @xmath435 which is retract of @xmath83 and still there is homomorphism @xmath436 iff there is homomorphism @xmath437 . the set @xmath80 is given by construction in theorem @xmath438 . observe that every inclusion minimal covering set of every piece of tree is induced by single tuple and thus the class @xmath85 is defined by forbidden ( rooted ) homomorphism of structures covered by single tuple . this means that the generic graph @xmath83 has finite retract defined by all consistent combinations of new relations of its vertices . note that it is also possible to construct @xmath412 in a finite way without using the limit : for every possible combination of new relations on single vertex create single vertex of @xmath412 and then keep adding tuples as long as possible so @xmath412 is still in @xmath85 . finally , let us remark that one can prove that @xmath439 has a finite presentation in the sense of @xmath440 . [ aplikace ] the situation of course changes if we consider more complicated ( non - monadic and non - trees ) lifted classes @xmath81 and in fact it has been proved in @xcite that any np language is polynomally equivalent to a class @xmath441 for a finite set @xmath80 set of lifted structures . as these structures may be supposed to be connected ( by adding dummy edges ) we have the following corollary of theorem [ univthm ] : [ npu ] every np problem is polynomally equivalent to membership problem for a class @xmath63 where @xmath59 is shadow of an ultrahomogeneous structure @xmath439 . constraint satisfaction problems @xmath414 for countable templates @xmath351 were investigated in @xcite and it has been shown there that the basic results of so called `` algebraic method '' @xcite hold also for @xmath442-categorical templates @xcite . recently , @xcite proved that @xmath414 for general countable template @xmath351 encodes any problem in np . it is not known whether @xmath351 may be chosen @xmath442-categorical . corollary [ npu ] complements this : lifts and shadows enable to code any np language by membership problem for ages of shadows of ultrahomogeneous structures . we turn briefly our attention to binary lifts . this relates some of the earliest results on universal graphs with recently intensively studied urysohn spaces : [ urysohnsection ] we will consider a finite family @xmath78 consisting of graphs of odd cycles of lengths @xmath443 . as shown by @xcite ( see also @xcite ) those families have universal graphs in @xmath444 and as shown by @xcite those are the only classes defined by forbidding a finite set of cycles . these classes also form especially easy families of pieces . in fact each piece is an undirected path of length at most @xmath5 , where @xmath5 is the length of longest cycle in @xmath78 with both ends of the patch being roots . this allows particularly easy description of the lifted structure . we use the following definition which is motivated by metric spaces . when specialized to graphs , this definition is analogous to ( corrected form ) of an @xmath445-structure @xcite . however this approach also gives new easy description ( i.e. finite presentation ) of the urysohn space @xcite . [ evenoddp ] pair @xmath446 is considered to be _ even - odd pair _ if @xmath447 is even non - negative integer or @xmath17 and @xmath448 is odd non - negative integer or @xmath17 . for even - odd pairs @xmath446 and @xmath449 we say that @xmath450 iff @xmath451 and @xmath452 . consider @xmath453 and @xmath454 . put : @xmath455 for a set @xmath298 , function @xmath456 from @xmath298 to even - odd pairs is called _ even - odd distance function on @xmath298 _ if following conditions are satisfied : 1 . @xmath457 if and only if @xmath458 ; 2 . @xmath459 ; 3 . @xmath460 . finally pair @xmath461 where @xmath456 is even - odd distance function for @xmath298 is called _ even - odd metric space_. note that the even - odd metric spaces differ from usual notion of metric space primarily by the fact that the ordering of values of distance function is not linear , but forms a 2-dimensional partial order . some basic results about metric spaces are valid even in this setting . even - odd metric space can form a stronger version of the distance metric on the graph . for graph @xmath10 we can put @xmath462 where @xmath447 is length of the shortest walk of even length connecting @xmath226 and @xmath463 , while @xmath448 is the length of shortest walk of odd length . the even - odd distance metric specify length of all possible walks : for a graph @xmath10 and a even - odd distance metric @xmath456 we now have a walk connecting @xmath226 and @xmath463 of length @xmath447 iff @xmath464 such that @xmath465 for @xmath447 even or @xmath466 for @xmath447 odd . it is well known that universal and homogeneous metric space exists for several classes of metric spaces @xcite . analogously we have : [ ulema ] there exists a generic even - odd metric space @xmath59 . we prove that class @xmath467 of all even - odd metric spaces is an amalgamation class . to show that @xmath467 has the amalgamation property , take a free amalgam @xmath88 of even - odd metric spaces @xmath21 , @xmath61 over @xmath62 . this amalgam is not even - odd metric space , since some distances are not defined . we can however define a walk from @xmath289 to @xmath468 of length @xmath5 in @xmath88 as a sequence of vertices @xmath469 and distances @xmath470 such that @xmath471 and @xmath472 is present in the even - odd pair @xmath473 for @xmath474 . we produce even - odd metric space @xmath59 on same vertex set as @xmath88 where distance of some vertices @xmath475 is even - odd pair @xmath476 such that @xmath5 is the smallest even value such that there exists walk joining @xmath447 and @xmath448 of length @xmath5 in @xmath88 . @xmath477 is the smallest odd value such that there exists walk from @xmath447 to @xmath448 of length @xmath477 . it is easy to see that @xmath59 is even - odd metric space ( every triangular inequality is supported by the existence of a walk ) and other properties of amalgamation class follows from definition . the graphs omitting odd cycles up to length @xmath5 can be axiomatized by simple condition on their even - odd distance metric : there are no vertices @xmath478 such that @xmath462 where @xmath479 . denote by @xmath480 the class of all countable even - odd metric spaces such that there are no vertices @xmath478 such that @xmath462 with @xmath479 . the existence of universal and homogeneous even - odd metric space @xmath481 for class @xmath480 is simple consequence of lemma [ ulema ] . in fact @xmath482 is a subspace of @xmath59 induced by all those vertices @xmath282 of @xmath59 satisfying @xmath483 and @xmath484 . for metric space @xmath481 denote by @xmath485 a graph on vertex set @xmath486 where @xmath487 iff @xmath488 . for every choice of odd integer @xmath489 , @xmath490 is universal graph for class of all graphs omitting odd cycles of length at most @xmath5 . graph @xmath490 is omitting odd cycles up to length @xmath5 from the fact that any two vertices @xmath478 on a odd cycle of length @xmath14 have distance @xmath462 where @xmath491 is at most @xmath5 . now consider any countable graph @xmath492 omitting odd cycles of length at most @xmath5 . construct corresponding even - odd distance metric space @xmath493 . by universality argument @xmath493 is subspace of @xmath482 and thus also @xmath10 subgraph of @xmath490 . the explicit construction of the urysohn space , without using limit argument , is described in @xcite and same technique can be carried to even - odd metric . this is captured by the following definition . [ defu ] the vertices of @xmath494 are functions @xmath34 such that : 1 . the domain @xmath495 of @xmath34 is finite ( possibly empty ) set of functions and @xmath496 . 2 . the range of @xmath34 are even - odd pairs . 3 . for every @xmath497 and @xmath498 , we have @xmath499 . 4 . @xmath495 using metric @xmath500 defined bellow forms a even - odd metric space . @xmath34 defines an extension of even - odd metric space on vertices @xmath495 by adding a new vertex . this means that @xmath501 and for every @xmath502 we have @xmath503 and @xmath504 . the metric @xmath505 is defined by : 1 . if @xmath506 then @xmath507 2 . if @xmath508 then @xmath509 3 . if @xmath497 then @xmath510 4 . if none of above holds then @xmath511 + minimum is taken elementwise on the pairs . [ generic ] @xmath512 is the generic even - odd metric space . this may be seen as yet another incarnation of kattov functions @xcite , see also @xcite . we call pair @xmath513 a _ partition of structure _ @xmath514 if @xmath26 and @xmath155 are disjoint sets of vertices of @xmath514 and @xmath515 . we denote by @xmath21 the structure induced on @xmath26 by @xmath514 and by @xmath61 the structure induced on @xmath155 by @xmath514 . the structure @xmath514 is _ weakly indivisible _ if for every partion @xmath513 of @xmath100 for which some finite induced substructure of @xmath514 does not have copy in @xmath21 , there exists copy of @xmath514 in @xmath61 . for finite minimal family of finite structures @xmath78 , we call structure @xmath21 a _ minimal homomorphic image _ of @xmath86 iff @xmath21 is homorphic image of @xmath198 and every proper substructure of @xmath21 is in @xmath77 . the construction of universal structures as shadows of ultrahomogeneous makes this result particularly easy to apply to obtain indivisibility results for universal structures for classes @xmath77 . this leads to the following partial classification of classes @xmath78 that do admit weakly indivisible structure universal for @xmath77 . 1 . the class @xmath77 contains universal structure that is weakly indivisible if every vertex minimal cut @xmath96 of every homomorphic image @xmath21 of @xmath86 is of size at least 2 and additionally the structure induced by @xmath21 on @xmath96 is connected and has no cuts of size 1 . all universal structures @xmath59 are divisible if there is a structure @xmath21 which is a minimal homomorphic image of @xmath86 such that @xmath21 contains cut @xmath96 of size 1 . to prove @xmath516 , fix @xmath21 , a minimal homomorphic image of @xmath86 with vertex cut @xmath96 of size 1 . denote by @xmath517 all pieces of @xmath21 generated by cut @xmath96 . fix @xmath59 , universal structure for @xmath77 . denote by @xmath518 , @xmath519 set of all vertices @xmath282 of @xmath59 such that there is rooted homomorphism of @xmath203 to @xmath59 mapping root of @xmath203 to @xmath520 . structure induced on @xmath518 on @xmath59 is not universal as it does not contains a homomorphic image of @xmath130 . similarly structure induced on @xmath521 is not universal since there is no homomorphic image of @xmath109 . as a result @xmath59 is divided to finitely many substructures such that none is universal resulting in divisibility of @xmath59 . @xmath522 follows from weak indivisibility of class @xmath85 . to apply theorem [ sauerthm ] we only need to show that class @xmath85 admits free vertex amalgamation this follows directly from the construction of amalgam in proof of theorem @xmath523 : the amalgam constructed is not free in general , every new tuple added has the property that there is homomorphism from the structure induced by cut generating the corresponding piece into the vertices of tuple . but since we have free amalgam of the shadow and since all cuts of all homomorphic images do not have cuts of size 1 , we have free vertex amalgamation property . 99 a. atserias : on digraph coloring problems and tree with duality . in : lics05 , ieee ( 2005 ) , 106115 . m. bodirsky , m. grohe : non - dichotomies in constraint satisfaction complexity ( in preparation ) . m. bodirsky , j. neetil : constraint satisfaction with countable homogeneous templates , ( m. baaz , j. a. makowsky , ed . ) computer science logic , 2003 , 4457 . a. bulatov , p. jeavons , a. krokhin : classifying the complexity of constraints using finite algebras . siam j. comput . 34 ( 2005 ) , 720742 . j. cameron : the random graph , in : the mathematics of paul erds ( ed . r. l. graham , j. neetil ) , springer verlag ( 1998 ) , 333351 . g. cherlin , s. shelah , n. shi : universal graphs with forbidden subgraphs and algebraic closure , advances in applied mathematics 22 ( 1999 ) , 454491 . g. cherlin , n. shi : graphs omitting a finite set of cycles , j. graph theory , 21 ( 1996 ) , 351355 . g. cherlin , n . shi : forbidden subgraphs and forbidden substurctures , j. symbolic logic , 66(3 ) ( 2203 ) , 13421352 . g. cherlin , n. shi : graphs omitting sums of complete graphs , j. graph theory , 24 ( 1997 ) , 237247 . g. cherlin , p. komjth : there is no universal countable pentagon free graph , j. graph theory , 18 ( 1994 ) , 337341 . d. delhomm , c. laflamme , m. pouzet , w. sauer : divisibility of countable metric spaces , european j. combin . 28 ( 6 ) ( 2007 ) , 17461769 . r. frass ' e : th ' eorie des relations , , north holland , 1986 . j. foniok , j. neetil , c. tardif : generalised dualities and finite maximal antichains , graph - theoretic concepts in computer science ( f.v . fomin , ed . ) , lecture notes in comput . ( 4271 ) , 2736 . a. hajnal , j. pach : monochromatic paths in infinite graphs , finite and infinite sets , coll . j. bolyai , 37 ( 1984 ) , ( eger , hungary , 1981 ) , 359369 . p. hell , j. neetil : graphs and homomorphisms , oxford university press ( 2004 ) . w. hodges : model theory , _ cambridge university press _ , 1993 . j. hubika , j. neetil : finite presentation of homogeneous graphs , posets and ramsey classes , israel j. math 149 ( 2005 ) , 2144 a. s. kechris , v. g. pestov , s. todorcevic : frass ' e limits , ramsey theory , and topological dynamics of automorphism groups . , 15 ( 2005 ) , 106 - 189 . p. komjth , a. mekler , j. pach : some universal graphs , israel j. math . 64(1988 ) , 158168 . komjth : some remarks on universal graphs , discrete math . 199 ( 1999 ) , 259265 . g. kun , j. neetil : np by means of lifts and shadows , mfcs07 , lecture notes in computer science 4708 , springer ( 2007 ) , 171181 . a. h. lachlan : on countable stable structures which are homogeneous for finite relational language , israel j. math . ( 1984 ) , 69153 . a. h. lachlan , r. e. woodrow : countable ultrahomogeneous undirected graphs , trans . 262 ( 1980 ) , 5194 . j. matouek , j. neetil : invitation to discrete mathematics , oxford univ . press 1998 , 2nd edition 2007 . j. neetil : for graphs there are only four types of hereditary ramsey classes , j. comb . b , 46(2 ) , ( 1989),127132 . j. neetil : ramsey theory . in : handbook of combinatorics ( ed . r. l. graham , m. gr " otschel , l. lov ' asz ) , elsevier , 1995 , 13311403 . j. neetil j. hubika : a finite presentation of the rational urysohn space , topology and applications 155 ( 14 ) ( 2008 ) , 14831492 . j. neetil , a. pultr : on classes of relations and graphs determined by subobjects and factorobjects , discrete math . 22 ( 1978 ) , 287300 . j. neetil , v. rdl : chromatically optimal rigid graphs , j. comb . b , 46(2 ) , ( 1989 ) , 133141 . j. neetil , v. rdl : simple proof of the existence of restricted ramsey graphs by means of a partite construction , combinatorica 1 ( 2 ) , ( 1981 ) 199202 . j. neetil , v. rdl : strong ramsey theorem for steiner systems , trans . ( 1987 ) , 183192 . j. neetil , c. tardif : duality theorems for finite structures ( characterizing gaps and good characterizations ) , j. comb . b , 80 , ( 2000 ) , 8097 . j. neetil , c. tardif : short answers to exponentially long questions : extremal aspects of homomorphism duality , siam j. discrete math . 19 , no . 4 ( 2005 ) , 914920 . l. nguyen van th : thorie de ramsey structurale des espaces mtriques et dynamique topologique does groupes disomtries , universit paris 7 , 2006 . b. rossman : existential positive types and preservation under homomorphisms . in : lics05 , ieee ( 2005 ) , 467476 . n. sauer : age and weak indivisibility , european j. combin . , submitted . v. v. uspenskij : on subgroups of minimal topological groups , topology and its applications 155 ( 14 ) ( 2008 ) , 15801606 .

processes in ocean physics , air sea interaction and ocean biogeochemistry span enormous ranges in spatial and temporal scales , that is , from molecular to planetary and from seconds to millennia . identifying and implementing sustainable human practices depend critically on our understandings of key aspects of ocean physics and ecology within these scale ranges . the set of all ocean data is distorted such that three- and four - dimensional ( i.e. , time - dependent ) in situ data are very sparse , while observations of surface and upper ocean properties from space - borne platforms have become abundant in the past few decades . precisions in observations of all types vary as well . in the face of these challenges , the interface between statistics and oceanography has proven to be a fruitful area for research and the development of useful models . with the recognition of the key importance of identifying , quantifying and managing uncertainty in data and models of ocean processes , a hierarchical perspective has become increasingly productive . as examples , we review a heterogeneous mix of studies from our own work demonstrating bayesian hierarchical model applications in ocean physics , air sea interaction , ocean forecasting and ocean ecosystem models . this review is by no means exhaustive and we have endeavored to identify hierarchical modeling work reported by others across the broad range of ocean - related topics reported in the statistical literature . we conclude by noting relevant ocean - statistics problems on the immediate research horizon , and some technical challenges they pose , for example , in terms of nonlinearity , dimensionality and computing . , ,

1312.5904

the global ocean dominates the iconic image of earth viewed from space , leading to the now famous `` blue marble '' descriptor for our planet . the ocean covers more than 70% of the planetary surface and ocean processes are critical to life - sustaining ( and life - challenging ) events and processes occurring across broad ranges of temporal and spatial scales . understanding issues of ocean resource consumption ( e.g. , fisheries , coastal pollution , etc . ) lead to foci on ocean ecosystem dynamics and their coupling to physical processes ( e.g. , mixing , transports , upwelling , ice dynamics , etc . ) . understanding the ocean role in climate dynamics ( e.g. , sequestration of atmospheric @xmath0 , absorption of atmospheric heat , impacts on the hydrologic cycle , teleconnections , etc . ) lead to foci on massive and complex simulations and forecasts based on equations of geophysical fluid dynamics . in all instances , the broad range of scales , the energetic exchanges across them , and the associated uncertainties drive innovations that involve methods of modern statistical modeling.=1 oceanography has historically been a `` data poor '' science and the need to use advanced statistical methodology to perform inference and prediction has been paramount throughout its development . although it is the case that there are too few in situ observations of the ocean to characterize its evolution and its interaction with marine ecosystems , in an ironic twist , the discipline also suffers from having an abundance of particular data types when one factors in the satellite observations that have become available in the last couple of decades . the need for statistical collaboration in oceanography results from both the situation of not having enough observations in some parts of the system and having huge amounts of data in other parts of the system . the physical ocean is governed by basic laws of physics ( see , e.g. , @xcite ) . the primitive equations consist of the following : three equations corresponding to the conservation of momentum ( for the two horizontal and one vertical components of velocity ) , a continuity equation representing the conservation of mass , an equation of state ( relating density , pressure , temperature and salinity ) , and equations corresponding to the conservation of temperature and salinity . there are seven state variables ( three velocity components , density , pressure , temperature and salinity ) . this system of equations is nonlinear and exhibits a huge range of spatial and temporal scales of variability . given that many of these scales of variability are not resolved in data or in deterministic or `` forward '' ocean models , the equations are typically simplified by scale analysis arguments and the small scale ( turbulent ) structures are parameterized . these parameterizations involve relationships between the mean of the state variables and their gradients . in this way the eddy viscosity and diffusivity terms serve as so - called `` sub - grid scale '' parameterizations , representing the unresolved processes that are sinks of momentum and heat at the grid scale of a given model , for example , @xmath1 km in global ocean models and @xmath2 km in regional ocean models . the ocean system is nonlinearly coupled to the atmosphere across a broad range of scales . at the largest scales , air sea fluxes of heat and fresh water drive mostly vertical or `` thermohaline '' circulations while the surface shear stress and wind stress curl drive mostly horizontal or wind - driven gyre circulations ( e.g. , @xcite ) . at smaller scales , the vertical - horizontal separation breaks down and the ocean response to external forcing and internal instabilities results in a broadband spectrum of vigorous eddy circulations ( e.g. , @xcite ) . on all scales , the circulation provides the context for ocean processes affecting other components of the ocean system such as those related to ocean biology and chemistry . there are significant nonlinear interactions between the ocean chemistry , biology and its physical state . ocean biogeochemistry is concerned with the interaction of the biology , chemistry and geology of the ocean ( e.g. , @xcite ) . this is a very complex system that contains many interactions across a variety of scales . the system can be simply illustrated by thinking about the interactions of broad classes of its components . for example , the presence of nutrients near the ocean surface , where there is light , allows for the growth of phytoplankton , which deplete the nutrients as their population expands . the increased abundance of phytoplankton then provides a food source for zooplankton , which leads to growth in the zooplankton population . the consumption of phytoplankton leads to waste products from the zooplankton that settles as detritus to the ocean floor . as the zooplankton deplete the phytoplankton , the zooplankton population decreases due to the lack of a sufficient food source . eventually , the detritus at depth is transferred to the surface through upwelling and mixing , providing the nutrients that lead to another bloom in phytoplankton , etc . this simple four component system is a vast oversimplification , as there are many different species interacting at any one time . more critically , this lower trophic ecosystem is also coupled to higher levels of the food web , for example , with foraging fish predating the zooplankton , which are in turn predated by higher trophic fish , marine mammals , commercial fishing , etc . the chemical component of the cycle is critical in several respects . it provides a way for carbon to be removed from the atmosphere , as the phytoplankton remove carbon from the ocean water and are consumed by the zooplankton . some of that carbon is contained in the detritus that sinks to the ocean floor , becoming buried in the sediment and leading to a ( temporary ) carbon sink in the global carbon cycle , which is very important in the context of sequestration of carbon relative to potential climate change from greenhouse gases . in addition , the biological cycle in the ocean is closely tied to the distribution of dissolved oxygen in the water and also influences the distribution of other chemicals , such as silicon , nitrates and phosphates , for example , in the shells of diatoms . given the complexities in the ocean system , it is not surprising that there are numerous sources of uncertainty . first , although the large - scale equations of motion are in some sense deterministic , the scale issues that lead to eddy viscosity / diffusivity parameterizations are inherently uncertain . furthermore , the forms of the linkages between system components ( e.g. , wind stress , heat and moisture fluxes between atmosphere and ocean ) are not known with certainty . the components of traditional biogeochemical models are even more uncertain , both in terms of the functional forms and parameters . the process and parameter uncertainty is compounded by the inherent data issues in the ocean system . in situ observations of the ocean are quite limited in terms of spatial and temporal resolution , and in terms of the variables measured . for example , it is a painstaking process to measure zooplankton abundance , often requiring a scientist or technician to literally count critters through a microscope . fortunately , many surface variables can be observed remotely , particularly through satellite proxies . in some cases , for example , near surface winds from scatterometers , sea surface height from altimeters and sea surface temperature ( sst ) from radiometers , the satellite observations are typically quite precise , albeit with gaps corresponding to orbital geometries , swath widths and fields of view . in other cases , for example , ocean color as a proxy for phytoplankton , the observational representation of the process is more uncertain , at least on fairly short time scales . thus , a key issue in state prediction , parameter estimation and inference is to deal with incomplete observations that vary in precision and in spatial and temporal support . this is particularly important when one considers ocean `` data assimilation , '' that is , the blending of prior information ( e.g. , a numerical solution of the deterministic representation of the ocean state ) with observations . another traditionally important component of uncertainty in ocean process modeling corresponds to the selection of reduced - dimensional representations of the process . given the assumption that much of the larger scale processes in the ocean can be represented in a lower dimensional manifold , with smaller scales corresponding to turbulent scales ( that certainly may interact with the larger scale modes , or at least suggest the form of parameterizations or stochastic noise terms ) , there has been considerable attention given to different approaches to obtain the reduced - dimension basis functions . the choices vary depending on the part of the system being considered as well as whether one is looking at the system diagnostically or predictively . in the context of statistical models used to describe or predict portions of the ocean system , the nature of the error structures is important . given the nonlinearity that is inherent in the system , many process distributions are not well represented bygaussian errors . in addition , in some cases ( e.g. , biological abundance variables ) the distributions can only have positive support . the ocean and atmospheric sciences have benefitted from a strong tradition in applying fairly complex statistical methods to deal with many of the uncertainty issues described above . in particular , general monographs such as @xcite , @xcite , and @xcite provide comprehensive descriptions of traditional methods used to analyze such data . in addition to overviews of basic statistical concepts , these books describe multivariate methods ( e.g. , principal components empirical orthogonal functions , canonical correlation analysis , discriminant analysis , etc . ) , spectral methods ( e.g. , cross - spectral analysis ) and dynamically - based reduction methods ( e.g. , principal oscillation patterns ) to facilitate analysis of high - dimensional data that has inherent dependence in time and space . this is in addition to more focused monographs such as @xcite(@xcite ) , which gives a comprehensive overview ofeigen - decomposition methods , and several books and review papers devoted to various aspects of data assimilation ( see section [ sec3.1 ] ) . statistical presentations of many of these methods can be found in @xcite and @xcite . recognizing the challenges related to uncertainty in the ocean system and the need to foster more collaborative research between oceanographers and statisticians , the u.s . national research council(nrc ) commissioned a panel to write a report on `` statistics and oceanography '' @xcite ; see also the accompanying article by @xcite and published comments . this report contains a very nice review of physical oceanography for nonoceanographers and outlined the need for research in several key areas , including the change of support problem and the indirect nature of satellite observations , non - gaussian random fields , the incorporation of lagrangian and eulerian data , data assimilation , inverse modeling , model / data comparison and feature identification , to name some of the most prominent . the report focused on the physical component of the ocean and did not address biogeochemistry nor many issues of current interest , such as climate change and reduced - dimensional representations . our goal with this review is to provide an overview of some of the advancements that have occurred at the interface of statistics and oceanography since the @xcite report . in particular , we believe strongly that the hierarchical statistical perspective has played a significant role in this development and will focus our review from that perspective . section [ sec2 ] presents a brief discussion of hierarchical modeling , both empirical and bayesian , with some discussion of the need for computational tools . we note that although this paper is in a statistics journal , we hope that it will generate interest from both statisticians and oceanographers . for the same reason that we gave a brief and general overview of oceanography above , we will also give a brief and general overview of hierarchical modeling for those readers with little exposure to these ideas . in section [ sec3 ] we focus in more depth on examples related to data assimilation and inverse modeling , long lead forecasting and uncertainty quantification in biogeochemical models . we will follow this review with a brief discussion of current and future challenges in section [ sec4 ] . the idea of hierarchical modeling of scientific processes arose largely out of @xcite , when mark berliner was the director of the geophysical statistics project at the national center for atmospheric research . the idea , although fundamentally quite simple , was revolutionary in that it provided a probabilistically consistent way to partition uncertainty in systems with complicated data , process and parameter relationships , and coincided with the development and popularization of markov chain monte carlo ( mcmc ) methods in bayesian statistics . as described below , the key idea is to consider the joint model of data , process and parameters as three general linked model components , that is , the data conditioned on the process and parameters , the process conditioned on parameters , and the parameters . these ideas quickly spread into statistics and subject matter journals in climatology , meteorology , oceanography and ecology , to name a few . this particular perspective on hierarchical modeling is summarized in several books , including @xcite , @xcite and @xcite . more traditional bayesian presentations of hierarchical models can be found in many books on bayesian statistics ( e.g. , @xcite ; @xcite ) . statistical modeling and analysis are about the synthesis of information . this information may come from expert opinion , physical laws , previous empirical results or various observations both direct and indirect . consider the case where we have a scientific process of interest , denoted by @xmath3 . as an example , say that this process corresponds to the near - surface north / south and east / west wind components over a portion of the ocean ( i.e. , a multivariate spatio - temporal process ) . we also have observed data associated with this process , say @xmath4 , which might come from a satellite - based scatterometer ( i.e. , wind component observations derived from speed and direction that are incomplete in space and time ) . we assume that we have parameters associated with the measurement process , say @xmath5 , that might represent differences in support and representativeness between the satellite observations and the true wind process at the resolution of interest . in addition , we assume that there are some parameters , say @xmath6 , that describe the underlying wind process dynamics ( i.e. , the evolution operator and innovation covariances that propagate the joint spatial fields of the wind components through time ) . thus , using the total law of probability , it is natural to write the decomposition of the joint distribution of the data and process conditioned on the parameters as @xmath7 = [ z | y , \theta_z ] [ y | \theta_y ] , \label{eq : zyhier}\ ] ] where @xmath8 $ ] is the `` data distribution '' ( or `` data model '' ) and @xmath9 $ ] is the `` process distribution '' ( or `` process model '' ) , and we have assumed conditional independence of the parameters on the distributions of the right - hand side ( rhs ) of ( [ eq : zyhier ] ) . note , we are using brackets `` @xmath10 $ ] '' to refer to a distribution and the vertical bar `` @xmath11 '' to denote `` conditioned upon . '' clearly , we could also consider an alternative decomposition in which @xmath3 is conditioned on @xmath4 followed by the marginal distribution of @xmath4 . however , such a decomposition is less scientific as described below . in traditional statistics , one might think about the data @xmath4 given some specified distributional form and some associated parameters , @xmath12 ( e.g. , corresponding to a spatio - temporal mean and associated variances and covariances , or their parameterization ) . in the context of ( [ eq : zyhier ] ) , such a distribution arises from integrating out the random @xmath3 process , yielding the distribution @xmath13 $ ] . we are then typically interested in estimating these parameters given the data . such estimation ( e.g. , maximum likelihood estimation ) does not include an explicit representation of a model for the underlying dynamical wind process , @xmath3 , but rather includes it implicitly through the first and second moments ( as a consequence of the integration ) . in addition , this distribution accounts for the uncertainty that is due to sampling and measurement . the question is then why might we be interested in @xmath3 ? first , in many such applications , one is actually interested in predicting the true , but unobserved process , @xmath3 , rather than just accounting for its ( co)variability . second , given the complexity of most ocean and atmospheric processes , the multivariate spatiotemporal dependence structures associated with @xmath3 can be very complicated ( e.g. , nonlinear in time , nonstationary in space and/or time ) and potentially very high - dimensional . this seriouslycomplicates the likelihood - based inference , as it puts much of the modeling burden on the realistic specification of complicated dependence structures . rather , by focusing attention on the process @xmath3 directly , one can incorporate scientific insight ( e.g. , markovian approximations to mathematical representations of the process , spatially or time - varying parameters , etc . ) and , critically , disentangle the measurement uncertainty ( which can also be quite complicated ) and the process ( co)variability and uncertainty . that is , marginal means and covariances contain potentially complicated functions of @xmath5 and @xmath6 , which can be difficult to specify without explicitly doing the integration of the random process . thus , one is effectively trading the complexity of specifying very complicated marginal dependence structures with a more scientific specification of the conditional mean as a random process at the next level of the hierarchy . this also allows one to focus effort on modeling the conditional error structure in the data stage , without having to try to disentangle the measurement uncertainty with the process ( co)variability . statisticians will recognize this as just a manifestation of the issue that one faces in traditional mixed - model analysis where one has a choice of considering a marginal model , whereby the random effects are integrated out , or a conditional model , whereby the random effects are predicted and the conditional covariance structure in the data model is simpler ( e.g. , @xcite ) . applying bayes rule to the hierarchical decomposition in ( [ eq : zyhier ] ) gives @xmath14 \propto[z | y , \theta_z ] [ y | \theta_y ] , \label{eq : zybayes}\ ] ] with the normalizing constant obtained by the integral of ( [ eq : zyhier ] ) with respect to the process @xmath3 . note that this implies that we are updating our knowledge of the process of interest given the data we have observed . this is the goal of prediction and confirms that the hierarchical decomposition on the rhs of ( [ eq : zybayes ] ) is the more plausible scientific decomposition of the joint distribution of data and process described above in ( [ eq : zyhier ] ) . clearly , ( [ eq : zybayes ] ) assumes that the parameters are known without uncertainty . this is seldom the case in reality , although one may have estimates of these parameters from some source and be comfortable with substituting them into ( [ eq : zybayes ] ) , leading to what @xcite refer to as an empirical hierarchical model ( ehm ) . alternatively , and more realistically for most ocean processes , at least some of the parameters are typically not known , and we are interested in learning more about them or , at least , would like to account for their uncertainty . in this case , we consider the fully - hierarchical or bayesian hierarchical model ( bhm ) : @xmath15 \propto[z | y , \theta_z ] [ y | \theta _ y ] [ \theta_z , \theta_y ] , \label{eq : zybhm}\ ] ] where one must specify a prior distribution for the parameters @xmath16 $ ] and we note that the normalizing constant integrates over the parameters in addition to the process . it is often the case that we have information to inform these parameter distributions . for example , in the case of the wind process described above , we have knowledge about the quality of scatterometer observations of wind components , and recognize that certain turbulent scaling laws must be followed , which can be incorporated through restrictions and informative prior distributions ( e.g. , @xcite ) . schematically , it is helpful to think about the rhs of ( [ eq : zybhm ] ) by using the following representation of berliner ( @xcite ) : @xmath17 \nonumber \\ & & \quad = [ \mbox{data } | \mbox{process } , \mbox{parameters } ] \\ & & \qquad { } \times [ \mbox{process } | \mbox{parameters } ] \times [ \mbox{parameters } ] . \nonumber\end{aligned}\ ] ] the strength of this hierarchical representation is that it provides a probabilistically consistent way to think about modeling complex systems while quantifying uncertainty . critically , each of the stages on the rhs of ( [ eq : zybhm ] ) or ( [ eq : bhm ] ) can be split into sub - models . for example , multiple data sets with differing supports can be accommodated with a model such as @xmath18 \nonumber \\[-8pt ] \\[-8pt ] \nonumber & & \quad = \bigl[z^{(1 ) } | y , \theta_{z^{(1 ) } } \bigr ] \bigl [ z^{(2 ) } | y , \theta_{z^{(2 ) } } \bigr],\end{aligned}\ ] ] where @xmath19 and @xmath20 correspond to data sets ( 1 ) and ( 2 ) , respectively , with associated parameters@xmath21 and @xmath22 . we note that ( [ eq : z1z2 ] ) makes the assumption that , conditioned on the true process , both data sets are independent . this is often a reasonable simplifying assumption and greatly facilitates the combination of data sets ( although this assumption should be verified in specific applications ) . the parameters and distributional form for the two distributions on the rhs of ( [ eq : z1z2 ] ) can be quite different , perhaps accommodating differing types of spatial or temporal support , and/or measurement and sampling errors . examples of this in the context of the wind example discussed previously are data from satellite scatterometers as well as data from ocean buoys or even weather center analysis winds ( e.g. , @xcite ) . it is also important to note that the process model component on the rhs of ( [ eq : zybhm ] ) or ( [ eq : bhm ] ) can also be decomposed into subcomponents . this could correspond to a hierarchical markov decomposition in time , for example , @xmath23 = \prod_{t=1}^t [ y_t | y_{t-1}][y_0]$ ] , where @xmath24 . or , it could correspond to a multivariate decomposition , @xmath23 = [ y^{(2 ) } | y^{(1)}][y^{(1)}]$ ] , where @xmath25 . in both cases , there would also be process model parameters . clearly , combinations of these types of distributions and other types of dependencies ( e.g. , spatial , spatiotemporal , etc . ) could be considered . in the context of the wind example , it would be natural for the wind components to be represented by @xmath26 and these could be conditioned on near surface atmospheric pressure fields , say @xmath27 , both of which would be spatiotemporal processes with further conditioning possible . last , we recognize that a huge advantage of hierarchical models in complicated systems is that the parameters are themselves endowed with potentially quite complex distributions . that is , they exhibit multivariate dependence between parameters or in terms of space and time , or may themselves be dependent on exogenous information . such complex dependencies in parameters are very difficult to accommodate in the classical paradigm . as discussed above , in the wind example , it is critical to specify parameter distributions that adhere to known empirical and theoretical turbulent scaling properties ( e.g. , @xcite ) . there is no free lunch ! although the hierarchical modeling paradigm is extremely powerful , it often comes at a substantial computational cost . in particular , the normalizing constant in ( [ eq : zybhm ] ) involves the integration over all random quantities in the model , which can correspond to a very high - dimensional integration in many applications . given that analytical solutions to these integrals are almost never available in complex models , one has to resort to numerical methods . in the fully bayesian context this is typically some type of markov chain monte carlo algorithm ( e.g. , @xcite ) . in the ehm case represented schematically in ( [ eq : zybayes ] ) , one may be able to work out alternative computational solutions [ e.g. , expectation - maximization ( e m ) or numerical maximum likelihood or method - of - moments estimation ; e.g. , see chapter 7 of @xcite ] . it is critical to note that the complexity of these calculations often leads to modifications in model structure to facilitate practical computation . there have been many examples of hierarchical modeling in the ocean sciences since the late 1990s . in this section we briefly review some of that work at the interface of statistics and oceanography . we focus most attention on two data assimilation examples ( ocean vector winds and ocean tracer state estimation ) , long lead forecasting of sst , and uncertainty quantification and assimilation of biogeochemical models . for these topics , we present vignettes to illustrate the power of hierarchical modeling from problems we have worked on . we also briefly describe some of the work at the interface related to other important oceanographic problems . @xcite summarize data assimilation ( da ) as an approach for optimally blending observations with prior information concerning the system ( i.e. , mathematical representations of mechanistic relationships , model output , etc . ) to obtain a distributional characterization of the true state of the system . relevant to this overview paper , these concepts originated in the atmospheric and ocean sciences and there is an extensive literature describing various methodologies ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . in essence , da can be thought of as an inverse problem , and one can derive algorithms from numerous perspectives , including optimal estimation theory , variational analysis and bayesian statistics . the bayesian approach to da ( e.g. , @xcite ; @xcite ) is helpful because the problem is inherently hierarchical and , thus , it provides a coherent probabilistic approach that can be used to describe the various approaches . in addition , if one is willing to consider the bhm perspective , one can often obtain a more realistic characterization of uncertainty in various components of the data and mechanistic models ( e.g. , @xcite).=-1 most oceanographic processes of interest in the da context concern spatial fields that vary with time . a general dynamical spatiotemporal model(dstm ) can be written in the bhm paradigm ( e.g. , @xcite ) . the data model is given by @xmath28 where @xmath29 represents the data at time @xmath30 , @xmath31 the associated process , where the mapping function , @xmath32 , may be linear or nonlinear , the error process , @xmath33 , may be additive or multiplicative , and the model depends on parameters given by @xmath34 that may be spatial or time - varying . the process model is given by @xmath35 where the evolution operator @xmath36 may be linear or nonlinear for the oceanographic process of interest , the error process , @xmath37 , may be additive or multiplicative , and the parameters , @xmath38 , may be spatial or time - varying . note that this model is assumed to be valid beyond the maximum data time - period ( @xmath39 ) , so that forecasting is appropriate . finally , the model is completed by the specification of distributions for the parameters in the previous stages , @xmath40[{{\bolds{\theta}}}_p(t)]$ ] , where we have assumed that the parameters from the two stages are independent for convenience . note , this hierarchical framework would also include a specification of the initial process distribution , @xmath41 $ ] . the sequential modeling perspective is a natural framework in which to consider the da problem , as one can interpret the prior distribution from above as a forecast distribution , which gets updated given new observations ( e.g. , see @xcite ) . surface winds directly transfer momentum to the ocean and surface wind speed modulates the exchanges of heat and fresh water to and from the upper ocean as modeled by bulk transfer formulae ( e.g. , @xcite ) . the advent of space - borne scatterometer instruments in the 1990s provided the first global wind fields , on daily timescales , from observations . prior to the scatterometer era , ocean winds were inferred from global weather forecast models and reanalyses ( e.g. , @xcite ) that depended upon a very sparse global network of in situ wind observations from buoys and ships of opportunity . reliable resolutions in the pre - scatterometer wind fields were limited to ocean basin scale features ( e.g. , associated with large - scale high and low pressure systems ) and monthly averages . the wind fields retrieved from scatterometer observations are not direct measures of the wind , but rather the observations are of the roughness imparted on the ocean by surface capillary waves in response to the shear stress vector at the air sea interface . the amplitudes and orientations of surface capillary waves are in equilibrium with the surface shear stress , and these amplitudes and orientations are retrievable from measures of how the capillary waves scatter impinging radar pulses of known frequency and polarization , that is , so - called microwave backscatter cross - sections ; for details , see @xcite . radar backscatter cross - sections are spatially averaged over wind - vector cells and related to a surface vector wind ( svw ) via a geophysical model function . the svw retrievals from scatterometers are accurate to within at least 2 ms@xmath42 in speed and @xmath43 in direction . depending on the sensor system and agency providing the data , resolutions are on the order of 12.550 km for up to @xmath44 global coverage on daily timescales . the svw retrievals occur in swaths along the polar - orbitting satellite ground track . swath widths vary by system from 6001800 km , that is , between about 20 and 100 wind - vector cells across a given swath depending on resolution . because of the polar orbit ( about 14 polar orbits per day ) the swaths overlap at high latitudes and are separated by gaps in coverage at low latitudes , with the largest swath gaps occurring at the equator . again , depending on the system , the swath gaps at the equator are on the order of @xmath45 degrees longitude and can take up to three days to fill . the svw from scatterometers resolve features of the synoptic storms ( e.g. , fronts , convergences in rain bands , closed circulations , etc . ) that form , propagate and dissipate every day , over the world ocean . to fill the gaps in the scatterometer winds , one could make use of the complete ( yet lower resolution ) wind fields from the operational meteorological centers . however , these wind fields have different properties than the scatterometer observations . differences in the true resolutions of weather - center analyses and scatterometer winds are efficiently described in terms of surface wind kinetic energy spectra , that is , kinetic energy as a function of spatial wavenumber . @xcite showed a power - law relation for surface winds from multiple data sources in the tropical pacific and the power law behavior in surface winds from scatterometry has been documented for the globe @xcite . @xcite compare kinetic energy between weather center analyses and svw retrievals and find that the kinetic energy drops off unrealistically in weather center analyses at smaller scales . for example , at spatial scales on the order of @xmath46 km , the weather center kinetic energy content is more than an order of magnitude weaker than the svw observations . the goal of a statistical data assimilation is then to blend the complete , but energy - deficient , weather center analyses with the incomplete , yet energy - realistic , svw in order to provide spatially complete wind fields at sub - daily intervals while managing the uncertainties associated with the different data sources and the blending procedure . @xcite implemented a spatiotemporal bhm for tropical winds in a region of the equatorial western pacific . as noted above , the inter - swath gaps in scatterometer coverage on daily timescales are largest in the tropics . a bhm formulation to blend weather - center analyses and scatterometerwinds requires space time properties that account for the greater intermittency in the scatterometer data . this is achieved in the process model stage in @xcite , which is posed in terms of three scientifically - motivated components as @xmath47 where @xmath48 is a vector of spatially - indexed zonal ( east west ) wind velocity components ( typically denoted by @xmath49 ) . analogous terms apply for the meridional velocity ( @xmath50 ) . here , @xmath51 is the mean zonal wind process that accounts for the prevailing wind and its variance in the tropical domain , @xmath52 is a basis function matrix corresponding to the large - scale modes of the equatorial beta - plane approximation ( i.e. , a linear , thin fluid approximation ) to the momentum equations , and @xmath53 are nested wavelet basis functions used to model fine - scale winds according to observed kinetic energy - wavenumber properties for the region . the use of scientifically - motivated basis functions is crucial here , as we have information from the thin - fluid analytical approximation to the wind field that is appropriate over the tropics . in particular , the leading equatorial normal modes [ @xmath54 are large - scale wave signals in the tropics , that is , westward propagating rossby waves , eastward kelvin waves and mixed rossby gravity waves ( e.g. , @xcite ) . based on data analyses ( e.g.,@xcite ) , the equatorial normal mode basis functions were limited to the leading modes defined by two zonal wavenumbers and four wavenumbers in the meridional direction @xcite . these are sufficient to support the important basin - scale kelvin , rossby and mixed rossby gravity waves that can be used to describe much of the tropical dynamics within the pacific basin . the amplitude coefficients , @xmath55 and @xmath56 , are treated as random variables in the bhm , with first - order markov models . the priors on the parameters associated with these models correspond to time series of the theoretical amplitudes for each mode for the @xmath55 coefficients and the slopes of the kinetic energy spectrum for the @xmath56 coefficients as obtained from @xcite . data stage inputs were obtained from nscat svw retrievals and from sea level pressure and surface winds taken from reanalyses of the national centers for environmental protection ( ncep ) . the posterior mean surface wind and surface convergence / divergence exhibited variability on scales not achievable in the ncep reanalyses . this is particularly evident in a comparison of the ncep reanalysis wind field with the posterior mean winds from the tropical wind bhm for the period when tropical cyclone dale crossed the study area ( figure [ 2paneltrop ] ) . the posterior mean winds are organized into strong convergence regions that spiral into the center of the tropical cyclone . these are consistent with rain band features of tropical cyclones and this was confirmed by comparing the posterior mean winds with cloud - top temperatures from an independent satellite observation nearly coincident with the snapshot from the posterior distribution of the bhm ( see @xcite ) . uncertainty management properties of svw bhm based on mechanisitic models ( i.e. , leading order terms and/or approximations of the primitive equations ) are particularly relevant in ocean forecast settings . the mediterranean forecast system ( mfs ) is an operational system producing 10-day forecasts for upper ocean fields every day . the mfs ocean forecast models resolve synoptic variability in the upper ocean . the uncertain parts of the forecast fields are at ocean mesoscales , that is , hourly and 1050 km scales . these are the scales of upper ocean hydrodynamic instabilities driven by the surface wind . so , modeling uncertainty in the surface wind field is a useful means of quantifying uncertainty in the mfs ocean forecasts on the scales that are most important to daily users . @xcite describe the details of the svw bhm for mfs , and @xcite review the impacts of bhm svw fields in an ensemble forecast methodology built around realizations from the posterior distribution for svw from the bhm . the mechanistic process model in @xcite involves the leading - order terms in a rayleigh friction equation approximation at synoptic scales and , again , a nested wavelet basis model to represent turbulent closure at the finest spatial scales . thus , again it is critical to incorporate scientific information into the specification of the process rather than try to model such behavior through the marginal covariance structure . this then also allows specification of the measurement uncertainties associated with the data conditioned upon the true process . specifically , data stage inputs are obtained from the quikscat scatterometer svw retrievals and from weather center sea level pressure and surface wind fields . ten realizations of the posterior distribution for svw were used to drive ensemble data assimilation and ensemble forecasts in the mfs . figure [ 2panelmed ] depicts the data assimilation system response in sea - surface height ( ssh ) at the forecast initial condition time . the mean ssh initial condition ( figure [ 2panelmed ] , top ) reflects the accurate spatial scales for mfs forecasts where synoptic variability overlies the general circulation patterns for the mediterranean sea . sub - basin scale cyclonic gyres are represented by blue ( depressed ) ssh and anticyclones by reds . the gradients between blue and red signals are proportional to surface current speeds . the spread in the ssh initial condition ( figure [ 2panelmed ] , bottom ) demonstrates the value of managing uncertainty in the svw . the largest amplitude signals in the spread are localized in a few places only where surface wind is driving hydrodynamic instabilities and pulsing the energy in ocean mesoscale response . inferring the structure of the circulation in the world s oceans is a significant part of our quest for understanding the climate . direct measurements of the circulation are difficult to obtain ; however , in the past decades , scientists have collected large amounts of hydrographic data ( hydrostatic pressure , temperature , salinity , oxygen concentration , etc . ) . such data have been used to produce climatological maps that exhibit the large scale structure ( lozier , owens and curry @xcite ) . based on these maps and data , one can infer some constituents of the ocean circulation , such as boundary currents , wind - driven gyres and abyssal interior flow . yet , hydrographic data are typically available at very sparse locations in space and time . for example , the world ocean circulation experiment lasted from 1990 until 1998 . in figure [ fig1 ] we illustrate these data that were available for the south atlantic ocean . consequently , statistical smoothing methods are required to provide estimates at space time locations of interest . to that end , one would typically combine data sets collected at different times . this seems reasonable only under the assumption that the distribution of dynamically - passive tracers is representative of a mean circulation and that such tracers are sufficiently stable on a fixed neutral density layer . a classical approach to inverting hydrographic data into circulation structure is the _ box inverse method _ starting with the thermal wind equations , the problem reduces to estimation of a reference velocity field . this is achieved via a large system of linear equations expressing conservation of mass within a collection of connected boxes . due to the size of the problem , a high - dimensional regression approach is subsequently employed . traditionally , ridge regression estimators have proved useful , however , modern statistical / machine learning methods can provide more effective alternatives . other approaches are based on more complex inverse modeling and nonlinear optimization ( @xcite ; @xcite ; @xcite ; @xcite ) . the inverse problem described here is ill - posed and some type of regularization is required @xcite . we describe the bayesian inversion approach of @xcite ( see also @xcite ) , which connects the data ( tracer concentrations ) to parameters ( ocean circulation ) using a system of partial differential equations . in this case , regularization is imposed through a prior distribution over the parameters . the _ solution _ is the posterior distribution . it can be used to select `` representative '' values and the associated uncertainty for any aspect of the circulation . formally , let @xmath57 denote the concentration of a tracer of interest at a location @xmath58 in a rectangular domain @xmath59 representing the layer of the ocean being studied . the problem is to estimate the ( horizontal ) water velocities and diffusion coefficients based on noisy measurements of @xmath60 at a sparse set of locations in @xmath61 . in the right panel of figure [ fig1 ] we display an interpolated map of oxygen concentration for a @xmath62 m deep layer in the south atlantic ocean . the connection between the tracer concentration @xmath60 and the velocities and diffusion coefficients is modeled by the steady - state advection - diffusion equation @xmath63 where @xmath64 is the horizontal water velocity , the diagonal diffusivity matrix @xmath65 does not vary with location , and the sink term @xmath66 is present only when @xmath60 represents oxygen concentration . equation ( [ ad ] ) is augmented with dirichlet boundary conditions @xmath67 when @xmath68 , where @xmath69 denotes the boundary of @xmath61 . the statistical model assumes additive observational error @xmath70 here , @xmath71 indexes a particular tracer and @xmath72 indexes a spatial location where data for tracer @xmath73 are available . the underlying tracer concentration @xmath74 is obtained as a solution of the advection diffusion equation ( [ ad ] ) . as this solution is not available in closed form , one uses a numerical ( grid - based ) approximation . we collect all quantities of interest ( velocities , diffusion coefficients , boundary values ) in a high - dimensional vector @xmath75 . under the that the measurement errors @xmath76 are unbiased gaussian variables with constant ( yet tracer - dependent ) variance , the posterior distribution is written as @xmath77 where @xmath78 represents the selected prior distribution . the posterior probability model ( [ post0 ] ) is explored via monte carlo methods ( e.g. , @xcite ) . the ill - posedness mentioned above is resolved by specifying a proper prior distribution @xmath78 . this implies that the posterior distribution is proper . however , an efficient mcmc approach is still required . in addition , one can design specific sampling strategies as described in @xcite and @xcite . it is important to understand that each component described above ( physical model , prior distribution , likelihood function , data ) plays a crucial role in determining the solution . under the bayesian approach , a sensitivity analysis , although possible , is hampered by the immense computational cost associated with the mcmc sampler . in addition , the under - determination problem is present in this case . there are(roughly ) 300 data points , while there are thousands of parameters to estimate ( velocities , diffusions , boundary values ) . in the results given in @xcite and @xcite , not all estimated velocities are significantly different from zero , which is the prior mean . however , the data are informative about the large - scale features ( zonal jets , gyres ) . the posterior mean velocities are compared with velocities determined from float data @xcite , showing a comforting consistency ( @xcite and @xcite ) . tropical pacific sst exhibits some of the most important variability on inter - annual time scales for the ocean / atmosphere system ( e.g. , see the overview in @xcite ) . this variation arises from complicated interactions , across a large range of spatiotemporal scales , between the ocean and the atmosphere . the most prominent signal on these time scales is the el nio - southern oscillation ( enso ) phenomenon . this is the well - known quasi - periodic ( 35 year period ) warming ( el nio ) and cooling ( la nia ) in the tropical pacific . these warming and cooling events lead to dramatic effects in weather across the globe due to teleconnections with the global atmospheric circulation . because of these significant weather related impacts ( e.g. , droughts , floods , etc . ) , it is critical to be able to forecast several months in advance the possible development and transition of these events . increasingly , such``long lead '' forecasts have shown useful skill and correspond to one of the few situations in ocean science where a purely statistical forecast methodology is competitive with , and in many cases better than , equivalent deterministic model forecasts ( @xcite ; @xcite ) . at typical spatial resolutions , there can easily be several thousand gridded spatial locations corre - sponding to the tropical pacific region of forecasting interest . complicated spatiotemporal statistical models are difficult or impossible to implement at these dimensions . for this reason , statistically - based models for tropical sst have been `` spectral '' in the sense that they are based on coefficients associated with a projection of the sst fields on spatial basis functions . the associated projection coefficients that are evolved are typically a reduced set , usually corresponding to larger modes . that is , let @xmath48 correspond to an @xmath79-dimensional vector of the true sst process at @xmath79 spatial locations and time @xmath30 and consider the decomposition of this process vector @xmath80 where @xmath81 , is an @xmath82 matrix of spatial basis functions , and @xmath83 and @xmath84 are the associated expansion coefficients . a first - order linear markov assumption on the evolution of the coefficients @xmath83 , for example , @xmath85 , for appropriate time increment , @xmath86 , and with gaussian errors , @xmath87 , was shown in the early 1990s to be a model with reasonable skill at long - lead forecasting ( e.g. , @xcite ) . as with the tropical wind case presented previously , the specification of the process in terms of two basis sets is critical from a scientific perspective . in particular , it is thought that the active dynamical process is driven by a lower dimensional manifold ( corresponding to the first set of basis functions ) and , yet , the residual spatially - dependent structures captured by the second set of basis functions remain an important source of variability . although dimension - reduced linear markov models showed reasonable skill , the enso phenomenon is better characterized as a nonlinear process ( e.g. , @xcite ; @xcite ) . nonlinear statistical models typically are better at capturing the magnitude of the predicted el nio and la nia events ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . most nonlinear stochastic methods have not successfully characterized the uncertainty in the forecasts . an exception to this is the model of @xcite , who use a dimension - reduced threshold markov model with certain components governed by the onset of intra - seasonal oscillations ( so - called `` westerly wind bursts '' ) . there are several key components of the model in @xcite that illustrate the power of hierarchical modeling . first , the data model is constructed @xmath88 where the data vector , @xmath89 , is very high - dimensional ( say , @xmath90 ) , and @xmath52 is a reduced - dimension set of spatial basis functions as described above ( an @xmath91 matrix ) . it is important to note that the remaining basis functions ( e.g. , @xmath53 from above ) associated with the basis expansion are used to parameterize the observational spatial covariance matrix , @xmath92 , which now accounts for both observation error and the truncation . the assumption is that the active dynamics exist on the lower dimensional manifold represented by @xmath52 and , like in typical turbulence parameterizations , the small scale structures accounted for by the @xmath53 portion of the expansion are associated with nonpredictive covariability . critically , the evolution of the dynamical process is nonlinear , but conditionally linear . specifically , @xmath93 where prior distributions are also specified for @xmath94 . the important modeling assumption is that @xmath95 and @xmath96 depend on both the current ( time @xmath30 ) and future ( time @xmath97 ) climate `` regimes , '' that is , @xmath98 and @xmath99 , where @xmath100 classifies the current sst regime as `` cool , '' `` normal '' or `` warm , '' and @xmath101 anticipates one of these three regimes at time @xmath97 . specifically , @xmath100 is based on the southern oscillation index and @xmath101 is based on a latent variable that is modeled , hierarchically , in terms of the east - west component of the near surface wind in the western pacific ocean ( for details , see @xcite ) . the point is that this leads to nine potential mean states and nine potential dynamical operators , accommodating the nonlinear transition between enso phases . because this model was developed in a bayesian hierarchical framework , it was able to capture the uncertainty characterized by the data , process and parameters . however , despite the fact that this model used physical notions ( i.e. , westerly wind bursts ) in the lower stages of the model hierarchy , it was unable to directly account for quadratic interactions across spatial scales . as mentioned previously , nonlinearity is important in most atmosphere and ocean processes . @xcite demonstrate the effectiveness of a quadratic nonlinear model for long - lead prediction of enso from a classical regression perspective . @xcite demonstrate the implementation of a quadratic nonlinear model in the context of a bayesian hierarchical framework in which the random nonlinear process is `` hidden '' and parameters are random as well . they illustrate that simple arguments from turbulence scaling support the form of this model and further suggest a dimension reduction strategy . as defined in @xcite , general quadratic nonlinearity ( gqn ) with respect to the @xmath83 process can be written @xmath102 for @xmath103 , where @xmath104 is some transformation of @xmath105 that depends on parameters @xmath106 and gives the process more generality than the simple dyadic interactions alone . this framework is exceptionally flexible in accommodating many real - world mechanistic processes , but it comes at the cost of a curse of dimensionality in the parameter space ; that is , there are @xmath107 parameters corresponding to the nonlinear coefficients ( @xmath108 ) , in addition to the linear coefficients ( @xmath109 ) and @xmath106 . one can use scale analysis to help with the dimensionality concerns , and also to motivate components of the hierarchical model . specifically , as before , assume we can decompose the spectral coefficients into large - scale components ( @xmath83 ) and small - scale coefficients ( @xmath84 ) corresponding to the spatial basis functions @xmath52 and @xmath53 discussed above . consider all of the possible dyadic interactions of the elements of this vector that is , there are small - scale / small - scale , small - scale / large - scale and large - scale / large - scale interactions . loosely motivated by `` reynolds averaging '' in turbulence theory ( e.g. , @xcite ) , @xcite suggest considering the large - scale / large - scale pairwise interactions explicitly , with the small - scale / small - scale interactions contributing to a correlated structure in the additive error , and the small - scale / large - scale interactions corresponding to the linear term in the large - scale coefficients with random parameters ( i.e. , the small - scale coefficients play the role of random coefficients in the ) . such an argument leads to a quadratic nonlinear model on the large - scale coefficients , which can be written in matrix form as @xmath110 \\[-8pt ] \eqntext{{{\bolds{\eta}}}_t \sim \operatorname{gau}({{\mathbf{0}}},{{\mathbf{q}}}).}\end{aligned}\ ] ] prior distributions are then given to the parameters in @xmath111 and @xmath112 as well as the covariance matrix @xmath113 ( e.g. , @xcite ) . this approach was shown to quite reasonably account for the uncertainties so that the prediction error bounds covered the extreme enso events , even when the forecasts did not adequately capture the true magnitude . critically , these quadratic nonlinear implementations still suffer from a fairly high curse of dimensionality in the parameters . in this sense , various model reduction approaches have to be implemented that necessarily fail to account for as much model uncertainty as is likely present . it is difficult to know a priori which quadratic interactions are important . to alleviate this curse of dimensionality and to provide a natural framework in which to `` average '' across the various model specifications , @xcite employed a stochastic search variable selection methodology ( e.g. , @xcite , @xcite ) . in general , although the gqn statistical models are very flexible in representing oceanographic processes , these models can easily experience finite - time blow up ( i.e. , explosive growth ) when fit to data @xcite . this is seldom an issue when one is using these models for data assimilation , given the presence of observations to act as a control ( e.g. , @xcite ) , nor is it typically a problem when one is only forecasting out one time step ( e.g. , as in the sst example ) . it can be a problem when multiple time steps are forecast that require some form of constraint , either statistical or physical @xcite . analysis of marine ecosystem dynamics involves various sources of uncertainty in the observations , the underlying scientific process and the parameters that describe the process dynamics . critically , it is often the case that the observation errors are non - gaussian and that the process being modeled is nonlinear , that is , there is an explicit system of coupled nonlinear differential equations that describe the complex ecosystem dynamics . as a result of these uncertainties and complexities , the bhm framework is natural , but can be difficult to implement . soon after the introduction of mcmc methods into bayesian computation , the bhm approach was used in data assimilation for marine ecosystem models . anticipating the forthcoming increase in re - motely - sensed ocean color observations , @xcite implemented an mcmc - based sampling protocol that explored the ability to recover various model parameters in a seven - compartment marine ecosystem model with and without added model noise . they noted the computational difficulties required to adequately account for correlation in these parameters . while identifying correlation in only ten parameters may no longer be a computational issue , there are still issues related to the sampling methodology that adequately generates reasonable block proposals for the entire parameter vector . adaptive metropolis hastings algorithms , which update the covariance of the proposal distribution , may be useful when trying to generate adequate proposals for a nonlinear marine ecosystem model ( see , e.g. , @xcite ) . current research has taken into account advances in sampling methodology , including the following : sequential importance resampling , particle filters , ensemble kalman filters and state - augmentation approaches ( @xcite , @xcite;@xcite , @xcite ; @xcite ; @xcite ) . an innovative practice in statistical modeling of ocean ecosystems is the inclusion of information from deterministic , mechanistic models into the statistical framework , typically in the process stage of a bhm . however , the necessary estimation procedures often require iteratively running the mechanistic model , which poses a problem when the model is computationally expensive and can only be run a very limited number of times . in certain situations where the computer model is too computationally expensive to run a sufficient number of times for the desired analysis , statistical surrogates ( i.e. , emulators ) are used . in its simplest form , an emulator is simply the resulting estimated statistical model when computer model output is used as data . then , this model is used to predict the output of the computer model under untried input settings ( e.g. , initial conditions , parameter values , forcings ) . traditionally , so - called `` computer model '' emulation has been done using gaussian process ( gp ) models ( e.g. , @xcite ; @xcite ) . in our case , these `` computer models '' are deterministic ocean forward models . gp emulators are related to spatial gps , which use a correlation function such that the model output is more highly correlated for inputs that are `` nearer '' to one another in a given sense than those that are `` farther apart . '' however , because ocean ecosystem models are nonlinear , a gp emulator of the joint distribution of the output may be inappropriate ( as a nonlinear process can not be specified by only two moments ) . in this case , one could consider the use of a dynamic gp emulator , which considers the value of the process at the current time step ( i.e. , the initial conditions ) as an input to the ocean forward model ( and gp emulator ) . this offers several benefits over the traditional gp emulation approach ( e.g. , @xcite ; @xcite ) . rather than modeling the dynamics through a covariance function in a gp , it may be more appropriate to model the output using first - order characteristics ( as the computer models themselves are written ) . the use of so - called `` first - order emulators '' has appeared in several applications , for example , @xcite and @xcite use neural networks and @xcite use random forests , a machine learning algorithm . there is also the potential to use other nonlinear statistical models , polynomial chaos expansions ( mattern , fennel and dowd , @xcite ) or gp models in a dynamical context @xcite . these approaches use dimension - reduction techniques to overcome the curse of dimensionality , with the emulator describing the dynamics of the reduced - dimensional process . most of these implementations use nonparametric approaches . alternatively , @xcite use an emulator based on a reduced - rank multivariate quadratic nonlinear statistical model as described above . related to the use of emulators for dynamic models , there is also a growing body of work considering the use of emulators for spatial or spatiotemporal forward model output . marine ecosystem models can vary from nonspatial , 0-d models , to models that have spatial structure . @xcite used a 1-d ( in the vertical ) four - component model that included nutrients , zooplankton , phytoplankton and detritus ( e.g. , a `` 1-d nzpd '' model ) with iron limitation ( i.e. , `` 1-d npzdfe '' ) to model a process by creating a forest of 1-d models ( more specifically , a forest of emulators of the 1-d model ) . these 1-d models resolve vertical processes , but do not account for horizontal diffusion and advection . variability resulting from horizontal advection and diffusion is accommodated by putting a spatial gp model on the parameters ( inputs ) to the 1-dnpzdfe model . in certain circumstances , the forward model emulator is developed simply to learn about the behavior of the forward model itself . however , in other situations , the forward model emulator may be useful in a data assimilation context . in those circumstances , the emulator may be used in a two - stage approach , where the emulator is fit `` off - line , '' and then is used inside a bayesian hierarchical model in the place of the forward model itself . @xcite and @xcite developed bhms using forward model output as data . @xcite performed data assimilation using seawifs ocean color observations , as well as sea surface height and sst output from the regional ocean model system ( roms ) forward model coupled with the npzdfe model ( @xcite ) . they consider a reduced - dimension quadratic nonlinear process model similar to ( [ eq : yphiphi ] ) and ( [ eq : gqnvec ] ) , but where @xmath114 , representing the three state variables of interest . basis functions @xmath52 and @xmath53 were based on output from the roms - npzdfe model for a different time period using a singular value decomposition and the priors for @xmath115 and @xmath116 were developed using the remaining right singular vectors . @xcite show that this approach was sufficient to accommodate nonlinear dynamics even in the absence of observations over a substantial portion of the domain as shown in figure [ fig.bio ] . space limitations prevent us from giving complete overviews of all hierarchical statistical models inoceanography . however , we mention here some notable and important papers that have been published in recent years in areas outside of those mentioned above . by necessity , this list is not exhaustive , neither in terms of topics nor citations within topics , but it does give some sense of the breadth and depth of work being done in the area . as mentioned above , the amount of in situ data for ocean state variables is fairly limited and there is a need to construct complete spatial fields representing climatological features of the ocean state , as well as the uncertainty in those states using spatial and spatiotemporal models . there are many notable examples from a bhm perspective ( e.g. , @xcite ; @xcite ; lemosand sans , @xcite ; @xcite(@xcite ) ) . it has long been known that there are strong teleconnections between the ocean and the atmosphere and this has led to useful statistical models for prediction ( e.g. , @xcite ; @xcite ; @xcite ) . there is an increasingly large number of papers that have used the bhm perspective to model teleconnections as well as their impacts . for example , @xcite used such a model when evaluating the impacts of sst on tornado report counts in the eastern two - thirds of the us ; @xcite ( and references therein ) have shown great success in modeling hurricane activity and forecasts ( see also @xcite ) ; @xcite use a bhm to model precipitation considering ocean conditions . in addition , there are recent studies showing linkages between the physical ocean and ecological impacts ( e.g. , @xcite ; @xcite ) . bayesian hierarchical models have long been used to model the higher trophic levels of the marine ecosystem . for example , stock - recruitment models have been an important tool in marine fisheries management ( e.g. , @xcite ; @xcite ; @xcite(@xcite ) ; @xcite ; @xcite(@xcite ) ; @xcite ) . in addition , bhms have been used extensively in recent years to model distributions and movement of marine mammals ( e.g. , @xcite ; ver hoef andjansen ( @xcite ) ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . clearly , a critical problem of vast societal interest concerns climate change . given the role of the ocean in the climate system and its direct connections to weather events and ecological impacts , oceanic climate change and its uncertainty characterization are extremely important . this is a vast research area with many papers that have taken a hierarchical bayesian approach . this topic is beyond the scope of this review , but a few notable examples include @xcite , @xcite , @xcite , @xcite and @xcite . as demonstrated throughout this review , the ocean system is quite complex with numerous interacting subcomponents and external processes , and with substantial uncertainty in terms of knowledge and data . statistical methods have been critically important to improve our understanding of this system and to characterize uncertainty . in recent years , the hierarchical statistical modeling paradigm has proven to be an exceptionally useful tool to manage the various sources of uncertainty . we have only just scratched the surface in terms of our presentation of the work that has been done in this area , but it is clear that the use of hierarchical modeling in oceanography has blossomed . that being said , in addition to continued development in the areas mentioned above , there are still many important problems at the interface of statistics and oceanography to be considered and methodologies to be explored . one of the biggest challenges for statisticians working in oceanography is to develop statistical models that can effectively parameterize the complex nonlinear interactions that are associated with the ocean system . in particular , models must be developed to account for potentially high - dimensional state - processes , and yet effectively manage uncertainty with varying data quality and non - gaussian errors . in general , to be effective in this context , dimension reduction should not be independent of the physical and biological environment of the problem under consideration . one component of nonlinearity that is important concerns the coupling of subsystems , whether that be the atmosphere / ocean interface , the ocean / ice interface , the physical / biological interface or interactions between trophic levels in the ocean . many of these couplings require parameterizations for fluxes across boundaries , and there is often only limited observational data available to help inform the process . this presents an extreme challenge and one in which we must use the large amounts of remotely sensed observations along with the sparse in situ data to build these nonlinear relationships . not unrelated is the notion of accounting for and describing model error . the precise characterization of uncertainty that is fundamental in bhm can be used to help identify and characterize model error in deterministic models of the ocean system , for example , ocean forecast models . as with all model abstractions , deterministic approaches accept trade - offs in resolution and approximations of the ocean variability to gain affordability in simulations and forecasts . because the ocean system and the models are inherently nonlinear , the errors introduced by acceptable approximations can grow and lead to model error properties that are difficult to diagnose . parameters of posterior distributions from independent bhm analyses for ocean model response or control variables can be used as a standard against which deterministic model error can be quantified . for example , if the incremental adjustment of the surface momentum flux control vector in a variational data assimilation procedure pushes the momentum flux at a point outside the reasonable bounds of a posterior distribution for momentum flux from a bhm , the variational procedure is probably compensating for forecast model error as opposed to uncertainty in the control vector . to gain understanding of the impact of physical processes on biota as well as the interaction of biological organisms , there is increasing reliance on individual ( or agent - based ) models . these models have become more prevalent in the ecological realm and are starting to be considered from the bhm perspective ( e.g. , @xcite ) . the use of these models in a statistical context across the spectrum of scales and processes involved in oceanography is a growing area of interest . for example , @xcite link physical ocean models to individual - based bioenergetic models for fish . the issues described above will almost certainly require new computational strategies , particularly in the bayesian paradigm . as the models get more and more complex and larger data sets become available , standard mcmc methods begin to fail to provide meaningful results . among the many challenges associated with bayesian computing , two stand out : ( 1 ) there are very high - dimensional distributions to be explored , and ( 2 ) the models are complex , which leads to inexact and sometimes impossible likelihood evaluations . novel mcmc methodology will address these issues , while maintaining feasibility . for example , particle mcmc ( pmcmc ) methods are designed to address the first issue . for high - dimensional posterior distributions ( thousands ofstate variables and parameters ) , it is nearly impossible to design good metropolis hastings proposal distributions and in this case , even adaptive mcmc may fail . @xcite propose to use sequential monte carlo ( smc ) methods combined with importance sampling to design near optimal proposal distributions . the resulting algorithm , which has similar features to a particle filter , will update the entire collection of state variables at once and can be extremely useful for space time models . @xcite use pmcmc for state and parameter estimation as well as state forecasting in the context of a marine biogeochemical model . in addition , hamiltonian mcmc methodologies may provide an attractive alternative for situations for which samples from complicated high - dimensional processes and parameter distributions are required ( e.g. , @xcite ) . approximate bayes computing ( abc ) methods are a new and emerging class of likelihood - freemcmc algorithms . they address the second issue above that is , cases when evaluation of the likelihood function involves integrals over very large spaces that are impossible to calculate ( e.g. , @xcite ) . abc relies on the ability to simulate from the selected model without much computing effort . consequently , the user is forced to select a relevant ( multivariate ) statistic and `` posterior samples '' are defined as parameter values that result in a statistic similar to the one observed . this algorithm , in fact , explores the distribution of the variables of interest _ conditional on the selected statistic _ , not the data . although this may raise some concern regarding the interpretation of the results , @xcite describe a semi - automatic way of selecting good summary statistics in a general setting . in general , it seems likely that for inference and prediction of complicated oceanographic processes in the foreseeable future , one will have to continue to make compromises between model complexity and computational feasibility . indeed , one must be willing to accept some reasonable lack of `` optimality '' in the solution in order to make headway on many of these problems . this is true regardless of whether one takes a bayesian or frequentist perspective . the hierarchical paradigm helps to some extent as it allows one to consider trade - offs between approximate computational strategies , incorporation of scientific information and model specification for the different model components ( data , process and parameters ) separately . in conclusion , there is a long history of activity at the interface of statistics and oceanography . in recent years , the hierarchical statistical paradigm has proven to be very helpful for managing the uncertainty associated with data , process and parameters in modeling the ocean and its related systems . there are increasingly more oceanographers with advanced statistical training and more statisticians with oceanographic backgrounds and this is sure to lead to even more innovative and exciting methodological developments in years to come . the authors wish to thank the editor , ae and reviewer who provided substantial suggestions to improve the manuscript . in addition , wikle acknowledges the support of national science foundation ( nsf ) grants dms-10 - 49093 , oce-0814934 and office of naval research ( onr ) grant onr - n00014 - 10 - 0518 . herbei acknowledges support from nsfgrant dms-12 - 09142 and onr grant n00014 - 10 - 1 - 0488 . milliff acknowledges early support for bhm research from the physical oceanography program , onr ; from the nasa , international ocean vector winds science team , and recent support from us globec under grant number nsf oce-0815030 . it is a pleasure to acknowledge our bhm mentor l. mark berliner and the inspirations he provides dating back to his days as director , geophysical statistics project at ncar through the present . we are also pleased to acknowledge our colleagues and collaborators who participate in vigorous cross - disciplinary discussions at summer confabs in boulder , co. in recent years , these include , professors berliner , mevin b. hooten , andrew m. moore , nadia pinardi and thomas m. powell , and doctors jeremiah brown , manny fiadeiro and jerome fiechter .

in this work[multiblock footnote omitted ] the curvature of weinhold ( thermodynamical ) metric is studied in the case of systems with two thermodynamical degrees of freedom . conditions for the gauss curvature to be zero , positive or negative are worked out . signature change of the weinhold metric and the corresponding singular behavior of the curvature at the phase boundaries are studied . cases of systems with the constant , including ideal and van der waals gases , and that of berthelot gas are discussed in detail .

math-ph0505010

usage of geometrical methods in homogeneous thermodynamics started in the works by j.gibbs ( @xcite ) and c.caratheodory ( @xcite ) was further developed in the works of r.hermann , r.mrugala , in the dissertation of h.heemeyer ( @xcite ) and in other works . thermodynamical metrics also have their source in the works of j. gibbs ( @xcite ) . explicitly a thermodynamical metric ( td - metric ) was introduced by f.weinhold ( @xcite ) and , later , from a different point of view , by g.ruppeiner ( @xcite ) . deeper studies by p.salamon and his collaborators , by p.mrugala and h.janyszek ( see @xcite ) clarified principal properties of thermodynamical metrics , relations between different td - metrics and their place in relation to the contact structure of equilibrium thermodynamical phase space ( @xcite ) . g. ruppeiner ( see review @xcite and the bibliography cited there ) has developed a covariant thermodynamical fluctuation theory based on the riemannian metric @xmath2 defined by the second momenta of entropy with respect to the fluctuations and related the curvature of this metric to the correlational volume near the critical point . he applied this scheme to a variety of models : ideal gas , ideal paramagnetic , ising model , takahashi gas , van der waals gas . p.mrugala , h. janyszek , p.salamon studied the role of thermodynamical metrics in the statistical thermodynamics , one of examples being the model magnetic system ( @xcite ) . these and some other developments , including an interest to thermodynamical metrics in the study of black holes led us to the question of a more systematic study of the curvature of thermodynamical metrics in the case of systems of smallest dimension where such curvature is present - the case where the phase space of td - system is one with variables @xmath3 . degeneracy of a td metric @xmath4 due to the homogeneity of a constitutive law ( see @xcite or see below ) is removed by considering 1 mole of the media ( or by the reduction procedure , @xcite ) . as a result , a td - metric is defined on the 2d constitutive surface ( containing all equilibria states of the system ) . all the intrinsic curvature properties of this metric are defined by the scalar ( gauss ) curvature @xmath0 . in section 2 we introduce the thermodynamical metrics @xmath5 defined by a thermodynamical potential @xmath6 on the corresponding `` constitutive surfaces '' in the phase space of a thermodynamical system endowed with the contact structure ( @xcite ) . in section 3 we introduce the elementary thermodynamical system ( of two td degrees of freedom ) and calculate its weinhold and ruppeiner metrics . in addition we present examples of td - metrics of a chemical system and the systems with magnetic and electrical properties . in section 4 we calculate levi - civita connection and the curvature tensor of the metric @xmath5 in the general n - dim case(see also @xcite ) . in section 5 we introduce and study the hessian surface @xmath7 of the td potential @xmath6 in the space of symmetrical @xmath8 matrices and express the curvature properties of the metric @xmath5 in terms of geometrical properties of this surface . in particular , we determine the condition for the positivity , negativity and nullity of the scalar curvature @xmath9 in terms of this surface . in section 6 we continue the study in of curvature started in section 4 by calculating the determinant @xmath10 ( characterizing the signature of the metric ) and the scalar curvature of the weinhold metric @xmath11 ( defined by the internal energy @xmath12 ) of an elementary thermodynamical system in terms of @xmath13 , heat capacity @xmath1 , volume coefficient of expansion @xmath14 , isothermal compressibility @xmath15 and their derivatives . in sec.7 we consider the case of * constant @xmath1 * where calculations simplify greatly . we define the form of the fundamental constitutive relations @xmath16 in such a case and determine the systems of this type for which @xmath17 and the regions where scalar curvature is positive or negative . in sec.8 , example of the ideal gas is considered . in sec.9 , we study the weinhold metric for the van der waals gas , curve of the signature change ( degeneracy curve of the @xmath18 ) , separating the domain of ( stable ) equilibria from the unstable region , scalar curvature behavior near this curve , calculation of critical values of physical variables in terms of the degeneracy curve . in sec.11 the case of berthelot gas is considered . in the last section 12 the geodesic equations of the weinhold metric @xmath11 for the general elementary thermodynamical system are obtained . in this section we introduce , following @xcite , some geometrical structure of on the ( energy - phase ) state space of homogenous thermodynamics and , in particular , the thermodynamical metrics ( @xcite ) on the constitutive surfaces of this system - the principal object of our study . we introduce the * thermodynamical phase space * @xmath19 of a thermodynamical system with @xmath15 `` thermodynamical degrees of freedom '' as a @xmath20-dimensional manifold endowed with a contact structure - 2d - distribution @xmath21 on @xmath19 which is totally non - integrable , @xcite . this structure is presented ( locally , and in many cases globally ) by a contact ( pfaff ) form @xmath22 such that @xmath23 at each point of @xmath19 . the fundamental geometrical structure of a thermodynamical system is a choice of a canonical chart ( darboux chart ) @xmath24 of variables @xmath25 for the 1-form @xmath26 , realizing it in the form @xmath27 couples of variables @xmath28 have the meaning of * extensive * ( @xmath29 ) and * intensive * ( @xmath30 ) variables , corresponding to the different processes that may undergo in the system and @xmath6 is the chosen thermodynamical potential ( internal energy , entropy , helmholtz free energy or entalpy , see @xcite ) . examples of couples @xmath28 are : 1 ) temperature and entropy @xmath31 ; 2 ) pressure and volume @xmath32 ; 3 ) mole number of @xmath33-th component and corresponding electrochemical potential @xmath34 ; etc . horizontal distribution @xmath21 of contact structure - subbundle of the tangent bundle @xmath35 @xmath36 is endowed with the symplectic form @xmath37 having , in a canonical chart @xmath24 , the standard form @xmath38 admissible processes described by the system are * horizontal * curves @xmath39 of structure @xmath22 ( i.e. such that @xmath40 for all @xmath41 ) . in the case where @xmath6 is interpreted as the internal energy of the system , @xmath42 is the work applied ( or produced ) during the process @xmath43 , see ( @xcite ) , and condition of admissibility ensures the fulfillment of the energy conservation law during the process . * constitutive surfaces * are defined as maximal integral surfaces of structure @xmath22 ( legendre submanifolds ) , ( @xcite ) . standard way to present such a surface @xmath44 is to define it as the first jet of a function @xmath45\backslash i ) $ ] , defined in the open subset @xmath46 . here we choose a subset @xmath47 $ ] of indices and take @xmath6 to be a function of @xmath29 with @xmath48 and of @xmath49 with @xmath50 in the complemental set of indices @xmath51\backslash i$ ] . relation @xmath52 defines the fundamental constitutive relation of td system , ( see @xcite ) , from which all others equations of state follows ( @xcite ) . in the usually considered case where @xmath53 $ ] corresponding constitutive surface is given by @xmath54,\ ( x^{i } ) \in d\subset r^{k}\}.\ ] ] from now on we will denote arguments of a function @xmath6 defining a fundamental constitutive relation and the corresponding equilibria surface by @xmath55 $ ] . constitutive surface @xmath44 defined by a function @xmath56 contains the subsets @xmath57 of * equilibria states * of the homogeneous thermodynamical system @xmath58 subject to the ( possible ) constraints imposed on the system in the situation @xmath59 ( see @xcite or other standard text in the thermodynamics for a variety of such situations ) . subset @xmath60 is determined , due to the second law of thermodynamics , by the condition of extremum of the thermodynamical potential @xmath6 subject to the constraints @xmath59 , see @xcite . submanifold @xmath44 contains also unstable states as well as _ locally stable _ states ( @xcite ) . partition of the constitutive surface @xmath44 corresponding to a given family of constraints was studied in many cases ( see @xcite ) . geometry of the constitutive surface @xmath44 , of its equilibria region(s ) @xmath61 in relation to the admissible thermodynamical processes is one of the primary object of study of `` geometrical thermodynamics '' , see @xcite ) . thermodynamical metric ( td - metric ) defined by the constitutive relation @xmath58 on the constitutive surface @xmath44 of the contact structure @xmath26 has the form @xmath62 for the case where @xmath6 is the internal energy @xmath12 , metric @xmath11 is called the weinhold metric . for the case where @xmath6 is the entropy @xmath63 , instead , metric @xmath2 is called the ruppeiner metric . notice that this form of the metric is not invariant under the diffeomorphisms of domain @xmath21 ( except linear ones , see @xcite for more ) . thus , though is not a drawback , td - metric is defined by a physically meaningful choice of canonical chart @xmath24 and transformation of this metric under the transition from one physically meaningful canonical chart to another one can not be simple . as an example , we mention that the legendre transformation of the space @xmath19 leading to the replacement of internal energy @xmath12 by the entropy @xmath63 applied to the td - metric @xmath11 gives , as is shown in @xcite , the td - metric , conformal to the metric @xmath2 on the equilibria surface @xmath64 , @xmath65 metric @xmath5 with @xmath66 being the * internal energy * of a thermodynamical system , was introduced by f.weinhold in series of papers @xcite . it was further studied by p.salaman , s.berry and their collaborators ( see @xcite and the bibliography in @xcite ) . a td - metric @xmath5 is induced on the surface @xmath44 of the form ( 2.1 ) by the following symmetrical tensor @xmath67 this tensor is the only ( up to a conformal factor ) symmetrical tensor in @xmath19 , annihilating the reeb vector field @xmath68 of structure @xmath22 ( here @xmath69 ) , invariant under the substitution of indices @xmath33 , and obtained as the sum of symmetrical tensors in the 2d - elementary subspaces @xmath70 of @xmath71 spanned by couples of covectors @xmath72 of thermodynamically conjugated variables . a very interesting way to construct all the td - metrics @xmath5 in the framework of the contact geometry was suggested by r.mrugala ( @xcite ) . he has defined ( starting from some arguments of statistical mechanics ) the contact metric @xmath73 on the phase space @xmath19 inducing td metric ( 2.5 ) on each equilibria surfaces . geometrical properties of this metric are studied in the forthcoming work of the second author with j.vargo @xcite . function @xmath6 defining constitutive relation and the constitutive surface @xmath44 is assumed to be homogeneous of the first order : @xmath74,@xcite . as a result , weinhold metric is degenerate - its kernel is generated by the radial vector field in the space @xmath75 @xmath76 this can be remedied either restricting all considerations to the subspace of @xmath75 ( and the section of cone @xmath44 ) with fixed value of one of variables @xmath29 ( mole number or volume are typical examples , see @xcite ) or using the geometrical reduction of the surface @xmath77 by the action of this dilatation group ( @xcite ) . after this reduction of the td - system its weinhold metric is in general non - definite - the constitutive surface @xmath44 is the union of domains of where this metric has different signature separated by the submanifolds ( generically of codimension one ) of states where this metric is degenerate . submanifold of thermodynamical equilibria in the processes with a given set of constrains @xmath59 - @xmath61 of the constitutive surface @xmath44 lays the region of the * definite td - metric * due to the second law of thermodynamics which requires the entropy to be ( locally ) maximal and energy to be ( locally ) minimal in the stable equilibria state to which system tends . gauss curvature @xmath0 of thermodynamical metrics ( in energy or entropy form ) was calculated for several td systems with two degrees of freedom ( i.e. 5-dim phase space @xmath19 ) - ideal gas ( @xmath78 ) , takahashi gas , multi - component ideal gas , paramagnetic ideal gas , van - der - waals gas , ideal quantum gases(see review @xcite ) . g. ruppeiner ( see review @xcite and the bibliography cited there ) related the curvature @xmath79 of the metric defined by entropy to the correlational volume near the critical point . the fact that for the ideal gas the curvature of weinhold metric @xmath11 and that of ruppeiner metric @xmath2 is zero and the calculations of curvature for systems mentioned above allows to suggest some relation between this curvature ( or at least its sign or nullity ) and the interactions undergoing in the system on the microscopic level . clarification of these questions requires deeper study of intrinsic and especially extrinsic geometry of constitutive submanifolds @xmath44 of the thermodynamical phase space @xmath19 . here we present several examples of td - metrics including the basic * elementary thermodynamical system * with 5-dim phase space @xmath19 in variables @xmath80 as the first example of a thermodynamical system and its weinhold and ruppeiner metrics we consider the system with 5-dimensional phase - energy space of variables @xmath81 which we will call from now on an * elementary thermodynamical system*. this system will be the principal object of our study here , so we will consider it in more details . contact form @xmath22 of such a system can be chosen in a form @xmath82 consider a 2d integral submanifold @xmath83 of this system defined by a constitutive relation @xmath84 with @xmath63 and @xmath85 being our extensive variables , @xmath86 and @xmath87 the relative conjugate intensities and @xmath12 the internal energy function of the extensive variables . then @xmath88 notice that homogeneity condition for any constitutive relation in u - system ( @xcite ) leads to the relation @xmath89 that should be fulfilled for all integral surfaces of contact structure @xmath22 given by a constitutive equation . correspondingly , contact condition for thermodynamical processes has two forms , i.e. @xmath90 where the second one ( gibbs - duhem equation , see @xcite ) leads to the second form of constitutive relation provided the constitutive relation in the first ( fundamental ) form is chosen . calculating the weinhold metric of u - system we will use some notations standard in the literature ( see @xcite ) . namely , we denote 1 . @xmath1 is the heat capacity at constant volume : @xmath91 2 . @xmath92 is the heat capacity at constant pressure : @xmath93 @xmath94 is the volume coefficient of expansion : @xmath95 4 . @xmath15 is the isothermal compressibility : @xmath96 these quantities are related by : @xmath97 calculating weinhold metric for u - system on the two - dimensional integral surface @xmath98 and using introduced notations we get @xmath99 characteristic equation of tensor @xmath100 ( with respect to the canonical euclidian metric @xmath101 on the @xmath102-plane ) is @xmath103 and @xmath104\ ] ] for the discriminant @xmath105 we have the following inequality @xmath106 which , in particular , implies that @xmath107 notice that since @xmath108 , discriminant @xmath109 is always positive . therefore eigenvalues @xmath110 are both real and distinct . moreover , since @xmath111 , and since also @xmath112 , we get the following let @xmath113 . then @xmath114 if @xmath115 and @xmath116 then @xmath117 and @xmath118 . @xmath119 if @xmath120 and @xmath121 then @xmath117 and @xmath122 . @xmath123 if @xmath1 and @xmath124 have different sign then @xmath125 and @xmath126 , @xmath127 . fulfillment of 1 ) and 2 ) guarantees the stability for the system even if the case 2 ) might look non physical at all . however , there are some cases when it is natural to assume that the heat capacity is negative ( in a cluster of sodium atoms or in the black holes , @xcite ) . there are also cases in which the isothermal ( and adiabatic ) compressibility seems to be negative ( in amino acid , @xcite ) . these are examples of cases away from stability . in case of an ideal gas , @xmath128 and @xmath129 , and the metric @xmath11 is given by @xmath130 applying legendre s transformation from variables @xmath81 to the variables @xmath131 to the contact form @xmath22 and rewriting the constitutive relation correspondingly , we get @xmath132 = 0\rightarrow \omega = ds-\frac{1}{t}du-\frac{p}{t}dv,\ ] ] and @xmath133 correspondingly , td - metric on the two - dimensional integral surface @xmath134 ( ruppeiner metric ) takes the form , see also @xcite , @xmath135 in case of an ideal gas , metric ( 3.17 ) takes the form @xmath136 due to the gibbs - duhem form of constitutive relation @xmath137 . here we consider a ( 2m+5 ) homogeneous thermodynamical system in energy - phase space defined by @xmath138 with @xmath139 being the number of molecules of type @xmath33 and @xmath140 being their conjugate intensities - chemical potential per molecule i - th . consider ( m+2)-dim integral surface defined by an entropy constitutive relation @xmath141 we get @xmath142 @xmath143 then we calculate @xmath144 constitutive relation should now include the work of magnetic and electric field on the ( homogeneous ) system . let s consider the work of magnetic field first . fundamental result of the thermodynamics of magnetic system defines the differential of the magnetic work to be ( @xcite ) : @xmath145 where h is the external magnetic field , m is the magnetization ( magnetic momentum ) vector . first term represents the change of energy of the magnetic field , second , the work of the magnetic field on the magnetic momentum of the system . since we consider here only homogeneous systems , we take h to be constant . let also assume that the magnetization @xmath146 is homogeneous . then @xmath147 where @xmath148 is the total magnetic dipole moment of the system . considering just the contribution to the magnetic work due to the magnetic dipole moment @xmath149 , let s define the ( internal ) energy to be @xmath150 where @xmath151 is the component of the total magnetic moment parallel to the external field . then @xmath152 @xmath153 where @xmath154 is the permittivity of free space . then the weinhold metric for the system with constitutive relation ( 3.26 ) in the space of variables @xmath155 is @xmath156 similarly , for the work of a homogeneous external electrical field @xmath157 in homogeneous case we have ( @xcite ) @xmath158 where @xmath159 is the total electric dipole moment . similarly to the previous case , if we consider just the contribution to the electric work due to the electric dipole moment , we can define the internal energy as follows @xmath160 then @xmath161 where @xmath162 @xmath163 is the component of the electric moment parallel to the external field . thus , the system consisting of several chemical components in the presence of an electrical filed takes the form @xmath164 in this section , we consider a thermodynamical potential @xmath165 as function of the extensive variables @xmath166 and calculate the curvature of the corresponding thermodynamical metric defined on the integral ( constitutive ) surface @xmath44 , by the equation ( 2.3 ) @xmath167 christoffel symbols for this metric are given by @xmath168 where @xmath169 after some calculations , it can be shown that the curvature tensor of metric @xmath5 is given by @xmath170 therefore , ricci tensor of metric @xmath5 is @xmath171 and its scalar curvature @xmath172 is , see also @xcite , @xmath173 in this section we study the hessian representation of a metric @xmath5 of a 5-dim thermodynamical system with two `` degrees of freedom '' ( using the terminology of mechanics ) @xmath174 in terms of an oriented 2d surface in 3d space of symmetrical @xmath175 matrices . we find a geometrical conditions on this surface determining regions where scalar curvature of @xmath5 is positive , negative , zero , singular . in later sections we demonstrate , on examples of ideal , van der waals and berthelot gases - to what physical consequences these properties of metric @xmath11 lead to . for @xmath176 and a metric @xmath177 , we can use the known formula for scalar curvature ( @xcite , ch.viii , sec.2 ) @xmath178 , \ ] ] which , for the weinhold metric reduces to the first term only : @xmath179 this expression shows that the scalar curvature goes to infinity as @xmath180 notice that the condition on a function @xmath6 that the metric @xmath5 is * degenerate everywhere * , i.e. @xmath181 is the simplest * monge - ampere equation*,@xcite . geometry on the surfaces defined by such a function is very interesting and well studied ( see , for instance @xcite and the bibliography therein ) but has little relation to the case where @xmath6 is a thermodynamical potential of a physically interesting system . scalar curvature @xmath0 is zero if and only if the determinant in the numerator of expression ( 5.2 ) is zero . to determine where it can take place , consider the * hessian mapping * @xmath182 of a function @xmath6 mapping a 2d ( canonically oriented ) domain @xmath21 of variables @xmath183 into the 3d vector space of 2x2 symmetrical matrices @xmath184 with the coordinates @xmath100 : @xmath185 image of this mapping is the 2d oriented surface @xmath7 in the vector space @xmath186 this surface is endowed with the frame formed by the coordinate tangent vectors to this surface @xmath187 at the nonsingular points of the surface where these vectors are linearly independent . this determines orientation of the surface @xmath7 , or , what is the same , field of normal vectors @xmath188 where @xmath189 is the standard vector product in @xmath190 . notice that , given a function @xmath6 , the physically relevant domain @xmath191 where mapping @xmath192 is defined can be smaller then the natural domain @xmath193 of the function @xmath6 . as an example , let us note the condition of * positivity * of some physical quantities ( volume , pressure ) or other lower boundary ( kelvin temperature ) condition . sometimes it is reasonable to assume that function @xmath6 is defined on the boundary of domain @xmath21 . correspondingly , not all the surface @xmath7 is interesting from the point of view of physics . example of such a situation is given below . condition of degeneracy of a matrix defines in the space @xmath184 the standard conic ( nilpotent cone ) @xmath194 subset of degeneracy @xmath195 of the metric @xmath5 is the pre - image under the @xmath192 of the intersection of surface @xmath7 and the cone ( 5.4 ) . as the first example of the @xmath196 consider the ideal gas whose internal energy @xmath12 as a function of variables @xmath197 and the mole number @xmath198 has the form @xmath199 see @xcite , chapter 6 . for one mole , taking @xmath200 we find @xmath201 excluding variables @xmath197 from the formula for weinhold metric , it is easy to get equation for the surface @xmath202 @xmath203 thus the surface @xmath204 is trivially isomorphic to the standard * conic * @xmath205 in @xmath206 . conic @xmath207 defines the homogeneous nonlinear second order ( monge - ampere ) pde @xmath208 satisfied by the energy function @xmath12 of the ideal gas ( see theorem 2 below ) . comparing equations ( 5.4 ) and ( 5.6 ) we see that the signature degeneration may happen on the hessian surface of ideal gas only in * nonphysical * points @xmath209 or under the condition @xmath210 , i.e. when @xmath211 . as the second example , consider the van der waals gas with molar internal energy ( 9.6 ) and the weinhold metric ( 9.8 ) . excluding @xmath212 and @xmath85 from the equation ( 9.8 ) we get equation for the hessian surface @xmath202 for the vdw gas : @xmath213 expression on the right is the ( linear ) correction to the ideal gas equation for the hessian surface ensuring nonzero curvature almost everywhere on this surface . multiplying by the denominator we get the algebraic equation for @xmath7 @xmath214 given by the * polynomial of the fifth order*. intersection of this surface with the surface @xmath215 determines the curve of the signature change ( critical curve ) for the vdw gas ( in this representation ) @xmath216=0\ ] ] it is interesting to notice that in these two examples the hessian surface @xmath7 is turned to be * algebraic*. in general case of @xmath1-const ( see the constitutive equation ( 7.3 ) ) this is not so anymore . 0.4 cm equality of the determinant in the numerator of the formula for curvature to zero is equivalent to the linear dependency of radius vector @xmath217 of a point of the surface and two tangent vectors @xmath218 . endow the space @xmath184 with the canonical metric @xmath219 induced by the killing metric on the lie algebra @xmath220 , @xcite . choosing the orthonormal basis : @xmath221 allows to identify @xmath184 with @xmath206 @xmath222 and to introduce scalar product with the standard euclidian scalar product @xmath223 . then , the determinant in the numerator of formula ( 5.2 ) takes the form @xmath224 where @xmath225 is the normal vector to the surface @xmath7 introduced above . thus , we can use alternatively picture of surface @xmath7 in @xmath184 and the standard euclidian @xmath190-representation of this surface and other objects . we will use standard cartesian coordinates @xmath226 in @xmath206 . let @xmath227 be an equation of a surface @xmath228 in @xmath229 satisfying , on some open set @xmath230 where the surface @xmath231 is smooth ( nonsingular ) , to the condition above : radius vector @xmath232 at each point of the surface is parallel ( and belongs ) to the tangent plane of the surface at this point . this means , in particular , that tangent planes at each point of the surface pass through the origin . since radius vector @xmath232 at the point of the surface @xmath233 belongs to the tangent plane of the surface at that point , then @xmath234 at each point of the domain @xmath21 . from this it follows that @xmath235 on the subset @xmath21 of the surface @xmath227 . in other words , radial vector field @xmath236 is tangent to the surface @xmath227 . therefore , subset @xmath21 of this surface is invariant under the ( local ) action of the group of dilations - phase flow of vector field @xmath237 . thus , in an open subset @xmath21 where @xmath78 , surface @xmath238 is * conical * and vice versa . consider now the projectivization @xmath239 of the 3d vector space @xmath186 image of the conical surface @xmath238 in this projective space is one - dimensional and , near nonsingular points ( on an open subset of this image ) is a smooth curve . in homogeneous coordinates @xmath226 of @xmath240 , this curve is ( locally ) given by an equation @xmath241 with a * homogeneous function * @xmath242 . this equation is also the equation of the surface @xmath243 near almost all points . vice versa , if a surface in @xmath184 is given by a homogeneous equation of the type ( 5.5 ) , it is conical . as a result we see that components of hessian of function @xmath12 must satisfy to the following homogeneous ( nonlinear in general ) second order pde @xmath244 these arguments prove the following : let @xmath176 , and let @xmath245 be a smooth function . following statements are equivalent 1 . weinhold metric @xmath5 has zero scalar curvature on an open subset @xmath246 2 . image of subset @xmath21 under the hessian mapping @xmath247 is a * conical * surface . function @xmath6 satisfies to a * homogeneous * second order partial differential equation @xmath248 on the open set @xmath21 . 0.4 cm consider in more details the case where the conical surface @xmath7 is a plane . this case has no , as far as we know , a direct physical application but is interesting as allowing to get the explicit necessary and sufficient condition on the thermodynamical potential @xmath6 to have zero scalar curvature @xmath249 in this case there is a constant vector @xmath250 such that radius vector @xmath217 of the surface is orthogonal to @xmath251 at all points @xmath252 . in these terms , condition of zero curvature above takes the form @xmath253 for all @xmath254 . vector @xmath251 is defined up to a multiplication by a nonzero factor . last equation can be rewritten as the 2nd order linear pde for the function @xmath255 : @xmath256 linear nondegenerate transformation in the plane @xmath257 transforms linear differential operator of the second order with constant coefficients @xmath242 to one of three canonical forms : 1 . if @xmath258 , to the ( elliptic ) form @xmath259 2 . if @xmath260 , to the ( degenerate ) form @xmath261 3 . if @xmath262 , to the ( hyperbolic ) form @xmath263 correspondingly , solutions of equation @xmath264 can be described as the composition of an arbitrary solution of one of these three model equations with a linear non - degenerate transformation in the plane @xmath257 . this finishes the proof of the first part of the following theorem . for a thermodynamical system with two degrees of freedom and a constitutive relation @xmath265 the following statements are equivalent : 1 . scalar curvature @xmath0 of the ( weinhold ) metric @xmath266 equals zero in an open set @xmath267 and image of @xmath21 under the hessian mapping is the ( part of ) plane . each point @xmath268 has a neighborhood @xmath269 such that the generating function @xmath245 satisfies to an equation @xmath270 with some @xmath271 and 1 . ( elliptic case ) if @xmath258 , then @xmath6 is the composition of a harmonic function on @xmath272 with a nondegenerate linear transformation of the plane @xmath257 . ( degenerate case ) if @xmath260 , then @xmath6 is the composition of an arbitrary function @xmath273 of variable @xmath274 with a nondegenerate linear transformation of the plane @xmath257 . hyperbolic case ) if @xmath262 , @xmath6 is the composition of a function of the form @xmath275 with arbitrary functions @xmath276 of one variable with a nondegenerate linear transformation of the plane @xmath257 . 3 . statement of part two is valid in the whole open connected set @xmath21 with the same vector @xmath251 . to prove the last statement of theorem @xmath277 , notice that , unless surface @xmath7 degenerates to a straight line , @xmath6 can not satisfy to two equations with nonparallel vectors @xmath278 . as a result , local statement of the theorem can be extended to the whole connected set @xmath21 with the same vector @xmath251 . generically , the set of points @xmath257 where @xmath279 represent the curve ( possibly singular i.e. consisting of isolated points ) on the plane @xmath257 - pre - image under the mapping @xmath192 of the set - points that are singular for the mapping @xmath280 of projectivization . if at some point @xmath257 , @xmath281 , then the sign of @xmath282 is invariant under the relabelling of variables @xmath274 , namely , when we change @xmath283 , in the expression @xmath284 for curvature , denominator is not changing while the two last columns of the determinant of the matrix in the numerator are permuted . but at the same time , first and last rows of the matrix in the numerator are permuted too restoring the sign of the scalar curvature . let @xmath285 be an oriented surface in @xmath286 . we call the surface @xmath285 ( with the normal @xmath287 ) , * radially convex * ( correspondingly , * radially concave * ) ) if @xmath288 for all @xmath289 ( correspondingly @xmath290 for all @xmath291 ) . then , as the arguments above shows , the following statement is true , let @xmath292 be an open subset of the surface @xmath7 . following statements are equivalent : 1 . @xmath293 ( correspondingly @xmath294 ) for @xmath295 , 2 . surface @xmath296 is radially convex ( correspondingly radially concave ) . 0.4 cm recalling ( @xcite ) that thermodynamical metrics @xmath11 and @xmath2 are related by contact transformations , generated by the mapping @xmath297 of canonical `` internal energy '' chart @xmath298 to the entropy chart @xmath299 . more specifically , we have , for the contact form in these two charts @xmath300 correspondingly , equilibrium surface @xmath83 is mapped onto the corresponding surface @xmath301 . for * metrics @xmath302 on these surfaces * we have the following conformal equivalence relation ( @xcite : @xmath303 noticeably with the same conformal factor @xmath304 as the contact form @xmath305 recall ( @xcite , ch.18 ) that if two metrics on a smooth manifold @xmath146 are conformally equivalent : @xmath306 with some function @xmath307 , then , their scalar curvatures are related by @xmath308 where @xmath309 is the dimension of the manifold @xmath146 and @xmath310 , with @xmath311 thus after conformal transformation one get in the expression for scalar curvature the additional term @xmath312 first term on the right in the last part of this formula is the covariant divergence with respect to the original metric @xmath313 of the gradient ( with respect to @xmath313 ) of function @xmath314 : @xmath315 thus , @xmath316 for @xmath317 , the second term vanishes and we get for scalar curvatures the following relation @xmath318 in our case above @xmath319 , so we get relation between curvatures of ruppeiner @xmath2 and weinhold @xmath11 ( expressed in coordinates @xmath102 ) : @xmath320 it follows from this that scalar curvatures of both metrics are zero simultaneously if and only if the logarithm of temperature @xmath321 is the * harmonic function * of @xmath102 . in this section we study the scalar curvature of a general 2-dim elementary thermodynamical system . we found in section 3 the weinhold metric of any two - dimensional elementary system of entropy s and volume v to be , @xmath322,(@xcite ) , : @xmath323 determinant of the tensor @xmath11 ( with respect to the standard euclidian metric ) is @xmath324 using definition of k as above , we finally get @xmath325 excluding t from being zero , the metric @xmath11 is degenerate along the curve @xmath326 which is usually presented in one of two forms : @xmath327 or / and @xmath328 . the critical triple point of the system is the extremum point of these functions . note that it is possible to write the isothermal compressibility in terms of the determinant : @xmath329 determinant of our matrix @xmath330 is the denominator of the expression ( 5.2 ) for the scalar curvature @xmath0 for such a system . this implies that if @xmath331 , then @xmath332 therefore , along this curve the weinhold metric of the thermodynamical phase space changes signature and its curvature is singular . as we will see , in the example of the van der waals gas , this is also the curve along which phase transition occurs . thus , this curve determines the boundary of the equilibria region @xmath333 on the constitutive surface @xmath83 . * remark : speed of sound . * let @xmath334 and @xmath335 respectively be the speed of sound in terms of isothermal and adiabatic compressibility @xmath15 and @xmath336 . it is known that @xmath337 . so , the speed of sound in a thermodynamical system is given by @xmath338 or @xmath339 where r is the scalar curvature of the system . in region where @xmath340 the inverse of @xmath330 is given by @xmath341 let s calculate now the third derivatives of the energy e. we get @xmath342 @xmath343 @xmath344 @xmath345 @xmath346 now , since @xmath347 , where @xmath348 we get the following identity i. @xmath349 similarly , since @xmath350 , where @xmath351 we have identity ii @xmath352 now , for the components of the ricci curvature we have @xmath353 @xmath354 @xmath355 it is easy to show that @xmath356 and so , scalar curvature is given by , @xmath357 after some calculation , we get @xmath358\ ] ] where @xmath359 @xmath360 @xmath361 and @xmath362 then , from the @xmath363 component of the ricci curvature we derive that @xmath364 and @xmath365 therefore , ricci tensor have the form @xmath366 scalar curvature is equal to @xmath367 and using ( 6.19 ) we get the following result , scalar curvature of a general two - dimensional elementary system is given by @xmath368\ ] ] where @xmath369 @xmath370 and @xmath371 here we discuss the class of systems where @xmath1 is constant - class where both the ideal gas and van der waals gas are included . in this case we can easily get the general form of the constitutive law @xmath16 , calculations of the previous section simplifies and we will be able to discuss curvature of metric @xmath11 in more details . notice that if molar internal energy @xmath12 is given as a function of volume @xmath85 and the temperature @xmath86 , then the helmholz equation @xmath372 leads ( see @xcite , ch . 6 ) to the formula @xmath373 relating the heat capacity of any 2td - system to one for ideal or vdw gas . from this formula it follows that , for a 2td - system for which the integral above converges , heat capacity @xmath1 is constant if and only if @xmath374 ( i.e. when @xmath87 is the * linear * function of @xmath86 ) . let us find possible constitutive laws @xmath375 for which @xmath376 recall that @xmath377 . rewrite this as @xmath378 and integrate : @xmath379 with arbitrary function @xmath380 . on the other hand , @xmath381 . substituting this in the definition of @xmath1 in the form @xmath382 , we get for @xmath12 equation @xmath383 integrating once we get @xmath384 with an arbitrary function @xmath385 . setting @xmath386 and @xmath387 and solving this equation we get the fundamental constitutive law in the form @xmath388 with arbitrary functions @xmath389 . case of the ideal gas is obtained here if we take @xmath390 since @xmath391 with @xmath392 and @xmath393 .(see section @xmath394 ) . for the van der waals gas we take @xmath395 since @xmath396 with @xmath392 and @xmath393.(see section 10 ) . taking derivatives by both variables and excluding @xmath63 from these relations we find the following state equation @xmath397 calculating derivative of ( 7.3 ) and denoting @xmath398 for i=1,2 , we get @xmath399 @xmath400 @xmath401 @xmath402 as a result we get the metric @xmath11 in the form @xmath403 with the determinant @xmath404 @xmath405 consider , now , a special case of a constitutive relation ( 7.9 ) with the same @xmath406 and with @xmath407(ideal gas ) , and mark this case with @xmath408 . for this case @xmath409 this determinant is zero if and only if @xmath410 , i.e. if @xmath411 with positive constants @xmath412 . now we have the important relation @xmath413 calculate third derivatives of @xmath12 of the form ( 6.3 ) : @xmath414 @xmath415 @xmath416 @xmath417 using these derivatives in formula ( 6.19 ) for the component @xmath418 of the ricci tensor we get : @xmath419 which can be written as @xmath420 so , finally we get scalar curvature to be @xmath421}{2[t(f_{1}f^{''}_{1}-(f^{'}_{1})^{2})-f^{2}_{1}f^{''}_{2}]^{2}}\ ] ] this leads to the following statement : let @xmath422 and let the internal energy @xmath423 be given by the fundamental constitutive relation @xmath388 then the curvature of weinhold metric is zero in a domain @xmath267 , @xmath424 if and only if one of three cases holds : 1 . @xmath425 for some positive constants @xmath426 , 2 . @xmath427 with some constants @xmath428 , that includes the case of ideal gas . 3 . @xmath429 degenerate case . it is interesting to see how the scalar curvature is expressed , in the case @xmath430 , in terms of isothermal compressibility @xmath15 . we calculate @xmath431 @xmath432 @xmath433 moreover @xmath434 reduces to the following identity iii @xmath435 from which we also get that @xmath436 becomes @xmath437 scalar curvature of metric @xmath11 , in the case of a constant @xmath1 , is given by @xmath438 since @xmath1 is constant , we can use @xmath439 and @xmath440 to get @xmath441 therefore , after some calculation and considering @xmath442 and @xmath443 , we get @xmath444 curvature @xmath445 is * negative * if and only if @xmath446 here we consider the ideal gas , the simplest model example of a td system with two degrees of freedom . for more detailed discussion of weinhold metric(s ) for ideal gas in different representations we refer to @xcite . except its illustrative interest , ideal gas will serve us as a reference point in the further study of curvature of metric @xmath5 for non - ideal gases . internal energy @xmath12 of the ideal gas as a function of @xmath197 and the mole number @xmath198 has the form @xmath447 with some reference constants @xmath448 , see @xcite , chapter 6 . other equations of state are obtained from ( 8.1 ) in a standard way : @xmath449 we will take @xmath200 . then it is known that @xmath450 from which we can solve for @xmath12 obtaining @xmath451 weinhold metric for an ideal gas is given by ( 3.14 ) . setting @xmath393 , we get @xmath452 in this case @xmath1 and @xmath92 are positive constants . from @xmath207 we get @xmath453 which is always positive . thus , weinhold metric is * positive definite * on the constitutive surface @xmath83 . inverse of @xmath330 is given by @xmath454 now , @xmath455 through @xmath456 become @xmath457 @xmath458 where @xmath459 for the component @xmath418 of the ricci tensor @xmath460 of ideal gas we get @xmath461 using ( 6.26 ) , we get the known result ( see @xcite ) : @xmath462 and , therefore , @xmath463 zero curvature of weinhold metric ( as well as of ruppeiner metric based on entropy @xcite ) is positively correlated with the known suggestion to use thermodynamical curvature as the characteristic of interactions on the microscopic level of media description . let s consider now a van der waals gas whose equation of state is given by @xmath464 where a and b are positive constants . this equation represents the behavior of real gases more accurately than the ideal gas by introducing two additional positive constants a and b characteristic ( of molecule interaction in the gas and of the part of the volume occupied by the molecules correspondingly , see @xcite ) of the particular gas under consideration . for the coefficients @xmath94 and @xmath15 we get @xmath465 and @xmath466 therefore the relation between the coefficient of expansion and the isothermal compressibility is given by @xmath467 the entropy of vdw gas per one mole ( @xmath200 ) is given by,@xcite , @xmath468 the internal energy of van der waals gas as a function of s and v is obtained from the last equation , namely @xmath469 then , the weinhold metric of this gas is given by @xmath470 @xmath471 with @xmath1 being constant . in the limit @xmath472 to the ideal gas we get exactly the metric @xmath473 . the determinant is then given by @xmath474 @xmath475\ ] ] which is zero along the curve @xmath476 @xmath477\ ] ] or , when @xmath478 , supposing of course that our system is at non - zero temperature . the determinant of the metric @xmath11 is positive if entropy is large enough @xmath479\ ] ] and negative if the opposite is true . the inverse of tensor @xmath330 is given by @xmath480 . now , we calculate the third derivatives of the energy u. we get @xmath481 @xmath482 then , the component @xmath418 of the ricci tensor is given by @xmath483 from @xmath484 , we get the scalar curvature @xmath445 of the weinhold metric to be @xmath485 the scalar curvature goes to zero as @xmath486 or as @xmath487 . since the quantity @xmath488 characterizes the attractive interaction within a system , scalar curvature seems to be a measure of the attraction among particles while its dependence on the parameter @xmath489 is more quantitative then qualitative . on the other hand , as we have seen before if @xmath490 then @xmath491 i.e. * at the curve @xmath476 of the signature change , scalar curvature @xmath0 has the singularity inverse quadratic by the distance to the curve*. it is interesting to note that , combining @xmath492 and example 5 ( see sec . 7 ) , we get @xmath493 . so , it seems reasonable to consider the quantity @xmath494 as a measure of non - interaction of the system as long as @xmath1 is constant . in the case of the ideal gas , @xmath495 . for the van der waals gas , @xmath496 . notice that the parameter @xmath489 is not present in @xmath497 . in the limit where @xmath486 , with a fixed volume , van der waals gas becomes ideal even if @xmath498 . let s now look at what happens when @xmath491 . as we discussed earlier , when r goes to infinity , the system is described by a _ degeneracy curves _ @xmath476 on which phase transition seems to happen . now , since the critical point is obtained whenever both @xmath499 and @xmath500 are satisfied , it is evident that , the idea of phase transition is given by the degeneracy of the weinhold metric . what follows are the derivations of the critical point for the van der waals gas through the zero - determinant of the metric matrix @xmath330 . indeed , considering the denominator of r being zero , we obtain : @xmath501 then , we get the curve @xmath476 in the p - v plane @xmath502 taking derivative with respect to v , we get @xmath503 which is zero when @xmath504 this is exactly the critical value of the volume for the van der waals gas . now , substituting this value back into the equation @xmath505 , we get the critical value for pressure @xmath506 naturally , if we consider the denominator of the scalar curvature with t and v in it and set it to zero we get the degeneracy curve @xmath476 in the t - v plane @xmath507 and since @xmath508 then we get , @xmath509 consider , now , @xmath510 , @xmath511 and @xmath512 . then @xmath513 and @xmath514 naturally , the two curves intersect at the critical point . solving @xmath515 for @xmath516 , we have the three roots of volume in terms of pressure , namely @xmath517 @xmath518 @xmath519 where @xmath520 now , solving @xmath521 for @xmath516 , we similarly get the roots of volume in terms of temperature , namely @xmath522 @xmath523 @xmath524 where @xmath525 and @xmath526 therefore , substituting these last three solutions in @xmath527 , we get @xmath528 with @xmath529 . in particular , for @xmath530 , we get the interesting _ coexistence _ curve(fig.1 ) , considering @xmath531 , instead , we get ( fig.2,3 ) now , differentiating @xmath527 and @xmath532 , then we get @xmath533 the important conclusion we gain by looking at the van der waals gas is that the singular locus @xmath534 of weinhold metric carries important information on the critical behavior of the one component system . in particular , along the curve @xmath535 phase transitions happens and exterma of coordinate functions along this curve corresponds to the critical points of the system . consider berthelot s gas ( see @xcite , chapter 6 ) whose conventional equation of state is @xmath536 where a and b are positive constants having the same meaning as for the van der waals gas . for this gas , @xmath1 is not constant . in particular , @xmath537 moreover , @xmath538 and @xmath539 therefore the relation between @xmath15 and @xmath94 is given by @xmath540 it is also interesting to look at the rates of change of the heat capacity at constant volume . in particular , denoting @xmath541 by @xmath1 and considering @xmath542 , we get @xmath543 and @xmath544 the weinhold metric @xmath100 of the berthelot s gas given by @xmath545 the determinant is then given by @xmath546 if we set the determinant of @xmath330 equal zero we get the singular locus @xmath534 of the bertellot gas @xmath547 and solving the first equation for @xmath86 we get @xmath548 taking derivative with respect to v , we get @xmath549 which is zero when @xmath504 considering @xmath550 . this implies that @xmath551 and , therefore , since @xmath552 whenever @xmath553 , we have @xmath554 we could normalize both p and t as function of v. then @xmath555 and @xmath556 let s calculate the inverse matrix . we have @xmath557 it is convenient , now , to calculate the third derivatives without grouping , in particular , @xmath558 @xmath559 @xmath560 @xmath561\ ] ] @xmath562\ ] ] @xmath563-\frac{a}{tv^{4}}[\frac{4a^{3}}{c^{3}_{v}t^{6}v^{3}}-\frac{5a^{2}}{c^{2}_{v}t^{4}v^{2}}+6]\ ] ] then , from @xmath484 , we get the scalar curvature to be @xmath564 where @xmath565 @xmath566 and @xmath567 @xmath568 notice that , as for vdw gas , as @xmath569 goes to zero then scalar curvature @xmath0 goes to zero as well , while it is not true in the case of @xmath489 being zero . we found the christoffel coefficients for a general 2-dimensional thermodynamical phase space to be @xmath570 now , the geodesic equations in a general form are given by @xmath571 for our case in which @xmath572 , the geodesic equations are of the form @xmath573 @xmath574 let s look at the christoffel coefficients in more details . @xmath575 @xmath576 @xmath577 considering the following relations : @xmath361 @xmath362 @xmath578 and @xmath579 then , we can calculate the coefficients explicitly , @xmath580\ ] ] @xmath581\ ] ] @xmath582\ ] ] @xmath583\ ] ] @xmath584\ ] ] @xmath585\ ] ] therefore we have the following proposition : the geodesic equations for a general two dimensional thermodynamical system are given by @xmath586(\frac{ds}{dt})^{2}-[\frac{1}{c_{v}}d+\frac{tv\alpha^{2}}{k^{2}c_{v}}f]\frac{ds}{dt}\frac{dv}{dt}+\frac{1}{2}[\frac{\alpha}{kc_{v}}d-\frac{tv\alpha^{3}}{k^{3}c_{v}}f-\frac{1}{k}b](\frac{dv}{dt})^{2}=0\ ] ] @xmath587(\frac{ds}{dt})^{2}+[\frac{tv\alpha}{kc_{v}}f]\frac{ds}{dt}\frac{dv}{dt}-\frac{1}{2}[\frac{c_{p}}{kc_{v}}f+\frac{1}{\alpha}b](\frac{dv}{dt})^{2}=0\ ] ] in the case in which @xmath1 is constant , then @xmath588 and @xmath589 are the same while the other relations become @xmath590 @xmath591 @xmath592 so , now , our christoffel s coefficients become @xmath593 @xmath594\ ] ] @xmath595\ ] ] @xmath596 @xmath584\ ] ] @xmath585\ ] ] so , finally , our geodesic equation with @xmath1 constant are given by @xmath597\frac{ds}{dt}\frac{dv}{dt}+\frac{1}{2}[\frac{\alpha^{2}}{k^{2}c_{v}}-\frac{tv\alpha^{3}}{k^{3}c_{v}}f-\frac{1}{k}b](\frac{dv}{dt})^{2}=0\ ] ] @xmath598\frac{ds}{dt}\frac{dv}{dt}-\frac{1}{2}[\frac{c_{p}}{kc_{v}}f+\frac{1}{\alpha}b](\frac{dv}{dt})^{2}=0\ ] ] in this work we have studied the scalar ( gauss ) curvature of weinhold metric for a thermodynamical systems with two thermodynamical degrees of freedom . we get criteria for the positivity , nullity and negativity of scalar curvature in terms of _ hessian surface _ of the thermodynamical potential , found scalar curvature for a general thermodynamical systems with two thermodynamical degrees of freedom . we have studied relation of the signature change of weinhold metric and the scalar curvature to the curves of phase transition of these systems . as examples we have considered the systems with the heat capacity @xmath1 constant , in particular the ideal and van der waals gases , and the berthelot gas . results obtained here suggest a kind of duality relation between the constitutive surface of a 2d thermodynamical system in the gibbs space ( space with coordinates @xmath599 in the case of internal energy ) and its hessian surface . relations between the convexity properties of both surfaces , curvature and signature of thermodynamical metric , extremal properties of corresponding thermodynamical potential and the phase transitions in the thermodynamical system present interesting and , in our opinion , highly promising direction of the future work . 99 m. akvis , v. goldberg , _ differential geometry of varieties with degenerate gauss maps _ , n.y . , berlin , springer verlag , 2004 . v. arnold , _ mathematical methods of classical mechanics _ , pringer verlag . r.s.berry , p.salamon , e.ihrig , _ a group of coordinate transformations which preserve the metric of weinhold _ , j.math.phys . 24(10 ) , 1983,2515 - 2520 . einstein manifolds _ , springer verlag , berlin , 1987 . h.b.callen , _ thermodynamics _ , whiley , 1985 . c. caratheodory,_unterschungen uber die grundlagen der thermodynamyk _ , gesammelte mathematische werke , b.2 . , munchen , 1955 , s. 131 - 177 . j. gibbs , _ the collected works _ , v.1 , thermodynamics , yale univ . press , 1948 . h. heemeyer , _ pseudoriemannsche faserbundel und ihre anwendung in der allegemeinrealitivistischen nichtgleichgewichtsthermodynamik _ , diss . berlin , wissenschaft und technik verlag , berlin , 1995 . r. hermann , _ geometry , physics and systems _ , dekker , n.y . , 1973 . h. janyszek , _ on the riemannian metric structure in the classical statisticsl equilibrium thermodynamics _ , reports of mathematical physics , v.24 , no.1 , 1986 , pp . 1 - 10 . h. janyszek , _ riemannian geometry and stability of thermodynamical equilibrium systems _ , journal of physics , a : math . v.23 , 1991 , pp . 477 - 490 . h. janyszek , r. mrugala _ geometrical structure of the state space in classical statisticsl and phenomenological thermodynamics _ , reports of mathematical physics , v.27 , no.2 , 1989 , pp . 145 - 159 . h. janyszek , r. mrugala _ riemannian geometry and the thermodynamics of model magnetic systems _ , physical review , a , v.39 , no.12 , 1989 , pp . 6515 - 6523 . a. knapp , _ lie groups beyond the introduction _ , birkhauser , boston , 1996 . k. kaviani , a.d.rezaie , _ riemannian geometry of the pauli paramagnetic gas _ , preprint , arxiv : cond - mat/9812054v1 , 3dec . d. kondepudi , i. prigogine,_modern thermodynamics _ , wiley and sons , 2001 . a. mlodziewskii , _ geometrical thermodynamics _ , mgu , moscow , 1956 ( in russian ) . r. mrugala , _ geometrical formulation of equilibrium phenomenological thermodynamics _ , reports of mathematical physics , v.14,no.3 , 1978 , pp.419 - 427 . r. mrugala , _ on equivalence of two metrics in classical thermodynamics _ , physica a , v.125,1984,pp.631 - 639 . r. mrugala , j.d . nulton , j.c . schoen , p. salamon , _ contact structure in thermodynamic theory _ , reports of mathematical physics , v.29 , no.1 , 1991 , pp.109 - 121 . r. mrugala , _ on a riemannian metric on contact thermodynamic spaces _ , reports of mathematical physics , v.38 , no.3 , 1996 , pp.339 - 348 . f.j.millero , a.lo surdo , c.shin , _ the apparent molal volumes and adiabatic compressibilities of aqueous amino acids at 25 c. _ , j. phys . chem , vol . 82 , no . 7 , 1978 , 784 - 792 . j. nulton , p.salamon,_geometry of ideal gas _ , physical review a , v.31,n.4,pp . 2520 - 2524 , 1985 . e. paci , m. marchi , _ instrinsic compressibility and volume compression in solvated proteins by molecular dynamics simulation at high pressure _ , proc . natl.acad.sci . m. postnikov , _ geometry vi . riemannian geometry _ , encyclopedia of mathematical sciences , v.91 , springer , 2001 . a.pogorelov , _ differential geometry _ , groningen , p.noordhoff , 1962 . s. preston , _ notes on the geometrical structures of homogeneous thermodynamics _ , manuscript , 2004 . s. preston , j.vargo , _ geometrical properties of mrugala metric _ , in preparation . g.ruppeiner , _ thermodynamics : a riemannian geometric model _ , phys . review a. 20(4 ) , 1979 , 1608 - 1613 . g. ruppeiner,_riemannian geometry in thermodynamic fluctuation theory _ , reviews of modern physics , v.67,n.3,1995 , pp.605 - 659 . g. ruppeiner,_riemannian geometry approach to critical points : general theory _ review e , v.57,n.5 , 1998 , pp.5135 - 5145 . p.salamon , j.nulton , e. ihrig,_on the relation between entropy and energy versions of thermodynamic length _ phys . , 80 , 436 ( 1984 ) . f. weinhold , _ metric geometry of equilibrium thermodynamics_,p . i - v , journal of chemical physics , v.63 , n.6,2479 - 2483 , 2484 - 2487 , 2488 - 2495,2496 - 2501,1976 , v.65,n.2,pp.559 - 564,1976 .

in this paper , we prove results concerning the large scale geometry of connected , simply connected nilpotent lie groups equipped with left invariant riemannian metrics . precisely , we prove that there do not exist quasi - isometric embeddings of such a nilpotent lie group into either a metric space or an alexandrov metric space with curvature bounded below . the main technical aspect of this work is the proof of a limited metric differentiability of lipschitz maps between connected graded nilpotent lie groups equipped with left invariant carnot - carathodory metrics and complete metric spaces . [ section ] [ pro]lemma [ pro]sublemma [ pro]theorem [ pro]definition [ pro]corollary

math9903089

in this paper we investigate the large scale geometry of connected nilpotent lie groups equipped with left invariant riemannian metrics by studying their quasi - isometric embeddings into various metric spaces . let @xmath1 be a connected nilpotent lie group with a left invariant riemannian metric and @xmath2 be the induced distance function on @xmath3 . if @xmath4 is a complete metric space , then @xmath5 is an @xmath6-quasi - isometric embedding if , for all @xmath7 , @xmath8 after studying certain invariants of these maps , we prove two main applications : [ nocat0 ] there do not exist quasi - isometric embeddings of a connected nonabelian nilpotent lie group equipped with a left invariant riemannian metric into a @xmath0 metric space . [ nocbb0 ] there do not exist quasi - isometric embeddings of a connected nonabelian nilpotent lie group equipped with a left invariant riemannian metric into a @xmath9 metric space . @xmath10 ( resp . @xmath11 ) metric spaces are spaces of curvature bounded above ( resp . below ) by @xmath12 is the sense of topanogov triangle comparison . thus , a @xmath0 metric space is a generalized space of nonpositive curvature while a @xmath9 metric space is a generalized space of nonnegative curvature . these include simply connected riemannian manifolds of nonpositive and nonnegative curvature respectively . in the literature , @xmath0 spaces are also called hadamard spaces and @xmath11 spaces are called alexandrov spaces ( with curvature bounded below ) . we retain the earlier notion for simplicity and consistency . we will discuss these in detail later in the paper . in @xcite , wolf showed that a connected nonabelian nilpotent lie group equipped with a left invariant riemannian metric must contain 2-planes of both positive and negative curvatures . theorems [ nocat0 ] and [ nocbb0 ] can be viewed as a large scale analogue of this theorem of j. wolf . while the large scale characteristics of such nilpotent lie groups may be interesting in their own right , we shall see that such investigations reduce to problems which are motivated by the now standard arguments used in the proof of mostow s rigidity theorem and its many extensions ( see @xcite , @xcite , @xcite , @xcite , @xcite as well as many others ) . in the proofs of these results , one attempts to show that two candidate spaces ( e.g. two compact constant negative curvature spaces with isomorphic fundamental groups ) are isometric by exhibiting an equivariant quasi - isometry between their universal covers and showing that the existence of such a quasi - isometry must imply the existence of a true equivariant isometry . a common element in many such proofs is to reduce to an examination of the ideal boundaries of the two spaces in question . the quasi - isometries between the two spaces induce quasiconformal maps on the boundaries which then become the focus of the study . in the cases of the papers mentioned above , the tangent cones at points on the ideal boundaries are isometric to connected simply connected graded nilpotent lie group equipped with left invariant carnot - carathodory metrics . thus , a local analysis of the ideal boundaries involves an examination of the geometry of graded nilpotent lie groups with left invariant carnot - carathodory metrics . this provides a natural motivation for considering the geometry of such spaces and , in particular , of studying the quasi - conformal maps between them . it is a natural generalization to instead consider the quasi - conformal embeddings of such spaces . unfortunately , the study of such embeddings seems quite intractable but does provide motivation for various lines of study - for example the study of bilipschitz embeddings of such spaces . as we shall see , the proofs of theorems [ nocat0 ] and [ nocbb0 ] rest on a local analysis of bilipschitz embeddings of graded nilpotent lie groups with left invariant carnot - carathodory metrics into complete metric spaces which are _ locally _ @xmath10 or @xmath11 . for metric spaces which are locally @xmath10 , i.e. for each point there is a closed ball about that point which is itself @xmath10 , we use the notation @xmath13 ( for `` curvature bounded above '' ) . note that , for a lower curvature bound , these distinctions are not necessary . in particular , we prove an intermediate theorem from which theorems [ nocat0 ] and [ nocbb0 ] follow : [ inter ] let @xmath14 be a connected simply connected graded nilpotent lie group equipped with a left invariant carnot - carathodory metric and @xmath15 be an open set . then @xmath16 does not admit a bilipschitz embedding into any @xmath13 metric space or into any @xmath11 metric space . to prove the main theorems from this intermediate theorem , we study some bilipschitz embedded invariants of graded nilpotent lie groups with left invariant carnot - carathodory metrics and use them to build local obstructions to quasi - isometric embeddings of nonabelian nilpotent lie groups with left invariant riemannian metrics into various metric spaces . next we give a brief outline of the argument and the paper and show how theorems [ nocat0 ] and [ nocbb0 ] follow from theorem [ inter ] . let @xmath3 be a connected simply connected nilpotent lie group with a left invariant riemannian metric @xmath17 and let @xmath2 be the induced distance function on @xmath3 . if @xmath5 is some @xmath6-quasi - isometric embedding of @xmath3 into a complete metric space @xmath18 , we first consider asymptotic cones of @xmath3 and @xmath18 and a map between them , denoted @xmath19 , derived from @xmath20 and the coning procedure . the asymptotic cone of a metric space @xmath4 is a limit metric space of the pointed dilated spaces @xmath21 for some sequence @xmath22 . since such spaces do not necessarily converge in the gromov - hausdorff topology , we use gromov s ultrafilter construction , choosing a nonprincipal ultrafilter @xmath23 ( see section [ tangcones ] for a definition ) to form the asymptotic cone @xmath24 . for precise details of the construction , we refer the reader to either @xcite or @xcite . for our purposes , there are a few keys pieces of information . first , if @xmath1 is a nilpotent lie group with a left invariant riemannian metric , then pansu ( @xcite ) proved that the asymptotic cone is unique and isometric to @xmath25 , a graded nilpotent lie group equipped with a left invariant carnot - carathodory metric ( see section [ cc ] for definitions ) . second , an asymptotic cone of a @xmath0 space is also @xmath0 and an asymptotic cone to a @xmath9 space is also @xmath9 . these facts ( and references to their proofs ) are reviewed in section [ curv ] . third , the quasi - isometric embedding @xmath20 gives rise to an l - bilipschitz map @xmath19 between the cones @xmath25 and @xmath26 . the construction of this map depends on the choice of ultrafilter . thus , to prove theorems a and b , we must prove instead theorem [ inter ] . the paper is devoted to proving this intermediate theorem . the main technical result in the paper is that , in an appropriate sense , such an l - bilipschitz map is differentiable in certain directions almost everywhere . this is a generalization of a classical theorem of rademacher . our proof follows the same lines as a covering argument used in kleiner s proof of the differentiability of lipschitz maps from @xmath27 into metric spaces ( in @xcite ) , which is a special case of differentiability theory of section 1.9 in @xcite . in @xcite , kirchheim shows the metric differentiability of lipschitz maps @xmath28 where the targets are complete metric spaces . our differentiability result is an extension of kirchheim s work . the interested reader should also consult @xcite and @xcite for results concerning the differentiability of quasiconformal maps between carnot - carathodory spaces . to prove theorem [ inter ] , we construct a tangent cone of @xmath19 at a point of differentiability . using this tangent cone , we can compare the local geometry of the two asymptotic cones . the local geometry of connected graded nilpotent lie groups with left invariant carnot - carathodory metrics is well understood and the differentiability allows us to `` push forward '' this structure to a tangent cone to the asymptotic cone of @xmath18 . when this asymptotic cone has additional structure , this allows us to measure the compatibility of these two objects . in the case of theorem [ inter ] , this structure provides estimates on the rate of growth of the spread between two tangent cone geodesics which show that they spread apart sublinearly but do not remain a bounded distance from one another . comparing this to the spread of geodesics in the tangent cone to either a @xmath0 or @xmath9 space , we derive a contradiction which proves theorem [ inter ] . sections [ nilrev ] and [ cc ] are reviews of the constructions mentioned above and discuss , respectively , nilpotent lie groups and carnot - carathodory metrics . section [ secdiff ] contains the proof of the limited differentiability of bilipschitz embeddings of connected graded nilpotent lie groups with left invariant carnot - carathodory metrics into complete metrics spaces . the main goal of this section is to prove theorem [ kscc ] . section [ tangcones ] reviews the tangent cone construction and interprets theorem [ kscc ] as a statement about maps between tangent cones . section [ curv ] reviews the definition and some properties of @xmath0 and @xmath9 spaces . in section [ end ] we prove theorem c , that there do not exist bilipschitz embeddings of a nonabelian connected graded nilpotent lie group equipped with a carnot - carathodory metric into either a @xmath13 or @xmath11 space . the author wishes to thank bruce kleiner for suggesting this line of work as well as for many helpful discussions . thanks are also due to chris croke for helpful discussions of this work and for many hours of help with the preparation of this document . in this section , we recall some basic definitions of nilpotent lie groups and some associated structures . let @xmath14 be a connected lie group and @xmath29 be its lie algebra equipped with @xmath30 $ ] a bracket . for the rest of this exposition , we assume that all lie groups are connected and simply connected . let @xmath29 be a lie algebra over a field @xmath31 , we define the descending central sequence , @xmath32 as follows : @xmath33 \end{split}\ ] ] we use the descending central sequence to define nilpotency : a lie group @xmath14 is called nilpotent if there exists an integer @xmath34 such that @xmath35 . the smallest such @xmath34 is called the degree of nilpotency of @xmath14 . recall that the exponential map , @xmath36 , is a diffeomorphism for simply connected nilpotent lie groups . for this paper , we will use the following notation : if @xmath37 and @xmath38 are exponential images of the lie algebra elements @xmath39 and @xmath40 , then th e product , @xmath41 is also the exponential image of a lie algebra element which we denote @xmath42 . in other words , @xmath43 . for this work , pansu s theorem tells us that after taking asymptotic cones , graded nilpotent lie groups are our main objects of study . a nilpotent lie group is graded if its lie algebra comes with a grading . a grading for a lie algebra @xmath29 is a decomposition @xmath44 where the subspaces @xmath45 satisfy the condition @xmath46 \subset \mathcal{v}^{i+j}\ ] ] for all @xmath47 . _ example : _ the three dimensional heisenberg group is defined by giving a three dimensional lie algebra spanned by vectors @xmath48 with the only nontrivial bracket relation given by @xmath49=z$ ] . thus , the lie algebra has a grading @xmath50 where @xmath51 and @xmath52 . every graded nilpotent lie group possesses a self - similarity . [ htdef ] let @xmath53 be an element in a graded nilpotent lie group of nilpotency degree @xmath54 such that @xmath55 where @xmath56 . for @xmath57 , consider the @xmath58-parameter group of automorphisms , @xmath59 , of @xmath14 given by @xmath60 we may sometimes analyze this action of lie algebra elements . thus we define @xmath61 . if we wish to consider @xmath62 we use the convention that @xmath63 ( please note that this convention is not standard in the literature ) . for general lie groups , the campbell - baker - hausdorff formula provides a method for computing the product of two nearby group elements . we state it in the special case when @xmath3 is a simply connected graded nilpotent lie group . for a more general version and , for that matter , a proof , see @xcite section 2.15 . [ cbh ] let @xmath3 be a simply connected graded nilpotent lie group of nilpotency degree @xmath54 , let @xmath64 be its lie algebra and let @xmath65 denote the exponential map at the identity . then , given @xmath66 @xmath67 where @xmath68 $ ] and if @xmath69 then the last term in the sum is @xmath70 . in other words , we have the formula : @xmath71+\frac{1}{12}[x,[x , y]]- \frac{1}{12}[y,[x , y]]+ ... } \ ] ] in the case that @xmath3 is a graded nilpotent lie group , we use the following convention . let @xmath72 be the pieces of @xmath73 which lie in @xmath45 . thus , if @xmath74 , @xmath75 , @xmath76 $ ] and so on . thus , the campbell - baker - hausdorff formula yields @xmath77 . using the campbell - baker - hausdorff formula , we easily confirm that @xmath78 is a group automorphism . in subsequent sections , we will use repeatedly several properties of @xmath78 . we group them here as a lemma for easy reference : [ ht ] let @xmath3 be a graded nilpotent lie group and let @xmath78 be the automorphism described above . * for @xmath79 , @xmath80 . * for @xmath81 , @xmath82 . * if @xmath83 is the derivative of the left translation map @xmath84 , then @xmath85 . * @xmath86 _ proof : _ the first two are simple exercises using the campbell - baker - hausdorff formula . the last two are restatements of the fact that @xmath78 is a group automorphism in terms of @xmath87 . carnot - carathodory metrics arise in a variety of settings and have been the object of a great deal of study . in this work , we examine graded nilpotent lie groups equipped with carnot - carathodory metrics and hence the definitions and propositions will be tailored for this small set of spaces . for a more general and comprehensive introduction to the theory , the interested reader is encouraged to consult @xcite , @xcite , and @xcite . the rough idea of a carnot - carathodory metric is to measure the distance between two points by taking the infimum of the length of piecewise @xmath89 paths connecting the points over a restricted set of paths , those tangent to some subbundle of the tangent bundle . more precisely ( in our setting ) , if @xmath3 is a connected graded nilpotent lie group with lie algebra grading @xmath90 , we define a left invariant subbundle of the tangent bundle , @xmath91 , by letting the fiber at a point be the left translate of the subspace @xmath92 . in the rest of the paper , we will , by abuse of notation , refer to the subbundle and the fiber by @xmath91 . if we place a norm @xmath93 on each fiber of @xmath91 , we define the carnot - carathodory distance between @xmath94 by @xmath95 if no such paths exist , we set @xmath96 . we call a path whose tangent is in the subbundle @xmath91 at each point a horizontal path . note that @xmath97 defines a left invariant metric on @xmath3 . in general , if @xmath98 is merely a norm , we call the quadruple @xmath99 a carnot - carathodory group or a sub - finsler group ( if we wish to emphasize that @xmath93 is only a norm ) and @xmath97 a carnot - carathodory or sub - finsler metric . if @xmath93 arises from an inner product , @xmath100 , on @xmath91 , we denote the resulting space by @xmath101 and we call it a sub - riemannian group . this nomenclature and notation is somewhat different from the literature which often interchanges the terms carnot - carathodory metric , sub - riemannian metric , and singular riemannian metric freely . in this exposition , any object adorned with a `` cc '' will be constructed to be with respect to @xmath97 . for example , @xmath102 is the @xmath97-ball of radius @xmath103 with center @xmath104 . it is a consequence of chow s theorem ( @xcite or see @xcite for a exhaustive discussion of this fact ) that any two points in @xmath105 can be connected with a smooth horizontal path . this leaves open the question of geodesics in carnot - carathodory spaces . we call a segment a geodesic if it is a local length minimizer . again , in the literature , there are many different terms / meanings for geodesic paths and/or length minimizing paths depending on whether the geodesics satisfy certain differential equations ( see , for example , @xcite and @xcite ) . for this exposition , these issues are not relevant and hence we will use the simplified language . in a sub - riemannian group @xmath101 , there is a nice class of geodesics which will be useful in later sections . let @xmath101 be a left invariant sub - riemannian metric where @xmath3 is a connected graded nilpotent lie group and fix @xmath79 , @xmath106 . let @xmath107 be a smooth left invariant vector field on @xmath3 with @xmath108 . let @xmath109 be an integral curve of @xmath107 such that @xmath110 . we call @xmath109 a radial geodesic eminating from @xmath34 . if a basepoint is understood , we shall simply call @xmath109 a radial geodesic . note that radial geodesics are actually geodesics of @xmath97 . indeed , considering the quotient @xmath111 $ ] endowed with the left invariant metric induced by the @xmath112 inner product on @xmath91 , the quotient map @xmath113 $ ] is @xmath58-lipschitz and takes radial `` geodesics '' to geodesics in the quotient and preserves their length . thus , these are geodesics of @xmath97 as well . in addition , we see that @xmath78 is a homothety of @xmath97 . since @xmath78 is an automorphism of @xmath3 and @xmath87 preserves @xmath91 and acts as a homothety on @xmath91 , we see that @xmath78 is a homothety on @xmath114 . in @xcite , mitchell describes the local structure of more general carnot - carathodory spaces . if @xmath101 is a sub - riemannian group , the local structure is much easier to determine given the existence of a left invariant metric , a homothety of that metric , and the uniqueness of the haar measure . the following results can either be thought of as mitchell s results in this special case or as consequences the extra structure in this special case . [ mitchell ] let @xmath101 be a sub - riemannian group . then , 1 . for any @xmath79 , the tangent cone at @xmath34 , @xmath115 exists , is unique , and is isometric to @xmath114 . 2 . let @xmath116 . then @xmath117 is the hausdorff dimension of @xmath114 . 3 . let @xmath118 be a riemannian completion of the inner product @xmath100 on all of @xmath64 such that the grading is orthogonal . let `` dvol '' denote the riemannian volume measure associated to this left invariant riemannian metric . then , if @xmath119 denotes the @xmath117 dimensional hausdorff measure , @xmath119 and @xmath120 are commensurable . in other words , there exist constants @xmath121 such that @xmath122 one should note that in mitchell s work on more general carnot - carathodory spaces , the constants @xmath121 must be recalculated depending on a choice of compact set and and the the hausdorff and riemannian measures are only commensurable on compact sets . however , we now see that in our setting , the left invariance of the metric and the presence of a homothety allows us have global constants for the volume and hausdorff measure comparisons . first , we start with an illustrative example : _ example : _ let @xmath123 denote the three dimensional heisenberg group with a left - invariant sub - riemannian metric . it is well known that the haar measure on @xmath123 is , in fact , the four dimensional hausdorff measure with respect to @xmath97 on @xmath123 ( see @xcite ) . this confirms the formula from mitchell s theorem . indeed , @xmath124 in addition , we can quickly see the volume estimate inequality ( [ ccvrball ] ) using the homothety @xmath125 . consider what @xmath125 , with @xmath126 , does to the riemannian volume . denote by @xmath17 the riemannian metric given by the inner product @xmath118 described above . at the identity , the metric @xmath127 , written in terms of the left invariant vector fields @xmath128 and @xmath129 looks like @xmath130 thus , the riemannian volume element determined by the metric @xmath17 is multiplied by @xmath131 under @xmath132 . thus , picking @xmath133 to work for a particular value of @xmath134 , we can then use the dilation @xmath125 to scale the picture so that the same constant works for any @xmath103 value larger than zero . this argument actually works for more general graded nilpotent lie groups yielding the same comparative scaling between the carnot - carathodory metric and the riemannian metric . in the proof of the limited differentiability of bilipschitz maps , we will need to be able to manipulate covers of sets in graded nilpotent lie groups while using the measure @xmath119 . in particular , we want to be able to take the analogue of a vitali cover of a measurable set @xmath135 and find a countable disjoint subcover which still covers almost all of @xmath135 with respect to @xmath119 . for this section , we will use the language of @xcite ( section 2.8 ) . assume that @xmath3 is a connected graded nilpotent lie group with a sub - riemannian metric , @xmath97 . let @xmath119 be the @xmath117 dimensional hausdorff measure associated to @xmath97 as in last section . let @xmath136 be the collection of closed cc - balls in @xmath3 . then , mimicing the standard proof in @xmath27 we prove the following lemma . [ vitali ] @xmath136 is a vitali cover . in other words , given a measurable set @xmath137 covered by a fine subcollection of @xmath138 , we may find a countable dijoint subcover , @xmath139 such that @xmath140 . let @xmath101 be a graded nilpotent lie group of nilpotency degree @xmath54 with a sub - riemannian metric . to set up the notation , we let @xmath141 be an l - bilipschitz map of @xmath3 into a complete metric space . let @xmath2 be @xmath142 , the metric @xmath143 pulled back through @xmath19 . we want to show that , appropriately defined , @xmath2 is differentiable in @xmath91 directions at most points . as discussed in the introduction , to prove this , we follow the style of the covering argument in the appendix of @xcite used to show a modified version one of korevaar and schoen s results in @xcite . one can also consider this a version of kirchheim s metric differentiability in @xcite although the argument is different . in examining differentiability , we consider an analog of the directional derivative at @xmath104 in the direction of @xmath144 by looking at difference quotients along the paths @xmath145 for @xmath146 : @xmath147 we consider whether such a `` directional derivative '' exists ( i.e. the limit as @xmath148 exists ) by examining the `` lower '' and `` upper '' limits . if @xmath16 is some open set in @xmath3 , we define these limits as follows . [ lowerd ] the lower derivate of @xmath2 is a function @xmath149 defined by : @xmath150 [ upperd ] the upper derivate of @xmath2 is a function @xmath151 defined by : @xmath152 note that both @xmath153 and @xmath154 are measurable functions . properly viewed , they are just the @xmath155 and @xmath156 of measurable functions of more than one variable ( see @xcite page 152 ) . to understand and estimate the lower and upper derivates , we want to estimate the quotients for small values of @xmath157 . to do this , we construct special boxes that aid in the estimates . for each @xmath106 , denote by @xmath107 the left invariant vector field on @xmath3 determined by @xmath39 . choose @xmath118 to be a riemannian completion of the sub - riemannian metric on @xmath91 such that the grading is orthogonal . [ boxend ] fix @xmath106 and let @xmath158 be the orthogonal complement of @xmath39 in @xmath91 with respect to @xmath100 . let @xmath159 . now , let @xmath160 @xmath161 forms the end of our box . note that @xmath162 is the exponential image of a carefully constructed rectangle which should be thought of as a square of side length @xmath163 . recalling theorem 3.6 in @xcite or one of the various formulations of the `` ball - box '' estimates ( see , for example @xcite ) , if @xmath164 then @xmath165 where @xmath137 is a constant depending only on the choice of @xmath118 and the structure constants of the lie group @xmath3 . clearly , the constant @xmath137 is invariant under scaling by @xmath78 : if @xmath166 and @xmath167 are the components of @xmath168 in the various pieces of the grading , then @xmath169 . [ box ] define the estimate boxes by @xmath170 } zh_se^v \bigr ) \ ] ] as in kleiner s argument , one first thickens @xmath171}$ ] into an estimate box , @xmath172 , and then uses smaller boxes , @xmath173 with @xmath174 close to @xmath104 , to cover it and help estimate the distance between the ends of the estimate box . to use this argument , we must ensure that the boxes we have constructed have metric properties ( with respect to @xmath97 ) similar to the boxes in @xmath27 , at least on the small scale . one important technical detail is to control the distances between points in the `` far '' end of the box with respect to corresponding distances in the `` near '' end . let @xmath175 with @xmath176 and @xmath106 . then , @xmath177)+ ... + ( ( { t { \varepsilon}})^kz_k + 2t^k { \varepsilon}c_{k-1}[z_{k-1},v]+ ... ) } ) \\ & = t { d_{cc}}(e^0,e^{{\varepsilon}z_1 + ( ( { \varepsilon})^2z_2 + 2 { \varepsilon}c_2[z_1,v])+ ... + ( ( { \varepsilon})^kz_k + 2 { \varepsilon}c_{k-1}[z_{k-1},v]+ ... ) } ) \\ & \le tc({\varepsilon } ) \end{split}\ ] ] here , @xmath178 is a constant that depends on @xmath179 and goes to zero with @xmath179 . this constant exists by the compactness of the closed unit cc - ball . the first equality is due to the left invariance of @xmath97 . the second is a use of the campbell - baker - hausdorff formula . the third follows from the fact the @xmath78 is a homothety of @xmath97 . as in kleiner s argument , we will prove a relation between the upper and lower derivates by covering boxes which thicken the segment @xmath171}$ ] with well controlled smaller boxes . to do this , we make the following estimates . let @xmath180 are boxes of some variable height @xmath181 such that the difference quotients of the endpoints of the middle segment are infimal up to an error of @xmath179 . in fact , we get a specific estimate on the difference quotient between points on the ends . [ endest ] let @xmath182 be points on opposite ends of @xmath183 . then , @xmath184 where @xmath185 is a constant depending on @xmath186 which goes to zero with @xmath186 . _ proof : _ this is just a computation : @xmath187 let @xmath188 . @xmath88 this end estimate provides the needed precision to make the later estimates . the `` well controlled smaller boxes '' mentioned above will be members of @xmath189 . other than the end estimates , the other technical point we need is a measure comparison . we wish to use the estimate boxes to prove a relation between @xmath153 and @xmath154 almost everywhere with respect to @xmath119 . thus , we need to know the relative measure of the boxes in the cc balls . in light of the discussion concerning the comparison of hausdorff and riemannian measures in section [ cc ] , this is fairly straightforward . [ density ] fix @xmath190 , @xmath191 and @xmath192 . there is a number , @xmath193 , such that @xmath194 for all @xmath195 . then , if @xmath195 , the estimate boxes , @xmath196 , have bounded density variation in cc balls . i.e. there exist constants @xmath197 depending only on @xmath198 and @xmath199 such that @xmath200 _ proof : _ since @xmath97 is a left invariant metric , the constant @xmath193 exists and is independent of the point @xmath174 . it can be estimated by equation ( [ spread ] ) and the construction of the boxes . since @xmath117 is the hausdorff dimension of @xmath114 and @xmath201 , @xmath202 where @xmath203 . thus we only need estimate @xmath204 . by construction , @xmath205 is fibered by paths ( radial geodesic segments ) @xmath206}$ ] where @xmath207 . now , @xmath208 ( note this is independent of @xmath209 by the left invariance of the hausdorff measure ) . here @xmath210 depends on the normalization of the hausdorff measure and @xmath211 . let @xmath212 be a left invariant riemannian metric on @xmath3 making the grading orthogonal and matching the sub - riemannian metric s inner product on @xmath91 . let @xmath213 be the volume form on @xmath3 associated to @xmath118 and let @xmath214 be the volume form of @xmath118 restricted to a submanifold of one less dimension . since @xmath119 bounded above and below by a constant multiple of @xmath120 , there exists a constants @xmath215 such that @xmath216 . by construction , the riemannian arclength of @xmath217 for @xmath207 is equal to @xmath218 . let @xmath219 be the ( riemannian ) tangent space to @xmath3 at @xmath174 and let @xmath220 be the set of vectors such that @xmath221 . now , @xmath222 $ ] is diffeomorphic to @xmath223 via the map @xmath224 . note that for any @xmath225 we let @xmath226 be the vectors such that @xmath227 , the same map restricted to @xmath228 $ ] is a diffeomorphism onto @xmath205 . using lebesgue measure on @xmath229 we see that @xmath230}\int_{r_t}d\mathscr{l}^{n-1}d\mathscr{l}^1 = ( 2t\beta)^{q-1}t{d_{cc}}(e^0,e^v)\ ] ] using the jacobian of the diffeomorphism to compute @xmath231 we get the estimate : @xmath232 where the @xmath233 are constants which take into account the diffeomorphism s riemannian distortion of @xmath228 $ ] . note that @xmath234 because the jacobian of the diffeomorphism can be taken to be bounded below by a positive constant due to the compactness of @xmath222 $ ] . similarly , @xmath235 is finite . thus , taking @xmath236 and @xmath237 , we have the desired result . @xmath88 _ remark : _ lemma [ density ] and [ vitali ] show that we may refine covers of the boxes defined above in the same way that we refine covers of closed balls . in other words , if @xmath137 is a measurable set in @xmath3 and @xmath238 is a cover of @xmath137 by boxes which is fine at every point of @xmath137 , then we may find a countable disjoint subcover @xmath239 such that @xmath240 . this follows from standard covering arguments ( i.e. a countable number of iterations of the `` greedy '' algorithm used to cover at least a fixed portion of @xmath137 in one iteration ) . following federer , we say that such a collection of boxes is a _ vitali covering realtion_. the goal of this discussion is to show : [ kscc ] let @xmath101 be a sub - riemannian group of nilpotency degree @xmath54 . suppose @xmath2 is a distance function on @xmath241 , @xmath16 an open subset of @xmath3 , which is l - lipschitz to @xmath97 . then , there exists a measurable function @xmath242 defined for all @xmath243 . further , there is a subset @xmath244 of @xmath16 of full measure such that for every @xmath245 , * for @xmath106 , @xmath246 * for @xmath247 . * for @xmath106 , @xmath248 . * for @xmath249 such that @xmath250 , @xmath251 . we will also see that if , for some @xmath252 , @xmath253 exists , we can interpret this proposition as a statement about @xmath254 . one should view the function @xmath255 as being something like an infinitesimal sub - finsler metric and think of the last three claims of the theorem as giving a limited homothety , symmetry and left invariance for the function @xmath255 . it would be interesting to find conditions under which this construction actually converged to a sub - finsler metric . _ remark : _ this theorem leaves open the question of complete metric differentiaibilty in this setting and other settings . in a subsequent paper , the author will address such issues . we prove theorem [ kscc ] through several lemmas . [ diff1 ] if @xmath106 and @xmath104 an approximate continuity point of @xmath256 and @xmath257 , then latexmath:[${\overline{\rho}}(x,\tau v ) \le @xmath104 , @xmath259 . _ proof : _ fix @xmath260 . we aim to show that for small enough @xmath146 and @xmath261 , @xmath262 for simplicity , we will first consider only @xmath263 . the nonpositive case is an easy consequence of the positive case and will be discussed at the end of the proof . to prove the positive case , we will consider special boxes which will easily estimate the middle segment of @xmath264 ( for some appropriate @xmath186 ) and still have their endpoint quotients close to the infimum @xmath153 . first , pick @xmath265 . since @xmath104 is an approximate continuity point of @xmath256 , the set @xmath266 has density in @xmath102 approaching @xmath58 as @xmath103 approaches @xmath267 . considering @xmath261 and @xmath268 such that @xmath269 , we observe that the left invariance of @xmath119 implies that @xmath270 for any such @xmath174 and @xmath271 . since @xmath129 has density approaching 1 ( with @xmath157 ) in @xmath272 then @xmath273 where @xmath274 as @xmath148 thus , since @xmath275 , @xmath276 . combining this with the density estimate of lemma [ density ] yields that @xmath129 has density approaching @xmath58 in boxes of the form @xmath277 with @xmath278 as @xmath279 . for the rest of the proof , fix @xmath280 such that for @xmath281 , @xmath278 , @xmath282 to get estimate ( [ aim ] ) , we fix a box @xmath283 with @xmath281 and @xmath278 . let @xmath284 . next , let @xmath285 . recalling the definition of @xmath153 , we see that for @xmath286 and if @xmath157 is sufficiently small ( i.e. @xmath157 such that @xmath287 must be close to infimal and @xmath288 ) , then there are infinitely many boxes in @xmath289 with height @xmath181 . using lemma [ density ] , we see that @xmath290 is a vitali covering relation ( see the remark after [ density ] ) of @xmath291 . thus , we may select from the cover a disjoint finite subcover such that @xmath292 which , by abuse of notation , we denote this by @xmath290 as well . consider the picture at this point , we have fixed a particular box and covered it with smaller boxes , all contained in the fixed box , such that the end estimates on the small boxes are approximately infimal . next , we will extract a particular curve from the fixed box and approximate the difference quotient of the endpoints of its middle segment using the cover . using the same idea as in lemma [ density ] , we realize @xmath293 as the diffeomorphic image of @xmath294 \subset { \mathbb{r}}^{n-1 } { \times}{\mathbb{r}}$ ] . we can calculate @xmath295 , up to bounded error , by calculating the lebesgue measure of @xmath294 $ ] . using this diffeomorphic identification , we denote by @xmath296 the image of the collection @xmath297 in @xmath298 $ ] . the fact that @xmath299 implies that @xmath300))}{\mathscr{l}^n(r { \times}[0,t\tau ] ) } \ge 1- 2 \widetilde{c}_0 { \varepsilon}$ ] where @xmath301 is a constant depending on the structure constants of the lie group and the jacobian of the diffeomorphism restricted to the compact set @xmath294 $ ] . thus , using fubini s theorem on @xmath302 , we can conclude that there exists a point @xmath303 such that @xmath304)}{\mathscr{l}^1(r_0 { \times}[0,t\tau ] ) } \ge 1 - 2 \widetilde{c}_0{\varepsilon}$ ] . let @xmath305 be the fiber which is diffeomorphically identified with @xmath306 $ ] . then , @xmath307 where @xmath308 is a ( potentially ) different constant depending on the same data . now , we will do end estimates on the smaller boxes and sum them to get an estimate for @xmath309 . let @xmath310 , @xmath311 be the finite list of points along @xmath109 , listed in order following the parameter , such that each @xmath312 lies in the end of one of the small boxes and @xmath313 lies in the other . for @xmath314 a pair which are in the ends of a single box , @xmath315 , the construction yields the following estimate . by lemma [ endest ] , we have : @xmath316 now using the triangle inequality for @xmath2 , the fact that the smaller boxes cover most of @xmath293 shows that @xmath317 . let @xmath318 . the error is the sum of the distances between the endpoints of the adjacent boxes . this error is less that the one dimensional d - hausdorff measure of @xmath137 , @xmath319 because for any connected set @xmath133 , @xmath320 ( see @xcite 2.10.12 ) . using the lipschitz property of @xmath2 , @xmath321 . next , we estimate @xmath322 . since @xmath323 , we have that @xmath324 and using the fact that @xmath321 and the lipschitz property of @xmath2 , @xmath325 letting @xmath326 we have : @xmath327 next , we note that @xmath328 . to see this , note that since the curve @xmath329 is rectifiable with respect to @xmath97 , we can pick time indices @xmath330 and @xmath331 such that @xmath332 by left invariance , this equals @xmath333 where @xmath334 depends only on @xmath179 . thus , picking @xmath179 so that @xmath335 allows us to conclude that @xmath336 @xmath337 taking the @xmath156 and letting @xmath338 tend to zero , we have the desired result . recall that we had assumed that @xmath339 was positive at the outset of the proof . to see that the same result holds true for negative @xmath339 , we make the following observations . first , @xmath340 since @xmath106 . second , for small enough @xmath157 , the density of @xmath129 in @xmath341 is still greater than @xmath342 . therefore , by the symmetry of the distance @xmath2 , the same argument goes through for @xmath343 . since the case of @xmath344 is trivial , the lemma is valid for all @xmath345 . @xmath88 [ itsanorm ] retaining the assumptions of theorem [ kscc ] , define @xmath346 . then , there exists a @xmath347 of full measure such that for @xmath245 , @xmath348 for all @xmath144 . we will call @xmath244 the set of * metric regular * points of @xmath2 . _ proof : _ let @xmath349 . let @xmath117 be a countable subset of @xmath91 so that @xmath350 is dense in @xmath351 . since @xmath153 is measurable function , it follows that for fixed @xmath106 , the set @xmath352 has full measure in @xmath16 . therefore , @xmath353 is a borel set of full measure in @xmath16 . on this set , we have the desired properties . using lemma [ diff1 ] , we see that for @xmath354 and @xmath355 , @xmath356 therefore , @xmath357 given the monotonicity in @xmath339 of @xmath153 and @xmath154 , this holds for all @xmath339 in @xmath358 and @xmath359 . note in particular that @xmath360 and therefore we have symmetry of this type . to extend these properties to all of @xmath91 we check that @xmath361 is continuous . fix @xmath106 and consider a sequence @xmath362 which converges to @xmath39 . then , @xmath363 @xmath88 [ lirho ] suppose @xmath364 is a metric regular point . then , if @xmath365 such that @xmath366 then @xmath367 _ proof : _ this fact follows from the covering argument in lemma [ diff1 ] . fix @xmath260 . using the same notation , picking @xmath186 sufficiently small , the density of @xmath291 in @xmath368 is bigger than @xmath342 . now , we can run the same covering argument to produce the estimate @xmath369.@xmath88 _ proof of theorem [ kscc ] : _ let @xmath370 this clearly satisfies the first part of the claim with @xmath244 being the set of metric regular points of @xmath2 . we now prove the other items in reverse order . by lemma [ lirho ] , we see that @xmath371 for @xmath372 proving the fourth part of the theorem . lemma [ itsanorm ] implies that @xmath373 and the symmetry of the distance function @xmath2 immediately shows that @xmath374 , proving the third part of the claim . lemma [ itsanorm ] also shows that for @xmath106 , @xmath375 proving the second part of the theorem . @xmath88 the differentiability result of the last section allows us to compare the local structures of a sub - riemannian group and some metric space to which it is bilipschitz equivalent . presuming the existence of tangent cones to image points of @xmath19 in @xmath18 , we would like construct a `` derivative '' mapping between the tangent cone at a metric regular point in @xmath3 to the tangent cone at its @xmath19 image in @xmath18 . to do this , we will first review the tangent cone construction and apply it to this setting , interpreting the results of theorem [ kscc ] . the goal of this section is to prove the following : [ conemap ] let @xmath101 be a sub - riemannian group , @xmath4 be a metric space and @xmath376 be an l - bilipschitz map between them . letiing @xmath377 we let @xmath378 be the function constructed in theorem [ kscc ] . suppose there exists an @xmath379 in the image of the metric regular points of @xmath19 such that a tangent cone , @xmath380 , exists . then , * there exists an l - bilipschitz map @xmath381 such that for @xmath382 that is in the exponential image of @xmath91 , @xmath383 . * @xmath384 takes radial geodesics to geodesic lines . we begin by considering the structure of the tangent cone to @xmath114 at some point . by proposition [ mitchell ] , we know that the tangent cone at any point is unique and isometric to @xmath114 . we will consider the tangent cone at @xmath385 for simplicity . given a complete metric space @xmath386 , one usually constructs the tangent cone at @xmath104 as the gromov - hausdorff limit of the sequence of dilated spaces @xmath387 for some sequence @xmath388 tending towards infinity as @xmath389 . for @xmath114 , one may construct the dilated spaces using the homothety . thus one considers the convergence of @xmath390 . now , since @xmath114 is isometric to @xmath390 via the isometry @xmath391 , the sequence is , in fact , the constant sequence and thus converges in the gromov - hausdorff topology for any choice of sequence @xmath388 . ( see @xcite for an explanation for the convergence of pointed spaces . the reader may also want to consult @xcite which provides a somewhat shorter explanation for compact spaces ) . before continuing , we reiterate the assumptions about @xmath4 : first , there exists @xmath392 and the tangent cone to x at x exists , i.e. there exists a sequence @xmath388 such that @xmath393 ( the gromov - hausdorff limit of @xmath394 ) . to construct the map @xmath384 , we first need to review the concept and usage of ultrafilters and ultralimits . a nonprincipal ultrafilter is a finitely additive probability measure @xmath23 on subsets of @xmath395 such that * @xmath23 is zero on any finite subset of @xmath395 . * for every subset @xmath396 , @xmath397 is either zero or one . it is a consequence of the axiom of choice that nonprincipal ultrafilters exist ( see @xcite chapter 10 , theorem 7.3 ) . we can use an ultrafilter to choose a distinguished `` convergent '' point for a sequence . as an illustrative example , let @xmath358 be the line with its usual metric and let @xmath398 be a bounded sequence in @xmath358 . think of @xmath398 as the image of a map @xmath399 . then there is a unique element @xmath400 ( @xmath401 ) such that for any neighborhood @xmath16 of @xmath402 , @xmath403 . in this way , the nonprincipal ultrafilter picks a limit point of the sequence . next we point out a useful fact about ultralimits of sequences in @xmath358 . if @xmath404 and @xmath405 are bounded sequences and @xmath406 then @xmath407 . indeed , if @xmath408 then we can find open neighborhoods @xmath409 and @xmath410 of @xmath411 and @xmath412 which separate the ultralimits . since @xmath413 , if the inverse images are disjoint , this violates the additivity of the ultrafilter while if they intersect they must intersect in an infinite set by the additivity of the ultrafilter . thus , for indices in this intersection , @xmath414 , contradicting the assumption that @xmath415 for all @xmath416 . let @xmath23 be a nonprincipal ultrafilter and define a limit metric space @xmath417 as follows . let @xmath418 . then @xmath419 where @xmath420 if @xmath421 ( note that here @xmath422 is a bounded map from @xmath395 to @xmath358 ) and @xmath423 . note that the choice of ultrafilter forces convergence of the dilated spaces and allows us to group the many choices we otherwise have to make into the choice of the ultrafilter . also note that our assumption that the gromov - hausdorff limit of @xmath394 exists allows us to find an isometry between @xmath424 and @xmath425 . for a proof of this fact , as well as a more general discussion of ultrafilters and ultralimits , see @xcite section 2.4.2 and , in particular , lemma 2.4.3 . using left invariance , we may assume without loss of generality that @xmath426 . we define @xmath427 by @xmath428 . note that @xmath384 is l - bilipschitz if @xmath19 is : @xmath429 @xmath430 as mentioned above , lemma 2.4.3 in @xcite shows that @xmath380 is isometric to @xmath431 and therefore we may view @xmath384 as an l - bilipschitz map between @xmath432 and @xmath380 . in addition , by construction , @xmath433 which by theorem [ kscc ] applied to @xmath142 converges to @xmath434 when @xmath435 is in the exponential image of @xmath91 . thus we have proved the first part of proposition [ conemap ] . for the second part , consider a radial geodesic , @xmath436 in @xmath3 where @xmath106 . using the definition of @xmath384 , let @xmath437 . we claim @xmath329 minimizes between any two time indices @xmath438 . assume for now that @xmath439 . indeed , @xmath440 here we used the left invariance of @xmath255 and lemma [ ht ] . for @xmath441 , we simply recall our convention regarding @xmath442 for negative values of @xmath157 . when @xmath443 we observe that @xmath444 thus , @xmath329 minimizes between any two time indices . in addition , by left invariance , all radial geodesics are taken to geodesics under the map @xmath384 . the reader may wonder why such trouble is taken to use an ultrafilter construction to produce convergence . in fact , if @xmath18 is a separable and locally compact metric space , then one can use more standard ( e.g. arzela - ascoli type ) arguments to produce uniform convergence on compact sets . however , since asymptotic cones are not always locally compact , we use the ultrafilter construction to resolve convergence issues . again , we follow the notation and definitions of @xcite . to determine curvature bounds of metric spaces , we will need to compare to model spaces . for @xmath445 , let @xmath446 be the two dimensional constant curvature @xmath12 model space ( i.e. @xmath447 , or @xmath448 ) . denote by @xmath449 the diameter of the model space @xmath450 . let @xmath386 be a complete metric space . given a triangle , @xmath451 , in x with minimizing geodesic sides @xmath452 , we call the triangle @xmath453 with the same side lengths as @xmath451 a comparison triangle for @xmath451 . @xmath386 is a @xmath10 space if * for every pair of points @xmath454 with @xmath455 , there is a geodesic segment joining @xmath104 to @xmath174 . * let @xmath451 be geodesic triangle in @xmath18 with sides @xmath456 of lengths @xmath457 such that @xmath458 . for any two points @xmath459 , @xmath460 where @xmath461 and @xmath462 are the corresponding points on @xmath463 . another way of saying the second condition in the definition is that the triangles in @xmath18 are thinner than their counterparts in @xmath450 . as mentioned in the introduction , we sometimes will consider metric spaces which have this triangle property locally . hence , we say a metric space , @xmath18 , is a @xmath13 metric space if for each point @xmath379 , there exists an @xmath103 such that the closure of the ball of radius @xmath103 about @xmath104 is a @xmath10 metric space in its own right . in the next several lemmas , we outline the properties of @xmath10 spaces that will be of use below . a nicely self - contained discussion of these facts can be found in @xcite . let @xmath18 be a simply connected complete + @xmath0 metric space . then , for any two points @xmath454 , there exists a unique geodesic connecting @xmath104 to @xmath174 . [ catgrowth ] let @xmath386 be a @xmath0 space and suppose @xmath464 and @xmath465 are unit speed geodesic rays in @xmath18 . then the function @xmath466 either grows linearly in @xmath157 or is bounded . _ proof : _ the key fact is that in a @xmath0 space , the distance function is convex ( for a proof , see @xcite proposition 5.4 ) . if the function @xmath467 is linear or bounded , we are done . assuming that the function is not linear but is unbounded , we can find an increasing sequence @xmath468 such that @xmath469 for all @xmath416 . one can now easily show that @xmath470 must grow at least linearly in this case . but , @xmath470 is bounded above by @xmath471 simply by measuring along the two geodesics . thus , @xmath470 grows linearly or is bounded . @xmath88 for the sake of completeness , we quote some results concerning the two coning operations we use in this argument . [ catcone ] let @xmath386 be a @xmath10 space . 1 . for all @xmath379 , @xmath424 exists and is a @xmath0 space . if @xmath472 and @xmath23 is a nonprincipal ultrafilter , then @xmath473 is a @xmath0 space . _ proof : _ for 1 ) this is proved in @xcite , 2 ) see @xcite or @xcite . also note that the lemma implies that the tangent cone to a @xmath13 space is also a @xmath0 space . curvature bounded below for metric spaces is defined in much the same way as for @xmath10 spaces . a standard resource for this material is @xcite . the definition and first uses of these spaces were due to a. d. alexandrov and are often referred to as `` alexandrov spaces '' . using the notation of the previous section , we make the following definitions . let @xmath386 be a complete metric space . then , @xmath386 is a @xmath11 space if * @xmath386 is a locally compact geodesic space . * let @xmath451 be geodesic triangle in @xmath18 with sides @xmath456 of lengths @xmath457 such that @xmath458 . for any two points @xmath459 , @xmath474 where @xmath461 and @xmath462 are the corresponding points on @xmath463 . as with @xmath0 spaces , there is a natural picture , @xmath9 triangles are `` fatter '' than euclidean triangles . as mentioned in the introduction , by the usual arguments using toponogov s theorem the localized version of this definition curvature is equivalent to the global and , for the purposes of this exposition , do not distinguish between the two . we end this section with some useful theorems concerning the geometry of @xmath11 spaces . [ cbtcone ] let @xmath386 be a @xmath11 space . 1 . any tangent cone to @xmath18 at @xmath104 is a @xmath9 space . if @xmath475 and @xmath23 is a nonprincipal ultrafilter then @xmath476 is a @xmath9 space . _ proof : _ the first claim is discussed in @xcite . the discussion in section 2.4.2 in @xcite reduces the second claim to the claim that the tangent cone to a @xmath9 space is @xmath9 as well . the latter fact is also discussed in @xcite . [ gp ] let @xmath18 be a locally compact , complete , noncompact @xmath9 metric space . if @xmath18 contains a line ( i.e. an unbounded geodesic which minimizes for all time ) then @xmath18 splits isometrically as @xmath477 where @xmath478 is some totally convex subset of @xmath18 and @xmath358 is the real line with its usual metric . for our purposes , we use this theorem to prove an analogous result to lemma [ catgrowth ] . [ cbtgrowth ] let @xmath386 be a @xmath9 space and let @xmath464 and @xmath465 be two lines in @xmath18 . then , @xmath464 and @xmath465 are either parallel or diverge linearly . _ proof : _ since these are lines , theorem [ gp ] says that @xmath18 splits isometrically as @xmath479 where @xmath358 is the geodesic @xmath480 . now , @xmath465 is a geodesic line in the product and hence its projection to each factor must be a geodesic line as well . therefore , the projection of @xmath465 in @xmath481 is either a point or a line . if it is a point , then @xmath480 and @xmath482 are parallel . if its projection in @xmath481 is not a point , then since it is a line , this implies that @xmath466 must be linear in @xmath157 . the purpose of this section is to use the differentiability result ( albeit weakly ) to examine the possible quasi - isometric embeddings of @xmath483 , a nonabelian simply connected connected nilpotent lie group equipped with a left invariant riemannian metric . as discussed in the introduction , we wish to prove that quasi - isometric embeddings of such a lie group into @xmath0 or @xmath9 can not exist . after taking asymptotic cones , denoting by @xmath3 the resulting nonabelian simply connected connected graded nilpotent lie group equipped with a left - invariant sub - riemannian metric , we reduce to proving the following theorem . * theorem c. * _ let @xmath14 be a connectes simply connected grapded nilpotent lie group equipped with a left invariant carnot - carathodory metric and @xmath15 be an open set . then @xmath16 does not admit a bilipschitz embedding into any alexandrov metric space with curvature bounded above ( resp . bounded below ) . _ [ diverge ] suppose @xmath484 is an l - bilipschitz map such that radial geodesics in @xmath3 are mapped to @xmath2-geodesic lines in @xmath18 . then , there exists a pair of geodesics @xmath464 and @xmath465 in @xmath18 which are images of radial geodesics and constants @xmath485 and @xmath486 such that , @xmath487 for @xmath488 . _ proof : _ this is an exercise in using the campbell - baker - hausdorff formula . suppose @xmath489 and @xmath490 for @xmath491 such that @xmath492 \neq 0 $ ] . using the bilipschitz property , we reduce immediately to consider the rate of growth of the function @xmath493 for @xmath494 . @xmath495+h(w , v , t)})\\ \intertext{where $ h(v , w , t)$ are the higher order bracket terms in the campbell - baker - hausdorff expansion . note that the order $ n$ bracket terms in the expansion each have a coefficient which of the form $ bt^m$ where $ b$ is a constant from the campbell - baker - hausdorff expansion and $ m < n$. therefore , for $ 0 < b < 1 $ this is equal to , } & { d_{cc}}(e^0 , h_{{|t|^{b}}}e^{\frac{1}{|t|^{b}}w+ \frac{t}{|t|^{2b}}[w , v]+ h'(w , v , t)})\\ & = |t|^b { d_{cc}}(e^0,e^{\frac{1}{|t|^{b}}w+ \frac{t}{|t|^{2b}}[w , v]+ h'(w , v , t)})\\ \intertext{again , $ h'(w , v , t)$ are higher order bracket terms but now , the terms or order $ n$ each have a coefficient of the form $ \pm bt^{m-{n b}}$ , where $ b$ a constant . now , picking $ 0 < \alpha < \beta < 1 $ such that $ m - n\alpha > 0 $ for at least one value of $ n$ and $ m - n \beta \le 0 $ for all values of $ n$ , we have that : } c_1 |t|^\alpha & \le { d_{cc}}(h_te^v , e^wh_te^v ) \le c_2 |t|^\beta\\ \end{alignat*}\ ] ] where @xmath496 is the minimum of the function @xmath497+h'(w , v , t)})$ ] on @xmath498 \cup [ 1,\infty)$ ] and @xmath499 is the maximum of the function @xmath500 + h'(w , v , t)})$ ] on the same domain . @xmath496 exists because the function in question is nonnegative and goes to infinity as @xmath501 while @xmath499 exists because the function in question is nonegative and goes to zero as @xmath502 . thus , using the l - bilipschitz property of the map , we have the desired result with the constants @xmath503 adjusted by either @xmath504 or @xmath505 . @xmath88 _ proof of theorem c : _ suppose @xmath5 is the l - bilipschitz map we wish to investigate and @xmath18 is either @xmath13 or @xmath11 . then , proposition [ conemap ] lets us reduce to an examination of a bilipschitz map @xmath506 where @xmath114 and @xmath507 are the appropriate tangent cones and @xmath384 takes radial geodesics to geodesic lines . lemma [ catcone ] or lemma [ cbtcone ] implies that the tangent cone @xmath424 is either @xmath0 or @xmath9 . now we satisfy the assumptions of lemma [ diverge ] and hence have a pair of geodesics with the specified divergence rate . however , this contradicts lemma [ catgrowth ] or lemma [ cbtgrowth].@xmath88 m. gromov . asymptotic invariants of infinite groups . in _ geometric group theory , vol . 2 ( sussex , 1991 ) _ , volume 182 of _ london math . lecture note ser . _ , pages 1295 . cambridge univ . press , cambridge , 1993

previous research on semiflexible polymers including cytoskeletal networks in cells has suggested the existence of distinct regimes of elastic response , in which the strain field is either uniform ( affine ) or non uniform ( non affine ) under external stress . associated with these regimes , it has been further suggested that a new fundamental length scale emerges , which characterizes the scale for the crossover from non affine to affine deformations . here , we extend these studies by probing the response to localized forces and force dipoles . we show that the previously identified nonaffinity length [ d.a . head _ et al . _ pre * 68 * , 061907 ( 2003 ) . ] controls the mesoscopic response to point forces and the crossover to continuum elastic behavior at large distances .

cond-mat0505662

semiflexible polymers such as filamentous proteins resemble elastic rods on a molecular scale , while exhibiting significant thermal fluctuations on the scale of micrometers or even less . this has made them useful as model systems allowing for direct visualization via optical microscopy . but , semiflexible polymers are not just large versions of their more well - studied flexible cousins such as polystyrene . filamentous proteins , in particular , have been shown to exhibit _ qualitatively _ different behavior in their networks and solutions . a fundamental reason for this is the fact that the thermal _ persistence length _ , a measure of filament stiffness as the length at which thermal bending fluctuations become apparent , can become large compared with other important length scales such as the spacing between polymers in solutions , or the distance between chemical crosslinks in a network . one of the most studied semiflexible polymers in recent years has been f - actin , a filamentous protein that plays a key structural role in cells @xcite . these occur in combination with a wide range of specific proteins for crosslinking , bundling and force generation in cells . these composites , together with other filamentous proteins such as microtubules , constitute the so - called _ cytoskeleton _ that gives cells both mechanical integrity and structure . this biopolymer gel is but one example of a large class of polymeric materials that can store elastic energy in a combination of bending and extensional deformations of the constituent elements . such systems can be called semiflexible gels or networks . one of the important lessons from recent experimental and theoretical studies is that the shear modulus of cross - linked semiflexible networks bears a much more complex relationship to the mechanical properties of the constituent filaments and to the microstructure of the gel than is the case for flexible polymer gels @xcite . recently , it has been shown that semiflexible gels exhibit a striking cross - over @xcite between a response to external shear stress that is characterized by a spatially heterogeneous strain ( a _ non - affine _ regime @xcite ) and a uniform strain response ( an _ affine _ regime @xcite ) . this crossover is governed primarily by cross - link density and molecular weight ( filament length ) . the bulk shear modulus of the network simultaneously increases by about six orders of magnitude at this nonaffine - to - affine cross - over . the underlying mechanism responsible for this abrupt cross - over appears to be the introduction of a new , mesoscopic length scale in the problem that is related to both the bending stiffness of the constituent polymers and the mean spacing between consecutive cross - links along the chain @xcite . one can associate this mesoscopic length with the length below which the deformation of the network departs from the standard affinity . the nonaffinity length @xmath0 , introduced in refs.@xcite , can be qualitatively understood as the typical length over which one finds nonaffine deformation in the network . in this previous work we presented a scaling analysis that relates this mesoscopic length scale to the network density and the stretching and bending moduli of the constituent filaments . the macroscopic shear response of the network is then controlled by a competition between @xmath0 ( the nonaffinity length ) and the filament length , @xmath1 . on the one hand , when the filament length is long , nonaffine corrections to the deformation field , which are localized to regions within @xmath0 of the filament ends , do not significantly affect the mechanical properties of the network ; the shear modulus of the macroscopic system is well - described by calculations based on affine deformation reflecting the fact that nonaffine deformations of the filament ends are subdominant corrections in this limit . moreover , the elastic energy is stored primarily in the ( homogeneous ) extension and compression of filaments . on the other hand , when the filaments are of a length comparable to , or shorter than the nonaffinity length , _ i.e. _ @xmath2 then the nonaffine deformations of the ends play a large , even dominant role in determining the mechanical properties of the network . the network is found to be generally more compliant , and the elastic energy under applied shear stress is stored in a spatially heterogeneous manner in the bending of filaments . the existence of these distinct regimes as a function of filament length reflects a fundamental difference of these semiflexible polymer networks with respect to their flexible counterparts : polymers can maintain their mechanical integrity and state of stress / strain across network nodes or crosslinks . these results naturally lead one to pose a number of basic questions regarding the elastic properties of semiflexible networks . while these random networks must on basic theoretical grounds appear as continuum , isotropic materials at the longest length scales , these considerations do not predict the length at which the continuum approximation applies . the previous identification of the nonaffinity length , @xmath0 as the only mesoscopic length associated with the nonaffine - to - affine cross - over in uniformly sheared semiflexible networks suggests that this length should more generally control the cross - over to continuum behavior @xcite . after all the affine deformation of the network under uniform stress at scales large compared to @xmath0 shows that in one case at least the nonaffinity length controls the cross - over to continuum behavior . one of the principal results of the present work is the demonstration that @xmath0 more generally controls this cross - over to continuum mechanics in semiflexible gel systems . prior work has focussed exclusively on simple shear and uniaxial extension . in order to better examine the universality of the previous results , we study the opposite limit of a highly localized external force in the form of a point force monopole or dipole . if one can show that the elastic ( displacement ) green s function of the system similarly depends on only one additional parameter @xmath0 then it would appear that this quantity completes the coarse - grained elastic description of the system on all length scales down to the mean distance between cross - links . it may be , however , that the deformation field of these semiflexible networks is much more complex and the simplification introduced by @xmath0 in the description of the network s response to uniform shear strain can not be generalized to deformations resulting from more general forcing conditions . while our results below , indeed , show that @xmath0 does largely control the crossover to ( the _ far field _ ) continuum elasticity , the observed elastic green s function is sensitive to the local structure of the network on length scales below @xmath0 . below we discuss how we quantify the structure of the green s function and its approach to the form required by continuum elasticity . the observed elastic green s function , however , depends not only on the @xmath0dependent lam coefficients of the material , but also on local properties of the displacement field immediately surrounding the the point force . in effect one may imagine that , upon the application of the point force , the network acts as a type of composite material : within a distance @xmath0 of the point force it deforms in a way not well described by continuum theories , while outside of that zone it does appear to act like an elastic continuum . the complete green s function depends , of course , on the material properties of both media . unfortunately , only one of those media ( the outer zone ) is well characterized by the simple continuum theory of an isotropic elastic solid , so the complete green s function remains complex and depends on the detailed network structure within the inner zone surrounding the applied force . there is another set of questions that may be addressed via the study of the network s response to point forces and force dipoles . such forces not only probe the material properties of the network in a manner complimentary to the uniform strains explored earlier , they also have direct physical implications for microrheology in f - actin networks and for the dynamics of the cytoskeleton in response to the activity of nanoscale molecular motors , _ e.g. _ myosin . fluctuation - based microrheology , an application of the fluctuation - dissipation theorem to the study of rheology via the statistical analysis of the thermally fluctuating position of sub - micron tracer particles embedded in the medium , requires one to understand the elastic green s function of the medium . thus understanding the response of semiflexible networks to localized forces has direct experimental implications and consequences for force production in the cytoskeleton . in the biological context , the semiflexible network making up the cytoskeleton is generally found in association with molecular motors that , to a good approximation , generate transient localized force dipoles in the material . to both understand force generation in the cell as well as the material properties of these cytoskeletal networks driven into nonequilibrium steady states by these molecular motors , one must determine the displacement field associated with such motor - induced forces . notwithstanding our biological motivation for this work , our findings also bear on the broader problem of elastic modes in amorphous materials . it has been shown that the vibrational modes of deep quenched lennard jones systems approach a continuum description only on scales exceeding some mesoscopic length @xmath3 @xcite ; for the protocols considered , a value @xmath4 particle dimensions was robustly found . this was physically identified with a length scale for non affinity , suggesting a direct correspondence with our @xmath0 ( although our @xmath0 can be controlled by varying the mechanical properties of the constituents ) . a comparable length was also found to control the self averaging of the green s function to the form expected by continuum elasticity @xcite . these findings for radially interacting particles are broadly in keeping with our own investigations for semiflexible polymer networks . we also mention here that the relationship between continuum elasticity and the green s function has also been discussed for mildly disordered spring networks @xcite . the remainder of this paper is organized as follows : in section [ s : model ] we develop our model of semiflexible , permanently cross - linked gels , summarize the numerical simulations used to study it , and discuss the expected structure of the displacement field when averaged over numerous realizations of the network . in section [ s : results ] we report our results for both point forces in [ ss : point - force ] and force dipoles in [ ss : dipole ] . we then discuss our studies of the bulk elastic properties of these networks in [ ss : bulk ] before concluding in section [ s : discussion ] . a highly successful continuum continuum model of individual semiflexible polymers is the _ worm like chain_. this treats the linear filaments as elastic rods of fixed contour length and negligible thickness , so that the dominant contribution to the elastic energy comes from bending modes and the hamiltonian linearized to small deviations from a straight configuration that is given by @xmath5 where @xmath6 is the transverse displacement of the filament relative to an arbitrary straight axis , @xmath7 is its arc length , and the elastic modulus @xmath8 gives the bending energy per unit length @xmath9 . the longitudinal response of the wormlike chain model is calculated from the increase in free energy due to an extensional stress applied along the mean filament axis @xcite . however , the numerical algorithm used in our simulations is based on the minimization of the hamiltonian of the system , and hence is fundamentally athermal . the reason for this choice is essentially one of efficiency : assuming there are no subtle convergence issues , minimization is expected to be faster than stochastic modelling and , it is anticipated , give better statistics for a given cpu time . this does , however , mean that the entropic mechanism governing the longitudinal response is absent , and an explicit energetic term is required . an unconstrained filament at @xmath10 forms a straight configuration , and thus elongation of its end to end distance must be accompanied by a change in the absolute contour length . it is therefore natural to incorporate longitudinal modes by adding a second elastic term to the hamiltonian for the extension or shortening of the filament backbone , @xmath11 where @xmath12 gives the strain or relative change in local contour length , and @xmath13 is the young s modulus of the filament ( essentially a spring constant normalized to 1/[length ] ) . of course , such modes also exist in thermal systems , but may be dominated by the entropic spring terms @xcite , except possibly for very short filament segments or very densely cross - linked gels . the connection between thermal / entropic and athermal longitudinal compliance is discussed in greater detail in ref . @xcite . the two elastic coefficients @xmath8 and @xmath13 together define a length scale @xmath14 , which will shall refer to as the intrinsic _ bending length _ by observing that an isolated filament constrained to have different tangents at its end points will deform with this characteristic length . to avoid potential confusion , however , we note that this is _ not _ the typical length scale for bending deformations of a semiflexible filament . rather , the bending energy of filaments tends to make the longest unconstrained wavelength bending mode the dominant one . thus , for instance , in a crosslinked gel , filaments are expected to be bent primarily on a length comparable to the distance between crosslinks . nonetheless @xmath15 is a useful measure of filament rigidity , in that large @xmath15 corresponds to rigid filaments , and small @xmath15 to flexible ones . although @xmath8 and @xmath13 have been introduced as fundamental coefficients , if the filament is regarded as a continuous elastic body with uniform cross section at zero temperature , then they can both be expressed in terms of the characteristic filament radius @xmath16 and intrinsic bulk modulus @xmath17 as @xmath18 and @xmath19 . thus @xmath20 , and thinner filaments are more flexible than thick ones ( as measured by @xmath15 ) , as intuitively expected . given the possibility of entropic effects in @xmath13 , however , we shall treat these as independent parameters of the theory @xcite the gel is constructed by depositing filaments of monodisperse length @xmath1 and zero thickness onto a two dimensional substrate . the center of mass position vector and orientation of the filaments are uniformly distributed over the maximum allowed range , so the system is macroscopically isotropic and homogeneous . whenever two filaments overlap they are cross - linked at that point . deposition continues until the required mass density , as measured by the mean distance between crosslinks @xmath21 , has been reached . the network thus constructed can be described by three lengths : @xmath22 , and @xmath21 and one modulus scale : @xmath13 . in two - dimensions @xmath21 characterizes both the mass density and cross - link density in spatially random , isotropic networks . the length @xmath15 characterizes the mechanical properties of the constituent filaments via the ratio of their bending to stretching compliance . the overall modulus scale @xmath13 will be absorbed into the point forces applied to the network . previously @xcite , we identified an additional length @xmath23 , where @xmath24 . this _ non affinity length _ characterized the crossover from non affine to affine network response . specifically , for filament lengths @xmath1 much larger than @xmath0 ( _ i.e. _ , high molecular weight ) , the bulk network properties could be understood quantitatively in terms of affine strains , while significant non affine effects were observed for @xmath25 . although this length arose naturally from considerations of bulk network properties @xcite , its dependence on @xmath21 and @xmath15 can be understood in terms of a balance of stretching / compression and bending energies of a single filament , treated in a self consistent way within a network @xcite . it is important to note that this ( material ) length is intermediate , between the ( geometric ) network length @xmath21 and the macroscopic scale . in fact , in dilute networks , for which @xmath26 , we shall argue below that the network can be thought of on scales @xmath27 as a quasi continuum : continuous , as opposed to discrete , but not necessarily described by macroscopic continuum elasticity . we shall find that this non affinity length will play a key role in our analysis of the displacement field on this intermediate / quasi continuum scale . in order to apply the point forces , a crosslink is chosen at random and identified as the origin of the system ; it is this crosslink that will later be perturbed . a fixed circular boundary at radius @xmath28 from the origin is imposed , and any filaments or filament segments that extend beyond the boundary are simply removed or truncated , respectively . filaments ending on the rigid boundary are fixed there by another freely rotating bond as are found at all cross - links in the system . the allowed free rotation at the boundary means that the boundary supplies arbitrary constraint forces on the network but can not support any localized torques . details of the simulation method have been presented elsewhere @xcite . here we briefly summarize the procedure , with particular attention on those aspects that are central to the problems studied in this paper . the system hamiltonian @xmath29 is constructed from discrete versions of ( [ e : h_wlc ] ) and ( [ e : h_hooke ] ) applied to the geometry generated by the random deposition procedure described above . the degrees of freedom @xmath30 , or ` nodes , ' are the position vectors of both crosslinks and midpoints between crosslinks , the latter so as to incorporate the first bending mode along the filament segments . additional nodes could be included at the cost of additional run time , but are expected to have a small effect , since the longest unconstrained wavelengths tend to dominate bending deformations . crosslinks are treated as constraints on the relative position of each connected filament segment , but not on their relative rotation . physically this corresponds to an inextensible but freely rotating linkage . as previously noted , constrained bending at crosslinks has a small effect except at high network concentrations ( specifically , when @xmath21 becomes comparable to @xmath16 ) @xcite . nodes on the boundary are immobile . note that , as in our earlier work @xcite , the network is assumed to be initially unstressed on both macroscopic and microscopic length scales . there are two ways in which the system may be perturbed . the first , which we call _ monopole _ forcing , is to apply an arbitrarily small external force @xmath31 to the crosslink at the origin . the network is then allowed to relax to a new configuration consistent with mechanical force balance at every node . the second , which we call _ dipole _ forcing , is to introduce a geometrical defect into the system by moving the central crosslink along the contour length of one of the filaments to which it belongs , but not the other . physically , this corresponds , _ e.g. _ , to a motor introducing relative motion of one filament with respect to another filament . in the simulations , this effect is incorporated by infinitesimally ` shifting ' the image of the central node with respect to other nodes on one filament . once the perturbation has been specified , the displacements of the nodes in the new mechanical equilibrium are calculated by minimizing the system hamiltonian @xmath29 using the conjugate gradient method @xcite . this generates a displacement field for the particular geometry under consideration , as shown in fig . [ f : eg_indiv_run ] . note that @xmath29 is linearized about small nodal displacements @xmath32 from their original positions @xmath33 , so linear response is assured . the bulk response to non linear strains has been recently studied by onck _ et al . _ @xcite . ( grey line segments ) perturbed by an external force , denoted by the large red arrow , applied to the crosslink at the origin of a circular system of radius @xmath34 . the arrow lengths are logarithmically calibrated to the magnitude of the displacement of at each crosslink . in this example , @xmath35 , @xmath36 and the force is perpendicular to one of the filaments that form the central crosslink ; forces can also be applied parallel to a filament . , width=302 ] as shown in fig . [ f : eg_indiv_run ] , the displacement field for a particular network is quite complex and generically shows anti - correlations between displacements and the local mass distribution . although these fluctuations reflect inherent and possibly interesting physical properties of the gels , a more basic and immediately applicable quantity to measure is the _ mean _ displacement field , found by averaging many individual runs with differing geometries but identical system parameters @xmath28 , @xmath1 , @xmath21 and @xmath0 ( or equivalently @xmath28 , @xmath1 , @xmath21 and @xmath15 ) . two examples for differing @xmath0 ( the radius of the green circle centered on the point of force application ) are given in fig . [ f : eg_averaged ] . these plots demonstrate one of the primary results of this paper , namely that the crossover between continuum ( or quasi continuum ) response at large lengths , to a more exotic displacement field at shorter lengths , happens at a length of order @xmath37 with a prefactor close to unity . networks with the same parameters as in fig . ( [ f : eg_indiv_run ] ) ( including one filament fixed perpendicular to the external force ) except for @xmath0 , which is _ ( a ) _ @xmath38 and _ ( b ) _ @xmath39 . for easy visualization , a green circle of radius @xmath0 has been inserted into the background of each plot . vectors near the center of each system have not been plotted for clarity . , title="fig:",width=302 ] networks with the same parameters as in fig . ( [ f : eg_indiv_run ] ) ( including one filament fixed perpendicular to the external force ) except for @xmath0 , which is _ ( a ) _ @xmath38 and _ ( b ) _ @xmath39 . for easy visualization , a green circle of radius @xmath0 has been inserted into the background of each plot . vectors near the center of each system have not been plotted for clarity . , title="fig:",width=302 ] in order to precisely describe the structure of the deformation field , we consider its most general possible form . for monopole forcing , the displacement field @xmath40 at position vector @xmath41 relative to the origin can be projected onto 3 other vectors , namely the direction of the external force @xmath42 , the unit position vector @xmath43 and the axis @xmath44 of one of the filaments to which the crosslink is attached , @xmath45 where @xmath46 is the magnitude of the external force , and @xmath47 is the shear modulus as predicted for affine deformation . this depends on @xmath1 and @xmath21 but _ not _ @xmath0 , and is included here to factor out the density dependence of the response . as defined , the @xmath48 ( and the @xmath49 below ) are dimensionless quantities in two dimensions . each @xmath48 can be further decomposed into angular modes in @xmath50 , where @xmath51 , @xmath52 terms in @xmath53 vanish since the ensemble averaged response must be invariant under @xmath54 , and @xmath55 terms with @xmath56 odd also vanish due to @xmath57 invariance . this latter symmetry may appear to violate the known polarity of typical semiflexible biopolymers such as f - actin and microtubules @xcite ; however , we are interested here in the _ mechanical _ properties of the filaments , which , within the approximations of our model , are indeed invariant under @xmath57 . for dipole forcing , the above procedure is followed with the additional simplification that @xmath58 is proportional to @xmath59 , since the displacement ( and hence force ) dipole induced by the motion of a motor will always be parallel to one filament axis . the decomposition is therefore somewhat simpler , @xmath60 the scalar @xmath46 is the magnitude of the force dipole ; since it is actually a displacement that is imposed , @xmath46 is unknown as will be treated as a fitting parameter . the angular decomposition is identical to before , @xmath61 later sections will refer to the continuum solution for each of the two forms of forcing . these are given in the appendix . for the monopole case , only two continuum modes are non zero , namely @xmath62 and @xmath63 . we shall refer to these components of the green s function as _ continuum modes _ in order to distinguish from those components ( _ non continuum modes _ ) that must vanish in the continuum . we do this even though for our finite systems the _ non continuum modes _ do not vanish . the dipole modes are slightly more subtle : after averaging of many dipole fields generated by the simulation , the resulting field is _ quadrupolar_. this is an immediate consequence of the means of forcing the system . recall that in an elementary step , a motor moves parallel to one filament axis . this has the effect of compressing the filament in front of the dipole , while stretching the trailing segment . thus two filament segments are perturbed , each of which can be treated as a force dipole which , for a particular network geometry , will be of different magnitudes and hence the resulting field is dipole . however , the net bias of this dipole is symmetrically distributed around zero , and thus vanishes after averaging , leaving a quadrupole field as shown in fig . [ f : eg_quadrupole ] . as derived in the appendix , the non zero modes for this field are @xmath64 , @xmath65 , @xmath66 and @xmath67 . individual fields induced by imposing a displacement dipole at the origin . the orientation of the dipole is given by the red arrow . as explained in the text , the mean dipole moment vanishes after averaging and the resulting field is quadrupole . a green circle at a radius @xmath68 from the origin is also shown . , width=302 ] the mechanical response to a localized perturbation depends on the distance from the point of perturbation . we divide the discussion of these results into the following parts . first , we examine the response to a point force using the decomposition of the displacement field outlined in previous section . we contrast the decay of the non - continuum modes of the displacement field with the behavior of the continuum modes of the displacement field and discuss the finite - size effects in the simulations . at large length scales , we find that the deformation field approaches a quasi continuum form , in which all but a small number of ensemble average modes decay rapidly toward zero . the remaining modes of the deformation field are the same as those predicted by continuum elasticity . in other words , the strain field about a point force reflects the expected tensorial character and rotational symmetries based on continuum elasticity theory . we find that @xmath0 again plays a central role in controlling the cross - over from the near - field to the quasi - continuum . we then turn to the spatial structure of the elastic energy density field around the point force paying particular attention to the partitioning of that energy density between stretching and bending modes of the filaments . we observe that the ratio of these energy contributions achieves the bulk value over much shorter distances from the point force than does the structure of the displacement field acquire its far - field , or bulk structure . we then extend our analysis to consider force dipoles in the medium . lastly , we extend our previous analysis of the bulk elasticity of the filament network by examining both the young s modulus and the poisson ratio of the network . in this section we focus on the short length scale behavior of the monopole response , for which the non continuum modes are non zero . we wish to distinguish two distinct forms of this non continuum behavior : _ ( i ) _ higher angular modes @xmath69 and @xmath70 with @xmath71 are non zero , and _ ( ii ) _ the @xmath72 modes do not vanish , _ i.e. _ the response depends on the orientation of the filament to which the force is applied . this latter observation gives a clear indication of how the response can ` see ' the microscopic structure of the gel on short length scales . an example demonstrating the appearance of non continuum modes at short lengths is given in fig . [ f : gf2_5p2_r1l ] , which shows the @xmath73 mode for systems with different crosslink densities @xmath74 but with the filament flexibility chosen to give the same @xmath75 in each case . this plot shows the decay of the @xmath76 amplitude of the component of the displacement field in the direction of the applied force ( at the origin ) . in all three systems one observes a rapid decay of this angular harmonic that must vanish for a continuum isotropic system . moreover , the characteristic length scale for this decay appears to be of order @xmath0 ( @xmath77 ) , although we examine this point more quantitatively below . for all network densities and values of @xmath0 studied it is clear that the magnitude of @xmath73 vanishes rapidly with distance from the point force . no data are shown for larger values of @xmath74 since at high network densities the numerical convergence of the strain field is so slow as to prevent attaining meaningful statistics . ( which is dimensionless ) versus distance from point of force application @xmath78 for systems with the same @xmath36 and crosslink densities @xmath74 as given in the key ( @xmath1 is the filament length ) . the system radius @xmath34 in all cases . , width=321 ] this picture of non continuum modes decaying to near zero at a length comparable to @xmath0 holds also for the @xmath79 modes of @xmath70 and for all the @xmath71 modes of @xmath69 . for reasons of space we do not show these data here . instead we present data in fig . [ f : gn0_5p2_r1l ] that demonstrate the decay of @xmath80 with @xmath78 for three networks densities such that in each case @xmath81 . this non continuum mode of the displacement field measures the circularly averaged amplitude of the component of displacement in the direction @xmath82 , _ i.e. _ along the axis of the rod to which the force has been applied . clearly , such behavior has no counterpart in an isotropic continuum elastic model . it is interesting to note that this amplitude also appears to decay exponentially with a decay length of order @xmath0 . versus @xmath83 for @xmath36 , @xmath34 and the @xmath74 given in the key . , width=321 ] in order to test quantitatively whether @xmath0 indeed controls the decay of those components of the displacement field that have no counterparts in the continuum theory , we examine , as an example , @xmath84 vs. @xmath78 more closely in fig . [ f : log - lin - gf2 ] . here we plot this for a range of values of networks density _ and _ of @xmath0 . if , as suggested above , the decay is exponential with characteristic length @xmath0 , then plotting these data log - linear with radial distances scaled by @xmath0 should cause all these curves to exhibit the same slope . we have shifted the data sets vertically in order to facilitate visual comparison . ( we note that , although a known force is applied to the origin of our sample , since we expect the unknown local constitutive relations in our system to depend on density and other parameters , the amplitudes can not be directly compared . here , we wish only to establish the nature of the decay of the non continuum modes . ) we have also introduced the solid lines corresponding to @xmath85 merely as guides to the eye . these are not fits , although fits to these data for various @xmath86 and @xmath0 demonstrate decay lengths of @xmath0 within 10% . versus the scaled distance from perturbation @xmath87 for systems with various values of @xmath0 as given in the key . the system radius @xmath34 in all cases . data within a boundary layer of width @xmath0 from @xmath88 is not shown ( see text for details).,width=321 ] thus , we observe a near field regime in which the decay of the @xmath84 appears to be generically more rapid than @xmath89 followed by the quasi - continuum regime where the exponential decay with decay length @xmath0 is observed . this demonstrates that the nonaffinity length @xmath0 , indeed , controls the approach toward the expected continuum behavior ( _ i.e. _ , vanishing of the non continuum modes ) . similar results ( not presented here for reasons of space ) are also found for the other non continuum modes . specifically , the decay lengths are also found to be @xmath0 within 10% . of course , the fixed zero - displacement boundary condition at @xmath88 requires all components of the displacement field to vanish there . the essential distinction between the continuum and non - continuum modes lies in the manner in which their amplitudes decay upon approach to the rigid boundary . the non - continuum modes examined above decay exponentially with a decay length proportional to @xmath0 . we demonstrate in the next section that the continuum mode amplitudes , in contrast , do not appear to decay in the same way . in particular , their approach to zero at the rigid boundary is _ not _ controlled by @xmath0 . in this section we address two related points : ( i ) the fundamental change in the structure of the displacement field as one moves away from the immediate vicinity of the applied force , and ( ii ) the role of finite size effects in our numerical simulation in two - dimensions . we can not address the former point without confronting the latter one for the following reason . due to the presence of the rigid boundary , all modes of the displacement field decay to zero at the boundary . we show in this section , however , that the decay of the non - continuum modes is controlled by internal , mesoscopic length scales in the network , while the decay of the continuum modes is determined by the macroscopic geometry of the system including the presence of the rigid boundary . since the distinction between the continuum and non - continuum modes depends essentially on the presence of the boundary and specifically on the separation of the system size @xmath28 from the internal length scales controlling the decay of the non - continuum modes ( @xmath90 ) , we consider this distinction along with a more general discussion of finite size effects in our simulation . versus @xmath83 for @xmath36 , @xmath34 and differing @xmath74 as shown in the legend . , width=321 ] to have a well - defined elastic response in two - dimensions ( meaning that the displacement field vanish at large distances from the applied point force ) one needs to impose a rigid boundary . in order to eliminate the differential influence of the boundary on the various angular modes of the displacement field we chose this boundary to be a circle of radius @xmath28 centered on the point of force application this rigid boundary forces all components of the displacement field to vanish exactly at @xmath88 . such a rigid boundary can be expected to introduce a boundary layer near the edges of our system , which we observe and discuss below . figure [ f : gr0-r = l ] shows the decay of the @xmath91 mode amplitude for systems of size @xmath34 and @xmath75 . we note that the decay of this _ continuum _ mode is qualitatively distinct from the decay of the non - continuum modes shown in figs . [ f : gf2_5p2_r1l ] , [ f : gn0_5p2_r1l ] , and [ f : log - lin - gf2 ] . we do not observe an exponential decay of this continuum mode amplitude . nevertheless , observing an exponential decay having a long decay length , or , more reasonably , the product of an algebraic function and such a weak exponential decay would be difficult to resolve in this plot . to study these issues further , we plot in fig.[f : gr0_syssize ] @xmath92 for various system sizes from @xmath93 to @xmath94 . we observe both an apparent convergence for the larger systems to a common curve when distances are measured relative to the size of the system , as well as a systematic down - turn near the boundary . the first of these observations suggests that the sample geometry controls the dependence of this continuum mode , as opposed to intrinsic lengths like @xmath0 . the down - turn in these data within a distance of order @xmath0 ( here , @xmath95 ) of the boundary is to be expected from the observations above concerning the role of the length scale @xmath0 . one can view this crossover length as the distance along a filament over which a force applied to the filament dissipates / expands into a ( quasi-)continuum stress / displacement field . thus , the effects of the boundaries on the individual filaments contacting the boundary are expected to propagate at least this distance into the system . because of this , we have also removed all data within a distance @xmath0 of the boundaries in fig . [ f : log - lin - gf2 ] above , which exhibit a similar down - turn near the boundary . in fig . [ f : lambda_collapse ] we examine the radial dependence of @xmath92 for three different values of @xmath96 for a large system where @xmath97 . in this case , we see a coincidence of the data for different @xmath0 , indicating that the long - range behavior of @xmath92 is not controlled by @xmath0 . although we were unable to obtain sufficient statistics on the non continuum ( and sub dominant ) modes for larger systems , the combination of the qualitatively different behavior from the continuum modes , together with the consistent dependence on the crossover length @xmath0 lead us to conclude that the principal results of the prior section are not strongly influenced by finite - size effects . thus , @xmath0 controls the disappearance of the non - continuum modes of the system . versus @xmath83 for @xmath36 , @xmath98 and the system radii @xmath28 given in the key . the continuum response for the bulk elastic moduli corresponding to these parameter values ( as given by ( [ e : ctm_gr0 ] ) ) is shown as a solid line . , width=321 ] versus @xmath83 for @xmath99 on log linear axes , with @xmath96 and @xmath74 as given in the key . each curve has been shifted vertically by a @xmath0dependent scale factor to attempt data collapse . since @xmath0 varies by over a factor of 4 , the rough collapse is enough to rule out any significant @xmath0dependence on the decay . error bars are not shown for clarity.,width=321 ] we also see a large discrepancy between the expected continuum solution and the observed displacement of the simulated filament network as shown in fig . [ f : gr0_syssize ] . this difference can not be simply attributed to finite - size effects ; as can be seen in fig . [ f : gr0_syssize ] , there is no observable convergence of the larger system data for the continuum modes to the predictions of continuum elasticity for a material having the appropriate lam coefficients . nevertheless , we clearly observe the more rapid decay of the non - continuum modes on a length - scale controlled by @xmath0 . one is left with the following puzzle : the tensorial structure and the rotational symmetry of each component of the displacement field approaches the form required by the continuum theory , but the continuum model appears to never quantitatively agree with the numerical data . we may speculate about the underlying cause of this discrepancy . it is clear that that the deformation of the material in the immediate vicinity of the point force ( this `` near zone '' extends out to a distance @xmath90 from the point force ) and in a boundary region at the rigid wall is not well described by any continuum theory . we suggest that under the application of the point force at the origin , the network effectively partitions itself into three different elastic materials . in the near field region @xmath100 around the point force the deformation response to the point force is quite complex . similar complexities appear to exist near the rigid wall , where @xmath101 . these regions apply a complex set of tractions on the circles @xmath102 and @xmath103 bounding the intermediate region that deforms in a manner consistent with some continuum elastic material . because these tractions are not themselves determined by a simple , continuum model , the resultant deformation of the intermediate region is also not simply derivable from an analysis of the elastic green s function for a continuum . one might imagine that for very large systems having a consequently larger intermediate region , the complexities of the tractions in the transition zones become less significant . because of the rigid boundary at @xmath88 and the low - dimensionality of the system , this may not be the case . the continuum green s function contains logarithmic terms and , due to the rigid boundary , growing polynomial terms as well . thus , we do not expect convergence to the continuum green s function even for significantly larger systems . one may ask whether three dimensional systems will show similarly poor convergence to the continuum solutions . further research here is needed . we suggest that we do indeed observe the approach of the structure of the elastic green s function to that predicted by continuum theory . we term the region surrounding the point force where the deformation field has the expected form the quasi - continuum . based on our numerical data , we do not expect to find a region in which the deformation field agrees quantitatively with the predictions of continuum elasticity using the lam coefficients appropriate to the medium as determined by uniform stress measurements . this suggests that semiflexible networks admit a highly complex point force response that can not be fully captured by continuum elasticity even in the far - field . the full implications of this complexity have not been explored . another instructive measure of the monopole response is the density of elastic energy @xmath104 at a given distance from the point perturbation , which measures gradients of the displacement field and therefore complements the analysis of @xmath105 given above . in addition , since the simulation contains explicit terms for the transverse and longitudinal filament deformation modes , it is straightforward to measure the partitioning of the energy between these modes . based on previous work @xcite , under homogeneous shear strain the partitioning of elastic energy between the bending and stretching modes of the filaments is determined entirely by the affine - to - nonaffine crossover . the ratio of @xmath106 was found to control this energy partitioning at a macroscopic or average level . it remains to be seen how this partitioning of the elastic energy occurs in vicinity of a point force . the freedom to choose the angle between the point force and the direction of the filament to which that force is applied allows one to determine locally the partitioning of the elastic energy between bending and stretching modes of the filaments . forces directed along the filament axis generate primarily stretching deformations in the immediate vicinity of the origin ( where the force is applied ) . forces directed perpendicular to the filament axis , however , locally create a large bending deformation . as seen in the previous homogenously imposed strain deformation calculations , for any given value of @xmath106 the network responds very differently to the bending or stretching deformations . thus it is not surprising that the decay of the energy density from the point of force application to the boundary is strongly dependent on whether the force is applied parallel or perpendicular to one of the filaments at the crosslink . [ f : toted_fixedlam_par ] shows @xmath104 for parallel forces , where @xmath0 is fixed but the filament density @xmath74 varies . there is an approximate data collapse onto a single curve as shown . the most notable discrepancy occurs near the boundary , where the low density data remains higher than those for larger @xmath74 values . this is most likely due to the boundary layer already discussed above for the displacement modes . ( in arbitrary units ) versus distance @xmath78 from an externally applied force that is directed along the filament to which it is applied . @xmath74 is varied as given in the key , and @xmath34 and @xmath36 in all cases . , width=321 ] and the data has been collapsed in the small @xmath107 regime . the solid line is proportional to @xmath108 , showing @xmath21 controls the initial decay of elastic energy . in the lower panel the same energy densities are plotted against distance scaled by @xmath0 . this shows the energy density decays with that characteristic length scale at longer ranges . error bars are not shown for reasons of clarity . , width=321 ] in contrast , @xmath104 for forces perpendicular to a filament exhibits much richer behavior , as shown in fig . [ f : toted_fixedlam_perp ] , which shows the same data plotted against @xmath107 and @xmath87 . for large distances , the data for different @xmath74 appear to differ by only a scale factor . since there is one arbitrary multiplicative factor for each curve ( as only the magnitude of the applied force is fixed , not its displacement and hence work done on the network ) , these portions of the curves can be made to collapse after scaling , as in the parallel force case , suggesting that beyond some near - field regime , the decay of elastic energy density is once again exponential and governed by @xmath0 see fig . [ f : toted_fixedlam_perp ] lower panel . however , in this near - field regime for perpendicular forces there is clearly no possibility of such a collapse . instead , it appears that the more rapid decay of elastic energy density in this near - field regime is governed by the length @xmath21 , as can be seen in the upper panel of fig . [ f : toted_fixedlam_perp ] where the ( rescaled ) energy density data are plotted vs. @xmath107 and shown to decay as @xmath109 for distances @xmath110 ( ignoring finite size effects ) . based on the two different regimes of data collapse shown in fig . [ f : toted_fixedlam_perp ] , we conclude that there is a near - field region in which the decay of elastic energy is governed by a microscopic length scale the mean distance between cross - links and a longer - range regime in which the spatial decay of elastic energy density is controlled by the mesoscopic length @xmath0 . to better study this cross - over , we examine the partitioning of elastic energy between bending and stretching modes of the filaments . recall that in the affine limit , the elastic energy is stored primarily in the stretching and compression of the filaments . in the nonaffine regime , on the other hand , the elastic energy is stored almost entirely in bending modes of the filaments . plotted in fig . [ f : fraced_fixedlam_perp ] is the proportion of elastic energy due to longitudinal filament deformation for the perpendicular force case . three data sets are displayed having different values of @xmath111 and @xmath0 see figure caption . the distance from the point of force application has been scaled by the geometric mean of @xmath111 and @xmath1 leading to the observed coincidence of the crossover between regimes in the data . close to the point of the application of the force , the displacement response is clearly dominated by bending modes of the filaments . this is to be expected since the perpendicularly directed force directly injects bending energy into the system at the origin the point of force application . the energy partitioning is inconsistent with that which is expected based on previous homogeneous shear measurements . from that work , one expects the network to determine the energy partitioning based only on the ratio : @xmath106 . we observe the fraction of stretching energy to rapidly increase towards the expected value based on @xmath106 @xcite and we may characterize each curve has having a ` knee ' separating the region of varying energy ratio from a nearly constant intermediate regime where the partitioning of elastic energies corresponds well with the previously identified fraction of stretch / compression energy under macroscopic strain . this correspondance is demonstrated in fig . [ f : fraced_fixedlc_perp ] . the horizontal lines show the fraction of stretching energy in periodic systems subjected to macroscopic shear , and clearly coincide with the plateau reached beyond the knee . at the largest distances one notices the vanishing of bending energy as the fraction of stretching energy approaches unity . this last effect is due to the fact that each filament has a freely rotating bond at the outer , rigid wall . as this wall can not support torques , the bending energy vanishes in a boundary layer whose width is determined by the mean distance between cross - links . we return to this point below . , for three systems having differing values of @xmath111 ( see key ) but the same value of @xmath0 . , width=321 ] ( in fact , the exact same geometries ) , @xmath34 but different @xmath0 . the different values of the stretching energy fraction in the intermediate regime reflect the differing values of the ratio @xmath106 for the networks . for comparison , the macroscopic energy fraction under bulk shear is shown as horizontal dashed lines , in the same order as the data points ( _ i.e. _ @xmath38 , 0.191 and 0.412 from top to bottom ) . , width=321 ] the observed coincidence of the knees of all three curves under the rescaling of distances by @xmath112 suggests that this length sets the scale over which injected bending energy is redistributed into the combination of bending and stretching appropriate for long - length scale deformations . based on these observations we note that in all networks the strain field acquires the structure of the point force response based on continuum elasticity over a ( typically mesoscopic ) length scale of @xmath0 . when the networks is subjected to large , local bending deformations , however , it readjusts the partitioning of bending to stretching energy over a generally much shorter length scale , @xmath111 . thus the system is able to repartition the local elastic energy storage to the value appropriate for its @xmath106 ratio over smaller length scales than does the system recover the expected long length scale structure of its continuum elastic response . we have already shown that in the intermediate field regime the fraction of stretching elastic energy in the system approaches its far - field value as determined by the ratio : @xmath106 . nevertheless , we in fact find that at the edges of our sample the network energy becomes stored solely in stretching modes regardless of the value of @xmath106 . as mentioned above we attribute this final redistribution of the elastic energy density between bending and stretching modes to a boundary effect imposed by the freely - rotating nature of coupling of the network filaments to the rigid boundary . to further test that this final redistribution is indeed a boundary effect , we consider a few larger system sizes as shown in fig . [ f : fraced_syssize_perp ] . in that figure the force remains perpendicular to the direction of the filament to which it is applied while the system size is varied from @xmath113 to @xmath114 for a network of fixed @xmath0 . when the data are plotted against the radial distance from the point force scaled by system size , we find an excellent collapse in the intermediate regime and in the putative boundary layer . for the smallest system size considered , @xmath34 , we note poorer data collapse in the near field region suggesting that one must study systems that are are least larger than a single filament length to access the bulk behavior of the network with quantitative accuracy . clearly , all three curves taken together are consistent with the notion of an elastic boundary layer that is produced by the freely - rotating boundary at the wall and that extends distance approximately equal to @xmath0 into the sample . term of the hamiltonian for perpendicular forces , as a function of distance from the origin @xmath78 for system sizes @xmath34 , @xmath115 and @xmath94 . in all cases @xmath98 and @xmath75 . , width=321 ] we now consider the mechanical response of the network to localized force dipoles at the origin . understanding this response function is central to elucidating the effect of molecular motor activity in the cytoskeleton . many of the general features observed in the response of the system to point force monopoles are also in evidence . for example in fig . [ f : dip1 ] we see the rough collapse in the far field of the mode amplitude @xmath116 to the continuum solution for three different networks having the same value of @xmath0 . note that the amplitude has been scaled by @xmath117 since the magnitude of the imposed force dipole for each realization of the network will depend on the distance to the next constraint , _ i.e. _ cross - link , which is @xmath21 . since the displacement field when averaged over network realizations is quadrupolar , being the difference in two dipole , that length must enter squared . the analogous displacement field amplitude , @xmath116 from the continuum solution ( solid line in fig . [ f : dip1 ] ) was calculated using unit forces so this amplitude is known only up to an overall scale factor ; that scale factor was adjusted to best fit the data . a similar comparison can be made for the other continuum modes of the displacement field ( see the appendix ) . [ f : dip2 ] , for example , shows @xmath118 with the same arbitrary prefactor . the agreement to the continuum theory scaled as discussed above is quite poor . as is also seen in the point - force response of the strain field , higher angular modes , which vanish in the continuum theory such as @xmath119 are significantly non zero in the data . the amplitude of the non - continuum modes in the dipolar response is even more dramatic than in the point - force response of the network examined earlier . the simulation data for the response of the network to force dipoles can be broadly summarized as follows : the amplitude of the continuum modes of the dipolar displacement field from the naive continuum theory do not agree with simulation data . in light of the discussion regarding such disagreements between the monopole data and the continuum calculation , this is perhaps not surprising . more interestingly , the observed displacement field has significant amplitudes of higher order angular modes . in short the universality of the response of semiflexible networks to localized point forces does not appear to extend to their response to localized force dipoles . we speculate that the principal difference between these two cases stems from the fact that the network s response to the force dipole probes the more detailed microscopic structure of the network in the immediate vicinity of the point of the force dipole application . the amplitude of the lowest order force multipole communicated from the near field region to the far field where we expect a continuum based theory to apply is not constrained by elementary force balance . neither are the amplitudes of any higher order force multipoles generated within the near field region . thus , upon reaching the inner edge of the far field region the force dipole imposed at the origin has generated a highly complex set of tractions on the rest of the material whose structure depends on local details of the connectivity of the network near the origin . in contrast for the case of the force monopole , the dominant term in those tractions is the fixed total monopolar force acting on the intermediate region . the higher order force multipoles created in the inner region decay rapidly with distance from the origin leaving one with highly reproducible results for the response of the network to applied forces . the amplitude of the force dipole communicated from the near field region to the intermediate region , on the other hand , is not similarly constrained . it appears that one generically generates large amplitude higher order force multipoles in addition to whatever force dipole is communicated to the intermediate region making convergence to a simple dipolar form slow and difficult to observe in our finite samples . for dipole forcing for fixed @xmath36 compared to the continuum solution ( smooth line ) , which has one free fitting parameter , namely the overall magnitude of the dipole forcing . both data and curve are negative , so the magnitudes have been plotted to allow use of a logarithmic axis . each data curve has been scaled by @xmath120 to ensure the same mean dipole magnitude ( see text ) . the system radius was @xmath34 . , width=321 ] for same systems as in fig . [ f : dip1 ] . the only fitting parameter for the continuum solution is the same as used previously for @xmath121 , so there are no remaining free fitting parameters in this plot . much of the data for @xmath122 is negative and hence not visible on these axes . , width=321 ] lastly , we present new results on homogeneous deformations of semiflexible networks . it has already been shown that the macroscopic elastic moduli of this class of model networks depend in a crucial way on the ratio of @xmath0 to the filament length @xmath1 @xcite . for @xmath123 , the deformation is approximately affine , whereas non affine deformation modes dominate when @xmath124 . previously this was demonstrated only for the shear modulus ; here we can now confirm that the young s modulus @xmath125 behaves in an identical manner . [ f : y ] shows @xmath125 measured from uniaxial extension of a rectangular cell , scaled by the prediction for an affine strain , plotted against @xmath96 for the range of @xmath74 considered in this paper . there is a clear data collapse , as for the shear modulus . furthermore the deviation from the affine prediction is small for @xmath123 , but becomes increasingly pronounced as @xmath96 increases . this confirms that @xmath96 controls the macroscopic elastic response of these systems . [ f : nu ] gives the poisson ratio @xmath126 for the same systems as in fig . [ f : y ] . it is striking to observe that , within error bars , @xmath126 is consistent with the value @xmath127 , which is the value expected for an affine deformation @xcite . however , it is apparent from this figure that the measured values are consistent with @xmath127 for _ all _ data points , even those well into the non affine regime . the mechanism behind this striking robustness currently evades us ( is has nothing to do with incompressibility , which fixes @xmath128 in two dimensions ) . note that @xmath126 at the rigidity percolation @xmath129 ( at which the elastic moduli vanish ) is @xmath130 @xcite , which is clearly inconsistent with the data in fig . [ f : nu ] and confirms our earlier claims that the non affine regime is distinct from the scaling regime of the transition . on log the affine prediction , which depends only on @xmath74 , can be found in @xcite . the symbols are larger than the error bars . , width=321 ] for the same systems as in fig . [ f : y ] . the solid line corresponds to @xmath127 . , width=321 ] in this paper we have presented the results of numerical studies on the response of semiflexible networks to both point forces and to homogeneously imposed strain . the data presented on the young s modulus taken in combination with previous work on the static shear modulus shows that the mechanical response of the system can be understood in terms of lam coefficients that depend on the ratio of the filament length to the nonaffinity length : @xmath106 . the poisson ratio of the material , however , appears to be remarkably insensitive to this ratio . we can offer no explanation for this insensitivity at this time . the data presented on the point force response form a necessary compliment to previous work on the development of a long length scale elastic theory of such materials . based on these data , it appears that the storage of elastic energy and the structure of the strain field is rather complex in the immediate vicinity of the applied point force . we characterize these quantities by considering three qualitatively different regimes as a function of radial distance from the applied point force . immediately surrounding the point force in the near - field regime we find that the partitioning of elastic energy into bending and stretching is determined primarily by the angle between the applied force and the direction of the filament to which that force is applied . the disorder - averaged strain field is rather complex having higher angular harmonics that predicted by continuum elasticity theory . in the intermediate field regime farther from the point force the partitioning of elastic energy is determined solely by the ratio @xmath106 as found in the homogeneous shear measurements . the higher angular harmonics present in the strain field appear to decay exponentially with a decay length proportional to @xmath0 with a constant of proportionality near unity . these results taken in combination with our previous work suggests that one may think of @xmath0 as setting the minimum length scale for the applicability of continuum modeling quite generally . in the quasi - continuum regime ( @xmath131 ) , we find a strain field consistent with structure of that predicted by continuum model . the strain itself , however , can not be simply computed from a knowledge of the effective lam coefficients . we believe the source of this discrepancy is the fact that the intermediate region , being poorly described by the continuum theory , applies a much more complex set of tractions to the material in the far - field where the continuum theory must apply . if these boundary conditions were known , we suspect that one could in fact calculate the resulting displacement field using the continuum theory . this belief is supported by the fact that in the far - field regime the displacement field for different networks having the same @xmath0 collapse onto the same curve as one would expect for any effective continuum theory in which the lam coefficients depend on @xmath0 . to summarize these results , we suggest that there is an emerging description of the mechanics semiflexible networks . there mechanical behavior over length scales longer than @xmath0 appears to be described by a modified elastic theory in which the effective lam constants depend on the ratio @xmath106 . for point forces and presumably any force applied over regions having a characteristic length scale that is small compared to @xmath0 , the local response of the network is quite complex and the material appears to be anomalously compliant ; at longer length scales , however , the structure of the deformation field appears to be consistent with the predictions of continuum elasticity . we believe that it should be possible to construct an elastic continuum theory of these networks that is applicable in both the intermediate and far - field and that is based on a modified gradient expansion of the strain field incorporating explicitly the mesoscopic length @xmath0 . the behavior of the strain field on scales much smaller than @xmath0 appears to depend on other , more microscopic length scales . understanding the response of semiflexible networks to localized forces is a necessary for both microrheological investigations of semiflexible networks and for understanding the effect of molecular motors in the cytoskeleton . clearly , further numerical investigations are required as well as a theoretical examination of the development of elastic continuum models applicable to the intermediate and far - field regimes . moreover , additional investigations are required to examine the analogous questions in three dimensional semiflexible networks . while we expect the basic physics outlined above , including the existence of a mesoscopic length @xmath0 , to persist in three dimensions , the expected @xmath132 decay of the elastic green s function should alter the results . moreover the more rapid decay of the displacement field and energy density in the continuum three dimensional system may further simplify the structure of the analogous green s functions for the semiflexible network . ajl and fcm acknowledge the hospitality of the isaac newton institute for mathematical sciences where part of this work was done . dah was jointly funded by the european community marie curie and the japanese society for the promotion of science programs . ajl was supported in part by nsf - dmr0354113 . here we derive the displacement field @xmath133 predicted by continuum elasticity for the cases of both monopole and dipole forcing . for the monopole case , a point force @xmath134 is applied to the origin of an elastic sheet with lam coefficients @xmath135 and @xmath0 , or equivalently @xmath135 and the poisson ratio @xmath136 . for an isotropic elastic body , the stress obeys @xmath137 and the resulting equation for force balance is @xcite @xmath138 solving this in polar coordinates @xmath139 with the boundary condition @xmath140 at a radius @xmath28 from the origin eventually leads to @xmath141 \nonumber\\ & & \left.\mbox{}-\left(r / r\right)^{2 } \left[2\hat{r}_{i}\hat{r}_{j}\hat{f}_{j}-c_{2}\hat{f}_{i}\right ] \right\}\,,\end{aligned}\ ] ] where @xmath142 and @xmath143 . then the only two non zero modes according to the definition of the @xmath48 given in ( [ e : g_modes ] ) are @xmath144 a naive calculation of the corresponding dipole solution would simply superpose the above monopole solution for two point forces , @xmath145 and @xmath146 with @xmath147 , and then take the limit @xmath148 . however , this ignores the boundary at radius @xmath28 , which should be kept fixed but is shifted a distance @xmath149 by the above procedure . an exact calculation would require the monopole solution for force applied near to ( but not at ) the center of a circular system , which , since it no longer obeys radial symmetry , is likely to be highly complex . here we ignore such issues and simply use the above monopole solution , in the expectation that it will closely approximate the exact case except possibly near the boundary . the displacement field induced by the force dipole is then @xmath150 \nonumber\\ & & \mbox { } + 8\hat{r}_{i}({{\bf\hat}{r}\cdot{\bf\hat}{f}})^{2 } \left[c_{2}(r / r)^{-2}-c_{1}\right ] \nonumber\\ & & \mbox { } + 2\hat{f}_{i}({{\bf\hat}{r}\cdot{\bf\hat}{f } } ) \left[2(c_{1}-1)-2c_{2}(r / r)^{-2}\right . \nonumber\\ & & \mbox{}\hspace{2 cm } \left.-(c_{2}-1)(r / r)^{2}\right ] \bigg\}\,.\end{aligned}\ ] ] as explained in sec . [ ss : decomposition ] , the dipole solution above applies for a single realization on large length scales , but after averaging over many networks on short lengths ( as in the simulations ) the dipole moment vanishes , leaving a quadrupole displacement field . the required quadrupole is one consisting of two parallel dipoles of equal and opposite magnitude , aligned along their axes , _ @xmath151 \nonumber\\ & & \mbox { } + 24\hat{r}_{i}({{\bf\hat}{r}\cdot{\bf\hat}{f } } ) \left[c_{2}(r / r)^{-2}-c_{1}\right ] \nonumber\\ & & \mbox { } + 8\hat{f}_{i}({{\bf\hat}{r}\cdot{\bf\hat}{f}})^{2 } \left[1 - 2c_{1}+3c_{2}(r / r)^{-2}\right ] \nonumber\\ & & \mbox { } + 16\hat{r}_{i}({{\bf\hat}{r}\cdot{\bf\hat}{f}})^{3 } \left[2c_{1}-3c_{2}(r / r)^{-2}\right ] \bigg\}\,.\end{aligned}\ ] ] this gives the displacement field in response to known forces of magnitude @xmath46 . since it is rather the displacement that is controlled , @xmath46 is a free parameter . finally , the non zero modes are @xmath152

alma observations show a non - detection of carbon monoxide around the four most luminous asymptotic giant branch ( agb ) stars in the globular cluster 47 tucanae . stellar evolution models and star counts show that the mass - loss rates from these stars should be.23.5 10 m yr . we would navely expect such stars to be detectable at this distance ( 4.5 kpc ) . by modelling the ultraviolet radiation field from post - agb stars and white dwarfs in 47 tuc , we conclude co should be dissociated abnormally close to the stars . we estimate that the co envelopes will be truncated at a few hundred stellar radii from their host stars and that the line intensities are about two orders of magnitude below our current detection limits . the truncation of co envelopes should be important for agb stars in dense clusters . observing the co ( 32 ) and higher transitions and targeting stars far from the centres of clusters should result in the detections needed to measure the outflow velocities from these stars . [ firstpage ] stars : mass - loss circumstellar matter infrared : stars stars : winds , outflows globular clusters : individual : ngc 104 stars : agb and post - agb

1508.05240

[ introdrivingsect ] lcccccccccccccc & & & & & & & & & & + & & & & & & & & & & + notes & ( 1 ) & ( 1 ) & ( 2 ) & ( 3 ) & ( 3 ) & ( 3 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 4 ) & ( 5 ) & ( 5 ) + v1 & 00 24 12.65 & 72 06 39.9 & 1.87 & 14.7 & 10 & 4.03 & 221 & 4824 & 4760 & 3623 & 3410 & 0.90 + v2 & 00 24 18.56 & 72 07 59.0 & 3.26 & 12.7 & 12 & 2.78 & 203 & 3031 & 4470 & 3738 & 3620 & 0.35 + v3 & 00 25 15.96 & 72 03 54.8 & 5.49 & 35.2 & 11.5 & 4.15 & 192 & 2975 & 4590 & 3153 & 3540 & 0.18 + v8 & 00 24 08.59 & 72 03 54.9 & 0.99 & 32.7 & 8 & 1.6 & 155 & 3583 & @xmath5 & 3578 & @xmath5 & @xmath5 + + lccccccccccccc & @xmath6 & @xmath7 & @xmath8 & dust & @xmath9 & @xmath10 + & ( m@xmath3 yr@xmath4 ) & km s@xmath4 & ( k ) & composition & ( m@xmath3 yr@xmath4 ) & ( m@xmath3 yr@xmath4 ) & + notes & ( 1 ) & ( 1,3 ) & ( 1 ) & ( 1 ) & ( 2 ) & ( 2 ) & + v1 & 2.1 @xmath1 10@xmath11 & 4.0 & 900 & silicate , metallic iron & 2.9@xmath12 @xmath1 10@xmath2 & 1.1@xmath13 @xmath1 10@xmath11 & + v2 & 1.2 @xmath1 10@xmath11 & 3.8 & 900 & silicate , metallic iron & 2.1@xmath14 @xmath1 10@xmath2 & 6@xmath15 @xmath1 10@xmath2 & + v3 & 0.9 @xmath1 10@xmath11 & 3.2 & 1000 & metallic iron only & 2.4@xmath16 @xmath1 10@xmath2 & 7@xmath15 @xmath1 10@xmath2 & + v8 & 1.5 @xmath1 10@xmath11 & 4.0 & 900 & silicate , alumina , metallic iron , oxides & 1.9@xmath17 @xmath1 10@xmath2 & 1.0@xmath18 @xmath1 10@xmath11 & + + our understanding of how baryonic matter is recycled from metal - poor stars back into their host environment currently faces two major problems : how is mass actually lost from the stars and what happens to it when it is returned to the interstellar medium ( ism ) ? mass loss from giant stars follows a two - stage process . chromospheric or magneto - acoustically - driven mass loss occurs during a star s red giant branch ( rgb ) and early asymptotic giant branch ( agb ) evolution , when warm ( @xmath06000 k ) plasma is ejected from the surface ( e.g. @xcite ) . although initially faster , this wind slows to @xmath010 km s@xmath4 at the luminosity of the rgb tip @xcite . later , @xmath19-mechanism pulsations can levitate material from the star , enhancing the mass - loss rate @xcite . strongly linked to this pulsation is the formation of dust in the denser stellar atmosphere ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? radiation pressure on this dust drives the stellar wind in a cooler outflow , also of typically @xmath010 km s@xmath4 , rising with stellar luminosity to @xmath020 km s@xmath4 ( e.g. @xcite ) . although dust production begins at luminosities below the rgb tip @xcite , sufficient radiation pressure to drive the wind is only achieved once the star reaches @xmath010005000 l@xmath3 ( e.g. @xcite ) . however , the expansion velocities of metal - poor stellar winds are poorly known . the stars are smaller compared to metal - rich stars at the same luminosity , so pulsations are thought to be weaker @xcite . the stellar surface resides deeper in the gravitational potential , so material must be levitated further before reaching temperatures where dust can condense . we also expect the dust : gas ratio will decrease , meaning that the associated mass - loss rate and velocity enhancements will also decrease . the transition between a chromospheric and pulsation - enhanced , dust - driven wind should consequently happen at a higher luminosity @xcite . without measurement of these expansion velocities , however , these theories remain untested . the wind expansion velocity is an important parameter in determining mass - loss rates from radiative transfer modelling of infrared excesses , which is a valuable method of determining the mass of both dust and gas ejected by giant stars , and the only method currently feasible outside our galaxy ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? once material leaves the star it is reprocessed , first by the interstellar radiation field ( isrf ) and then by interstellar shocks . other unknowns then become important in determining how quickly dust is destroyed , molecules are dissociated , and atoms ionised . contributing factors include the star - formation history of the population ( setting the uv flux from high - mass main - sequence , low - mass horizontal - branch and post - agb stars ) , the amount of interstellar dust shielding , and the local stellar density . there is a growing realisation that the enrivonment in which stars lie can have profound effects on their molecular and mineralogical yields @xcite . nowhere is this more strongly felt than globular clusters . the preparation of this work sparked the theory that ionisation should exert a powerful control over the interstellar environment of globular clusters @xcite . very strong isrfs should be generated by post - agb stars and the hottest white dwarfs , but the short - lived nature of these objects means the isrf is constantly varying . not only should these sources be capable of dissociating and ionising the intra - cluster medium ( icm ) , but they also provide the icm with enough thermal energy that it overflows the globular cluster , preventing further star formation . we now turn our attention to the circumstellar environments of globular cluster stars . many of the above issues can be addressed by observing co lines emanating from the winds of stars in globular clusters . co is one of the first molecules to form as the extended atmosphere of the star is ejected , and it survives until its dissociation by the isrf . this dissociation is mainly by photo - absorption in molecular lines , hence even in the absence of dust , self - shielding can be important in protecting co in stellar winds . observationally , the widths of co lines provide a direct measurement of the wind - expansion velocity , while their strength provides an independent estimate of mass - loss rate . their strengths are also affected by the characteristic radius from the star at which co is dissociated by the isrf @xcite . hence , by comparing the strength of co lines from stars with known mass - loss rates to model predictions , we can quantify the strength of the isrf to which they are subjected . in this paper , we report on observations with the atacama large millimetre array ( alma ) of the four brightest stars in 47 tucanae . section [ topobssect ] describes these observations , and defines upper limits to the non - detections we find . section [ codisssect ] models the expected mass - loss rate from the star , the size of the co envelope , and calculate the expected strength of the co transitions . this is done under the assumption that the isrf is strong dissociating the co envelope . results are discussed in section [ discsect ] . alternative scenarios which might remove the co envelopes of our target stars are explored in the appendix . among the closest and most populous globular clusters is the intermediate - metallicity cluster 47 tuc ( ngc 104 ; 4.5 kpc , [ fe / h ] = 0.72 dex , @xmath20 = 1.5 @xmath1 10@xmath21 m@xmath3 ; @xcite ) . this cluster was chosen partly because its stars show a relatively small spread in chemical abundances compared to other clusters of similar size @xcite , and partly because it contains a well - studied set of luminous , dusty agb stars . the cluster s variable stars have a unique , long and well - covered history @xcite . infrared photometry and spectra have been published at many wavelengths over many epochs @xcite . mid - infrared spectroscopy of the brightest variables @xcite show a mixture of silicate and alumina dust features , and a potential featureless contribution from metallic iron dust ( cf . this combination has led to several estimates of their dust - production rates @xcite . high - resolution optical spectroscopy of moderately bright giants suggest relatively slow outflows from the stellar chromospheres of @xmath010 km s@xmath4 @xcite . near - infrared spectroscopy @xcite shows that the brightest stars have abundances similar to other stars in the cluster . they have not experienced substantial third dredge - up , and perhaps have not experienced any third dredge - up episodes . we present sub - mm co observations of the four most luminous , most evolved stars in 47 tuc . they are denoted by @xcite as v1 , v2 , v3 and v8 . previously published parameters of these stars from the above references are given in tables [ starstable ] and [ windstable ] . in particular , table [ windstable ] gives the mass - loss rate ( @xmath6 ) and wind velocity ( @xmath7 ) predicted for a purely radiation - driven wind . @xcite derived these on the basis of radiative transfer modelling of their infrared spectra using the dusty code @xcite with a radiation - driven wind ( density type = 3 ) . in all four cases , the spectral energy distributions were best modelled with a mostly metallic iron wind , with some stars having contributions from amorphous and crystalline silicates , aluminium oxide and iron oxide . table [ windstable ] also gives the mass - loss rate expected for a purely magneto - acoustically driven wind , as derived from semi - empirical scaling laws applied to the cluster s other stars ( @xmath9 , @xmath10 ; see section [ mdotsect ] ) . lccc & co(2@xmath221 ) & r.m.s . noise & @xmath23 + & ( mjykms@xmath4 ) & ( mjykms@xmath4 ) + v1 & @xmath24183 & 85 & 2.1 + v2 & @xmath24168 & 114 & 1.4 + v3 & @xmath24170 & 99 & 1.7 + v8 & @xmath24162 & 70 & 2.3 + + observations of 47 tuc v1 , v2 , v3 and v8 ( see figure [ mapfig ] ) were carried out using alma in receiver band 6 on the night of 2013 nov 05 over a 1.05 h integration . observations were centred on the @xmath25c@xmath26o @xmath27 transition at 230.538 ghz , and were carried out at high precipitable water vapour ( 2.83.5 mm ) but otherwise good conditions . the correlator output consisted of 3840 channels of 244.141 khz each . the data were processed following the standard alma quality assurance calibration and imaging processes . neptune was used for amplitude calibration and the quasars j1924291 and j01027546 for bandpass and phase calibration , respectively . during the alma observations the majority of antennae ( 26 of 29 ) were positioned within @xmath0250 m of the array centre ; the remaining three telescopes were at larger distances ( @xmath0530 m , @xmath0790 m and @xmath01014 m ) . at the imaging stage of data reduction the visibilities from these three antenna were downweighted . this was accomplished using the uvtaper parameter set to true and the outaper option set to 1.5@xmath28 in casa s clean task @xcite . in addition to this a restoring beam of @xmath29 was also applied at the cleaning stage . these techniques were used to generate images with a beam more representative of the antennae in the centre of the array . imaging additional to that provided with the alma data release was conducted in casa using the clean routine to create two data cubes for each source , one with a binning of ten spectral channels ( channel width : 3.2 km s@xmath4 bin@xmath4 ) and one with two - channel binning ( 0.64 km s@xmath4 bin@xmath4 ) . the pixel scale in both cubes is 0.15@xmath28 pixel@xmath4 . no line or continuum source is visible in any of the maps to the depth of the observations . the standard deviation of the noise in the final images is @xmath014 mjy beam@xmath4 per 0.64 km s@xmath4 channel pair for v1 , v3 and v8 , and @xmath016 mjy beam@xmath4 per channel pair for v2 . the shortest baseline ( 17 m ) corresponds to a maximum detectable object size of 15@xmath307 , though the co shells are expected to be unresolved ( @xmath311@xmath28 ; section [ coexpectsect ] ) . on - source spectra were extracted using a 3@xmath13 pixel box surrounding the central source . figure [ specfig ] shows the two - channel - averaged spectra around the velocity of the cluster . the maximum two - channel flux for each source is 30 , 33 , 36 and 37 mjy beam@xmath4 for v1 , v2 , v3 and v8 , respectively . each spectrum was examined and found to have noise following a gaussian distribution , both on each source and at various points away from each source . on - source standard deviations in the spectra are 10.2 , 11.9 , 11.6 and 11.2 mjy beam@xmath4 per 0.64 km s@xmath4 bin for v1 , v2 , v3 and v8 , respectively . additional data were obtained from the atacama pathfinder experiment ( apex ) telescope . a four - hour integration on 47 tuc v1 was made in the 345 ghz @xmath25c@xmath26o @xmath32 transition ( project identifier : o-092.f-9327a ) . standard reduction of the apex data was performed at onsala space observatory before receipt of the data . no line was detected , with a limiting peak flux of @xmath00.1 jy over a typical 10 km s@xmath4 bin from the 17@xmath303 beam . neither the alma nor the apex data showed any convincing indication of spectrally resolved co emission ( e.g. departure from gaussian noise centred around zero flux ) . the only previous significant co observations of the cluster are by @xcite , who claim a weak co ( 1@xmath220 ) detection around the cluster s velocity with a signal - to - noise ratio of 3 . we interpret their peak fluxes as 1.6 jy , assuming an aperture efficiency for the swedish eso sub - millimetre telescope ( sest ) of 27 jy k@xmath4 . origlia et al.s observing position is not quoted , but assumed to be @xmath33 24 22 , @xmath34 05 48 , following their offset from @xcite . this is 68@xmath28 from v1 , compared to a full - width half - maximum sest beam width of 45@xmath28 . in neither dataset do we recover any source comparable to the co ( 10 ) detection of @xcite , though the beam footprint of the sest observations extends considerably beyond both the alma primary beam of 15@xmath307 and the apex primary beam of 17@xmath303 . the co emission is expected to be spatially unresolved by alma and the mass - loss rate is expected to be sufficiently low that the lines will be optically thin . an unresolved , optically thin line will have a rectangular ( ` boxcar ' ) spectral profile . to identify whether a low - contrast , spectrally resolved feature is present in our data , we smoothed our data using a boxcar function ( running average ) . the boxcar was given different central velocities , from @xmath35 to 0 km s@xmath4 , and different widths , from 1.27 km s@xmath4 to 85 km s@xmath4 ( 4133 channels ) . typical expansion velocities for dust - producing stars in the milky way are @xmath01020 km s@xmath4 , and are typically @xmath010 km s@xmath4 for stars producing little dust ( e.g. * ? ? ? * ; * ? ? ? we presume the expansion velocity is @xmath36 km s@xmath4 , and that the centre of the emission line is doppler shifted by no more than 5 km s@xmath4 from the average radial velocity of the star . the limits to the velocity - integrated co line strength are typically below @xmath24200 mjy km s@xmath4 , with a typical 1@xmath23 uncertainty being @xmath0100 mjy km s@xmath4 ( see table [ winds2table ] ) . the boxcar - smoothed spectra producing these maximum fluxes are shown in figure [ specfig ] . the mean photospheric radial velocities and their pulsation - induced variation are listed in table [ starstable ] has been applied . ] . the highest flux in the v1 and v8 spectra is within 10 km s@xmath4 of the mean photospheric velocities of the stars as published by @xcite . these may indicate a tentative detection of the co line , but we do not consider them reliable as they are only at the 1.3 and 2.1@xmath23 confidence levels , respectively . our observations were carried out under the nave assumption that the envelopes of our target stars behave as galactic stars do . in this section , we show that we could expect a similar galactic star to be observable at the distance of 47 tucanae . this provides a baseline to which we can compare our final model . @xcite provides an estimate of the peak co(21 ) line flux from an evolved star , namely : @xmath37 where @xmath38 is the mass - loss rate in m@xmath3 yr@xmath4 , @xmath39 is the expansion velocity in km s@xmath4 , @xmath40 is the ratio of co to h@xmath41 , and @xmath42 is the distance in kpc ( @xmath43 kpc ) . from table [ windstable ] , we can modestly expect @xmath44 m@xmath3 yr@xmath4 , depending on the star and the method used . if we assume @xmath45 m@xmath3 yr@xmath4 ( appropriate for v1 ) , a canonical @xmath46 km s@xmath4 and @xmath47 ( based on scaling a canonical 2 @xmath1 10@xmath48 @xcite by the metallicity of 47 tuc , which is a fifth of solar metallicity ) , we obtain a peak flux of @xmath4914 mjy . assuming a boxcar profile gives an integrated intensity of @xmath49280 mjy km s@xmath4 , which is still greater than our upper limit . @xcite also predicts that the co(32 ) flux should be around twice as strong . a less conservative value of @xmath50 m@xmath3 yr@xmath4 produces 1200 mjy km s@xmath4 for the co(21 ) line , implying 2400 mjy km s@xmath4 for the co(32 ) line , or 1.2@xmath23 in our apex observation . much depends on the exact values used in the equation above , as well as the accuracy of the formula itself . stars in globular clusters exist in environments unlike the well - studied stars in the solar neighbourhood . in particular , it has recently been shown that the isrf is considerably stronger and harder @xcite . this will have a substantial impact on the size of the co envelope and the line flux emitted from it . in this section , we re - evaluate the expected co line fluxes from these stars by creating a photochemical dissociation model for v1 . we consider other factors peculiar to globular clusters in the appendix , but find that they have little impact . in general , mass - loss rates of stars without dust - driven winds are comparatively well determined and have been parameterised into scaling relations . two common formalisms which are frequently adopted are the emprical formalism of @xcite and its semi - emprical modification by sc05 . in these formalisms , the mass - loss rate is scaled by an efficiency factor , @xmath51 . in terms of observable units , they can be written as : @xmath52 and : @xmath53 where the luminosity , mass and temperature ( @xmath54 , @xmath20 , @xmath55 ) are all scaled to solar units and the mass - loss rate is in m@xmath3 yr@xmath4 . throughout a large sample of milky way globular clusters , @xcite compared the mass lost between the main - sequence turn - off and the horizontal branch and compared it to theoretical isochrones at fixed @xmath51 . this fixes @xmath51 around the point where mass loss is strongest , close to the rgb tip . from these , the following efficiency parameters were derived for the median stars in globular clusters : @xmath56 where each value is given with its respective statistical and systematic error on the rgb ( @xmath57 ) , and a higher @xmath51 on the agb ( @xmath58 ) . this would increase the predicted mass - loss rate for our targets . private communications with heyl et al . have not identified the source of the discrepancy . @xcite also argue for a slightly smaller @xmath59 , on the basis of the stellar period luminosity diagram . ] . from these , we derive the minimum mass - loss rates listed in table [ windstable ] of @xmath60 m@xmath3 yr@xmath4 and @xmath61 m@xmath3 yr@xmath4 . these values are appropriate for ` stable ' stars , without extra energy from pulsations or radiation pressure on dust . while the interplay of pulsations on transfer of magneto - acoustic energy to the chromosphere is poorly determined , we would expect these values to provide a lower limit to the mass - loss rate from pulsating , dusty agb stars . the mass lost from an agb star during the last part of its life must not exceed the envelope mass of the star at the beginning of that period . low - luminosity agb stars are not easily seperable from rgb stars . agb stars can only be uniquely identified once they exceed the luminosity of the rgb tip . if we can determine both the remaining lifetime of an agb star passing the luminosity of the rgb tip , and the envelope mass it has when it does so , the quotient of the two should yield the average mass - loss rate of agb stars above the rgb - tip luminosity . the time spent above the rgb tip is relatively easy to compute . our four stars lie above the rgb of 47 tuc . for the purposes of these calculations , we also include the fifth - brightest star in the cluster . v4 is a probable agb star at the luminosity of the rgb tip ( see the hertzsprung russell diagram presented by @xcite ) . like its brighter counterparts , it shows long - period pulsations and an infrared excess , consistent with circumstellar dust production . assuming there are typically 5 @xmath62 @xmath63 stars above the cluster s rgb tip , and taking an evolutionary rate of one star passing the rgb tip every 80000 years @xcite , these five stars should represent the last 0.40 @xmath62 0.18 myr of agb evolution in the cluster . the envelope mass at the luminosity of the rgb tip is more difficult to calculate . @xcite determine zero - age horizontal branch ( zahb ) star masses in 47 tuc . their ground - based measurements suggest the zahb star mass ranges from 0.629 m@xmath3 to 0.666 m@xmath3 , with a median of 0.648 m@xmath3 , but they also include masses derived from _ hubble space telescope ( hst ) _ photometry , which range from 0.650 m@xmath3 to 0.691 m@xmath3 , with a median of 0.674 m@xmath3 . we assume these stars will end their lives as a 0.53 m@xmath3 white dwarf , based on measured masses of white dwarfs in other clusters @xcite . these masses imply that @xmath00.12 m@xmath3 or @xmath00.16 m@xmath3 is lost during the entire horizontal branch and agb evolution , including that below the rgb - tip luminosity . assuming the limiting case of losing 0.16 m@xmath3 over 221000 years , the _ average _ mass - loss rate for stars above the rgb tip of 47 tuc must be @xmath247 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . a more limiting case is reached if we can estimate the envelope mass at the luminosity of the rgb tip itself . calculating the mass lost between the zero - age horizontal branch up the agb to the luminosity of the rgb tip is not trivial . the evolutionary track used by @xcite was designed to reproduce the rgb and hb of 47 tucanae . this track was generated with the mesa ( modules for experiments in stellar astrophysics ) stellar evolution code @xcite . mass loss was included at the rate of @xmath64 , reproducing a zahb mass of 0.666 m@xmath3 after 12.02 gyr . the model loses a further 0.067 m@xmath3 between the zahb and the point where the star reaches the same luminosity as the rgb tip ( i.e. the same luminosity as v4 ) . taking the limiting case of the _ hst_-derived mass of 0.674 m@xmath3 , a total mass at the luminosity of the rgb tip of 0.607 m@xmath3 is implied , giving an envelope mass of @xmath00.077 m@xmath3 and limiting the average mass - loss rate to @xmath243.5 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . in contrast to the mass - loss rate in the previous paragraph , this mass - loss rate requires that reimers mass - loss formula is appropriate for early - agb stars . reimers scaling law8.5 @xmath62 4.3 @xmath1 10@xmath2 m@xmath3 yr@xmath4 for our stars . this is inconsistent with the @xmath243.5 @xmath1 10@xmath2 m@xmath3 yr@xmath4 provided by the evolutionary analysis . this suggests the sc05 law is not appropriate for this particular situation . ] provides a lower limit to the mass - loss rates for pulsating agb stars , implying an average mass - loss rate for our four stars of @xmath492.3 @xmath62 1.1 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . stellar evolution models , pinned by the mass of hb stars ( section [ maxmdotsect ] ) suggest an upper limit of @xmath243.5 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . we can therefore expect the mass - loss rate from our four observed stars to lie in the range 1.23.5 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . this predicted mass - loss rate is estimated to produce a co line flux within the noise limit of our observations . however , the average predicted mass - loss rate is much less than the average of 1.4 @xmath1 10@xmath11 m@xmath3 yr@xmath4 derived from these stars infrared spectra , although the infrared - based rates do come with a sizeable and poorly quantifiable uncertainty . these infrared - based mass - loss rates assumed slow ( 24 km s@xmath4 ) winds , accelerated only by radiation pressure on dust ; faster ( @xmath6510 km s@xmath4 ) winds would increase the calculated mass - loss rate . .modelled isrf incident on each star at 1076 , relative to the flux in the solar neighbourhood from ( * ? ? ? * photons @xmath66 s@xmath4 a@xmath4 ) . the centile probabilities approximate the best estimate , 1-@xmath23 , 2-@xmath23 and full ranges of the modelled fluxes . [ cols= " < , > , > , > , > , > , > , > " , ] the radiation environment within globular clusters differs from that in the solar neighbourhood @xcite . the ionised hydrogen and lack of interstellar dust within globular clusters reduces the opacity for ionising photons . ionising sources are mainly post - agb stars and white dwarfs , rather than the o and b stars of the solar neighbourhood this leads to a much harsher isrf than is typically assumed . the isrf of 47 tuc was modelled in detail by @xcite . the isrf can vary considerably , as the presence of very hot post - agb stars in the cluster is stochastic . many of the hottest post - agb stars and white dwarfs in the cluster remain unobserved . we must therefore rely on stellar evolution models to describe the typical isrf within 47 tuc and its temporal range . to compute the isrf impinging on each agb star , we have created a three - dimensional model of the cluster . we begin with the catalogue from @xcite , which contains the two - dimensional ( ra dec ) position of almost every post - main - sequence source in the cluster up to the early post - agb phase . we assume this is also representative of the distribution of more - evolved , hotter post - agb stars . for each catalogued source , we assign a physical depth within the cluster . assuming an earth - to - cluster distance of 4500 pc , we randomly select a source nearby in right ascension to each star and use the declination of the random source to assign a radial distance from the cluster centre for each target source . this creates a stellar distribution with the same radial density profile as the density profile in declination , approximating a spherically symmetric distribution . we then take the post - agb stellar evolution model of @xcite , following the emitted spectrum as it evolves through its upper - agb , post - agb and hot - white - dwarf evolution . we follow the evolution of each of our model sources in time , randomly forcing one of our sources to undergo this post - agb evolution with an average of once per 80000 years ( the stellar evolution rate according to @xcite ) . summing the modelled flux from these sources at the locations of our four target stars , we obtain the integrated interstellar radiation field impinging upon them . this model was run 100 times , in order to simulate different radial distributions of the radiation sources and observed targets . for each run , the evolution was followed for 10@xmath21 years to quantify the time variability of the radiation . figure [ isrffig ] shows the isrf , averaged for each source over all runs and all times . for comparison , the figure also shows the solar neighbourhood isrf used by ( * ? ? ? * originally from @xcite ) . the radiation field in the 9141120 region ( normally associated with co dissociation ) is typically many times higher than the @xcite value , and it is also considerably harsher . co in the circumstellar environment has some self - shielding , but most shielding comes from circumstellar dust and h@xmath41 , as shown by ( * ? ? ? * their figure 2 ) . at some depth the atmosphere becomes optically thick to incoming or outgoing radiation , forming the photosphere , or @xmath67 layer . in mass - losing stars , particularly at short wavelengths , the @xmath67 layer is wavelength dependent , depending on the shielding within the wind . dissociation can therefore take place in an unshielded environment , if the time to intercept a uv photon is large compared to the time taken to reach the @xmath67 layer , or a shielded environment if it is smaller . if shielding is not important , the co dissociation radius will be limited simply by the average time it takes a co molecule to encounter an interstellar uv photon , and the co dissociation radius around metal - poor stars should be very similar to metal - rich stars . if the shielding is important , co will be dissociated close to the @xmath67 layer , and the dissociation radius around metal - poor stars should be smaller . however , it can not become much smaller before h@xmath41 becomes the dominant shielding mechanism . in either case , we expect metallicity to only have a second - order effect on the co dissociation radius , and that the primary effect will be the strength of the isrf . for a solar - neighbourhood isrf , we would expect the 47 tuc stars have co envelopes filled to @xmath00.0130.023 pc ( section [ coexpectsect ] ) . a stronger isrf will dissociate co proportionally more quickly , so the dissociation radius should be inversely proportional to the isrf strength . the primary contribution to co dissociation is normally the 1076 band @xcite , though the harder radiation field may change this in our case . the potential range in isrf strength is large ( table [ isrftable ] ) and is considerably greater for the stars closer to the cluster centre ( v1 , v8 ) than stars on the cluster s outskirts ( v2 , v3 ) . taking the 1@xmath23 ranges for the potential variation in isrf , we compute a typical range of co envelope outer radii of @xmath04 @xmath1 10@xmath68 0.008 pc , or 81800 au . this does not for changes in the of co or the harder isrf , and implies co may be dissociated very close to the stellar surface ( @xmath6910 r@xmath70 ) at certain times . if the wind remains optically thin below these heights , there may be substantial co dissociation within the dust - formation shell ( 210 r@xmath70 ) around these stars . more typically , however , we would expect 90 per cent of the co to be dissociated by @xmath0100 au ( @xmath0120 r@xmath70 ) in v1 , and @xmath030 au ( @xmath040 r@xmath70 ) for v8 . to get a simple estimate of the abundance of co that could be expected in a high - uv - field , metal - poor environment we set up the following toy model , which we base on 47 tuc v1 . we used the moving - mesh code arepo @xcite . our version of the arepo code has been modified to include time - dependent chemistry as presented by @xcite ( for full details see @xcite and @xcite ) . the co abundance is calculated assuming that the co formation is limited by an initial radiative association step , and that the co destruction rate is primarily through photodissociation . the gas is shielded from the ambient radiation field by dust and gas self - shielding , which we calculate using the approach of @xcite . the physical model consists of an expanding shell of gas with a total mass of 9 @xmath1 10@xmath48 m@xmath3 , following a @xmath71 power law , starting from the assumed stellar surface at @xmath72 = 1.3 @xmath1 10@xmath73 cm = 0.87 au . the entire envelope corresponds to the mass ejected over 3000 yr13002200 yr ; section [ coexpectsect ] ) . the exact radius and time are not critical to the results . ] at a mass - loss rate of 3 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . the gas initially follows a @xmath74 power law ( e.g. @xcite ) with the inner surface corresponding to the temperature of the stellar photosphere , taken to be 3517 k. in this oxygen - rich environment , the fractional abundance of co is determined by the carbon abundance : [ c / fe ] = 0.13 @xmath62 0.20 dex @xcite . the carbon abundance varies between [ c / fe ] @xmath0 0.5 dex for a few cno - processed stars to [ c / fe ] = 0.0 dex for the majority of the stellar population @xcite . we use [ fe / h ] = @xmath75 dex , following @xcite , and [ c / h ] = 0.72 dex to conservatively model the carbon - rich end of the population . we assume a dust - to - gas ratio of 1:1000 @xcite , and [ o / h ] = 0.57 dex @xcite . the envelope is initially fully molecular , with all its carbon in the form of co. the co abundance relative to protons was tracked just inside the inner boundary of the expanding shell , representing an expanding region of gas in a dynamic envelope . the code assumes an external uv field with a spectrum equivalent to that seen in the solar neighbourhood @xcite , but with a flux that is 50@xmath1 higher ( i.e. 50 habing units erg s@xmath4 @xmath66 ] ; see section [ codisssect ] and table [ isrftable ] ) . this model includes several important simplifications , which generally lead to an _ overestimate _ of the co abundance , and an overestimate of the co dissociation radius . the most important of these assumptions are : * the adoption of an isrf from the solar neighbourhood , scaled upwards by a factor of 50 to the expected _ time - averaged _ flux of the isrf in 47 tuc at 1076 . the uv field in globular clusters also contains harder uv photons , which will lead to more rapidly dissociated co , and is likely to result in significant quantities of co@xmath77 , or even co@xmath78 . * a lack of radiative heating from the central star , and a lack of thermal heating by circumstellar dust ( collisional heating is included ) . this means the temperature of the co envelope is colder than we expect ( figure [ arepofig ] ) , protecting the co in lower ro - vibrational states . figure 2 of @xcite shows that halving the co excitation temperature has a small but noticeable effect on co shell size . figure [ arepofig ] shows the abundance of co relative to protons as a function of dynamic radius . at the stellar surface , the carbon is entirely bound in molecules , with a co : proton ratio of 2.67 @xmath1 10@xmath68 . the abundance drops rapidly with distance due to photodissociation by the ambient uv field . by a radius of 2.8 @xmath1 10@xmath79 cm ( 0.00092 pc , or 218 r@xmath70 ) , 90 per cent of the co has been dissociated . this represents a shell @xmath022@xmath1 smaller than predicted by @xcite for a similar galactic star . bearing in mind the simplifications above , we can expect the shell size to be smaller still , suggesting a close to inverse relation between isrf strength and co shell size . l@ c@ c@ c & & & + + gas temperature@xmath80}$ ] & ... & ... & @xmath81 + dust : gas ratio@xmath82}$ ] & ... & ... & 1:1000 + initial co / h@xmath41 & ... & ... & @xmath83 + + outflow velocity & ... & km s@xmath4 & 10 + wind density at @xmath72 & ... & @xmath84 & @xmath85 + wind density law & ... & @xmath84 & @xmath86 + mass - loss rate & ... & m@xmath3 yr@xmath4 & @xmath87 + dust condensation & @xmath88 & cm & @xmath89 + dust temp . at @xmath88 & ... & k & [ 4 ] + isrf & ... & habing & 50 + + stellar radius@xmath90}$ ] & @xmath72 & cm & @xmath89 + stellar spectrum & ... & ... & [ 2 ] + stellar mass & @xmath91 & m@xmath3 & 0.55 + stellar luminosity & @xmath92 & l@xmath3 & 4824 + stellar @xmath93 & @xmath55 & k & 3623 + dust condensation@xmath90}$ ] & @xmath88 & cm & @xmath94 + dust temp . at @xmath88 & ... & k & 850@xmath90}$ ] + dust composition & ... & ... & [ 2 ] + + millimetre / sub - mm co lines at these mass - loss rates are optically thin ( e.g. @xcite ) . given the @xmath95 density law of our wind , we can expect the co line flux to scale with the shell size , such that the co line flux should be @xmath022@xmath1 lower than predicted by @xcite for a star losing mass at 3 @xmath1 10@xmath2 m@xmath3 yr@xmath4 . other factors are important , notably the temperature structure of the atmosphere , which controls the rotational level populations of co. we have therefore predicted line strengths from our envelope using the smmol radiative transfer code @xcite . table [ modeltable ] lists the parameters used in the model . smmol is a spherical - geometry , non - lte radiative transfer code which solves level populations with the accelerated lambda iteration ( ali ) scheme , described by @xcite and @xcite . it relies on a discrete spatial grid and ray tracing for radiation transfer and is optimised for high - optical - depth regimes . smmol adopts dust properties from @xcite , which uses the dusty radiative transfer code @xcite to model the spectral energy distribution of v1 . this model uses a grain mixture of 88 per cent metallic iron @xcite and 12 per cent silicates @xcite , surrounding a star modelled by a bt - settl model atmosphere spectrum @xcite . the gas temperature structure was taken from our photo - chemical modelling results . an accelerating wind was used , with a velocity at the inner region of 8 kms@xmath4 , and a terminal velocity of 11 kms@xmath4 , to mimic radiative acceleration of dust . this lowers the optical depth of the profile but , as our stars have low mass - loss rates , the resulting line flux differs negligibly from a constant 10 km s@xmath4 wind . a turbulent velocity of 2kms@xmath4 was adopted . table [ modeltable ] lists a full set of physical parameters . smmol modelled the observation as observed by a single 250-metre telescope with a perfect gaussian beam . the co envelope ( 0@xmath3005 ) remains a point source at this resolution ( 1@xmath3007 ) , such that the conversion from antenna temperature to flux units ( janskys ) represents the flux from the entire source as observed by alma . table [ modelresultstable ] lists the resulting line fluxes from smmol . smmol clearly predicts line strengths below the upper limits of our observations , showing that photo - dissociation explains well our non - detection of the co emission from these stars . we stress that these fluxes are only order - of - magnitude estimates , as many wind parameters are not accurately known . the line flux is strongly affected by the temperature structure of the inner stellar wind . we have adopted a simple power law for our model , but the physical temperature will be significantly affected by the dust properties . the dust properties of our target stars are not well known , particularly the grain mineralogy , grain size and dust - production rate , though they are likely very different from agb stars in the solar neighbourhood @xcite . this uncertainty therefore translates into a considerable uncertainty on the co line strengths . when combined with the time - variability of the isrf ( section [ isrfsect ] ) , which can exceed an order of magnitude , it becomes clear that we are limited to stating that the line fluxes will scale very approximately with isrf strength , and that we would typically expect them to be below the sensitivity limit of our observations . lcc & & + & & + & & + @xmath96 = 4@xmath223 & @xmath5 & 16 + @xmath96 = 3@xmath222 & @xmath692000@xmath80}$ ] & 6.1 + @xmath96 = 2@xmath221 & @xmath24183 @xmath62 85@xmath90}$ ] & 2.6 + @xmath96 = 1@xmath220 & @xmath5 & 0.1 + + in this paper , we have presented new alma observations of the four brightest stars in the globular cluster 47 tucanae , examining the co @xmath97 transitions . no source was detected . having modelled the interaction between the isrf and the circumstellar enviroment , we expect that the co line fluxes from the giant stars in 47 tuc are around two orders of magnitude below our observation limit . for any given globular cluster giant , if the wind is optically thin , then the radius of co dissociation should exhibit an approximately linear relation with metallicity and an approximately inverse relation with isrf strength ( section [ codisssect ] ) . the radiation field should be highly stochastic , driven by the temporally variable population of hot white dwarfs . the population of young white dwarfs is not well determined in most clusters , therefore it is not immediately possible to identify which clusters will be the best candidates for observation . however , the isrf can be expected to vary by several orders of magnitude within a cluster , both temporally and spatially ( table [ isrftable ] ) . targets far from the cluster centre may experience radiation fields that are weaker than those in the solar neighbourhood ( though they will still contain harder radiation ) . giant stars at large distances from their host clusters centres may therefore provide the best metal - poor stars around which to detect stellar winds and measure their expansion velocities . table [ isrftable ] indicates that v2 and v3 may occasionally experience isrfs that are weaker than the solar neighbourhood . while this seems not to be the case at present , we can hope that a similar situation can be found in other milky way globular clusters . since the higher stellar density forces the co envelopes to be smaller than their galactic counterparts , the co(32 ) line ( or higher rotational states ) may be considerably brighter than co(21 ) , as these are better populated in the warmer regions close to the star ( table [ modeltable ] ) . we therefore suggest observations to detect co line strengths and widths from globular cluster stars focus on : 1 . stars at large radii from their host clusters and 2 . high co rotational states emitted closer to the star . we also advocate further observations of the circumstellar dust . unexpectedly large amounts of dust are seen around many globular cluster stars @xcite . their spectra have strong silicate features @xcite , but also an underlying continuum , possibly explained by metallic iron @xcite . this contrasts with the notable absence of interstellar dust within clusters @xcite , and suggests dust is being destroyed while still in the circumstellar environment . the highest - energy photons ( @xmath040 ev , @xmath0300 ) are likely to penetrate into the dust - forming regions . while they are of insufficient number to dissociate a large fraction of co in this region , they may still be important in the photo - chemical formation of dust . finally , we note that this situation is unlikely to be limited to globular clusters . the dissociation of co around agb stars by nearby post - agb stars and white dwarfs is largely a function of stellar density ( although stellar age , elemental composition and the density of the surrounding ism are also important ) . this source of co dissociation may also become important in old open clusters , nuclear star clusters , and gas - poor galaxies . the implications for the formation , chemistry , evolution and survival of dust in such environments are significant . this paper makes use of the following alma data : ads / jao.alma#2012.1.00333.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 . this publication is based on data acquired with the atacama pathfinder experiment ( apex ) . apex is a collaboration between the max - planck - institut fur radioastronomie , the european southern observatory , and the onsala space observatory . the project identifier for these data is o-092.f-9327(a ) . g.s . was supported by the nsf ( award 1108645 , an alma reconnaissance of distant dying stars ) . we are grateful to the anonymous referee for their constructive feedback on the manuscript . in the main text , we have shown that photo - dissociation of co by interstellar uv radiation is a significant mechanism which reduces the co envelope size . in this appendix , we consider whether alternative processes acting on globular cluster stars may also affect the size of the co envelope . in determining the effect of other processes , the relevant size - scales and time - scales must be compared to the extent and filling time expected for the co envelope . in this section , we determine what those would be for an agb star with the properties of those in 47 tuc but with a galactic isrf incident upon it . we also examine the possibility that the co might be resolved by the alma interferometer . the limiting size of a co envelope in a galactic environment is usually set by its dissociation by the isrf . * their figure 3 ) predict that 90 per cent of the co in the envelope of a galactic agb star with a mass - loss rate of a few @xmath1 10@xmath2 m@xmath3 yr@xmath4 will be dissociated by a radius of @xmath00.0130.023 pc . this equates to @xmath00.6@xmath281.1@xmath28 at the distance of 47 tuc . assuming a typical wind velocity for less - evolved stars of 10 km s@xmath4 @xcite , a typical co envelope would therefore be filled within 1300 to 2200 years . if co is more rapidly dissociated by interstellar uv , we can expect the radius to be smaller and the filling time to be correspondingly shorter . this would preclude a much larger envelope , which could be resolved by the alma beam . since the winds are optically thin , the dissociation radius is mainly set by the time taken for a co molecule to absorb a uv photon . a faster wind may therefore lead to a larger envelope , but the velocity and envelope size will not be directly proportional . larger envelopes will be less self - shielding , as the optical depth in the dissociating lines will decrease . it is unlikely that an envelope could reach the velocity needed to extend beyond the 2@xmath28 alma beam ( @xmath6520 km s@xmath4 ) , and virtually impossible for it to reach the @xmath6560 km s@xmath4 needed to be diluted by a sufficient factor that it becomes unobservable in the central beam . this would still be detectable by the larger @xmath98 box , so we consider it very unlikely that the large - scale structure of our envelopes are resolved by the alma interferometer . if mass loss is episodic or variable , it is feasible that we could see strong mid - ir emission from recent dust production near the star , while the co envelope further from the star traces a period of lower mass - loss rate . given the chromospheric and co mass - loss rates are discrepant by a factor of @xmath010 in each case , any variability or episodic nature must be correspondingly large , i.e. a mass - loss rate variation of 10@xmath1 or , alternatively , mass loss occurring only 10 per cent of the time . while the chances of finding any four stars going through this state is high , the chances of finding four particular stars ( in this case the most luminous ) simultaneously going through an episode of enhanced mass loss is therefore @xmath010@xmath99:1 against . this probability only holds if the enhancement in dust production is uncorrelated with stellar luminosity . such a correlation occurs , for example , during the bright phase of a thermal pulse . @xcite model the thermally - pulsating ( tp ) phase of a globular - cluster - like agb star to be @xmath01.21.8 myr in length . if a star leaves the agb every 80000 years @xcite , this lifetime equates to 1523 stars on the tp - agb , four of which we observe , the remainder of which will be at luminosities at or below the rgb tip . the bright phase of a thermal pulse is relatively short compared to the pulse cycle : @xmath00.3 per cent @xcite . the chances of a particular star being in the bright phase of a thermal pulse are therefore @xmath00.3 per cent , and the chances for four such stars are @xmath01 in 12 billion . although co bands are present in the near - ir spectra of these stars @xcite , it is possible that they could produce enough uv photons to destroy their own co envelopes . metal - poor stars are warmer than their solar - metallicity counterparts due to reduced metal - line opacity in their atmospheres . escaping stellar uv could lead to the destruction of co near the stellar surface . a bt - settl model atmosphere @xcite at 4000 l@xmath3 and 3300 k emits @xmath100 photons per second below the 1120- co photo - dissociation limit . by comparison , a mass - loss rate of @xmath101 m@xmath3 yr@xmath4 and efficient condensation ( i.e. a co : h@xmath41 mass ratio of 1:500 ) yields @xmath102 co molecules produced per second . many more co molecules are produced than there are uv photons to dissociate them . while the bt - settl atmospheres do not model chromospheric or shock heating in the outer atmosphere , the heating can not compensate for the factor of 10@xmath26 between the modelled number of ionising photons and number required to dissociate all the co the star produces . assuming a typical wavelength of 1105 , @xmath102 photons provide an energy of @xmath103 w , or @xmath104 w m@xmath105 as observed from earth . escaping photons would therefore be easily detectable in the _ galex _ fuv filter for all realistic spectral energy distributions . none of the four sources is present in catalogues from the satellite , which is complete over 47 tuc to a far - uv ab magnitude of @xmath3120.0 ( @xmath311 @xmath1 10@xmath106 w m@xmath105 ) , approximately two orders of magnitude more sensitive than would be necessary to detect such a uv field . reduced metal - line and dust cooling in pulsation - driven shocks presumably mean that the post - shocked gas further out in the atmosphere also cools more slowly , leading to more effective dissociation of co further from the stellar surface . however , observations show the expected column depths of co absorption lines in the @xmath107- , @xmath108- and ( archival ) @xmath54-band vlt / crires data @xcite , indicating that co must survive to at least a few stellar radii . the observed co absorption and lack of observed uv emission from these stars means that self - destruction of co by the stars that produce it appears unlikely . stellar encounters are common in globular clusters , and are thought to be responsible for a lack of planets and binary stars , and a prevalence of stellar exotica @xcite . while stellar encounter rates have previously been computed for 47 tuc , published values are not specifically applicable to giant stars envelopes ( e.g. @xcite ) . the encounter timescale ( @xmath109 ) between an agb star and any other star in the cluster can be computed as : @xmath110 where @xmath111 is the stellar number density , @xmath112 is the typical relative velocity at the start of the encounter , @xmath113 is the maximum distance at which an encounter is deemed to occur and @xmath114 is the typical stellar mass ( e.g. @xcite ) . we conservatively assume an envelope radius of 0.03 pc ( section [ coexpectsect ] ) , hence conservatively adopt @xmath115 pc . we assume @xmath112 = 22 km s@xmath4 , or twice the radial component of the central velocity dispersion of the cluster @xcite . given an average stellar density within the half - light radius of 1720 m@xmath3 pc@xmath116 and d@xmath117/d@xmath118 @xmath119 @xmath120 @xcite , bounded by a planetary mass ( 0.08 m@xmath3 ) and the maximum sub - giant mass ( @xmath00.8 m@xmath3 ) , we obtain @xmath114 @xmath65 0.2 m@xmath3 and @xmath111 @xmath69 8600 pc@xmath116 . we conservatively derive @xmath109 @xmath65 340000 years , though the possible range of values allow for @xmath121 myr . this is much longer than the 2300-year envelope filling timescale . while a fraction of stars may undergo envelope - disrupting stellar encounters at some point in their dust - producing phase , this is extremely unlikely to affect all four observed stars simultaneously . a circumstellar envelope will experience tidal disruption or ` spaghettification ' as it interacts with the cluster potential , which becomes significant once it experiences significant differential acceleration over the filling timescale ( which is considerably shorter than the @xmath01 myr orbital period ) . we can assume a 0.06-pc wide envelope in a radial orbit in a @xcite potential of characteristic radius 7.8 pc @xcite , and calculate the change in distance between opposite ends of the envelope . we find a maximum differential acceleration of @xmath122 m s@xmath105 , corresponding to a tidal elongation of one part in 10@xmath99 . tidal deformation of the envelope is therefore negligible . there is a general paucity of interstellar media in globular clusters ( e.g. @xcite ) , with interstellar media detected in only m15 and 47 tuc @xcite . this strongly suggests that globular clusters are efficiently cleared of their icm by some process , on timescales of @xmath691 myr @xcite . we model that this is because it is ionised by hot white dwarfs and post - agb stars , before being later cleared on slightly longer timescales ( a few myr ; @xcite ) . however , this model does not include ablation by the passage of the cluster through the hot gas of the surrounding galactic halo , which is another leading contender @xcite . the stand - off distance of a bow shock ( @xmath123 ) , i.e. the minimum distance from the bow shock to an isolated , mass - losing star , depends on four parameters : the mass - loss rate of the star and the expansion velocity of its wind ( @xmath38 , @xmath124 ) , the interstellar medium density ( @xmath125 ) and the relative velocity of the star with respect to the halo gas ( @xmath126 ) . @xmath123 is given by ( @xcite , and references therein about three times larger . ] ) : @xmath127 depending on the rotational coupling between the halo and milky way , @xmath126 lies between 57 km s@xmath4 ( co - rotating halo ) and 190 km s@xmath4 ( non - rotating halo ) @xcite . the internal velocity dispersion within 47 tuc is negligible in comparison ( 11 km s@xmath4 ; @xcite ) . no good determinations exist for @xmath125 ; estimates at 3.1 kpc from the galactic plane are typically @xmath128 @xmath84 , but vary by an order of magnitude @xcite . @xcite estimate an electron density of @xmath129 @xmath84 at this location ( we can expect @xmath130 in the ionised halo gas ) , and the more - recent spherical model by @xcite predicts a lower value of @xmath1310.002 @xmath84 . assuming @xmath38 @xmath31 @xmath132 m@xmath3 yr@xmath4 , @xmath124 = 415 km s@xmath4 and @xmath133 @xmath84 , @xmath123 = 0.0238 pc , always greater than the expected co shell radius . additionally , these stars are not isolated , but are surrounded by a hot , ionised icm @xcite , which should provide additional shielding out to @xmath13498 pc . we conclude that ram - pressure stripping by halo gas should not be effective at removing circumstellar co envelopes . m. l. , mcdonald i. , van loon j. t. , gordon k. d. , babler b. , block m. , bracker s. , engelbracht c. , hora j. , indebetouw r. , meade m. , meixner m. , misselt k. , oliveira j. m. , sewilo m. , shiao b. , whitney b. , 2009 , apj , 705 , 746 m. l. , mcquinn k. b. w. , barmby p. , bonanos a. z. , gehrz r. d. , gordon k. d. , groenewegen m. a. t. , lagadec e. , lennon d. , marengo m. , mcdonald i. , meixner m. , skillman e. , sloan g. c. , sonneborn g. , van loon j. t. , zijlstra a. , 2015 , apj , 800 , 51 m. l. , srinivasan s. , van loon j. t. , mcdonald i. , meixner m. , zaritsky d. , gordon k. d. , kemper f. , babler b. , block m. , bracker s. , engelbracht c. w. , hora j. , indebetouw r. , meade m. , misselt k. , robitaille t. , sewio m. , shiao b. , whitney b. , 2011 , aj , 142 , 103 m. l. , van loon j. t. , mcdonald i. , gordon k. d. , babler b. , block m. , bracker s. , engelbracht c. , hora j. , indebetouw r. , meade m. , meixner m. , misselt k. , sewilo m. , shiao b. , whitney b. , 2010 , apj , 711 , l99 n. l. j. , kerschbaum f. , van marle a .- j . , decin l. , ladjal d. , mayer a. , groenewegen m. a. t. , van eck s. , royer p. , ottensamer r. , ueta t. , jorissen a. , mecina m. , meliani z. , luntzer a. , blommaert j. a. d. l. , posch t. , vandenbussche b. , waelkens c. , 2012 , a&a , 537 , a35 r. m. , skrutskie m. f. , van dyk s. , beichman c. a. , carpenter j. m. , chester t. , cambrsy l. , evans t. , fowler j. , gizis j. , howard e. , huchra j. , et al . , 2003 , 2mass all sky catalog of point sources . the irsa 2mass all - sky point source catalog , nasa / ipac infrared science archive . r. l. , brown t. m. , guhathakurta p. , sarajedini a. , milone e. f. , albrow m. d. , baliber n. r. , bruntt h. , burrows a. , charbonneau d. , choi p. , cochran w. d. , edmonds p. d. , frandsen s. , howell j. h. , lin d. n. c. , marcy g. w. , mayor m. , naef d. , sigurdsson s. , stagg c. r. , vandenberg d. a. , vogt s. s. , williams m. d. , 2000 , apj , 545 , l47 j. p. , waters b. , schiebel d. , young w. , golap k. , 2007 , in astronomical society of the pacific conference series , vol . 376 , astronomical data analysis software and systems xvi , shaw r. a. , hill f. , bell d. j. , eds .

we show that a tev scale inverse seesaw model for neutrino masses can be realized within the framework of a supersymmetric model consistent with gauge coupling unification and observed neutrino masses and mixing . we present our expectations for non - unitarity effects in the leptonic mixing matrix some of which are observable at future neutrino factories as well as the next generation searches for lepton flavor violating processes such as . the model has tev scale and bosons which are accessible at the large hadron collider . umd - pp-09 - 055 + october , 2009 * tev scale inverse seesaw in so(10 ) and leptonic non - unitarity effects * +

0910.3924

a precise understanding of the origin of observed neutrino masses and mixing is one of the major goals of particle physics right now . a simple paradigm for understanding the smallness of the masses is the seesaw mechanism @xcite where one introduces three standard model ( sm ) singlet right - handed ( rh ) neutrinos with majorana masses @xmath4 , which mix with left - handed ( lh ) ones via the yukawa coupling @xmath5 . the resulting formula for light neutrino masses is given by @xmath6 , where @xmath7 is the dirac neutrino mass . since the sm does not restrict the majorana mass @xmath4 , we could choose this to be much larger than the weak scale thereby providing a natural way to understand the tiny neutrino masses . this is called the type i seesaw . there are several variations of this mechanism where one replaces the rh neutrino by either a sm triplet higgs field ( type ii seesaw ) @xcite or sm triplet of fermions ( called type iii seesaw ) @xcite . a great deal of attention has been devoted to testing these ideas . as far as the type i seesaw is concerned , the prospects of testing this depends on the scale @xmath4 as well as any associated physics that comes with it at that scale . it can be accessible to current and future collider experiments if the scale is not far above a tev . a different way to test the type i seesaw mechanism follows from the observation that this mechanism involves the mixing of the lh neutrinos with sm singlet heavy neutrinos as a result of which there would in general be violation of unitarity of the pontecorvo - maki - nakagawa - sakata ( pmns ) mixing matrix that describes only the mixing of the three light neutrinos . one could contemplate searching for such effects in oscillation experiments @xcite . however , in the type i seesaw case , the resulting mixing effects are of order @xmath8 and since neutrino masses are in the sub - ev range , such non - unitarity effects are too small to be observable for generic high scale seesaw models . this would also be true with tev mass rh neutrinos unless there are cancellations to get small neutrino masses from large dirac masses using symmetries ( see for example cases with @xcite ) . we note that the non - unitarity effects are also there in the type iii seesaw case but not in the type ii case even though they have other interesting effects such as lfv processes @xcite . since the testability of seesaw is intimately related to the magnitude of the seesaw scale , a key question of interest is whether there could be any theoretical guidelines for the seesaw scale . in such a case , the searches for seesaw effects in experiments could then be used to test the nature of physics beyond the standard model . it is well - known that @xcite the simplest grand unified theory ( gut ) realizations of the seesaw mechanism are based on the @xmath0 group which automatically predicts the existence of the rh neutrinos ( along with the sm fermions ) required by the seesaw mechanism . an advantage of gut embedding of the seesaw mechanism is that the constraints of gut symmetry tends to relate the dirac neutrino mass @xmath7 to the charged fermion masses thereby making a prediction for the seesaw scale @xmath4 from observations . for type i seesaw gut embedding , typical values for the @xmath4 are very high ( in the range of @xmath9 - @xmath10 gev ) . this makes both the collider as well as non - unitarity probes of seesaw impossible . the key feature that leads to such restrictions in type i seesaw case is the close link between the @xmath11 breaking rh neutrino mass and the smallness of the lh neutrino masses . a completely different realization of the seesaw mechanism is the so - called inverse seesaw mechanism @xcite , where instead of one set of three sm singlet fermions , one introduces two sets of them @xmath12 ( @xmath13 ) . in the context of @xmath0 models , since one of the two sets can be identified with the sm singlet neutrino in the @xmath0 * 16 * -representation containing matter , the others would have to be a separate set of three @xmath0 singlet fermions . due to the existence of the second set of singlet fermions ( and perhaps additional gauge symmetries e.g. @xmath0 ) , the neutrino mass formula in these models has the form @xmath14 where @xmath15 breaks the lepton number . because of the presence of this new mass scale in this theory , the seesaw scale @xmath4 can be very close to a tev even for `` large '' dirac masses . this makes the tests of this possibility in colliders much more feasible . in fact it has recently been argued that @xcite the inverse seesaw scenario can also lead to non - negligible non - unitarity effects which can be accessible at the future long - baseline neutrino oscillation experiments . there are also significant lepton flavor violation ( lfv ) effects in these models as noted many years ago in ref . these possibilities have generated a great deal of interest in the inverse seesaw models in recent days @xcite . our effort in this paper focuses on possible grand unification of inverse seesaw models . similar unification studies have been performed in ref . @xcite , but they have not addressed the non - unitarity issues . an interesting question is whether such models are necessarily compatible with grand unification when the seesaw scale is in the tev range and if so what kind of non - unitarity effects they predict . we find that it is indeed possible to embed the tev scale inverse seesaw models within a simple @xmath0 framework consistent with gauge coupling unification and realistic fermion masses . the @xmath0 symmetry helps to reduce the number of parameters in the inverse seesaw matrix , once we require degeneracy of the tev scale rh neutrinos to have successful resonant leptogenesis . within this set of assumptions , we present our expectations for the non - unitarity effects as well as consequences for lepton flavor violation which are in the testable range in future experiments . this paper is organized as follows : in section 2 , we describe the general framework of the inverse seesaw model and its embedding into a generic supersymmetric @xmath0 theory . in section 3 , we analyze the non - unitarity predictions of the inverse seesaw model . in section 4 , we investigate a specific @xmath0 breaking chain and obtain the gauge coupling unification with tev scale left - right symmetry with unification scale consistent with proton decay bounds . in section 5 , we analyze the renormalization group ( rg ) evolution of the yukawa couplings and obtain the running masses for quarks and leptons at the unification scale to check that our model leads to realistic fermion masses . in section 6 , we determine the dirac neutrino mass matrix using the results of section 5 . in section 7 , we study the implication of our model on non - unitarity effects and its phenomenological consequences . a brief summary of the results is presented in section 8 . in appendix a , we have given the expressions for the masses of the @xmath0 higgs multiplets in our model , and in appendix b , we have derived the rg equations for the quark and lepton masses and the ckm mixing elements in the context of a supersymmetric left - right model . the inverse seesaw scheme was originally suggested @xcite for theories which lack the representation required to implement the canonical seesaw , such as the superstring models . as noted in the introduction , the implementation of the inverse seesaw requires the addition of three extra sm gauge singlets @xmath16 coupled to the rh neutrinos @xmath17 through the lepton number conserving couplings of the type @xmath18 while the traditional rh neutrino majorana mass term is forbidden by extra symmetries . the lepton number is broken only by the self coupling term @xmath19 . the mass part of the neutrino sector lagrangian in the flavor basis is given by @xmath20 where @xmath15 is a complex symmetric @xmath21 matrix ( with dimension of mass ) , and @xmath7 and @xmath4 are generic @xmath21 complex matrices representing the dirac mass terms in the @xmath22 - @xmath23 and @xmath23 - @xmath24 sectors respectively . in the basis @xmath25 , the @xmath26 neutrino mass matrix becomes @xmath27 the lh neutrinos can be made very light ( sub - ev scale ) , as required by the oscillation data , even for a low @xmath4 , much smaller than the unification scale ( @xmath28 ) , provided @xmath15 is sufficiently small , @xmath29 , as the lepton number breaking scale @xmath15 is decoupled from the rh neutrino mass scale . assuming @xmath30 ( with @xmath31 tev ) , the structure of the light neutrino majorana mass term at the leading order in @xmath32 is given by eq . ( [ eq : neutrino ] ) , where @xmath33 is a complex @xmath21 matrix . we note that in the limit @xmath34 which corresponds to the unbroken lepton number , we have massless lh neutrinos as in the sm . in reality , a small non - vanishing @xmath15 can be viewed as a slight breaking of a global @xmath35 symmetry ; hence , the smallness of @xmath15 is natural , in the t hooft sense @xcite , even though there is no dynamical understanding of this smallness . the generic form of the inverse seesaw matrix in eq . ( [ eq : big ] ) has more parameters than the usual type i seesaw . however , if we embed this theory into a grand unified theory such as @xmath0 , that will help in reducing the parameters as we show below . in order to embed the inverse see - saw mechanism into a generic @xmath0 theory , we have to break the @xmath11 symmetry by using a @xmath36 pair rather than the @xmath37 pair of higgs representation . all the sm fermions are accommodated in a single @xmath38 representation of @xmath0 and we use three copies of @xmath39 for three generations . for each of them , we add a gauge singlet fermion @xmath40 to play the role of @xmath16 . we assume more than one copy of @xmath41 higgs multiplets in order to have a realistic fermionic spectrum . the @xmath0 invariant renormalizable yukawa superpotential is given by @xmath42 after the @xmath11 symmetry breaking , we get the neutrino mass matrix in eq . ( [ eq : big ] ) with @xmath43 and @xmath44 , where @xmath45 is the vacuum expectation value ( vev ) of one ( or , a linear combination ) of the @xmath41 s and @xmath46 the vev of the @xmath47 . in a typical tev - scale scenario with @xmath48 gev ( electroweak scale ) and @xmath49 tev , assuming @xmath50 , we find the lightest neutrino mass from eq . ( [ eq : neutrino ] ) in a one generation theory to be @xmath51 and the two other heavy eigenstates with mass of order @xmath52 . thus , we can get sub - ev light neutrino mass for @xmath53 kev . since this is a supersymmetric theory , such small values do not receive radiative corrections and keep the model natural . in the following section , we consider three generations which then results in the non - unitarity effect . it is important to note that in our model , we do not need to impose a discrete @xmath54-parity to our matter fermions , unlike the usual @xmath55 @xmath0 models discussed in literature , in order to prevent fast proton decay via dimension-4 operators of the type @xmath56 because these operators are already suppressed by a factor @xmath57 for a low - scale @xmath11 breaking with @xmath58 tev . the @xmath21 light neutrino mass matrix in eq . ( [ eq : neutrino ] ) can be diagonalized by a unitary transformation : @xmath59 where @xmath60 is the standard pmns matrix . since the above diagonalization of @xmath61 does not diagonalize the matrices @xmath4 and @xmath15 , there will be off - diagonal mixing between the different light neutrinos even after diagonalization of @xmath61 due to their mixing with the heavy neutrinos . in other words , in the basis where the charged - lepton mass matrix is diagonal , @xmath60 is only a part of the full mixing matrix responsible for neutrino oscillations . we have to examine the full @xmath26 unitary matrix @xmath62 which diagonalizes the mass matrix @xmath63 given by eq . ( [ eq : big ] ) : @xmath64 we can decompose @xmath62 into the blocks @xmath65 then the upper - left sub - block @xmath66 will represent the full ( non - unitary ) pmns mixing matrix . for a tev - scale @xmath4 and a reasonably small @xmath15 , it is sufficient to consider only up to the leading order in @xmath67 . then the new pmns matrix becomes @xcite @xmath68 in the commonly used parametrization @xcite , @xmath69 , and hence , all the non - unitarity effects are determined by the hermitian matrix @xmath70 which depends only on the mass ratio @xmath33 and not on the parametrization of the pmns matrix . the lh neutrinos entering the charged - current interactions of the sm now become superpositions of the nine mass eigenstates @xmath71 and at the leading order in @xmath67 , @xmath72 where @xmath73 and @xmath74 . then the charged - current lagrangian in the mass basis is given by @xmath75 this mixing between the doublet and singlet components in the charged - current sector has several important phenomenological consequences , as listed below : 1 . the flavor and mass eigenstates of the lh neutrinos are now connected by a non - unitary mixing matrix @xmath69 , where the non - unitarity effects entering different neutrino oscillation channels are measured by the parameter @xmath76 . in particular , the @xmath77-violating effects in the leptonic sector will now be governed by the pmns matrix @xmath78 instead of @xmath60 through the jarlskog invariant @xcite @xmath79 where the indices @xmath80 run over @xmath81 and @xmath82 , while @xmath83 can be 1 , 2 and 3 . in the standard pmns parametrization of @xmath60 by the three mixing angles @xmath84 and the dirac @xmath77-phase @xmath85 , one can expand eq . ( [ eq : jarl1 ] ) up to second order in @xmath86 and @xmath87 ( assuming those to be small ) to obtain @xmath88 where the first term governs the @xmath77-violating effects in the unitary limit and the second term gives the contribution coming from the non - unitarity effect : @xmath89 note that the unitary term @xmath90 vanishes if either @xmath91 or @xmath92 . however , @xmath93 depends on the off - diagonal elements of @xmath76 ( generally complex ) and does not necessarily vanish even if both @xmath94 and @xmath85 are zero ; in fact , it might even dominate the @xmath77-violating effects in the leptonic sector . 2 . the heavy neutrinos @xmath17 and @xmath95 entering the charged - current sector can also mediate the rare lepton decays , @xmath96 . hence , unlike in the canonical seesaw model where this contribution is suppressed by the light neutrino masses @xcite , in this case it is constrained mainly by the ratio @xmath33 . the lfv decays mediated by these heavy neutrinos have branching ratios @xcite @xmath97 where @xmath98 is the total decay width of @xmath99 and the function @xmath100 is defined by @xmath101 for degenerate rh neutrino masses , a reasonable assumption inspired by resonant leptogenesis @xcite , the amplitude is proportional to @xmath102 , and hence , for sizeable @xmath67 and tev - scale rh sector , one could expect appreciable rates in the lfv channels . on the other hand , in the conventional type i seesaw model , one has approximately @xmath103 , and therefore , the branching ratio , @xmath104 is strongly suppressed . 3 . the heavy neutrinos @xmath17 and @xmath95 also couple to the gauge sector of the sm and can be produced on - shell , if kinematically accessible , at hadron colliders via the gauge boson exchange diagrams . due to their pseudo - dirac nature , the striking lepton number violating lhc signature of the fine - tuned type i and type iii scenarios , namely @xmath105 jets , will be suppressed for heavy majorana states due to cancellation between the graphs with internal lines of the @xmath23 and @xmath106 type which have opposite @xmath77-quantum numbers . however , the lfv processes are insensitive to this effect and one can expect to get observable signals at the lhc . the most distinctive signature would be the observation of lfv processes involving three charged leptons in the final state plus missing energy , i.e. @xmath107 jets @xcite . thus we see that the phenomenology of the inverse seesaw mechanism depends crucially on the mass ratio @xmath33 . as noted earlier , we can choose the rh neutrino masses to be degenerate ( with eigenvalue @xmath108 ) , inspired by resonant leptogenesis . so we are left with a single mass parameter @xmath108 , together with the dirac mass matrix @xmath7 and the arbitrary small mass parameter @xmath15 . in what follows , we explicitly determine the form of @xmath7 in the context of a realistic supersymmetric @xmath0 model and then use the present experimental bounds on the elements of the non - unitary parameter @xmath109 to get a lower bound on the rh neutrino mass scale @xmath108 . finally we fit the observed lh neutrino mass and mixing parameters by the inverse seesaw formula to determine the structure of @xmath15 . we then study the phenomenological consequences of our results . as we have mentioned earlier , in order to embed the inverse seesaw mechanism into a supersymmetric @xmath0 theory , we have to break the @xmath11 symmetry by using a * 16 * @xmath110 pair rather than a * 126 * @xmath111 pair of higgs representation . in this context , there are two symmetry breaking chains that are particularly interesting : * @xmath112 @xcite * @xmath113 + @xmath114 @xcite where , as an example of our notation , @xmath115 means @xmath116 . in this paper , we consider only the former ( and simpler ) case of @xmath0 breaking chain . it was shown in ref . @xcite that it is possible to obtain the gauge coupling unification in this model with a low - energy ( tev - scale ) @xmath117 symmetry breaking scale @xmath118 . however , they considered only one @xmath119 higgs field which contains only a single bi - doublet ( corresponding to the ( 1,2,2,0 ) representation of @xmath120 ) . getting a realistic fermion mass spectrum in this model is difficult ( see , however , some recent ideas @xcite on how this could be done ) . instead , we consider a model with two @xmath119 at the tev scale . this requires that we reexamine the unification issue with two higgs bi - doublets . we show that we not only obtain the gauge coupling unification at a scale consistent with the proton decay bounds , but also successfully reproduce the observed fermion masses and mixing . to study the running of the gauge couplings and the possibility of their unification at a scale @xmath121 gev , we divide the whole energy range @xmath122 into three parts , according to the above mentioned symmetry breaking chain : * first , we have the well known sm from the weak scale @xmath123 to the susy - breaking scale @xmath124 ( which , for practical purposes , we assume to be a little higher than @xmath123 ) ; * then we have the mssm from @xmath124 to the @xmath11 breaking scale @xmath118 ( which is assumed to be of order tev , so that it is of interest for colliders ) ; * finally , we have the supersymmetric left - right ( susylr ) model from @xmath118 to the unification scale @xmath125 ( expected to be around @xmath126 gev ) . the running of the gauge couplings at one - loop level is determined by the rg equation @xmath127 where @xmath128 , @xmath129 is the energy scale , and @xmath130 s are the coefficients of the one - loop @xmath131-functions . the sm and mssm @xmath131-functions are well known @xcite : @xmath132 where @xmath133 stands for @xmath134 , and @xmath115 respectively . before calculating the @xmath131-functions for the susylr model , let us first discuss the particle contents of this model . here we consider only the doublet implementation of the susylr model @xcite , i.e. we use @xmath135 doublets of the @xmath55 higgs field to break the @xmath11 symmetry . in order to keep the model general , we allow for an arbitrary number of these doublet fields , to be denoted by @xmath136 and @xmath137 respectively for @xmath138 and @xmath117 doublets . likewise we have @xmath139 bi - doublets of the @xmath41 higgs field which , on acquiring vevs , give masses to the fermions through yukawa couplings . we also allow for an arbitrary number @xmath140 of singlet fields @xmath141 . this is the minimal set of particles in a generic susylr model . however , it turns out that with this minimal set of particles , it is not possible to obtain the gauge coupling unification at a scale higher than @xmath142 gev as required from current bounds on proton decay lifetime , @xmath143 years @xcite . as we have shown later in section 4.2 , unification is possible only after adding the contribution from the color triplets @xmath144 $ ] which come from the @xmath145 representation of higgs at the unification scale @xmath125 . it is justified in appendix a that it is indeed possible to have these color triplets at tev scale while all the other higgs multiplets are still naturally heavy at the gut - scale . .the representations of the particles under the @xmath120 gauge group in the doublet susylr model . here @xmath146 and @xmath147 . the @xmath11 quantum numbers given here are not gut renormalized ; to do so , we multiply a factor of @xmath148 ( not @xmath149 as mentioned in ref . @xcite ) . [ cols="^,^,^,^,^",options="header " , ] * ( 1,1,0 ) : we have three such states and the mass matrix is given by @xmath150 so the mass eigenvalues are @xmath151 \neq 0,\nonumber\\ m_3^{(1,1,0 ) } & = & \frac{e}{2\sqrt{15}}\left[\left(\lambda_2+\frac{3}{2}\lambda_1\right ) -\sqrt{\left(\lambda_2+\frac{3}{2}\lambda_1\right)^2 + 16\lambda_2 ^ 2}\right ] \neq 0\end{aligned}\ ] ] * @xmath152 $ ] : there is only one such multiplet and its mass is @xmath153 * @xmath154 $ ] : there are two such multiplets and the mass matrix is @xmath155 with the eigenvalues @xmath156 \neq 0\end{aligned}\ ] ] * @xmath157 $ ] : there are two of them and the mass matrix is @xmath158 with the same eigenvalues as the previous one : @xmath159 \neq 0\end{aligned}\ ] ] * ( 1,3,0 ) : there are also two of them and the mass matrix is @xmath160 so the mass eigenvalues are @xmath161 * @xmath162 $ ] : there is only one such multiplet whose eigenvalue is given by @xmath163 * @xmath164 $ ] : its eigenvalue is @xmath165 unless @xmath166 ( which we assume not to be the case ) . * ( 8,1,0 ) : there are two of them and the mass matrix is @xmath167 with the mass eigenvalues @xmath168\neq 0\end{aligned}\ ] ] thus we see that all the other multiplets have non - zero masses , and moreover , all these masses are of order @xmath169 . hence , none of these multiplets will contribute to the running of gauge coupling up to the unification scale @xmath125 except the color triplets since these color triplets have masses of order of the susy breaking scale . note that the @xmath119-higgs field also has a color triplet pair @xmath170 $ ] under the sm gauge group , apart from the tev - scale bi - doublet fields @xmath171 used in the susylr model in section 4 which reduce to @xmath172 under the sm gauge group . at the gut - scale , the @xmath173 field interacts with the @xmath174 field by the following term in the superpotential : @xmath175 after the @xmath176 acquires a vev , this gives rise to the color triplet mass @xmath177 while the doublet mass is @xmath178 we see that the @xmath172 field can be made light by fine - tuning @xmath179 tev which still leaves the @xmath180 field heavy ( of order @xmath125 ) . finally , let us discuss how only the right handed doublets fields @xmath181 from @xmath182-higgs fields ( @xmath183 ) remain massless at the gut scale . note that in the left - right language , the fields in * 16 * are @xmath184 and @xmath185 , and similarly for @xmath186 field . the superpotential involving these fields is @xmath187 the second coupling has been worked out explicitly in ref . @xcite . on substituting the vev of the * 45 * -higgs field ( @xmath188 ) , we get the following masses for the @xmath189 and @xmath185 fields : @xmath190 from this we see that to get only the @xmath191 fields light , we have to fine - tune @xmath192 tev . with this assumption , all other fields remain heavy at the gut scale . given the form of the bi - doublets vevs as in eq . ( [ eq : bivev ] ) , it immediately follows from the first two terms of the superpotential eq . ( [ eq : modsup ] ) that the fermion mass matrices can be written as @xmath193 henceforth , for clarity , we will denote the yukawa couplings as @xmath194 then using eqs . ( [ eq : ha ] , [ eq : hap ] ) and ( [ eq : vu ] , [ eq : vd ] ) the rges for the fermion mass matrices can be written as @xmath195\nonumber\\ & & + m_d\tan\beta\left[{\rm tr}\left(3h_d^\dagger h_u+h_e^\dagger h_n\right)+2h_d^\dagger h_u+c^\phi_{12}\right ] \label{eq : mu1}\\ 16\pi^2\frac{dm_d}{dt } & = & m_d\left[4h_d^\dagger h_d+2h_u^\dagger h_u-\sum_i \tilde{c}_i^{(q)}g_i^2\right]\nonumber\\ & & + \frac{m_u}{\tan\beta}\left[{\rm tr}\left(3h_u^\dagger h_d+h_n^\dagger h_e\right)+2h_u^\dagger h_d+c^\phi_{21}\right]\label{eq : md1}\\ 16\pi^2\frac{dm_e}{dt } & = & m_e\left[4h_e^\dagger h_e+2h_n^\dagger h_n+c^\chi-\sum_i \tilde{c}_i^{(l)}g_i^2\right]\nonumber\\ & & + \frac{m_d}{\tan\beta}\left[{\rm tr}\left(3h_u^\dagger h_d+h_n^\dagger h_e\right)+2h_n^\dagger h_e+c^\phi_{21}\right]\label{eq : me1}\\ 16\pi^2\frac{dm_d}{dt } & = & m_d\left[4h_n^\dagger h_n+2h_e^\dagger h_e+c^\chi-\sum_i \tilde{c}_i^{(l)}g_i^2\right]\nonumber\\ & & + m_e\tan\beta\left[{\rm tr}\left(3h_d^\dagger h_u+h_e^\dagger h_n\right)+2h_e^\dagger h_n+c^\phi_{12}\right]\label{eq : mnu1}\end{aligned}\ ] ] where @xmath196 , and for @xmath197 , @xmath198 note that the second line in each of the above mass rges , eqs . ( [ eq : mu1]-[eq : mnu1 ] ) , is characteristic of the left - right models , and does not appear in mssm . not all the parameters of the yukawa matrices are physical . under an arbitrary unitary transformation on the left(right)-handed fermion fields , @xmath199 ( where @xmath200 ) , the yukawa matrices undergo a bi - unitary transformation , @xmath201 and the charged current becomes off - diagonal , with the ckm mixing matrix @xmath202 . we will also have a leptonic counterpart of the ckm matrix that represents the mixing between the charged lepton and dirac neutrino sector . however , as the running of lepton masses is very mild and we are working only to the one - loop order , we can safely ignore this mixing in the leptonic sector . moreover , if we assume the @xmath77 phase in the higgs vev to be zero , then the mass matrices are hermitian and @xmath203 ( manifest left - right ) . thus we may perform scale - dependent unitary transformations @xmath204 on the fermion bases so as to diagonalize the yukawa matrices , and hence the mass matrices , at each scale : @xmath205 where @xmath206 and @xmath207 denote the diagonalized yukawa and mass matrices , respectively . the rges for the physically relevant quantities , namely the mass eigenvalues @xmath208 and the scale - dependent ckm matrix @xmath209 , are both contained in the rges of @xmath210 : @xmath211+\frac{1}{16\pi^2}\left[4\widehat{h}_u^2 + 2\widehat{h}_d^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right]2\widehat{m}_u^2\nonumber\\ & & + \frac{1}{16\pi^2}\tan\beta\left[\left\{{\rm tr}\left(3v_{\rm ckm}\widehat{h}_dv_{\rm ckm}^\dagger \widehat{h}_u\right ) + c^\phi_{12}\right\}\left(v_{\rm ckm}\widehat{m}_dv_{\rm ckm}^\dagger \widehat{m}_u\right)\right.\nonumber\\ & & \left.+2v_{\rm ckm}\widehat{m}_d\widehat{h}_dv_{\rm ckm}^\dagger \widehat{h}_u\widehat{m}_u+{\rm h.c.}\right]\label{eq : mu2}\\ \frac{d}{dt}\left(\widehat{m}_d^2\right ) & = & \left[\dot{l}_dl_d^\dagger,\widehat{m}_d^2\right]+\frac{1}{16\pi^2}\left[4\widehat{h}_d^2 + 2\widehat{h}_u^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right]2\widehat{m}_d^2\nonumber\\ & & + \frac{1}{16\pi^2}\frac{1}{\tan\beta}\left[\left\{{\rm tr}\left(3v_{\rm ckm}\widehat{h}_dv_{\rm ckm}^\dagger \widehat{h}_u\right ) + c^\phi_{12}\right\}\left(\widehat{m}_dv_{\rm ckm}^\dagger\widehat{m}_uv_{\rm ckm}\right)\right.\nonumber\\ & & \left.+2\widehat{m}_d\widehat{h}_dv_{\rm ckm}^\dagger\widehat{h}_u\widehat{m}_uv_{\rm ckm}+{\rm h.c.}\right]\label{eq : md2}\\ \frac{d}{dt}\left(\widehat{m}_e^2\right ) & = & \left[\dot{l}_el_e^\dagger,\widehat{m}_e^2\right]+\frac{1}{16\pi^2}\left[4\widehat{h}_e^2 + 2\widehat{h}_n^2+\re\left(c^\chi\right)-\sum_i \tilde{c}_i^{(l)}g_i^2\right]2\widehat{m}_e^2\label{eq : me2}\\ \frac{d}{dt}\left(\widehat{m}_d^2\right ) & = & \left[\dot{l}_nl_n^\dagger,\widehat{m}_d^2\right]+\frac{1}{16\pi^2}\left[4\widehat{h}_n^2 + 2\widehat{h}_e^2+\re\left(c^\chi\right)-\sum_i \tilde{c}_i^{(l)}g_i^2\right]2\widehat{m}_d^2\label{eq : mnu2}\end{aligned}\ ] ] where @xmath212 and @xmath213 denotes the real part of @xmath214 . the commutator @xmath215 $ ] has vanishing diagonal elements because @xmath216 is diagonal . thus the rges for the mass eigenvalues @xmath217 follow immediately from the diagonal entries of eqs . ( [ eq : mu2]-[eq : mnu2 ] ) . using dominance of yukawa couplings of the third generation over the first two , i.e. @xmath218 we obtain the following rges for the mass eigenvalues of the fermions : @xmath219\sum_{j = d , s , b}|v_{uj}|^2m_j\nonumber \\ 16\pi^2\frac{dm_c}{dt } & \simeq & \left(4h_c^2 + 2h_s^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right)m_c + \tan\beta\left[3|v_{tb}|^2h_bh_t+r_q\right]\sum_{j = d , s , b}|v_{cj}|^2m_j\nonumber\\ 16\pi^2\frac{dm_t}{dt } & \simeq & \left(4h_t^2 + 2h_b^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right)m_t + \tan\beta\left[\left(3|v_{tb}|^2 + 2\right)h_bh_t+r_q\right]|v_{tb}|^2m_b\nonumber\\ 16\pi^2\frac{dm_d}{dt } & \simeq & \left(4h_d^2 + 2h_u^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right)m_d + \frac{1}{\tan\beta}\left[3|v_{tb}|^2h_bh_t+r_q\right]\sum_{j = u , c , t}|v_{jd}|^2m_j\nonumber\\ 16\pi^2\frac{dm_s}{dt } & \simeq & \left(4h_s^2 + 2h_c^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right)m_s + \frac{1}{\tan\beta}\left[3|v_{tb}|^2h_bh_t+r_q\right]\sum_{j = u , c , t}|v_{js}|^2m_j\nonumber\\ 16\pi^2\frac{dm_b}{dt } & \simeq & \left(4h_b^2 + 2h_t^2-\sum_i \tilde{c}_i^{(q)}g_i^2\right)m_b + \frac{1}{\tan\beta}\left[\left(3|v_{tb}|^2 + 2\right)h_bh_t+r_q\right]|v_{tb}|^2m_t\nonumber\\ 16\pi^2\frac{dm_e}{dt } & \simeq & \left(4h_e^2 + 2h_{n_1}^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_e\nonumber\\ 16\pi^2\frac{dm_\mu}{dt } & \simeq & \left(4h_\mu^2 + 2h_{n_2}^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_\mu\nonumber\\ 16\pi^2\frac{dm_\tau}{dt } & \simeq & \left(4h_\tau^2 + 2h_{n_3}^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_\tau\nonumber\\ 16\pi^2\frac{dm_{n_1}}{dt } & \simeq & \left(4h_{n_1}^2 + 2h_e^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_{n_1}\nonumber\\ 16\pi^2\frac{dm_{n_2}}{dt } & \simeq & \left(4h_{n_2}^2 + 2h_\mu^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_{n_2}\nonumber\\ 16\pi^2\frac{dm_{n_3}}{dt } & \simeq & \left(4h_{n_3}^2 + 2h_\tau^2+r_l-\sum_i \tilde{c}_i^{(l)}g_i^2\right)m_{n_3 } \label{eq : rgmass}\end{aligned}\ ] ] where @xmath220 and @xmath221 . the vev rges , eqs . ( [ eq : vd ] ) and ( [ eq : vu ] ) , for third generation dominance become @xmath222,\\ 16\pi^2\frac{dv_d}{dt}&\simeq & v_d\left[\frac{3}{2}g_{2l}^2 + \frac{3}{2}g_{2r}^2 - 3h_b^2-h_\tau^2 - c^\phi_{11}\right]\end{aligned}\ ] ] the rge for the ckm matrix @xmath223 is given by @xmath224 however , the diagonal elements of @xmath225 are not determined by eqs . ( [ eq : mu2 ] ) and ( [ eq : md2 ] ) . this is because eq . ( [ eq : hat ] ) determines @xmath226 only up to right multiplication by a diagonal matrix of scale - dependent phases . these undetermined phases contribute arbitrary imaginary functions to the diagonal elements of @xmath225 . but the off - diagonal elements are unambiguously determined because they receive no contribution from the phases . we can , nevertheless , make the diagonal entries of @xmath225 , which are manifestly imaginary , vanish by an appropriate choice of phases . with this choice of phases , we can then obtain the rges for the ckm matrix elements using eq . ( [ eq : ckm1 ] ) : @xmath227v_{\gamma\beta } \right . \nonumber\\ & & \left . - \sum_{\stackrel{\scriptstyle \gamma = d , s , b}{\gamma\neq\beta}}v_{\alpha\gamma}\left[\frac{1}{\tan\beta(m_\gamma - m_\beta ) } \left\{{\rm tr}\left(3v\widehat{h}_dv^\dagger \widehat{h}_u\right)+r_q\right\}\left(v^\dagger \widehat{m}_u v\right)_{\gamma\beta } \right.\right.\nonumber\\&&\left.\left . ~~~~~~~~~~+ \frac{4}{v_u^2}\frac{m_\gamma^2+m_\beta^2}{m_\gamma^2-m_\beta^2}\left(v^\dagger \widehat{m}_u^2v\right)_{\gamma\beta}\right]\right)\end{aligned}\ ] ] as before , we use the third generation dominance and get the following rges for @xmath228 : @xmath229\nonumber\\ & & -\frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{uc}v_{cd } + \left(v\widehat{m}_d^2v^\dagger\right)_{ut}v_{td}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right ) \left[\frac{v_{us}\left(v^\dagger\widehat{m}_uv\right)_{sd}}{m_s } + \frac{v_{ub}\left(v^\dagger\widehat{m}_uv\right)_{bd}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left[v_{us}\left(v^\dagger\widehat{m}_u^2v\right)_{sd } + v_{ub}\left(v^\dagger \widehat{m}_u^2v\right)_{bd}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{us } & \simeq & -\tan\beta\left(3|v_{tb}|^2h_bh_t+r_q\right)\left[\frac{\left(v\widehat{m}_dv^\dagger\right)_{uc}v_{cs}}{m_c } + \frac{\left(v\widehat{m}_dv^\dagger\right)_{ut}v_{ts}}{m_t}\right]\nonumber\\ & & -\frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{uc}v_{cs } + \left(v\widehat{m}_d^2v^\dagger\right)_{ut}v_{ts}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right ) \left[-\frac{v_{ud}\left(v^\dagger\widehat{m}_uv\right)_{ds}}{m_s } + \frac{v_{ub}\left(v^\dagger\widehat{m}_uv\right)_{bs}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left[-v_{ud}\left(v^\dagger\widehat{m}_u^2v\right)_{ds } + v_{ub}\left(v^\dagger \widehat{m}_u^2v\right)_{bs}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{ub } & \simeq & -\tan\beta\left(3|v_{tb}|^2h_bh_t+r_q\right)\left[\frac{\left(v\widehat{m}_dv^\dagger\right)_{uc}v_{cb}}{m_c } + \frac{\left(v\widehat{m}_dv^\dagger\right)_{ut}v_{tb}}{m_t}\right]\nonumber\\ & & -\frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{uc}v_{cb } + \left(v\widehat{m}_d^2v^\dagger\right)_{ut}v_{tb}\right]\nonumber\\ & & + \frac{1}{m_b\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right ) \left[v_{ud}\left(v^\dagger\widehat{m}_uv\right)_{db } + v_{us}\left(v^\dagger\widehat{m}_uv\right)_{sb}\right]\nonumber\\ & & + \frac{4}{v_u^2}\left[v_{ud}\left(v^\dagger\widehat{m}_u^2v\right)_{db } + v_{us}\left(v^\dagger \widehat{m}_u^2v\right)_{sb}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{cd } & \simeq & -\tan\beta\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ -\frac{\left(v\widehat{m}_dv^\dagger\right)_{cu}v_{ud}}{m_c } + \frac{\left(v\widehat{m}_dv^\dagger\right)_{ct}v_{td}}{m_t}\right]\nonumber\\ & & -\frac{4}{v_d^2}\left[-\left(v\widehat{m}_d^2v^\dagger\right)_{cu}v_{ud } + \left(v\widehat{m}_d^2v^\dagger\right)_{ct}v_{td}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ \frac{v_{cs}\left(v^\dagger\widehat{m}_uv\right)_{sd}}{m_s } + \frac{v_{cb}\left(v^\dagger\widehat{m}_uv\right)_{bd}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left[v_{cs}\left(v^\dagger\widehat{m}_u^2v\right)_{sd } + v_{cb}\left(v^\dagger \widehat{m}_u^2v\right)_{bd}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{cs } & \simeq & -\tan\beta\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ -\frac{\left(v\widehat{m}_dv^\dagger\right)_{cu}v_{us}}{m_c } + \frac{\left(v\widehat{m}_dv^\dagger\right)_{ct}v_{ts}}{m_t}\right]\nonumber\\ & & -\frac{4}{v_d^2}\left[-\left(v\widehat{m}_d^2v^\dagger\right)_{cu}v_{us } + \left(v\widehat{m}_d^2v^\dagger\right)_{ct}v_{ts}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ -\frac{v_{cd}\left(v^\dagger\widehat{m}_uv\right)_{ds}}{m_s } + \frac{v_{cb}\left(v^\dagger\widehat{m}_uv\right)_{bs}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left [ -v_{cd}\left(v^\dagger\widehat{m}_u^2v\right)_{ds } + v_{cb}\left(v^\dagger \widehat{m}_u^2v\right)_{bs}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{cb } & \simeq & -\tan\beta\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ -\frac{\left(v\widehat{m}_dv^\dagger\right)_{cu}v_{ub}}{m_c } + \frac{\left(v\widehat{m}_dv^\dagger\right)_{ct}v_{tb}}{m_t}\right]\nonumber\\ & & -\frac{4}{v_d^2}\left[-\left(v\widehat{m}_d^2v^\dagger\right)_{cu}v_{ub } + \left(v\widehat{m}_d^2v^\dagger\right)_{ct}v_{tb}\right]\nonumber\\ & & + \frac{1}{m_b\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ v_{cd}\left(v^\dagger\widehat{m}_uv\right)_{db } + v_{cs}\left(v^\dagger\widehat{m}_uv\right)_{sb}\right]\nonumber\\ & & + \frac{4}{v_u^2}\left [ v_{cd}\left(v^\dagger\widehat{m}_u^2v\right)_{db } + v_{cs}\left(v^\dagger \widehat{m}_u^2v\right)_{sb}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{td } & \simeq & \frac{\tan\beta}{m_t}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ \left(v\widehat{m}_dv^\dagger\right)_{tu}v_{ud } + \left(v\widehat{m}_dv^\dagger\right)_{tc}v_{cd}\right]\nonumber\\ & & + \frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{tu}v_{ud } + \left(v\widehat{m}_d^2v^\dagger\right)_{tc}v_{cd}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ \frac{v_{ts}\left(v^\dagger\widehat{m}_uv\right)_{sd}}{m_s } + \frac{v_{tb}\left(v^\dagger\widehat{m}_uv\right)_{bd}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left[v_{ts}\left(v^\dagger\widehat{m}_u^2v\right)_{sd } + v_{tb}\left(v^\dagger \widehat{m}_u^2v\right)_{bd}\right]\nonumber \\ 16\pi^2\frac{d}{dt}v_{ts } & \simeq & \frac{\tan\beta}{m_t}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ \left(v\widehat{m}_dv^\dagger\right)_{tu}v_{us } + \left(v\widehat{m}_dv^\dagger\right)_{tc}v_{cs}\right]\nonumber\\ & & + \frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{tu}v_{us } + \left(v\widehat{m}_d^2v^\dagger\right)_{tc}v_{cs}\right]\nonumber\\ & & -\frac{1}{\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ -\frac{v_{td}\left(v^\dagger\widehat{m}_uv\right)_{ds}}{m_s } + \frac{v_{tb}\left(v^\dagger\widehat{m}_uv\right)_{bs}}{m_b}\right]\nonumber\\ & & -\frac{4}{v_u^2}\left[-v_{td}\left(v^\dagger\widehat{m}_u^2v\right)_{ds } + v_{tb}\left(v^\dagger \widehat{m}_u^2v\right)_{bs}\right]\nonumber\\ 16\pi^2\frac{d}{dt}v_{tb } & \simeq & \frac{\tan\beta}{m_t}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ \left(v\widehat{m}_dv^\dagger\right)_{tu}v_{ub } + \left(v\widehat{m}_dv^\dagger\right)_{tc}v_{cb}\right]\nonumber\\ & & + \frac{4}{v_d^2}\left[\left(v\widehat{m}_d^2v^\dagger\right)_{tu}v_{ub } + \left(v\widehat{m}_d^2v^\dagger\right)_{tc}v_{cb}\right]\nonumber\\ & & + \frac{1}{m_b\tan\beta}\left(3|v_{tb}|^2h_bh_t+r_q\right)\left [ v_{td}\left(v^\dagger\widehat{m}_uv\right)_{db } + v_{ts}\left(v^\dagger\widehat{m}_uv\right)_{sb}\right]\nonumber\\ & & + \frac{4}{v_u^2}\left [ v_{td}\left(v^\dagger\widehat{m}_u^2v\right)_{db } + v_{ts}\left(v^\dagger \widehat{m}_u^2v\right)_{sb}\right ] \label{eq : rgckm}\end{aligned}\ ] ] we have presented the results for these rges even though they look quite messy because we believe this is the first time such an analysis has been carried out in the susylr model , and these analytical results at the one - loop level may be useful later for future work in this direction . in order to solve these mass and mixing rges numerically , we need to know the initial values for all the 23 variables ( 12 masses , 9 ckm elements and 2 vevs ) . we know the experimental values at @xmath230 for all of them except for the dirac neutrino masses @xmath231 . we fix these values by iterations using the gut - scale predicted values , @xmath232 , which , in turn , are determined completely in terms of the other fermion masses at the gut - scale in @xmath0 gut models . here we note that adjusting the gut - scale values of @xmath231 to fit the @xmath0 model prediction do not change the other fermion masses at this scale significantly even though they are all coupled equations because of the mild running of the neutrino masses . hence the mass and mixing values given in eqs . ( [ eq : gutmass ] ) can be considered as generic and independent of the specific @xmath0 model chosen . we also have the free parameters @xmath233 and @xmath234 corresponding to the couplings @xmath235 and @xmath236 . assuming the couplings @xmath237 to be the same @xmath238 , we have @xmath239\nonumber\end{aligned}\ ] ] further assuming @xmath240 and @xmath241 , we have @xmath242 where @xmath243 and @xmath244 can take values between 0 and 1 ( for the theory to remain perturbative ) . for the running behavior shown in figures 2 , we have chosen @xmath245 and @xmath246 ( requiring @xmath247 unification ) and the initial values of the dirac neutrino masses @xmath248 such that the masses evaluated at the gut - scale , @xmath232 , agree with those predicted from the specific @xmath0 model described in section 6 . for consistency check , we note that the @xmath0 model predicted eigenvalues of @xmath7 given by eq . ( [ eq : mnuso10 ] ) , @xmath249 agree quite well with those obtained from the rges , @xmath250 99 p. minkowski , phys . * b 67 * , 421 ( 1977 ) ; m. gell - mann , p. ramond and r. slansky , _ supergravity _ ( p. van nieuwenhuizen _ et al . _ eds . ) , north holland , amsterdam , 1980 , p. 315 ; t. yanagida , in _ proceedings of the workshop on the unified theory and the baryon number in the universe _ ( o. sawada and a. sugamoto , eds . ) , kek , tsukuba , japan , 1979 , p. 95 ; s. l. glashow , _ the future of elementary particle physics _ , in _ proceedings of the 1979 cargse summer institute on quarks and leptons _ ( m. lvy _ et al . _ eds . ) , plenum press , new york , 1980 , pp . 687 ; r. n. mohapatra and g. senjanovi , phys . lett . * 44 * , 912 ( 1980 ) . g. lazarides , q. shafi and c. wetterich , nucl . * b 181 * , 287 ( 1981 ) ; r. n. mohapatra and g. senjanovi , phys . d * 23 * , 165 ( 1981 ) ; j. schecter and j. w. f. valle , phys . d * 22 * , 2227 ( 1980 ) . f. deppisch , t. s. kosmas and j. w. f. valle , nucl . * b 752 * , 80 ( 2006 ) ; j. garayoa , m. c. gonzalez - garcia and n. rius , jhep * 0702 * , 021 ( 2007 ) ; c. arina , f. bazzocchi , n. fornengo , j. c. romao and j. w. f. valle , phys . rev . lett . * 101 * , 161802 ( 2008 ) ; m. malinsk , t. ohlsson , z .- z . xing and h. zhang , phys . * b 679 * , 242 ( 2009 ) ; e. ma , mod . . lett . * a 24 * , 2491 ( 2009 ) ; m. b. gavela , t. hambye , d. hernandez and p. hernandez , jhep * 0909 * , 038 ( 2009 ) ; m. hirsch , t. kernreiter , j. c. romao and a. villanova del moral , arxiv:0910.2435 [ hep - ph ] . t. fukuyama , a. ilakovac , t. kikuchi and k. matsuda , jhep * 0506 * , 016 ( 2005 ) . g. t hooft in _ proceedings of the 1979 cargse summer institute on recent developments in gauge theories _ ( g. t hooft _ et al . _ eds . ) , plenum press , new york , 1980 . c. r. das and m. k. parida , eur . j. * c 20 * , 121 ( 2001 ) . c. h. albright and s. m. barr , phys . d * 62 * , 093008 ( 2000 ) ; k. s. babu , j. c. pati and f. wilczek , nucl . * b 566 * , 33 ( 2000 ) ; x. ji , y. li and r. n. mohapatra , phys . lett . * b 633 * , 755 ( 2006 ) ; and references therein . r. n. mohapatra and b. sakita , phys . d * 21 * , 1062 ( 1980 ) . s. antusch , j. p. baumann and e. fernndez - martnez , nucl . * b 810 * , 369 ( 2009 ) . a. van der schaaf , j. phys . * g 29 * , 2755 ( 2003 ) . y. kuno ( prime working group ) , nucl . * b 149 * _ ( proc . suppl . ) _ , 376 ( 2005 ) . g. l. fogli , e. lisi , a. marrone , a. palazzo and a. m. rotunno , nucl . phys . * b 188 * _ ( proc . suppl . ) _ , 27 ( 2009 ) . e. fernndez - martnez , m. b. gavela , j. lpez - pavn and o. yasuda , phys . lett . * b 649 * , 427 ( 2007 ) . g. altarelli and d. meloni , nucl . * b 809 * , 158 ( 2009 ) ; s. antusch , m. blennow and e. fernndez - martnez , phys . rev . d * 80 * , 033002 ( 2009 ) .

stuttering bisimulation is a well - known behavioral equivalence that preserves , namely without the next - time operator . correspondingly , the stuttering simulation preorder induces a coarser behavioral equivalence that preserves the existential fragment , namely without the next - time and globally operators . while stuttering bisimulation equivalence can be computed by the well - known groote and vaandrager s [ 1990 ] algorithm , to the best of our knowledge , no algorithm for computing the stuttering simulation preorder and equivalence is available . this paper presents such an algorithm for finite state systems .

0904.1488

[ [ the - problem . ] ] * the problem . * + + + + + + + + + + + + + + lamport s criticism @xcite of the next - time operator @xmath2 in @xmath1/@xmath6 arouse the interest in studying temporal logics like @xmath0/@xmath7 , obtained from @xmath1/@xmath6 by removing the next - time operator , and related notions of behavioral _ stuttering_-based equivalences @xcite . we are interested here in _ divergence blind stuttering _ simulation and bisimulation , that we call , respectively , stuttering simulation and bisimulation for short . we focus here on systems specified as kripke structures ( kss ) , but analogous considerations hold for labeled transition systems ( ltss ) . let @xmath8 be a ks where @xmath9 is a transition system and @xmath10 is a state labeling function . a relation @xmath11 is a stuttering simulation on @xmath12 when for any @xmath13 such that @xmath14 : ( 1 ) @xmath15 and @xmath16 have the same labeling by @xmath10 and ( 2 ) if @xmath17 then @xmath18 for some @xmath19 in such a way that the following diagram holds : = [ ->,>=latex ] ( 6,2 ) node @xmath15 ; ( 7,2 ) node @xmath20 ; ( 8,2.06 ) node @xmath21 ; ( 0,0 ) node @xmath16 ; ( 5,1 ) node @xmath22 ; ( 1,0 ) node @xmath20 ; ( 2,-0.06 ) node @xmath23 ; ( 3,0 ) node @xmath20 ; ( 4,0 ) node @xmath22 ; ( 5,0 ) node @xmath20 ; ( 6,-0.06 ) node @xmath24 ; ( 7,0 ) node @xmath20 ; ( 8,0.03 ) node @xmath19 ; ( 6,1.9 ) node[name=62 ] ( 8,1.9 ) node[name=82 ] ( 0,0.05 ) node[name=00 ] ( 2,0.1 ) node[name=20 ] ( 6,0 ) node[name=60 ] ( 8,0 ) node[name=80 ] ; \(62 ) ( 00 ) ; ( 62 ) ( 20 ) ; ( 62 ) ( 60 ) ; ( 82 ) ( 80 ) ; where a dotted line between two states means that they are related by @xmath25 . the intuition is that @xmath16 is allowed to simulate a transition @xmath26 possibly through some initial `` stuttering '' transitions ( @xmath27-transitions in case of ltss ) . @xmath25 is called a stuttering bisimulation when it is symmetric . it turns out that the largest stuttering simulation @xmath28 and bisimulation @xmath29 relations exist : @xmath28 is a preorder called the _ stuttering simulation preorder _ while @xmath29 is an equivalence relation called the stuttering bisimulation equivalence . moreover , the preorder @xmath28 induces by symmetric reduction the _ stuttering simulation equivalence _ @xmath30 . the partition of @xmath31 corresponding to the equivalence @xmath32 is denoted by @xmath33 . de nicola and vaandrager @xcite showed that for finite kss and for an interpretation of universal / existential path quantifiers over all the , possibly finite , prefixes , the stuttering bisimulation equivalence coincides with the state equivalence induced by the language @xmath0 ( this also holds for @xmath7 ) . this is not true with the standard interpretation of path quantifiers over infinite paths , since this requires a divergence sensitive notion of stuttering ( see the details in @xcite ) . groote and vaandrager @xcite designed a well - known algorithm that computes the stuttering bisimulation equivalence @xmath29 in @xmath34-time and @xmath35-space . clearly , stuttering simulation equivalence is coarser than stuttering bisimulation , i.e.@xmath36 . as far as language preservation is concerned , it turns out that stuttering simulation equivalence coincides with the state equivalence induced by the language @xmath3 , namely the existiential fragment of @xmath1 without next - time and globally operators @xmath2 and @xmath5 . thus , on the one hand , stuttering simulation equivalence still preserves a significantly expressive fragment of @xmath1 and , on the other hand , it may provide a significantly better state space reduction than simulation equivalence , and this has been shown to be useful in abstract model checking @xcite . [ [ state - of - the - art . ] ] * state of the art . * + + + + + + + + + + + + + + + + + + + to the best of our knowledge , there exists no algorithm for computing stuttering simulation equivalence or , more in general , the stuttering simulation preorder . there is instead an algorithm by bulychev et al . @xcite for _ checking _ stuttering simulation , namely , this procedure checks whether a given relation @xmath11 is a stuttering simulation . this algorithm formalizes the problem of checking stuttering simulation as a two players game in a straightforward way and then exploits etessami et al.s @xcite algorithm for solving such a game . the authors claim that this provides an algorithm for checking stuttering simulation on finite kss that runs in @xmath37 time and space . [ [ main - contributions . ] ] * main contributions . * + + + + + + + + + + + + + + + + + + + + + in this paper we present an algorithm for computing simultaneously both the simulation preorder @xmath28 and stuttering simulation equivalence @xmath32 for finite kss . this procedure is incrementally designed in two steps . we first put forward a basic procedure for computing the stuttering simulation preorder that relies directly on the notion of stuttering simulation . for any state @xmath38 , @xmath39 represents the set of states that are candidate to stuttering simulate @xmath40 so that a family of sets @xmath41 is maintained . a pair of states @xmath42 is called a refiner for @xmath43 when @xmath44 and there exists @xmath45 that can not stuttering simulate @xmath40 w.r.t . @xmath46 , i.e. , @xmath47 where @xmath48 is the set of all the states in @xmath49 that may reach a state in @xmath50 through a path of states in @xmath49 . hence , any such @xmath51 can be correctly removed from @xmath49 . actually , it turns out that one such refiner @xmath52 allows to refine @xmath43 to @xmath53 as follows : if @xmath54 then @xmath55 thus , our basic algorithm consists in initializing @xmath41 as @xmath56 and then iteratively refining @xmath43 until a refiner exists . this provides an _ explicit _ stuttering simulation algorithm , meaning that this procedure requires that for any explicit state @xmath38 , @xmath49 is explicitly represented as a set of states . inspired by techniques used in algorithms that compute standard simulation preorders and equivalences ( cf . henzinger et al . @xcite and ranzato and tapparo @xcite ) and in abstract interpretation - based algorithms for computing strongly preserving abstract models @xcite , our stuttering simulation algorithm @xmath57 is obtained by the above basic procedure by exploiting the following two main ideas . * the above explicit algorithm is made `` symbolic '' by representing the family of sets of states @xmath58 as a family of sets of blocks of a partition @xmath59 of the state space @xmath31 . more precisely , we maintain a partition @xmath59 of @xmath31 together with a binary relation @xmath60 a so - called partition - relation pair so that : ( i ) two states @xmath40 and @xmath46 in the same block of @xmath59 are candidate to be stuttering simulation equivalent and ( ii ) if @xmath61 and @xmath62 are two blocks of @xmath59 and @xmath63 then any state in @xmath62 is candidate to stuttering simulate each state in @xmath61 . therefore , here , for any @xmath38 , if @xmath64 is the block of @xmath59 that contains @xmath40 then @xmath65 . * in this setting , a refiner of the current partition - relation @xmath66 is a pair of blocks @xmath67 such that @xmath68 and @xmath69 , where @xmath70 is the existential transition relation between blocks of @xmath59 , i.e. , @xmath68 iff there exist @xmath71 and @xmath72 such that @xmath44 . we devise an efficient way for finding a refiner of the current partition - relation pair that allows us to check whether a given preorder @xmath25 is a stuttering simulation in @xmath73 time and @xmath74 space , where @xmath59 is the partition corresponding to the equivalence @xmath75 . hence , this algorithm for checking stuttering simulation already significantly improves both in time and space bulychev et al.s @xcite procedure . our algorithm @xmath57 iteratively refines the current partition - relation pair @xmath66 by first splitting the partition @xmath59 and then by pruning the relation @xmath76 until a fixpoint is reached . hence , @xmath57 outputs a partition - relation pair where @xmath77 and @xmath46 stuttering simulates @xmath40 iff @xmath78 , where @xmath79 and @xmath80 are the blocks of @xmath59 that contain , respectively , @xmath40 and @xmath46 . as far as complexity is concerned , it turns out that @xmath57 runs in @xmath81 space . it is worth remarking that stuttering simulation yields a rather coarse equivalence so that @xmath82 should be in general much less than the size @xmath83 of the concrete state space . [ [ notation . ] ] * notation . * + + + + + + + + + + + if @xmath11 is any relation and @xmath38 then @xmath84 . let us recall that @xmath25 is called a preorder when it is reflexive and transitive . if @xmath85 is a function defined on @xmath86 and @xmath38 then we often write @xmath87 to mean @xmath88 . a partition @xmath59 of a set @xmath31 is a set of nonempty subsets of @xmath31 , called blocks , that are pairwise disjoint and whose union gives @xmath31 . @xmath89 denotes the set of partitions of @xmath31 . if @xmath90 and @xmath91 then @xmath92 denotes the block of @xmath59 that contains @xmath15 . @xmath89 is endowed with the following standard partial order @xmath93 : @xmath94 , i.e. @xmath95 is coarser than @xmath96 , iff @xmath97 . for a given nonempty subset @xmath98 called splitter , we denote by @xmath99 the partition obtained from @xmath59 by replacing each block @xmath100 with the nonempty sets @xmath101 and @xmath102 , where we also allow no splitting , namely @xmath103 ( this happens exactly when @xmath104 is a union of some blocks of @xmath59 ) . if @xmath105 then we denote by @xmath106 ( or simply by @xmath107 ) the unique block in @xmath59 that contains @xmath61 ( this may possibly be @xmath61 itself ) . + a transition system @xmath108 consists of a set @xmath31 of states and a transition relation @xmath109 . the predecessor transformer @xmath110 is defined as usual : @xmath111 . if @xmath112 then @xmath113 iff there exist @xmath114 and @xmath115 such that @xmath116 . given a set @xmath117 of atomic propositions ( of some specification language ) , a kripke structure ( ks ) @xmath118 over @xmath117 consists of a transition system @xmath108 together with a state labeling function @xmath119 . @xmath120 denotes the state partition induced by @xmath10 , namely , @xmath121 . [ [ stuttering - simulation . ] ] * stuttering simulation . * + + + + + + + + + + + + + + + + + + + + + + + + let @xmath118 be a ks . a relation @xmath122 is a _ divergence blind stuttering simulation _ on @xmath12 if for any @xmath13 such that @xmath14 : * @xmath123 ; * if @xmath17 then there exist @xmath124 , with @xmath125 , such that : ( i ) @xmath126 ; ( ii ) for all @xmath127 , @xmath128 and @xmath129 ; ( iii ) @xmath130 . observe that condition ( 2 ) allows the case @xmath131 and this boils down to requiring that @xmath132 . with a slight abuse of terminology , @xmath25 is called simply a _ stuttering simulation_. if @xmath14 then we say that @xmath16 stuttering simulates @xmath15 and we denote this by @xmath133 . if @xmath25 is a symmetric relation then it is called a stuttering bisimulation . the empty relation is a stuttering simulation and stuttering simulations are closed under union so that the largest stuttering simulation relation exists . it turns out that the largest simulation is a preorder relation called _ stuttering simulation preorder _ ( on @xmath12 ) and denoted by @xmath28 . thus , for any @xmath134 , @xmath135 iff @xmath136 . _ stuttering simulation equivalence _ @xmath32 is the symmetric reduction of @xmath28 , namely @xmath137 , so that @xmath138 iff @xmath139 and @xmath140 . @xmath141 denotes the partition corresponding to the equivalence @xmath32 and is called stuttering simulation partition . + following groote and vaandrager @xcite , @xmath142 is defined as : @xmath143\end{gathered}\ ] ] so that a relation @xmath144 is a stuttering simulation iff for any @xmath145 , @xmath146 and if @xmath44 then @xmath147 . + it turns out @xcite that @xmath33 is the coarsest partition preserved by the temporal language @xmath3 . more precisely , @xmath3 is inductively defined as follows : @xmath148 and its semantics is standard : @xmath149 and @xmath150 . the coarsest partition preserved by @xmath3 is the state partition corresponding to the following equivalence @xmath151 between states : for any @xmath134 , @xmath152 for each state @xmath38 , the algorithm @xmath153 in figure [ algouno ] computes the stuttering simulator set @xmath39 , i.e. , the set of states that stuttering simulate @xmath40 . the basic idea is that @xmath49 contains states that are candidate for stuttering simulating @xmath40 . thus , the input partition of @xmath153 is taken as the partition @xmath154 determined by the labeling @xmath10 so that @xmath49 is initialized with @xmath155 , i.e. , with all the states that have the same labeling of @xmath40 . following the definition of stuttering simulation , a refiner is a pair of states @xmath52 such that @xmath156 and @xmath157 . in fact , if @xmath158 then @xmath51 can not stuttering simulate @xmath40 and therefore can be correctly removed from @xmath49 . conversely , if no such refiner exists then for any @xmath145 such that @xmath44 we have that @xmath159 so that any @xmath45 actually stuttering simulates @xmath40 . hence , @xmath153 consists in iteratively refining @xmath41 as long as a refiner exists , where , given a refiner @xmath52 , the refinement of @xmath43 by means of @xmath54 is as follows : @xmath160 it turns out that this procedure correctly computes the stuttering simulation preorder . [ basicssa ] @xmath153 is correct , i.e. , if @xmath43 is the output of @xmath153 on input @xmath154 then for any @xmath145 , @xmath161 . a _ partition - relation pair _ @xmath66 , pr for short , is given by a partition @xmath90 together with a binary relation @xmath162 between blocks of @xmath59 . we write @xmath163 when @xmath63 and @xmath164 and @xmath165 when @xmath166 and @xmath167 . our stuttering simulation algorithm relies on the idea of symbolizing the @xmath153 procedure in order to maintain a pr @xmath66 in place of the family of explicit sets of states @xmath168 . as a first step , @xmath169 induces a partition @xmath59 that corresponds to the following equivalence @xmath170 : @xmath171 hence , the intuition is that if @xmath172 then @xmath173 and @xmath174 are `` currently '' candidates to be stuttering simulation equivalent . accordingly , a relation @xmath76 on @xmath59 encodes stuttering simulation as follows : if @xmath91 then @xmath175 . here , the intuition is that if @xmath63 then any state @xmath176 is `` currently '' candidate to stuttering simulate any state @xmath177 . equivalently , the following invariant property is maintained : if @xmath133 then @xmath178 . thus , a pr @xmath66 will represent the current approximation of the stuttering simulation preorder and in particular @xmath59 will represent the current approximation of stuttering simulation equivalence . more precisely , a pr @xmath179 induces the following map @xmath180 : for any @xmath181 , @xmath182 note that , for any @xmath91 , @xmath183 , that is , @xmath184 represents the set of states that are currently candidates to stuttering simulate @xmath15 . a pr @xmath185 is therefore defined to be a stuttering simulation for a ks @xmath12 when the relation @xmath186 is a stuttering simulation on @xmath12 . recall that in @xmath153 a pair of states @xmath187 is a refiner for @xmath43 when @xmath188 and @xmath189 . accordingly , a pair of blocks @xmath67 is called a refiner for @xmath190 when @xmath68 and @xmath191 . thus , by defining @xmath192 the following characterization holds : [ st - char2 ] @xmath193 is a stuttering simulation iff @xmath194 and for any @xmath91 , @xmath195 . the algorithm @xmath153 is therefore made symbolic as follows : * @xmath196 is the input pr , where @xmath197 ; * find @xmath198 ; if @xmath194 exit ; * compute @xmath199 ; * @xmath200 , where @xmath201 and @xmath202 is modified in such a way that for any @xmath91 , @xmath203 ; * @xmath204 , where @xmath202 is modified to @xmath205 in such a way that for any @xmath206 : @xmath207 * @xmath208 and go to ( 2 ) . @xmath209 this leads to the symbolic algorithm @xmath57 described in figure [ fig - ssa ] , where : the input pr @xmath210 at line 1 is @xmath196 of point ( 1 ) ; point ( 2 ) corresponds to the call @xmath211 at line 3 ; point ( 3 ) corresponds to lines 4 - 5 ; point ( 4 ) corresponds to the call @xmath212 at line 6 ; point ( 5 ) corresponds to the call @xmath213 at line 7 . the following graphical example shows how points ( 4 ) and ( 5 ) refine a pr @xmath214,[2,3],[4,5],[6,7],[8,9]\},{\trianglelefteq}}$ ] w.r.t . the set @xmath215 , where if @xmath216 then @xmath61 is drawed below @xmath62 while if @xmath163 and @xmath217 then @xmath61 and @xmath62 are at same height and connected by a double line . = [ ->,>=latex ] ( 0,0 ) node[shape = rectangle , draw , name=01]0 1 ( 0,1.5 ) node[shape = rectangle , draw , name=23]2 3 ( 1.5,0 ) node[shape = rectangle , draw , name=45]4 5 ( 1.5,1.5 ) node[shape = rectangle , draw , name=67]6 7 ( 3,0 ) node[shape = rectangle , draw , name=89]8 9 ( 0.45,2.4 ) node[]@xmath104 ; \(01 ) ( 23 ) ; ( 01 ) ( 67 ) ; ( 45 ) ( 67 ) ; ( 89 ) ( 67 ) ; ( 4.2,1.5 ) node[]@xmath218 ; ( 6.1,0 ) node[shape = rectangle , draw , name=01m]0 1 ( 5.55,1.5 ) node[shape = rectangle , draw , name=2m]2 ( 6.65,1.5 ) node[shape = rectangle , draw , name=3m]3 ( 10.65,0 ) node[shape = rectangle , draw , name=9m]9 ( 9.55,0 ) node[shape = rectangle , draw , name=8m]8 ( 8.1,1.5 ) node[shape = rectangle , draw , name=67m]6 7 ( 8.1,0 ) node[shape = rectangle , draw , name=45m]4 5 ; ( 01 m ) ( 2 m ) ; ( 01 m ) ( 3 m ) ; ( 2 m ) to ( 3 m ) ; ( 01 m ) ( 67 m ) ; ( 45 m ) ( 67 m ) ; ( 8 m ) ( 67 m ) ; ( 9 m ) ( 67 m ) ; ( 8 m ) ( 9 m ) ; ( 11.5,1.5 ) node[]@xmath219 ; ( 13,0 ) node[shape = rectangle , draw , name=01n]0 1 ( 13,1.5 ) node[shape = rectangle , draw , name=2n]2 ( 13,3 ) node[shape = rectangle , draw , name=3n]3 ( 15.75,0 ) node[shape = rectangle , draw , name=9n]9 ( 15.75,1.5 ) node[shape = rectangle , draw , name=8n]8 ( 14.5,3 ) node[shape = rectangle , draw , name=67n]6 7 ( 14.5,0 ) node[shape = rectangle , draw , name=45n]4 5 ; ( 01n ) ( 2n ) ; ( 01n ) ( 67n ) ; ( 2n ) ( 3n ) ; ( 9n ) ( 8n ) ; ( 8n ) ( 67n ) ; ( 45n ) ( 67n ) ; ( 0.1,0.5 ) ( 0.1,2 ) ( 0.7,2 ) ( 0.7,0.73 ) ( 3,0.73 ) ( 3,-0.6 ) ( 0.8,-0.6 ) ( 0.8,0.5 ) ( 0.1,0.5 ) ; the correctness of this symbolic algorithm goes as follows . [ correctness ] @xmath57 is a correct implementation of @xmath153 , i.e. , if @xmath43 is the output function of @xmath153 on input @xmath154 and @xmath220 is the output pr of @xmath57 on input @xmath196 then for any @xmath38 , @xmath221 . the next step consists in devising an efficient implementation of @xmath57 . while it is not too hard to devise an efficient implementation of lines 2 and 4 - 7 of the @xmath57 algorithm , it is instead not straightforward to find a refiner in an efficient way . in groote and vaandrager s @xcite algorithm for computing stuttering bisimulations the key point for efficiently finding a refiner in their setting is the notion of _ bottom state_. given a set of states @xmath98 , a bottom state of @xmath104 is a state @xmath222 that can not go inside @xmath104 , i.e. , @xmath15 can only go outside @xmath104 ( note that @xmath15 may also have no outgoing transition ) . for any @xmath223 , we therefore define : @xmath224 bottom states allow to efficiently find refiners in kss that do not contain cycles of states all having the same labeling . following groote and vaandrager @xcite , a transition @xmath188 is called _ inert _ for a partition @xmath90 when @xmath225 . clearly , if a set of states @xmath104 in a ks @xmath12 is strongly connected via inert transitions for the labeling partition @xmath154 then all the states in @xmath104 are stuttering simulation equivalent , i.e. , if @xmath226 then @xmath227 . thus , each strongly connected component ( s.c.c . ) @xmath104 with respect to inert transitions for @xmath154 , called _ inert s.c.c . _ , can be collapsed to one single `` symbolic state '' . in particular , if @xmath228 is one such inert s.c.c . , i.e. @xmath229 , then this collapse simply removes the transition @xmath229 . it is important to remark that a standard depth - first search algorithm by tarjan @xcite , running in @xmath230 time , allows us to find and then collapse all the inert s.c.c.s in the input ks . we can thus assume w.l.o.g . that the ks @xmath12 does not contain inert s.c.c.s . the following characterization of refiners therefore holds . [ l2 ] assume that @xmath12 does not contain inert s.c.c.s . let @xmath179 be a pr such that for any @xmath100 , @xmath231 . consider @xmath67 such that @xmath68 . then , @xmath198 iff @xmath232 . if @xmath100 is any block then we define as _ local bottom states _ of @xmath61 all the bottom states of @xmath233 that belong to @xmath61 , namely @xmath234 also , we define @xmath235 as a _ bottom block _ for @xmath61 when @xmath62 contains at least a bottom state of @xmath233 and @xmath216 , that is : @xmath236 local bottoms and bottom blocks characterize refiners for stuttering simulation as follows : [ th - refiner ] assume that @xmath12 does not contain inert s.c.c.s . let @xmath179 be a pr such that @xmath76 is a preorder and for any @xmath100 , @xmath231 . consider @xmath67 such that @xmath68 and for any @xmath237 such that @xmath238 and @xmath239 , @xmath240 . then , @xmath198 iff at least one of the following two conditions holds : 1 . @xmath241 and @xmath242 ; 2 . there exists @xmath243 such that @xmath244 and @xmath245 . we will show that this characterization provides the basis for an algorithm that efficiently finds refiners . hence , this procedure also checks whether a given preorder @xmath25 is a stuttering simulation . this can be done in @xmath73 time and @xmath246 space , where @xmath59 is the partition corresponding to the equivalence @xmath75 . thus , this algorithm for checking stuttering simulation already significantly improves bulychev et al.s @xcite procedure that runs in @xmath37 time and space . @xmath57 is implemented by exploiting the following data structures . * a state @xmath15 is represented by a record that contains the list @xmath15.pre of its predecessors @xmath247 and a pointer @xmath15.block to the block @xmath92 that contains @xmath15 . the state space @xmath31 is represented as a doubly linked list of states . * the states of any block @xmath61 of the current partition @xmath59 are consecutive in the list @xmath31 , so that @xmath61 is represented by two pointers begin and end : the first state of @xmath61 in @xmath31 and the successor of the last state of @xmath61 in @xmath31 , i.e. , @xmath248 . moreover , @xmath61 contains a pointer @xmath61.intersection to a block whose meaning is as follows : after a call to @xmath249 for splitting @xmath59 w.r.t . a set of states @xmath104 , if @xmath250 then @xmath61.intersection points to a block that represents @xmath101 , otherwise @xmath61.intersection @xmath251 . finally , the fields localbottoms and bottomblocks for a block @xmath61 represent , resp . , the local bottom states of @xmath61 and the bottom blocks of @xmath61 . the current partition @xmath59 is stored as a doubly linked list of blocks . * the current relation @xmath76 on @xmath59 is stored as a resizable @xmath252 boolean matrix @xmath253 : @xmath254 iff @xmath63 . recall ( * ? ? ? * section 17.4 ) that insert operations in a resizable array ( whose capacity is doubled as needed ) take amortized constant time , and a resizable matrix ( or table ) can be implemented as a resizable array of resizable arrays . the boolean matrix @xmath253 is resized by adding a new entry to @xmath253 , namely a new row and a new column , for any block @xmath61 that is split into two new blocks @xmath255 and @xmath101 . * @xmath57 additionally stores and maintains a resizable integer table count and a resizable integer matrix bcount . count is indexed over @xmath31 and @xmath59 and has the following meaning : @xmath256 while bcount is indexed over @xmath257 and has the following meaning : @xmath258 . the table count allows to implement the test @xmath259 in constant time as @xmath260 , while bcount allows to implement in constant time the test @xmath261 as @xmath262 . the list @xmath59 is stored in reverse topological ordering wrt @xmath253 @xmath263 refiner ( , ) the algorithm @xmath211 in figure [ fig - refiner ] is an implementation of the characterization of refiners provided by theorem [ th - refiner ] . in particular , lines 8 - 10 implement condition ( i ) of theorem [ th - refiner ] and lines 11 - 12 implement condition ( ii ) . the correctness of this implementation depends on the following key point . given a pair of blocks @xmath67 such that @xmath68 , in order to ensure the equivalence : @xmath198 iff @xmath265 , theorem [ th - refiner ] requires as hypothesis the following condition : @xmath266 in order to ensure this condition @xmath267 , we guarantee throughout the execution of @xmath57 that the list @xmath59 of blocks is stored in reverse topological ordering w.r.t . @xmath76 , so that if @xmath268 then @xmath269 precedes @xmath61 in the list @xmath59 . the reverse topological ordering of @xmath59 initially holds because the input pr is the dag @xmath196 which is trivially topologically ordered ( whatever the ordering of @xmath154 is ) . more in general , for a generic input pr @xmath210 to @xmath57 the function @xmath270 in figure [ fig-2 ] in appendix [ app ] achieves this reverse topological ordering by a standard algorithm ( * ? ? ? * section 22.4 ) that runs in @xmath271 time ( cf.the call @xmath272 in the @xmath270 function ) . then , the reverse topological ordering of @xmath59 is always maintained throughout the execution of @xmath57 . in fact , if the partition @xmath59 is split w.r.t . a set @xmath104 and a block @xmath61 generates two new descendant blocks @xmath101 and @xmath102 then our @xmath273 in figure [ fig - sp ] modifies the ordering of the list @xmath59 as follows : @xmath61 is replaced in @xmath59 by inserting @xmath101 immediately followed by @xmath274 . this guarantees that at the exit of @xmath213 at line 7 of @xmath57 the list @xmath59 is still in reverse topological ordering w.r.t.@xmath253 . this is a consequence of the fact that at the exit of @xmath213 , by point ( 5 ) in section [ asa ] , we have that @xmath275 , i.e. , @xmath276 so that @xmath277 . the reverse topological ordering of @xmath59 w.r.t . @xmath76 ensures that if @xmath278 then @xmath279 is scanned by @xmath264 after the pair @xmath280 . since @xmath211 exits as soon as a refiner is found , we have that @xmath280 can not be a refiner , so that condition @xmath267 holds for @xmath279 . + when @xmath211 determines that a pair of blocks @xmath279 , with @xmath68 , is not a refiner , it stores this information in a local boolean matrix refiner that is indexed over @xmath257 and initialized to @xmath281 . thus , the meaning of the matrix refiner is as follows : if @xmath282 then @xmath283 . if @xmath284 then both ( i ) and ( ii ) do not hold , therefore @xmath211 executes the for - loop at lines 13 - 14 so that any @xmath285 with @xmath286 is marked as @xmath287 . this is correct because if @xmath283 and @xmath288 then @xmath289 : in fact , by lemma [ l2 ] , @xmath290 , and since @xmath286 implies , because @xmath76 is transitive , @xmath291 , we have that @xmath292 , so that , by lemma [ l2 ] , @xmath289 . the for - loop at lines 13 - 14 is therefore an optimization of theorem [ th - refiner ] since it determines that some pairs of blocks are not a refiner without resorting to the condition @xmath293 of theorem [ th - refiner ] . this optimization and the related matrix refiner turn out to be crucial for obtaining the overall time complexity of @xmath57 . @xmath294 @xmath295 ( @xmath104 , @xmath296 ) \ { @xmath297 @xmath25 given two lists of states @xmath104 and @xmath296 , we want to compute the set of states that belong to @xmath298 . this can be done by traversing once the edges of the transition relation @xmath20 provided that the list @xmath31 of states satisfies the following property : for all @xmath145 , if @xmath40 precedes @xmath46 in the list @xmath31 and @xmath299 then @xmath300 . we denote this property by @xmath301 . hence , this is a topological ordering of @xmath31 w.r.t . the transition relation @xmath20 that is local to each block of the labeling partition @xmath154 . as described in section [ sec : bs ] , as an initial pre - processing step of @xmath57 , we find and collapse inert s.s.c.s . after this pre - processing step , @xmath31 is then topologically ordered locally to each block of @xmath154 in @xmath302 time in order to establish initially @xmath301 . we will see in section [ sec - sp ] that while the ordering of the list @xmath31 of states changes across the execution of @xmath57 , the property @xmath301 is always maintained invariant . the computation of @xmath298 is done by the algorithm in figure [ fig - eu ] . the result @xmath25 consists of all the states in @xmath104 that are marked2 . we assume that all the states in @xmath104 have the same labeling by @xmath10 : this is clearly true when the function @xmath295 is called from the algorithm @xmath57 . the for - loop at lines 5 - 7 makes the states in @xmath303 marked2 . then , the for - loop at lines 8 - 10 scans backward the list of states @xmath104 and when a marked2 state @xmath46 is encountered then all the states in @xmath304 are marked2 . it is clear that the property @xmath301 guarantees that this procedure does not miss states that are in @xmath298 . split split split @xmath305 @xmath57 calls @xmath212 at line 6 with the precondition @xmath301 and needs to maintain this invariant property at the exit ( as discussed in section [ euc ] this is crucial for computing @xmath295 ) . this function must modify the current pr @xmath220 to @xmath306 as follows : * @xmath307 is the partition obtained by splitting @xmath59 w.r.t . the splitter @xmath104 ; * @xmath253 is modified to @xmath308 in such a way that for any @xmath38 , @xmath309 . recall that the states of a block @xmath61 of @xmath59 are consecutive in the list @xmath31 , so that @xmath61 is represented as @xmath248 . an implementation of the splitting operation @xmath310 that only scans the states in @xmath104 , i.e. that takes @xmath311 time , is quite easy and standard ( see e.g.@xcite ) . however , this operation affects the ordering of the states in the list @xmath31 because states are moved from old blocks to newly generated blocks . it turns out that this splitting operation can be implemented in a careful way that preserves the invariant property @xmath312 . the idea is rather simple . observe that the list of states @xmath313 can be ( and actually is ) built as a sublist of @xmath31 so that the following property holds : if @xmath40 precedes @xmath46 in @xmath104 and @xmath314 then @xmath315 . the following picture shows the idea of our implementation of @xmath310 , where states within filled circles determine the splitter set @xmath104 . = [ ->,>=latex ] ( -0.75,2 ) node @xmath31 ; ( -0.75,0 ) node @xmath316 ; ( 0.5,2.7 ) node @xmath317 ; ( 4.5,2.7 ) node @xmath318 ; ( 8.5,2.7 ) node @xmath319 ; ( 4.5,1 ) node @xmath310 ; ( -0.1,-0.7 ) node @xmath320 ; ( 1,-0.7 ) node @xmath321 ; ( 3,-0.7 ) node @xmath322 ; ( 6,-0.7 ) node @xmath323 ; ( 8,-0.7 ) node @xmath324 ; ( 9.1,-0.7 ) node @xmath325 ; ( 0,2 ) node[shape = circle , draw , name=0]0 ( 1,2 ) node[shape = circle , fill = blue!25,draw , name=1]1 ( 2,2 ) node[shape = circle , draw , name=2]2 ( 3,2 ) node[shape = circle , fill = blue!25,draw , name=3]3 ( 4,2 ) node[shape = circle , fill = blue!25,draw , name=4]4 ( 5,2 ) node[shape = circle , draw , name=5]5 ( 6,2 ) node[shape = circle , fill = blue!25,draw , name=6]6 ( 7,2 ) node[shape = circle , draw , name=7]7 ( 8,2 ) node[shape = circle , fill = blue!25,draw , name=8]8 ( 9,2 ) node[shape = circle , draw , name=9]9 ; ( 0,0 ) node[shape = circle , fill = blue!25,draw , name=00]1 ( 1,0 ) node[shape = circle , draw , name=11]0 ( 2,0 ) node[shape = circle , fill = blue!25,draw , name=22]3 ( 3,0 ) node[shape = circle , fill = blue!25,draw , name=33]4 ( 4,0 ) node[shape = circle , fill = blue!25,draw , name=44]6 ( 5,0 ) node[shape = circle , draw , name=55]2 ( 6,0 ) node[shape = circle , draw , name=66]5 ( 7,0 ) node[shape = circle , draw , name=77]7 ( 8,0 ) node[shape = circle , fill = blue!25,draw , name=88]8 ( 9,0 ) node[shape = circle , draw , name=99]9 ; ( 0.north west ) + + ( -0.18,0.18 ) node[name = b1a ] ( 1.south east ) + + ( 0.18,-0.18 ) node[name = b1b ] ; ( 2.north west ) + + ( -0.18,0.18 ) node[name = b2a ] ( 7.south east ) + + ( 0.18,-0.18 ) node[name = b2b ] ; ( 8.north west ) + + ( -0.18,0.18 ) node[name = b3a ] ( 9.south east ) + + ( 0.15,-0.15 ) node[name = b3b ] ; ( b1a ) rectangle ( b1b ) ; ( b2a ) rectangle ( b2b ) ; ( b3a ) rectangle ( b3b ) ; ( 00.north west ) + + ( -0.18,0.18 ) node[name = c1a ] ( 00.south east ) + + ( 0.18,-0.18 ) node[name = c1b ] ; ( 11.north west ) + + ( -0.18,0.18 ) node[name = c2a ] ( 11.south east ) + + ( 0.18,-0.18 ) node[name = c2b ] ; ( 22.north west ) + + ( -0.18,0.18 ) node[name = c3a ] ( 44.south east ) + + ( 0.18,-0.18 ) node[name = c3b ] ; ( 55.north west ) + + ( -0.18,0.18 ) node[name = c4a ] ( 77.south east ) + + ( 0.18,-0.18 ) node[name = c4b ] ; ( 88.north west ) + + ( -0.18,0.18 ) node[name = c5a ] ( 88.south east ) + + ( 0.18,-0.18 ) node[name = c5b ] ; ( 99.north west ) + + ( -0.18,0.18 ) node[name = c6a ] ( 99.south east ) + + ( 0.18,-0.18 ) node[name = c6b ] ; ( c1a ) rectangle ( c1b ) ; ( c2a ) rectangle ( c2b ) ; ( c3a ) rectangle ( c3b ) ; ( c4a ) rectangle ( c4b ) ; ( c5a ) rectangle ( c5b ) ; ( c6a ) rectangle ( c6b ) ; the property @xmath326 still holds for the modified list of states @xmath316 . in fact , from the above picture observe that it is enough to check that : if @xmath61 has been split into @xmath101 and @xmath102 by preserving the relative orders of the states in @xmath31 then if @xmath327 and @xmath328 then @xmath315 . this is true because if @xmath329 and @xmath330 then , since @xmath40 and @xmath46 are in the same block of @xmath59 and @xmath331 is a union of some blocks of @xmath59 , by definition of @xmath295 we would also have that @xmath332 , which is a contradiction . the functions in figure [ fig - sp ] sketch a pseudo - code that implements the above described splitting operation ( the @xmath333 function is in figure [ fig - bis ] in appendix [ app ] ) . the above point ( b ) , i.e. , the modification of @xmath253 to @xmath308 so that for any @xmath38 , @xmath309 is straightforward and is implemented at lines 18 - 20 of @xmath334 . @xmath335 } @xmath57 calls @xmath213 at line 7 with the precondition that @xmath104 is a union of blocks of the current partition @xmath59 . the function @xmath213 in figure [ fig - ref ] implements the point ( 5 ) of section [ asa ] . this function must modify the current pr @xmath337 to @xmath338 by pruning the relation @xmath253 in such a way that for any @xmath100 : @xmath339 this is done by the @xmath340 function at lines 5 - 7 by reducing the relation @xmath253 to @xmath308 as follows : if @xmath341 and @xmath342 then @xmath343 iff @xmath344 and @xmath345 , while the rest of the code updates the data structures count , bcount and bottomblocks accordingly ( note that localbottoms do not need to be updated ) . it is straightforward to implement the remaining functions @xmath270 and @xmath346 ( these are given in figure [ fig-2 ] in appendix [ app ] ) . it is just worth remarking that in @xmath270 , @xmath347 establishes initially the property @xmath301 , while the call @xmath272 provides an initial reverse topological order of @xmath59 w.r.t . @xmath253 when the input partial ordering @xmath253 is not the identity relation @xmath348 . time and space bounds for @xmath57 are as follows . in the following statement we assume , as usual in model checking , that the transition relation @xmath20 is total , i.e. , for any @xmath91 there exists @xmath349 such that @xmath188 , so that the inequalities @xmath350 and @xmath351 hold and this allows us to simplify the expression of the time bound . [ th - complexity ] @xmath57 runs in @xmath352-time and @xmath353-space . the algorithms @xmath57 computes the stuttering simulation preorder on kss , but it can be modified to work over ltss by following the adaptation to ltss of groote and vaandrager s algorithm @xcite for kss . due to lack of space the details are here omitted . we just mention that for any action @xmath354 , we have a parametric @xmath355 operator for any action @xmath354 so that the notions of splitting and refinement of the current pr are parameterized w.r.t . the action @xmath356 . we presented an algorithm , called @xmath57 , for computing the stuttering simulation preorder and equivalence on a kripke structure or labeled transition system . to the best of our knowledge , this is the first algorithm for computing this behavioural preorder . the only available algorithm related to stuttering simulation is a procedure by bulychev et al . @xcite that checks whether a given relation is a stuttering simulation . our procedure @xmath57 includes an algorithm for checking whether a given relation is a stuttering simulation that significantly improves bulychev et al.s one both in time and in space . [ [ acknowledgements . ] ] * acknowledgements . * + + + + + + + + + + + + + + + + + + + this work was partially supported by the prin 2007 project `` _ _ aida2007 : abstract interpretation design and applications _ _ '' and by the university of padova under the projects `` _ _ formal methods for specifying and verifying behavioural properties of software systems _ _ '' and `` _ _ analysis , verification and abstract interpretation of models for concurrency _ _ '' . 99 m.c . browne , e.m . clarke and o. grumberg . characterizing finite kripke structures in propositional temporal logic . _ theor . comp . _ , 59:115 - 131 , 1988 . bulychev , i.v . konnov and v.a . computing ( bi)simulation relations preserving @xmath7 for ordinary and fair kripke structures . _ mathemathical methods and algorithms _ , institute for system programming , russian academy of sciences , vol . 12 , 2007 . available from http://lvk.cs.msu.su/@xmath151peterbul . cormen , c.e . leiserson , r.l . rivest and c. stein . _ introduction to algorithms_. the mit press and mcgraw - hill , 2nd ed . , 2001 . r. de nicola and f. vaandrager . three logics for branching bisimulation . _ j. acm _ , 42(2):458487 , 1995 k. etessami , t. wilke , r.a . schuller . fair simulation relations , parity games , and state space reduction for buchi automata . _ siam j. comput . _ , 34(5):1159 - 1175 , 2001 . groote and f. vaandrager . an efficient algorithm for branching bisimulation and stuttering equivalence . in _ proc . 17@xmath357 icalp _ , lncs 443 , pp . 626 - 638 , springer , 1990 . henzinger , t.a . henzinger and p.w . computing simulations on finite and infinite graphs . in _ proc . 36@xmath357 focs _ , 453 - 462 , 1995 . l. lamport . what good is temporal logic ? in _ information processing 83 _ , pp . 657 - 668 , ifip , 1983 . p. manolios . _ mechanical verification of reactive systems_. phd thesis , university of texas at austin , 2001 . s. nejati , a. gurfinkel and m. chechik . stuttering abstraction for model checking . in _ 3@xmath358 ieee int . conf . on software engineering and formal methods ( sefm05 ) _ , pp . 311 - 320 , 2005 f. ranzato and f. tapparo . a new efficient simulation equivalence algorithm . in _ proc 22@xmath359 ieee symp . on logic in computer science ( lics07 ) _ , pp . 171180 , ieee press , 2007 . f. ranzato and f. tapparo . generalized strong preservation by abstract interpretation . _ j. logic and computation _ , 17(1):157 - 197 , 2007 . [ sitrans ] at the beginning of any iteration of @xmath153 , @xmath43 is a preorder . initially , @xmath43 is reflexive and transitive because @xmath360 is a partition . let us denote by @xmath361 the value of @xmath43 at the beginning of the @xmath362-th iteration of @xmath153 . then , @xmath363 then , by inductive hypothesis , @xmath43 is clearly reflexive . let us turn on transitivity . consider @xmath364 and @xmath365 . then , by inductive hypothesis , @xmath366 . if @xmath367 then @xmath368 and therefore @xmath369 . if , instead , @xmath370 then @xmath371 and therefore @xmath332 . hence , @xmath372 so that @xmath373 , i.e. @xmath374 . _ proof of lemma [ basicssa ] . _ the output relation @xmath43 is a stuttering simulation so that @xmath375 . thus , we need to prove that @xmath376 . let us denote by @xmath361 the value of @xmath43 at the beginning of the @xmath362-th iteration of @xmath153 . we show by induction on @xmath362 that @xmath377 . if @xmath387 then @xmath388 . if , instead , @xmath389 then @xmath390 . by inductive hypothesis , @xmath391 , therefore it is enough to show that @xmath392 . consider @xmath393 . since @xmath394 , there exists a path @xmath395 such that for any @xmath396 , @xmath397 and @xmath398 . it turns out that any transition @xmath399 can be lifted to a path @xmath400 where @xmath401 when @xmath402 and @xmath403 , and in particular @xmath404 . in fact , consider the first transition @xmath405 . since @xmath406 , there exists @xmath407 such that @xmath408 where @xmath409 for any @xmath410 and @xmath411 . thus , by a simple induction , any transition @xmath399 can be lifted to one such path . moreover , by induction , for any @xmath396 , @xmath412 , while @xmath413 . by lemma [ sitrans ] , @xmath361 is transitive so that from @xmath414 and @xmath415 we obtain that for any @xmath396 , @xmath416 and @xmath417 . the concatenation of the above paths therefore provides a path @xmath418 such that for any @xmath419 , @xmath420 and @xmath421 . consequently , @xmath422 and this concludes the proof . _ proof of theorem [ st - char2 ] . _ @xmath423 if @xmath424 is a stuttering simulation and @xmath425 then @xmath426 , i.e. , @xmath427 . moreover , if @xmath68 then there exists @xmath177 and @xmath428 such that @xmath26 , so that @xmath429 . since @xmath430 and @xmath431 , we have that @xmath432 . hence , @xmath433 . @xmath434 assume that @xmath26 and @xmath425 . therefore , @xmath427 , i.e. , @xmath426 . furthermore , @xmath435 , so that from @xmath433 we obtain that @xmath436 . since @xmath437 and @xmath438 , we have that @xmath439 , and therefore @xmath424 is a stuttering simulation . _ proof of theorem [ correctness ] . _ this is a consequence of the following two facts . let @xmath43 be the current relation in @xmath153 at the end of some iteration and let @xmath185 be the corresponding pr . * we have that @xmath440 is a refiner in @xmath153 iff @xmath441 is a refiner in @xmath57 . this is true because for any @xmath145 , we have that @xmath442 iff @xmath443 . * let @xmath440 be a refiner in @xmath153 and @xmath444 . let @xmath445 . consider @xmath446 @xmath447 where @xmath38 and @xmath206 . then , for any @xmath38 , @xmath448 . _ proof of lemma [ l2 ] . _ let @xmath449 and @xmath450 such that @xmath68 . + assume that @xmath451 and consider @xmath452 . then , @xmath453 , so that there exist @xmath454 , with @xmath125 , such that @xmath455 , for all @xmath127 , @xmath456 and @xmath457 , and @xmath458 . since @xmath459 , we have that @xmath460 and therefore necessarily either @xmath131 or @xmath461 . if @xmath131 then @xmath462 . if instead @xmath461 then @xmath463 . thus , @xmath464 . + conversely , assume that @xmath465 and consider @xmath466 . if @xmath467 then clearly @xmath468 . if instead @xmath469 then @xmath470 , so that there exists @xmath471 such that @xmath44 . again , if @xmath472 then @xmath473 and therefore we have that @xmath474 . if @xmath475 then we can go on with this construction . since @xmath31 is finite , in this way we would obtain a cycle of inert transitions inside @xmath476 , namely a contradiction . thus , it must exist some @xmath477 such that @xmath478 , and therefore @xmath474 . hence , @xmath451 . _ proof of theorem [ th - refiner ] . _ let us first observe that since @xmath479 is a preorder , and therefore transitive , if @xmath480 then @xmath481 . + @xmath423 let us assume that @xmath482 . if @xmath483 then both conditions ( i ) and ( ii ) trivially do not hold : for ( ii ) , @xmath484 implies @xmath485 , and therefore @xmath486 , which is in contradiction with @xmath487 . thus , assume that @xmath488 . since @xmath68 , by lemma [ l2 ] , we have that @xmath489 . hence , @xmath490 , because @xmath488 implies @xmath491 . moreover , if @xmath492 , then , again by lemma [ l2 ] , @xmath493 . if @xmath243 then @xmath494 and therefore @xmath495 , i.e. , @xmath496 . @xmath434 we prove that if ( i ) and ( ii ) do not hold then @xmath482 . by lemma [ l2 ] , let us show that @xmath489 . if @xmath497 then this is trivially true . thus , let us assume that @xmath488 . @xmath498\hspace*{20ex } } & \\ \cup \{d\cap\operatorname{bottom}(\mu_{{\mathcal{p}}}(b))~|~ b{\trianglelefteq}d\ } & = \\ \text{\small[by set theory]\hspace*{20ex } } & \\ ( b\cap\operatorname{bottom}(\mu_{{\mathcal{p}}}(b)))\cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b))~|~ b { \vartriangleleft}d , c\not{\trianglelefteq}d \ } \cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b))~|~ b { \vartriangleleft}d , c{\trianglelefteq}d \ } & = \\ \text{\small[by definition of $ \operatorname{bottomblock}$]\hspace*{20ex } } & \\ ( b\cap\operatorname{bottom}(\mu_{{\mathcal{p}}}(b)))\cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b))~|~d \in \operatorname{bottomblock}(b ) , c \not{\trianglelefteq}d\ } \cup&\\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b ) ) ~|~ b { \vartriangleleft}d , c{\trianglelefteq}d \ } & \subseteq \\ \text{\small[by conditions~(i ) and ( ii)]\hspace*{20ex } } & \\ \operatorname{pre}(\mu_{{\mathcal{p}}}(c ) ) \cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b ) ) ~|~ d \in \operatorname{bottomblock}(b),c\not{\trianglelefteq}d , d { { { \shortrightarrow}}^{\exists}}\mu_{{\mathcal{p}}}(c ) \ } \cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b ) ) ~|~ b { \vartriangleleft}d , c{\trianglelefteq}d \ } & \subseteq \\ \text{\small[because $ c{\trianglelefteq}d \;{\rightarrow}\ ; d\subseteq \mu_{{\mathcal{p}}}(c)$]\hspace*{20ex } } & \\ \operatorname{pre}(\mu_{{\mathcal{p}}}(c ) ) \cup & \\ \cup \ { d\cap \operatorname{bottom}(\mu_{{\mathcal{p}}}(b ) ) ~|~ d \in \operatorname{bottomblock}(b),c\not{\trianglelefteq}d , d { { { \shortrightarrow}}^{\exists}}\mu_{{\mathcal{p}}}(c ) \ } \cup&\\ \mu_{{\mathcal{p}}}(c ) . & \\ \end{aligned}\ ] ] consider now @xmath499 , @xmath487 and @xmath496 . then , @xmath500}\\ d \cap \neg \operatorname{pre}(\mu_{{\mathcal{p}}}(b ) ) & \subseteq \text{~~~\small[as $ d\subseteq \mu_{{\mathcal{p}}}(d ) \subseteq \mu_{{\mathcal{p}}}(b)$]}\\ \mu_{{\mathcal{p}}}(d ) \cap \neg \operatorname{pre}(\mu_{{\mathcal{p}}}(d ) ) & = \\ \operatorname{bottom}(\mu_{{\mathcal{p}}}(d ) ) & . \\\end{aligned}\ ] ] since @xmath496 , there exists @xmath501 such that @xmath238 . since @xmath502 and @xmath238 , by hypothesis , @xmath240 , so that , by lemma [ l2 ] , we have that @xmath503 . thus , summing up , it turns out that @xmath504 , so that , by lemma [ l2 ] , @xmath283 . * @xmath209 takes @xmath73 time . * @xmath505 takes @xmath506 time . this bound is computed as follows . line 4 takes @xmath271 time . lines 5 - 6 take @xmath507 time . note that lines 5 - 6 are actually implemented as follows : + lines 11 - 12 and 13 - 14 take @xmath508 time . the estimate of the overall cost of lines 7 - 10 deserves special care . at line 10 , it turns out that @xmath509 : if @xmath510 at line 10 then @xmath511 . however , as a consequence of the code at lines 13 - 14 , it turns out that when we are at line 10 , namely when refiner@xmath512 , it is true that @xmath513 so that @xmath514 . hence , the overall cost of lines 7 - 10 is @xmath515 takes @xmath516 time . * @xmath298 takes @xmath230 time . * @xmath212 takes @xmath517 time . in particular , @xmath310 takes @xmath311 time . * @xmath213 takes @xmath518 time . let us prove that the overall number of newly generated blocks by @xmath334 at line 6 of @xmath57 is @xmath519 . let @xmath520}$ ] be the sequence of partitions computed by @xmath57 where @xmath521 is the initial partition @xmath522 , @xmath523 is the final partition @xmath524 and for all @xmath525 $ ] , @xmath526 . the number of newly generated blocks by one splitting operation that refines @xmath527 to @xmath528 is given by @xmath529 . thus , the overall number of newly generated blocks is latexmath:[$\sum_{i=0}^{n-1 } 2(|p_{i+1}| - it turns out that the overall number of iterations of the main while - loop of @xmath57 is in @xmath531 . if at some iteration of @xmath57 it happens that line 7 of @xmath340 sets @xmath532 for some blocks @xmath61 and @xmath62 then for all the successive iterations of @xmath57 , for any block @xmath533 which is contained in @xmath61 ( namely , which is a descendant of @xmath61 ) and for any block @xmath534 which is contained in @xmath62 , and for all the successive iteratuons we will have that @xmath535 . moreover , at any iteration of @xmath57 , there exist at least two blocks @xmath341 such that the assignment @xmath536 at line 7 of @xmath340 is executed . since for any block @xmath61 , the assignment @xmath537 for some @xmath538 and for some @xmath62 may happen at most @xmath82 times , we obtain that the overall number of iterations is in @xmath531 . hence , the overall time complexities of the functions called within the main while - loop of @xmath57 are as follows : + @xmath211 : @xmath539 ; + @xmath540 : @xmath541 ; + @xmath298 : @xmath542 ; + @xmath212 : @xmath543 . + the analysis of the overall time complexity of @xmath213 needs the following observation . as observed above , if at some iteration of @xmath57 it happens that line 7 of @xmath340 sets @xmath544 for some blocks @xmath61 and @xmath62 then for all the successive iterations of @xmath57 , for any block @xmath533 which is contained in @xmath61 and for any block @xmath534 which is contained in @xmath62 , we will have that @xmath545 . thus , for a given block @xmath61 , if the test @xmath546 at line 6 of @xmath340 is true then for any block @xmath547 which is descendant of @xmath62 , the test @xmath548 will be false . this means that for any given block @xmath61 , the body at lines 7 - 16 of the if - then statement at line 6 will be executed at most @xmath82 times . therefore , the overall time complexity in @xmath57 of lines 3 and 17 of @xmath213 is @xmath549 . since the overall cost of lines 2 - 7 and 18 is @xmath550 , it turns out that the overall cost of @xmath213 is @xmath551 . * the pointers from any state @xmath556 to the block @xmath92 of the current partition are stored in @xmath557 space . * the lists localbottoms and bottomblocks globally take , respectively , @xmath558 and @xmath531 space . * the current partition @xmath59 is stored in @xmath559 space . * the current relation @xmath253 is stored in @xmath531 space . * the resizable tables count and bcount take , respectively , @xmath560 and @xmath561 space . * the local table refiner in function @xmath211 takes @xmath531 .

we determine all composition - closed equational classes of boolean functions . these classes provide a natural generalization of clones and iterative algebras : they are closed under composition , permutation and identification ( diagonalization ) of variables and under introduction of inessential variables ( cylindrification ) , but they do not necessarily contain projections . thus the lattice formed by these classes is an extension of the post lattice . the cardinality of this lattice is continuum , yet it is possible to describe its structure to some extent .

1102.4355

the goal of this paper is to describe composition - closed equational classes of boolean functions not necessarily containing projections , thereby generalizing post s description of boolean clones . first we recall the definition of a clone , and then we give an informal overview of the problem that we consider . for formal definitions and more background see section [ sect preliminaries ] and @xcite . we define the _ composition _ of an @xmath0-ary function @xmath1 by the @xmath2-ary functions @xmath3 as the @xmath2-ary function @xmath4 given by@xmath5 we say that @xmath6 is the _ outer function _ of the composition , and @xmath7 are the _ inner functions_. a _ clone _ on the set @xmath8 is a class @xmath9 of finitary functions that is closed under composition and contains the _ _ projections__@xmath10 ( here , and in the rest of the paper , @xmath11 , i.e. , we exclude @xmath12 from the set of natural numbers . ) although the above definition of composition is restrictive in the sense that the inner functions must have the same arity , by making use of projections one can see that clones are closed under compositions without restrictions on the arities . for example , let us suppose that @xmath6 is a ternary function in a clone @xmath13 , and @xmath14 are unary , binary , ternary functions in @xmath13 , respectively . if we would like to build the composite function @xmath15 using only compositions of the form ( [ eq comp ] ) , then we could proceed as follows : first construct ternary functions @xmath16 with the help of the projections:@xmath17 and then form the composition @xmath18 . as we can see from the above example , composing a function with projections allows us to add dummy variables to the function ( see @xmath19 ) , to permute the variables of the function ( see @xmath20 ) and to identify variables of the function ( see @xmath21 ) . function classes closed under the latter three operations are called _ equational classes _ , since they can be defined by functional equations ( see subsection [ subsect equational classes ] ) . the above discussion shows that every clone is an equational class , but , as we shall see in subsection [ subsect equational classes ] , there are equational classes that are not clones . all clones on a given finite base set @xmath8 form an algebraic lattice . this _ clone lattice _ has continuum cardinality if @xmath22 ( see @xcite ) , and it seems to be a very hard problem to describe its structure . the case @xmath23 , i.e. , the case of boolean functions was settled by e. l. post , who described all clones of boolean functions in @xcite . there are countably many such clones , and their lattice is known as the post lattice ( see figure [ fig post ] ) . we will generalize the notion of a clone by considering function classes that are closed under composition ( in the sense of ( [ eq comp ] ) ) but do not necessarily contain the projections . however , we would like to be able to identify and permute variables and introduce dummy variables , therefore we only consider composition - closed equational classes . these classes form a complete lattice that contains the clone lattice as the principal filter generated by the clone of projections ( see figure [ fig biglattice ] ) . the main result of this paper is a description of this lattice over a two - element base set . although the clone lattice is countable in this case , we will see that the lattice of composition - closed equational classes of boolean functions is uncountable . composition - closed equational classes subsume iterative algebras as well . a function class @xmath24 is an _ iterative algebra _ if @xmath25 , whenever @xmath26 and @xmath27for @xmath28 . clearly , every iterative algebra is a composition - closed equational class , and an iterative algebra is a clone iff it contains the projections . the difference between composition - closed equational classes and iterative algebras can be best understood by visualizing compositions as trees . as an example , let us consider the following two compositions : if @xmath24 is an iterative algebra and @xmath29 , then both @xmath8 and @xmath30 must belong to @xmath24 . however , if @xmath24 is only assumed to be a composition - closed equational class , then @xmath8 does not necessarily belong to @xmath24 , as it involves the composition @xmath31 , where one inner function is a projection . this problem does not arise with @xmath30 , hence @xmath32 is guaranteed , whenever @xmath24 is a composition - closed equational class . in general , we can say that composition - closed equational classes are closed under compositions whose tree satisfies the following condition : for any internal node , either all or none of its children are leaves . as an exercise in handling such compositions , we invite the reader to verify the following fact : the clone generated by the addition operation of a field ( or , more generally , of any additive commutative semigroup ) consists of functions of the form @xmath33 the iterative algebra generated by addition contains only those such functions where @xmath35 , while the composition - closed equational class generated by addition contains only those where @xmath36 is even . the paper is organized as follows : in section [ sect preliminaries ] we present the necessary background on equational classes , boolean clones , and galois connections between functions and relations . in section [ sect idempotents vs clones ] we make some basic observations about composition - closed equational classes of boolean functions , and we outline a strategy for constructing all of them . in section [ sect easy ] we carry out this strategy for the easy cases , and then we deal with the harder cases in sections [ sect w_infty ] and [ sect w_k ] . finally , in section [ sect concluding ] we put together all the information we found to get a picture about the lattice of composition - closed equational classes of boolean functions . let @xmath6 and @xmath37 be operations on a set @xmath8 of arity @xmath0 and @xmath38 , respectively . if there exists a map @xmath39 such that@xmath40 then we say that @xmath37 is a _ subfunction _ ( or identification minor or simple variable substitution ) of @xmath6 , and we denote this fact by @xmath41 . if @xmath42 is bijective , then @xmath37 is obtained from @xmath6 by permuting variables ; if @xmath42 is not injective , then @xmath37 is obtained from @xmath6 by identifying variables ; if @xmath42 is not surjective , then @xmath37 is obtained from @xmath6 by introducing inessential ( dummy ) variables . the subfunction relation gives rise to a quasiorder on the set of all finitary functions on @xmath8 ( see @xcite ) . the corresponding _ equivalence _ is defined by @xmath43 and @xmath41 , and it is clear that @xmath6 and @xmath37 are equivalent iff they differ only in inessential variables and/or in the order of their variables . we will not distinguish between equivalent functions in the sequel . for example , we will denote the set of all constant zero functions ( for any @xmath44 ) simply by @xmath45 , and @xmath46 will stand for the set of all projections , as these are the functions equivalent to the identity function . let @xmath47 denote the class of all boolean functions , i.e. , the class of finitary operations on @xmath48 . the subfunction relation induces naturally a partial order on @xmath49 ; the bottom of this poset is shown in figure [ fig subf ] . we can see ( and it is easy to prove ) that it has four connected components , namely @xmath50 , where@xmath51 let us observe that @xmath52 is nothing else but the clone of idempotent functions . for an arbitrary function class @xmath24 , we will abbreviate @xmath53 by @xmath54 , and we will later use the following ( hopefully intuitive ) notation as well:@xmath55 [ ptb ] subf.eps the minimal elements of @xmath56 are the unary functions : @xmath57 and @xmath58 ( negation ) . on the next level we can see the binary operations @xmath59 ( addition modulo @xmath60 ) , @xmath61 ( implication ) , @xmath62 ( disjunction ) , @xmath63 ( conjunction ) and the ternary functions @xmath64 ( majority operation ) , @xmath38 ( minority operation ) , @xmath65 ( @xmath66-minority operation , see @xcite ) together with their negations . here negation is taken from outside , e.g. , @xmath67 is a shorthand notation for the function @xmath68 . a class @xmath24 of operations on a set @xmath8 is an _ equational class _ if it is an order ideal in the subfunction quasiorder , i.e. , if @xmath69 and @xmath41 imply @xmath70 . the denomination is explained by the fact that these are exactly the classes that can be defined by functional equations @xcite . clearly every clone is an equational class , but there are other equational classes as well ; a natural example is the class of antimonotone ( order reversing ) boolean functions , which can be defined by the functional equation @xmath71 . another example is the class @xmath72 ( see subsection [ subsect subfunctions ] ) , which can be defined by the equation @xmath73 . equational classes can be characterized by relational constraints as well ; we will discuss this in more detail in subsection [ subsect relational constarints ] . the equational classes on a given set @xmath8 form a lattice @xmath74 with intersection and union as the lattice operations . this lattice has continuum cardinality already on the two - element set , and its structure is very complicated @xcite . if @xmath75 and @xmath76 are classes of functions , then their _ composition _ , denoted by @xmath77 , is the set of all compositions where the outer function belongs to @xmath75 and the inner functions belong to @xmath76:@xmath78 in general , associativity does not hold for function class composition , but it holds for equational classes @xcite , hence we obtain a monoid @xmath79 with the identity element @xmath46 . we will mostly consider boolean functions , and in this case we will drop the index @xmath8 , and denote the set of equational classes simply by @xmath80 . a class @xmath24 is _ closed under composition _ iff @xmath81 . as we shall see in proposition [ prop ( sub)idempotents ] , this is equivalent to the formally stronger requirement that @xmath24 is _ idempotent _ , i.e. , @xmath82 ( cf . ( let us note that this is a distinguishing feature of boolean functions : if @xmath8 has at least three elements , then one can construct a class @xmath83 such that @xmath84 . ) the goal of this paper is to describe the idempotent elements of @xmath85 . the usual notation for the set of idempotents of a semigroup @xmath86 is @xmath87 , but in our case this would lead to the somewhat awkward notation @xmath88 , therefore we will simply write @xmath89 for the set of composition - closed equational classes over @xmath90 . clearly @xmath89 is closed under arbitrary intersections ( we allow the empty class ) , hence it is a complete lattice . the lattice of clones appears in @xmath89 as the principal filter generated by @xmath46 , and we will see that the rest of @xmath89 is the principal ideal generated by @xmath72 ( see figure [ fig biglattice ] ) . we will also see that the lattice @xmath89 has continuum cardinality . there are countably many clones on the two - element set , and these have been described by e. l. post in @xcite . figure [ fig post ] shows the clone lattice on @xmath90 , usually referred to as the _ post lattice_. the top element is @xmath47 , the class of all boolean functions , and the bottom element is @xmath46 , the clone consisting of projections only . the other clones labelled in the figure are the following : [ ptb ] post6.eps * @xmath91 is the clone of @xmath12-preserving functions ; * @xmath92 is the clone of @xmath93-preserving functions ; * @xmath64 is the clone of monotone ( order preserving ) functions ; * @xmath94 is the clone of self - dual functions , i.e. , functions satisfying @xmath95 * @xmath96 is the clone of linear functions , i.e. , functions of the form @xmath97 with @xmath98 ; * @xmath99 consists of conjunctions @xmath100 @xmath101 and the two constants @xmath102 ; * @xmath103 consists of disjunctions @xmath104 and the two constants @xmath102 ; * @xmath105 is the clone of essentially at most unary functions ; * @xmath106 is the clone of functions preserving the relation @xmath107 ; it can be generated , e.g. , by the function @xmath108 of arity @xmath109 defined by in the post lattice are @xmath110 and @xmath111 . therefore , in order to verify that @xmath108 generates @xmath106 , it suffices to check that @xmath112 and @xmath108 is neither monotone nor idempotent , nor does it belong to @xmath111.]@xmath113{ll}0 , & \text{if } \left\vert \left\ { i : x_{i}=1\right\ } \right\vert = 2\text { and } x_{1}=1;\\ 1 , & \text{otherwise.}\end{array } \right.\ ] ] * @xmath114 is the clone generated by implication ; * @xmath115 is the dual of @xmath106 for @xmath116 . all other clones can be obtained as intersections of these clones . the different types of nodes and edges in figure [ fig post ] help the navigation in the post lattice as follows : * nodes representing clones of idempotent functions have a double outline ( others have a single outline ) , and a double edge connects a clone @xmath13 to @xmath117 ; * nodes representing clones of monotone functions are filled ( others have empty interior ) , and a thick edge connects a clone @xmath13 to @xmath118 ; * nodes representing clones of self - dual functions are squares ( others are circles ) , and a dashed edge connects a clone @xmath13 to @xmath119 ; * nodes representing clones of essentially at most unary functions are grey ( others are black ) , and all edges incident with unary clones are grey . if @xmath120 is a relation of arity @xmath38 , and @xmath121 is an @xmath122 matrix such that each column of @xmath123 belongs to @xmath124 , then we say that @xmath123 is a @xmath124__-matrix__. applying an @xmath0-ary function @xmath6 to the rows of @xmath123 , we obtain the column vector @xmath125 . a _ relational constraint _ of arity @xmath38 is a pair @xmath126 , where @xmath124 and @xmath127 are @xmath38-ary relations . an @xmath0-ary function @xmath6 _ satisfies _ the constraint @xmath126 if @xmath128 for every @xmath124-matrix @xmath123 of size @xmath122 . satisfaction of relational constraints gives rise to a galois connection that defines equational classes of functions . [ @xcite][thm pippenger]a class of functions on a finite set @xmath8 is an equational class iff it can be defined by relational constraints . as an illustration of this theorem let us consider our two examples from subsection [ subsect equational classes ] : the class of antimonotone functions can be defined by the constraint @xmath129 , and the class @xmath72 can be defined by @xmath130 . iterative algebras can be characterized by relational constraints as well . let us note that iterative algebras are usually defined with the help of the five operations @xmath131 introduced by malcev @xcite , but using function class composition we can give a very compact definition : a function class @xmath24 is an _ iterative algebra _ iff @xmath132 . [ @xcite]a class of functions on a finite set @xmath8 is an iterative algebra iff it can be defined by relational constraints @xmath126 with @xmath133 . a function @xmath6 _ preserves _ the relation @xmath124 iff @xmath6 satisfies the constraint @xmath134 . this induces the well - known pol - inv galois connection between clones and relational clones . [ @xcite]a class of functions on a finite set @xmath8 is a clone iff it can be defined by relations . now we present another galois connection that characterizes composition - closed equational classes . let us say that a function @xmath6 _ strongly satisfies _ the relational constraint @xmath126 , if @xmath6 satisfies both @xmath126 and @xmath135 ( i.e. , @xmath6 satisfies @xmath126 and preserves @xmath127 ) . the function class @xmath24 is _ strongly defined _ by relational constraints if there exists a set @xmath136 of relational constraints such that a function belongs to @xmath24 iff it strongly satisfies @xmath137 for all @xmath138 . [ thm galois connection]a class of functions on a finite set @xmath8 is a composition - closed equational class iff it can be strongly defined by relational constraints . first let us suppose that @xmath24 is strongly defined by a set @xmath139 of relational constraints , where @xmath140 and @xmath141 are relations of arity @xmath142 . by theorem [ thm pippenger ] , @xmath24 is an equational class . to verify that @xmath24 is closed under composition , let us consider arbitrary functions @xmath143 , where the arity of @xmath6 is @xmath0 , and the arity of @xmath7 is @xmath2 . then the composition @xmath144 is a @xmath2-ary function on @xmath8 . let @xmath138 , and let @xmath123 be a @xmath140-matrix of size @xmath145 . since @xmath146 satisfy the constraint @xmath137 , the @xmath147 matrix @xmath148 formed by the column vectors @xmath149 is a @xmath141-matrix . therefore , @xmath150 , as @xmath6 preserves the relation @xmath141 . this shows that the composition @xmath151 satisfies the constraint @xmath137 . noting that @xmath152 all preserve the relation @xmath141 , we see that @xmath151 preserves @xmath141 as well , hence @xmath151 strongly satisfies @xmath137 . this holds for all @xmath138 , thus @xmath153 , as claimed . for the other implication , let us assume that @xmath24 is a composition - closed equational class . let us write all elements of @xmath154 below each other ( as row vectors ) , and let @xmath155 denote the resulting @xmath156 matrix . let @xmath157 be the set of column vectors of @xmath155 , and let @xmath158 be the set of all column vectors of the form @xmath159 , where @xmath26 is of arity @xmath0 . let @xmath160 be the class of functions strongly defined by @xmath161 . we will prove that @xmath162 . if @xmath163 is a function of arity @xmath0 , then @xmath164 , hence , according to the definition of @xmath158 , there exists an @xmath0-ary function @xmath26 such that @xmath165 . since the rows @xmath155 contain every element of @xmath154 , this implies that @xmath166 , thus @xmath167 , hence we can conclude that @xmath168 . in order to prove that @xmath169 , we need to verify for an arbitrary @xmath0-ary function @xmath26 that @xmath6 strongly satisfies the constraint @xmath170 for every @xmath171 . if @xmath172 is a @xmath173-matrix of size @xmath174 , then @xmath175 for a suitable @xmath2-ary subfunction @xmath176 of @xmath6 . since @xmath24 is an equational class , we have @xmath177 , hence @xmath178 . this shows that @xmath6 satisfies the constraint @xmath170 . we need to check yet that @xmath6 preserves @xmath179 , i.e. , that @xmath180 for any @xmath179-matrix @xmath181 of size @xmath174 . by the definition of @xmath179 , the list of columns of @xmath181 is of the form @xmath182 for some @xmath2-ary @xmath183 , thus @xmath184 , where @xmath185 . since @xmath24 is closed under composition , we have @xmath153 , hence @xmath186 , as claimed . concerning our two examples , let us observe that the class of antimonotone functions is not closed under composition , but @xmath72 is closed . indeed , @xmath72 is strongly defined by the constraint @xmath187 , since the relation @xmath188 is just the equality relation , and it is preserved by every function . from now on we will restrict our attention to boolean functions . in this section we make some basic observations about the relationship between a composition - closed equational class @xmath24 and the clone @xmath13 generated by @xmath24 . these observations will make it possible to construct all such classes for a given clone @xmath13 . let us first briefly discuss the unary case : there are @xmath189 equational classes consisting of essentially at most unary functions , and the following @xmath190 are closed under composition:@xmath191 as we shall see in the following lemma , disregarding these 10 trivial cases , we can always assume that @xmath192 or @xmath193 is present in our composition - closed equational class . [ lemma + imp]if @xmath24 is a composition - closed equational class that is not a clone , then 1 . @xmath194 ; 2 . if @xmath195 , then @xmath196 . if @xmath24 is closed under composition , but @xmath24 is not a clone , then @xmath197 . since @xmath24 is an equational class , for each @xmath26 the unary subfunction @xmath198 of @xmath6 belongs to @xmath24 . if there is a function @xmath199 , then we can conclude that @xmath200 , a contradiction . if there is a function @xmath201 , then @xmath202 , and since @xmath24 is closed under composition , we have @xmath203 , which is a contradiction again . thus @xmath204 and @xmath205 , therefore @xmath206 . to prove the second statement of the lemma , let us observe that any equational class @xmath195 must contain at least one of the 14 functions shown on the second level of figure [ fig subf ] . since we have @xmath194 , we see that at least one of the functions @xmath207 belongs to @xmath24 . note that the first statement of the lemma shows that the non - clone composition - closed equational classes form a principal filter in @xmath89 ( see figure [ fig biglattice ] ) . hence the lattice @xmath89 has six coatoms : the five maximal clones ( @xmath208 ) and @xmath72 . next we prove that every composition - closed equational class is idempotent , as mentioned in subsection [ subsect equational classes ] . [ lemma elso]let @xmath209 be composition - closed equational classes such that @xmath210 and @xmath211 . then we have @xmath212 . one direction is obvious : @xmath213 . the containment @xmath214 is also clear if @xmath215 . if this is not the case , then one of the functions @xmath207 belongs to @xmath216 by lemma [ lemma + imp ] . if @xmath217 , then for any @xmath218 we have @xmath219 , thus @xmath220 . the same is true for @xmath221 and @xmath61 , while for @xmath222 we can use the composition @xmath223 . [ prop ( sub)idempotents]an equational class @xmath24 is closed under composition iff it is idempotent . the if part is obvious ; the only if part is also obvious if @xmath224 , otherwise it follows from the above lemma with @xmath225 . in light of the proposition above , in the following we will refer to a composition - closed equational class simply as an idempotent . in the next proposition we explore some relationships between an idempotent @xmath24 and the clone @xmath226 $ ] generated by @xmath24 that will play a crucial role in the sequel . [ prop ck]let @xmath195 be an idempotent , and let @xmath227 $ ] . then we have @xmath228 and @xmath229 . the second equality follows from lemma [ lemma elso ] . the containment @xmath230 is clear , since @xmath231 , so it remains to prove that @xmath232 . every element of @xmath233 is of the form @xmath185 , where @xmath234 is @xmath0-ary , and @xmath183 are @xmath2-ary functions . since @xmath234 , it is a composition of elements of @xmath24 and projections . we will prove @xmath153 by induction on the size ( number of nodes ) of the tree describing this composition ( see section [ sect intro ] ) . if the tree has only one node , then @xmath26 , and then @xmath235 follows since @xmath24 is closed under composition . if the tree has at least two nodes , then @xmath6 is of the form @xmath236 with appropriate functions @xmath237 and @xmath238 , where the composition tree of each @xmath239 is smaller than the tree corresponding to @xmath6 . now we can write @xmath151 as@xmath240 by the induction hypothesis , every one of the functions @xmath241 belongs to @xmath24 , therefore @xmath242 . for any equational class @xmath24 let us write @xmath243 for the smallest idempotent containing @xmath24 , and recall that @xmath226 $ ] denotes the smallest clone containing @xmath24 . the next proposition shows a relationship between these two closure operators . [ prop gen vs lgen]for any equational class @xmath24 we have @xmath226 = \left\lfloor \mathcal{k}\cup\left\ { \operatorname{id}\right\ } \right\rfloor $ ] and @xmath244 \circ\mathcal{k}.$ ] the first equality expresses the obvious fact that the clone generated by @xmath24 consists of those functions that can be obtained from elements of @xmath24 and from projections by means of composition . introducing the notation @xmath227 $ ] , the second equality takes the form @xmath245 . the right hand side contains @xmath24 , since @xmath231 , and it is closed under composition ( hence idempotent):@xmath246 to prove that @xmath233 is indeed the smallest idempotent containing @xmath24 , we have to show that @xmath247 . the case @xmath224 is trivial , otherwise we can apply proposition [ prop ck ] to the idempotent @xmath248 , and we get@xmath249 \circ\left\lfloor \mathcal{k}\right\rfloor = \left\lfloor \mathcal{k}\right\rfloor .\ ] ] therefore , @xmath250 , and this proves the proposition . let us give some simple examples of idempotents together with relational constraints strongly defining them . * @xmath251 + strongly defined by @xmath252 * @xmath253 + strongly defined by @xmath254 * @xmath255 + strongly defined by @xmath256 * @xmath257 + strongly defined by @xmath258 members of the class @xmath259 are called _ reflexive functions_. [ ex lgen+]one can determine @xmath260 using our observations about the composition - closed equational class generated by the addition operation of a field ( see section [ sect intro ] ) . alternatively , we can use proposition [ prop gen vs lgen ] : the elements of @xmath261 are exactly the functions of the form@xmath262 where @xmath263 = l_{0\ast}$ ] and @xmath264 are arbitrary ( not necessarily distinct ) variables . such a function is a sum of an even number of variables , hence @xmath265 similarly , we have @xmath266 proposition [ prop gen vs lgen ] gives the following general form for the elements of @xmath267:@xmath268 where @xmath269 = w^{\infty}$ ] , and @xmath264 are arbitrary ( not necessarily distinct ) variables . here , the lack of associativity makes it harder to decide whether a given function is of this form . for example , the function @xmath270 seems not to have the required form , but we may rewrite it as @xmath271 , therefore @xmath272 . on the other hand , @xmath273 does not belong to @xmath267 . to verify this , let us observe that @xmath274 , hence @xmath275 . however , @xmath276 as @xmath277 , therefore @xmath278 . in proposition [ prop bottom w_infty ] we will describe @xmath279 by relational constraints ; this description makes it easy to decide membership for @xmath267 . proposition [ prop gen vs lgen ] suggests the following strategy for finding all idempotents : let us fix a clone @xmath13 , and form all possible compositions @xmath233 , where @xmath24 is an equational class , such that @xmath226 = \mathcal{c}$ ] . this way we obtain all those idempotents that generate the clone @xmath13 let us denote the set of these idempotents by @xmath280:@xmath281 = \mathcal{c}\right\ } .\ ] ] clearly these sets form a partition of @xmath89 . for unary clones @xmath13 it is an easy task to determine @xmath282 : for @xmath283 we have @xmath284 ; for the other unary clones we have @xmath285 . [ thm interval]for any clone @xmath13 , the set @xmath282 is an interval in the lattice of idempotents . if @xmath280 has more than one element , then @xmath286 is its only coatom . we have settled the unary case above , so now let us suppose that @xmath287 . it is clear that the largest element of @xmath280 is @xmath13 itself , and it is also clear that @xmath280 is convex . therefore , in order to prove that it is indeed an interval , it suffices to show that @xmath280 is closed under arbitrary intersections , hence it has a least element . let @xmath288 , and let @xmath289 . then @xmath290 by proposition [ prop ck ] . on the other hand , @xmath231 implies that @xmath230 , so we have @xmath228 . composing by @xmath226 $ ] on the right , we get @xmath291 = \mathcal{k}\circ\left [ \mathcal{k}\right ] $ ] , and using proposition [ prop ck ] once more , we can conclude that @xmath292 = \left [ \mathcal{k}\right ] $ ] . the left hand side contains @xmath13 , since @xmath293 $ ] , so we have @xmath294 $ ] . the containment @xmath226 \subseteq\mathcal{c}$ ] is obvious , so we see that @xmath226 = \mathcal{c}$ ] , and therefore @xmath295 , as claimed . ( note that what we have proved is rather surprising : if each @xmath296 contains a generating set of the clone @xmath13 , then so does @xmath297 . ) the statement about the unique coatom follows from lemma [ lemma + imp ] : if @xmath24 is an element of @xmath280 that is different from @xmath13 , then @xmath24 is a composition - closed equational class that is not a clone , hence @xmath298 . the contents of the above theorem are represented in figure [ fig typint ] , which shows a picture of a typical nontrivial interval @xmath282 . as the next corollary shows , theorem [ thm interval ] allows us to prove that @xmath280 is trivial for many boolean clones . [ ptb ] typical_interval.eps if @xmath13 is a clone different from @xmath299 , then @xmath300 . if @xmath280 has at least two elements , then @xmath301 according to theorem [ thm interval ] . this implies that @xmath286 generates the clone @xmath13 . however , for many clones , @xmath302 is too small to generate @xmath13 . indeed , @xmath303 and @xmath304 therefore ( ignoring the unary clones ) @xmath282 is trivial unless @xmath13 is one of the clones represented in figure [ fig post ] by a black circle with a single outline and an empty ( non - filled ) interior . up to duality , only the cases @xmath305 remain . the first four cases are treated in section [ sect easy ] ; we will see that in these cases @xmath282 has at most @xmath306 elements . in section [ sect w_infty ] we give a characterization of the elements of @xmath307 , and we will prove that this interval is uncountable . we consider @xmath308 for finite @xmath2 in section [ sect w_k ] , and we will show that these intervals are finite , but their cardinalities do not have a common upper bound . as explained in the previous section , @xmath280 can be determined by computing all possible compositions of the form @xmath233 , where @xmath24 is an equational class generating the clone @xmath13 . in the following two lemmas we compute six such compositions for @xmath310 and @xmath311 , and then we will show that these suffice to determine the intervals @xmath312 and @xmath313 . [ lemma om.->]the following equalities hold : 1 . @xmath314 ; 2 . @xmath315 . \(1 ) we prove only that @xmath316 ; the other case is similar . the functions in @xmath317 are of the following form:@xmath318 where @xmath319 and @xmath320 . since @xmath321 , every function of this form belongs to @xmath72 . conversely , for any given @xmath322 we will construct a suitable @xmath6 so that ( [ om.- > ] ) holds . let us choose @xmath323 , and let @xmath324 . then we can rewrite ( [ om.- > ] ) as @xmath325 for all @xmath326 , where @xmath327 denotes the vector formed from all possible implications between entries of @xmath328 . now we can treat this as the definition of @xmath6:@xmath329{ll}h\left ( \mathbf{a}\right ) , & \text{if } \mathbf{b}=\widehat{\mathbf{a}}\text{;}\\ 0 , & \text{if } \nexists\mathbf{a}\in\left\ { 0,1\right\ } ^{k}:\mathbf{b}=\widehat{\mathbf{a}}. \end{array } \right.\ ] ] all we need to show is that @xmath6 is well - defined , i.e. , that @xmath330 determines @xmath331 uniquely . if @xmath332 , then @xmath333 , but we can not tell whether @xmath334 or @xmath335 . however , since @xmath322 , the value of @xmath331 is the same in both cases . now let us suppose that at least one entry of @xmath327 is @xmath12 , say , @xmath336 . then we can infer immediately that @xmath337 and @xmath338 . using this information we can recover the vector @xmath328 , since @xmath339 for all @xmath340 . thus in this case @xmath327 uniquely determines @xmath328 , hence @xmath331 is uniquely determined as well . \(2 ) we can proceed similarly as above , with @xmath341 and @xmath342 . [ lemma om.+]the following equalities hold : 1 . @xmath343 ; 2 . @xmath344 . we can use the same argument as in the previous proposition . for example , considering @xmath345 , we have@xmath346 in place of ( [ om.- > ] ) , and we define @xmath347 to be the vector formed from all sums of two entries of @xmath328 . now @xmath327 tells us which entries of @xmath328 are the same , but does not tell us which one is @xmath12 and which one is @xmath93 . therefore @xmath327 determines @xmath328 up to negation , and this is equivalent to the condition @xmath348 . if @xmath310 or @xmath349 , then @xmath280 is a three - element chain : @xmath350 . let us consider first @xmath310 . if @xmath351 , then @xmath352 by proposition [ prop ck ] . if @xmath24 is a clone , then clearly we have @xmath353 . if this is not the case , then @xmath354 , and at least one of the operations @xmath355 belongs to @xmath24 by lemma [ lemma + imp ] . if @xmath356 or @xmath357 , then @xmath358 by lemma [ lemma om.- > ] , hence @xmath359 . in the remaining cases the binary operations in @xmath24 are @xmath59 and/or @xmath221 , hence @xmath360 by lemma [ lemma om.+ ] . on the other hand , in these cases every binary operation in @xmath361 is reflexive . we will show that this implies @xmath362 . if @xmath363 and @xmath364 , then @xmath365 , where @xmath366 is obtained from @xmath367 by replacing @xmath368 by @xmath369 or @xmath370 depending on whether @xmath371 or @xmath372 . since @xmath37 is a subfunction of @xmath6 , we have @xmath373 , and then the reflexivity of @xmath37 implies @xmath374 . this is true for all @xmath375 , thus @xmath6 is reflexive , as claimed . the above considerations show that the only possible elements of @xmath312 are @xmath376 and @xmath259 . it is easily verified that each of these three classes indeed generate the clone @xmath47 , hence @xmath377 . if @xmath378 , then @xmath379 can not belong to @xmath361 . a similar argument as the above one shows that @xmath380 if @xmath381 , @xmath382 if @xmath383 but @xmath356 , and @xmath384 otherwise . thus we have @xmath385 . as one can expect , the case of linear functions is considerably simpler . if @xmath386 or @xmath387 then @xmath282 is a two - element chain : @xmath388 . if @xmath389 , then @xmath390 by theorem [ thm interval ] . however , @xmath391 ( see example [ ex lgen+ ] ) , so we can conclude that @xmath392 . on the other hand , since @xmath393 , it contains at least one of the functions @xmath379 by lemma [ lemma + imp ] . clearly @xmath394 , so we must have @xmath395 , and then @xmath396 . thus @xmath397 , hence @xmath398 . now let @xmath399 . then @xmath400 . similarly to the previous case we see that @xmath401 or @xmath402 . if only one of these held , then @xmath361 would be a subset of either @xmath403 or @xmath404 , contradicting @xmath405 = l$ ] . so we must have @xmath406 , hence @xmath407 . in the above cases for each given clone @xmath13 , the interval @xmath280 was finite . for @xmath409 the situation is more complicated : we will prove in this section that there are continuously many idempotents generating the clone @xmath408 . in the next section we will see that the intervals @xmath410 are finite , but their sizes tend to infinity as @xmath411 . in these two sections we will often work with the set of zeros ( set of false points ) of functions , so let us set up some notation for this . for an @xmath0-ary boolean function @xmath6 , let @xmath412 . we will often treat this set as a matrix : writing the elements of @xmath413 as row vectors below each other , we obtain an @xmath122 matrix , where @xmath414 . the order of the rows is irrelevant , moreover , since we do not need to distinguish between equivalent functions , we may rearrange the columns as well ( by permuting variables ) . let us observe that @xmath415 iff @xmath413 has a constant @xmath12 column , and @xmath416 iff every matrix formed from at most @xmath2 rows of @xmath413 has a constant @xmath12 column . next we introduce some operators on function classes that deal with zeros of functions . for @xmath417 and @xmath418 let @xmath419 denote the set of functions @xmath319 such that for every at most @xmath2-element subset @xmath420 there exists a function @xmath70 of the same arity as @xmath6 with @xmath421 . furthermore , let @xmath422 be the set of functions @xmath319 such that there exists a function @xmath70 of the same arity as @xmath6 with @xmath423 . clearly , @xmath424 are closure operators on @xmath47 , and for any @xmath418 we have@xmath425 proposition [ prop w_infty ] below describes @xmath307 with the help of the operator @xmath426 . [ lemma ize]if @xmath24 is an idempotent equational class such that @xmath226 = w^{k}$ ] for some @xmath427 , then @xmath428 . for @xmath429 the statement is clear , so let us suppose that @xmath430 . let @xmath431 be @xmath0-ary functions such that @xmath432 . we have to show that @xmath26 . we can suppose without loss of generality that @xmath433 , i.e. , @xmath6 and @xmath37 differ only at one position @xmath364 . ( otherwise we apply this several times changing the values of @xmath37 one by one until we reach the desired function @xmath6 . ) for notational simplicity let us also assume that @xmath434 and @xmath435 . then @xmath436 and @xmath437 . ( here , and in the rest of the proof all @xmath0-tuples are split into two parts of size @xmath438 and @xmath439 . ) from proposition [ prop ck ] we know that @xmath440 \circ\mathcal{k}=w^{k}\circ\mathcal{k}\supseteq w^{\infty } \circ\mathcal{k}$ ] , so it suffices to show that @xmath441 . we consider the following function of arity @xmath442 with @xmath443 , which clearly belongs to @xmath408:@xmath444 let us construct the following function @xmath445 , which is a composition of @xmath151 ( as outer function ) and several subfunctions of @xmath37 ( as inner functions):@xmath446 we claim that @xmath447 . in order to verify this fact , let us observe that since @xmath436 , we have @xmath448 , as @xmath449 . we also have @xmath450 , and then we must have @xmath451 , since otherwise @xmath452 would be a subfunction of @xmath37 , implying @xmath453 = w^{k}$ ] , contrary to our assumption . therefore @xmath454 iff @xmath455 and @xmath456 , hence the disjunction in the definition of @xmath457 equals @xmath12 iff @xmath458 and @xmath459 . thus @xmath457 differs from @xmath37 only at the position @xmath328 , where @xmath460 and @xmath461 . hence @xmath462 , as claimed . [ prop w_infty]let @xmath24 be an equational class such that @xmath226 = w^{\infty}$ ] . then@xmath463 @xmath464 : follows from the previous lemma with @xmath465 . @xmath466 : let us suppose that @xmath405 = w^{\infty}$ ] and @xmath428 . let @xmath26 be @xmath0-ary and @xmath183 be @xmath2-ary functions . we have to prove that @xmath185 belongs to @xmath24 . since @xmath415 , there exists an index @xmath467 such that for all @xmath468 we have @xmath371 . therefore , for all @xmath469 we have @xmath470 , i.e. , @xmath471 . this proves that @xmath472 . [ rem ideal]let us note that @xmath428 iff @xmath24 is a filter in the usual pointwise ordering @xmath473 of boolean functions . let @xmath474 be the transitive closure of @xmath475 . then @xmath474 is a quasiorder on @xmath47 , and a class @xmath24 is an ideal with respect to this quasiorder iff @xmath24 is an ideal w.r.t . @xmath476 ( i.e. , an equational class ) and a filter w.r.t . @xmath473 ( i.e. , satisfies @xmath428 ) . thus we can reformulate the above proposition as follows : @xmath477 iff @xmath226 = w^{\infty}$ ] and @xmath24 is an ideal w.r.t . this implies the somewhat surprising fact that a union of idempotents is idempotent in this case , i.e. , the lattice operations in the interval @xmath478 coincide with the set operations @xmath479 and @xmath480 . consequently , @xmath481 is a distributive lattice . in the next proposition we describe @xmath482 , the bottom element of the interval @xmath307 . [ prop bottom w_infty]the bottom element of the interval @xmath478 is @xmath267 , which is strongly defined by the relational constraints@xmath483 let @xmath484 denote the function class strongly defined by the above constraints . we will prove that@xmath485 the first containment follows immediately from the fact that @xmath486 , and for the second one it suffices to verify that @xmath487 , since @xmath484 is an idempotent according to theorem [ thm galois connection ] . for the third containment , let @xmath6 be an arbitrary @xmath0-ary function in @xmath484 , and let @xmath488 . since @xmath6 preserves the relation @xmath489 , the matrix @xmath490 has a constant @xmath12 column . moreover , since @xmath6 satisfies the relational constraint @xmath491 , the matrix @xmath413 has a constant @xmath93 column as well . we can suppose without loss of generality that the first column of @xmath413 is constant @xmath93 and the second column is constant @xmath12 . since @xmath492 = w^{\infty}$ ] , there is at least one nonmonotone function @xmath493 . for such a function , @xmath494 , thus , permuting variables , we may suppose that @xmath495 contains a vector of the form @xmath496 . applying the operator @xmath426 , we can obtain a function @xmath497 of the same arity as @xmath498 with @xmath499 . identifying appropriately the variables of @xmath500 , we get a binary function @xmath501 with @xmath502 ( @xmath503 is nothing else but the implication ) . adding @xmath504 dummy variables , we can obtain a function @xmath505 of arity @xmath0 , such that @xmath506 , i.e. , _ every _ @xmath0-tuple of the form @xmath507 belongs to @xmath508 . since @xmath413 consists of _ some _ of these tuples , we have @xmath509 . combining the previous two propositions and remark [ rem ideal ] , we get the following characterizations of the idempotents corresponding to @xmath408 . [ thm w_infty]for any class @xmath24 of boolean functions the following conditions are equivalent : 1 . @xmath510 ; 2 . @xmath511 , and @xmath24 is an equational class satisfying @xmath428 ; 3 . @xmath511 , and @xmath24 is an ideal with respect to the quasiorder @xmath474 . we conclude this section by proving that the interval @xmath481 is uncountable . the interval @xmath478 has continuum cardinality . let @xmath512 be the following @xmath513 matrix over @xmath90:@xmath514 and let @xmath515 be the @xmath516-ary boolean function such that @xmath517 consists of the rows of @xmath512 , and let @xmath157 be the @xmath0-ary relation consisting of the columns of @xmath512 . it is straightforward to verify that @xmath518 = w^{\infty}$ ] , hence @xmath519 . we claim that for all natural numbers @xmath520,@xmath521 it is clear that @xmath515 does not satisfy @xmath522 , since @xmath523 . now let us assume that @xmath38 is odd and @xmath524 . the matrix @xmath512 has a constant @xmath12 column , therefore @xmath515 preserves @xmath525 . suppose for contradiction that @xmath515 does not satisfy @xmath526 , i.e. , there exists a @xmath527-matrix @xmath123 of size @xmath528 such that @xmath529 . this means that every column of @xmath123 is a column of @xmath530 , and every row of @xmath123 is a row of @xmath512 . let us interpret the matrix @xmath123 as the incidence matrix of a graph @xmath531 ( with possibly multiple edges ) : the columns of @xmath123 correspond to the vertices @xmath532 of @xmath531 , and the rows of @xmath123 correspond to the @xmath38 edges of @xmath531 . since each row of @xmath123 is a row of @xmath512 , the vertex @xmath533 is isolated , and every edge connects two consecutive vertices in the cyclical ordering of @xmath534 . since every column of @xmath123 is a column of @xmath530 , every vertex has degree @xmath12 or @xmath60 . these properties imply that @xmath531 is either a cycle on the vertices @xmath534 ( together with the isolated vertex @xmath533 ) , or the components of @xmath531 are @xmath60-cycles ( double edges ) and isolated vertices : grafa.eps grafb.eps both cases lead to a contradiction : in the former case we have @xmath535 , while in the latter case we can conclude that @xmath38 is even . this contradiction proves our claim . to finish the proof of the theorem it suffices to observe that our claim implies that @xmath536 embeds the power set of the set of odd natural numbers into @xmath478 . we have shown that the power set of a countably infinite set embeds into @xmath478 , and it is obvious that @xmath478 embeds into the power set of @xmath47 . thus , @xmath478 is equimorphic to the power set of a countably infinite set . we will prove that the analogue of proposition [ prop w_infty ] holds for finite values of @xmath2 as well . first we need a technical lemma . for any @xmath418 and @xmath537 we have @xmath538 . let @xmath6 be an @xmath0-ary function in @xmath539 , and let @xmath540 . we need to show that there exists an @xmath0-ary function @xmath541 such that @xmath542 . since @xmath543 , for every @xmath544 we can find a function @xmath545 such that @xmath546 . now let us consider the following function @xmath151 of arity @xmath547:@xmath548{ll}0 , & \text{if } \left\vert \left\ { i : x_{i}=1\right\ } \right\vert \leq1;\\ 1 , & \text{otherwise.}\end{array } \right.\ ] ] it is easy to see that @xmath549 , therefore @xmath550 , and @xmath542 . this proves that @xmath551 . [ prop w_k]let @xmath24 be an equational class such that @xmath405 = w^{k}$ ] . then@xmath552 @xmath464 : let @xmath24 be an idempotent such that @xmath226 = w^{k}$ ] . lemma [ lemma ize ] shows that @xmath428 . for any @xmath553 we have @xmath554 by proposition [ prop ck ] , and using the previous lemma we get @xmath555 . the reversed containment @xmath556 is obvious , so we have @xmath557 in light of ( [ eq z_2 ... z_infty ] ) . @xmath466 : we just need to modify slightly the second part of the proof of proposition [ prop w_infty ] . let us suppose that @xmath24 is an equational class such that @xmath405 = w^{k}$ ] and @xmath24 is closed under the operator @xmath558 . let @xmath26 be @xmath0-ary , and @xmath183 be @xmath38-ary functions . we have to prove that @xmath144 belongs to @xmath559 . if @xmath560 , then the vectors @xmath561 belong to @xmath413 for @xmath562 . since @xmath416 , there exists an index @xmath467 such that @xmath563 for @xmath562 , i.e. , @xmath564 . this shows that @xmath565 . our next task is , just like in the previous section , to describe @xmath566 , the bottom element of the interval @xmath308 . the proof is very similar to the proof of proposition [ prop bottom w_infty ] . [ prop bottom w_k]the bottom element of the interval @xmath410 is @xmath567 , which is strongly defined by the relational constraint@xmath568 let @xmath569 denote the function class strongly defined by the above constraint . we will prove that@xmath570 the first containment follows immediately from the fact that @xmath571 , and for the second one it suffices to verify that @xmath572 , since @xmath569 is an idempotent according to theorem [ thm galois connection ] . for the third containment , let @xmath6 be an arbitrary @xmath0-ary function in @xmath569 , and @xmath123 be an at most @xmath2-element subset of @xmath490 . since @xmath6 preserves the relation @xmath489 , viewing @xmath123 as a matrix , it has a constant @xmath12 column . moreover , since @xmath6 satisfies the constraint @xmath573 , the matrix @xmath123 has a constant @xmath93 column as well . we can suppose without loss of generality that the first column of @xmath123 is constant @xmath93 and the second column is constant @xmath12 . since @xmath574 = w^{k}$ ] , there is at least one nonmonotone function in @xmath575 , and taking into account that @xmath575 is closed under the operator @xmath426 , we can apply the same argument as in the proof of proposition [ prop bottom w_infty ] to construct a function @xmath576 of arity @xmath0 , such that @xmath506 . then we have @xmath577 , and since we can construct such a function @xmath578 for any at most @xmath2-element subset @xmath123 of @xmath490 , we can conclude that @xmath579 . the previous two propositions yield the following characterization of the idempotents corresponding to @xmath106 . [ thm w_k]for any class @xmath24 of boolean functions the following conditions are equivalent : 1 . @xmath580 ; 2 . @xmath581 , and @xmath24 is an equational class satisfying @xmath582 . finally , we prove that the intervals @xmath308 are finite , but their sizes do not have a common upper bound . for any @xmath417 , the interval @xmath308 is finite and has at least @xmath583 elements . to obtain the lower bound , let us observe that for any @xmath584 we have @xmath585 , and this containment is proper , since @xmath586 and @xmath587 , where @xmath588 . thus we have the following chain of length @xmath583 in @xmath410 ( see figure [ fig intervals]):@xmath589 [ ptb ] intervals.eps in order to prove the finiteness of @xmath308 , we define the _ skeleton _ of an idempotent class @xmath590 as follows . for every @xmath26 , let us construct all matrices formed by at most @xmath2 rows of the matrix @xmath413 , and delete repeated columns , if there are any . the skeleton of @xmath24 is the collection of all such matrices . every matrix in the skeleton has at most @xmath2 rows , and at most @xmath591 columns , since there are no repeated columns . there are only finitely many such matrices , hence there are only finitely many possible skeletons . therefore , it suffices to prove that different idempotents in @xmath308 have different skeletons . so let us suppose that @xmath592 have the same skeleton @xmath593 . let @xmath176 be any function in @xmath216 , and let @xmath594 be any set with at most @xmath2 elements . deleting repeated columns of the matrix @xmath172 , we obtain a matrix @xmath595 . since @xmath593 is the skeleton of @xmath596 as well , there exists a function @xmath597 and a matrix @xmath181 formed by at most @xmath2 rows of @xmath598 , such that deleting repeated columns of @xmath181 we obtain the same matrix @xmath148 . identifying variables of @xmath599 we can construct a function @xmath600 such that @xmath601 . now adding dummy variables to @xmath500 , we can construct a function @xmath602 such that @xmath603 . since we can do this for any at most @xmath2-element subset @xmath172 of @xmath604 , we can conclude that @xmath605 . this proves that @xmath210 , and a similar argument yields @xmath606 , thus we have @xmath607 . assembling the results of the previous sections , we can draw the lattice of projection - free idempotents as shown in figure [ fig lattice ] . the bottom of the interval @xmath608 is denoted by @xmath609 ; it is the dual of @xmath569 , hence it can be strongly defined by the relational constraint @xmath610 for finite @xmath2 , and by all of these constraints for @xmath465 . to obtain the whole lattice @xmath89 , we have to put together this lattice with the post lattice , as shown schematically in figure [ fig biglattice ] . for the real picture we would have to connect @xmath286 to @xmath13 for each nontrivial interval @xmath280 , but this would make the figure incomprehensibly complex . [ t ] lattice.eps [ t ] biglattice.eps finally , let us mention a few directions for further investigations . our characterization of the intervals @xmath478 and @xmath308 is not explicit , hence a more concrete description would be desirable . in particular , it would be interesting to determine ( at least asymptotically ) the size of @xmath410 . to better understand the structure of @xmath481 , the quasiorder @xmath474 defined in remark [ rem ideal ] should be studied . although the lattice of clones over a base set with at least three elements is not fully described , it may be possible to get some results about composition - closed equational classes over arbitrary finite domains , e.g. , determine minimal and maximal closed classes . the description of @xmath89 obtained in this paper can be regarded as a first step in the study of the semigroup @xmath85 ; for further results in this direction see @xcite . the present project is supported by the national research fund , luxembourg , and cofunded under the marie curie actions of the european commission ( fp7-cofund ) , and supported by the hungarian national foundation for scientific research under grants no . t48809 , k60148 and k77409 . the author would like express his gratitude to miguel couceiro , pter hajnal , erkko lehtonen and the anonymous referee for helpful comments and suggestions . v. g. bodnaruk , l. a. kalunin , v. n. kotov , b. a. romov , _ galois theory for post algebras i - ii , _ ( russian ) , kibernetika ( kiev ) * 3 * ( 1969 ) , 110 ; * 5 * ( 1969 ) , 19 . translated in cybernetics and systems analysis * 3 * ( 1969 ) , 243252 ; * 5 * ( 1969 ) , 531539 .

confinement can modify the dynamics , the thermodynamics and the structural properties of liquid water , the prototypical anomalous liquid . by considering a general anomalous liquid , suitable for globular proteins , colloids or liquid metals , we study by molecular dynamics simulations the effect of a solvophilic structured and a solvophobic unstructured wall on the phases , the crystal nucleation and the dynamics of the fluid . we find that at low temperatures the large density of the solvophilic wall induces a high - density , high - energy structure in the first layer ( `` templating '' effect ) . in turn , the first layer induces a `` molding '' effect on the second layer determining a structure with reduced energy and density , closer to the average density of the system . this low - density , low - energy structure propagates further through the layers by templating effect and can involve all the existing layers at the lowest temperatures investigated . therefore , although the high - density , high - energy structure does not self - reproduce further than the first layer , the structured wall can have a long - range effect thanks to a sequence of templating , molding and templating effects through the layers . we find dynamical slowing down of the solvent near the solvophilic wall but with largely heterogeneous dynamics near the wall due to superdiffusive liquid veins within a frozen matrix of solvent . hence , the partial freezing of the first hydration layer does not correspond necessarily to an effective reduction of the channel section in terms of transport properties .

1406.1996

the study of the properties of liquids confined at the nanometer scale is a topic of high interest for its technological , experimental and theoretical implications . furthermore , confinement plays an important role in hydrated biological systems and organic solvents @xcite . structural , thermodynamical and dynamical properties of a liquid can change near an interface ( solid , liquid , etc . ) @xcite . when the surface - to - volume ratio is large , at least along one direction as for the slit pore geometry , the effect of the confining surfaces has to be taken into account . experiments and simulations on nanoconfined fluids show that molecules arrange in layers parallel to the surface . the effect becomes stronger for decreasing temperature or increasing density , until the fluid eventually solidifies . the nature of the solid , as an amorphous or a crystal , can depend on the interparticle potential and the confinement conditions @xcite . in different cases the role of the interfaces can results in complex behaviors , e.g. a persisting fluid mono - layer around a spherical impurity while the rest of the system is in a polycrystal or glassy state @xcite , or persisting amorphous water mono - layer near an hidrophilic disordered surface @xcite . in a recent experiment @xcite , kaya and coworkers found that a thin - film water on a @xmath0 surface remains in a high density liquid form for temperatures ranging from ambient ( 300k ) to supercooled ( 259k ) . the result is unexpected because , based on thermodynamics arguments @xcite , kaya and coworker would expect that the templating effect of @xmath0 on the structure of the water film should promote the tetrahedral structure of the low - density liquid . moreover , the presence of an interface can promote the heterogeneous nucleation of the crystal . however , recent experiments and simulations based on crystallographic analysis of liquids on crystals @xcite , or on colloidal self - assembly @xcite , show that the traditional theories need to be revised in these cases . confined fluids , under suitable conditions of density and temperature , can spontaneously develop patterns , e.g. stripes , and different mesophases or more complex structures @xcite . the confinement affects also the dynamics of the liquid . it has been found that near an interface there is a reduction of the local diffusivity of the liquid @xcite . computer simulations can help in interpreting the experimental results for nanoconfined fluids that are difficult to understand @xcite . different numerical approaches based on first principles simulations can give detailed informations , but are limited by their high computational cost . classical molecular dynamics of empirical fluid models employ parameters tested for the bulk case that not necessarily hold in confinement . the difficulty to adopt these bulk fluid models to the case of confinement leads to a variety of simulation results concerning the aggregation state of the fluid near the surfaces that in principle are model - dependent . it is , therefore , useful to develop coarse - grained models that allow for analytic calculations @xcite and more efficient simulations @xcite , like isotropic pairwise core - softened potentials , and that could allow us to better understand common features of fluids under confinement and the basic mechanisms of complex phenomena emerging in these systems , like pattern formation , e.g. stripes and different mesophases . this has been confirmed in a recent work @xcite in which the authors claim that the origin of quasi - crystals could be understood in the context of a coarse - grained model by the competing effect of the hard and soft core radius of interacting particles . here we focus on the study of nanoconfined anomalous fluids , relevant for biological and technological applications @xcite , by means of molecular dynamics simulations of a system of identical particles interacting through the continuous shouldered well potential ( csw ) , an isotropic pairwise core - softened potential with a repulsive shoulder and an attractive well @xcite . the csw fluid is confined in a slit pore obtained by a solvophilic structured wall , and a solvophobic wall with no structure , as described in sec.[sec : simulations ] . the csw model is suitable for studying globular proteins in solution @xcite , specific colloids @xcite , and liquid metals @xcite , and displays water - like anomalies @xcite . in particular the csw reproduces density , diffusion and structure anomalies following the water hierarchy @xcite and displays a liquid - gas and a liquid - liquid ( ll ) phase transition , both ending in critical points @xcite . here we focus on structural and dynamical properties of the csw model in confinement and show the role that the characteristic length scales of the inter - particle potential have in the structuring of the fluid . we consider a system of @xmath1 identical particles interacting by means of the csw potential confined between two parallel walls . the csw potential is defined as @xcite @xmath2+\left(\dfrac{a}{r}\right)^{24}\ ] ] where @xmath3 is the diameter of the particles , @xmath4 and @xmath5 are the distance of the attractive minimum and the repulsive radius , respectively , @xmath6 and @xmath7 are the energies of the attractive well and the repulsive shoulder , respectively , @xmath8 is the variance of the gaussian centered in @xmath4 and @xmath9 is the parameter which controls the slope between the shoulder and the well at @xmath5 . the parameters employed are the same as in refs.@xcite : @xmath10 , @xmath11 , @xmath12 , @xmath13 . in order to reduce the computational cost , we impose a cutoff for the potential at a distance @xmath14 . in the present simulations we use @xmath15 that allows to better evidence the anomalies in density , diffusion and structure @xcite . in our simulations we consider a @xmath16 ensemble system composed of @xmath17 particles at fixed temperature @xmath18 and volume @xmath19 . the temperature of the thermal bath @xmath18 is kept constant by rescaling the velocity of the particles at each time step by a factor @xmath20 , where @xmath21 is the instantaneous kinetic temperature ( allen thermostat ) @xcite . pressure , temperature , density and diffusion constant are all expressed in internal units : @xmath22 , @xmath23 , @xmath24 , and @xmath25 , respectively . the equation of motion are integrated by means of the velocity verlet method @xcite , using the time - step @xmath26 defined in units of @xmath27 ( that corresponds to @xmath28 for water - like molecules and to @xmath29 for argon - like atoms @xcite ) . we performed the same check as in ref . @xcite in order to verify that the value used for @xmath30 is small enough to satisfy the energy conservation of the system . the confining parallel walls are placed along the @xmath31 axis at a separation distance @xmath32 . the solvophilic wall is composed of a triangular lattice of csw particles quenched with the position of the centers placed at @xmath33 . the lattice constant is @xmath34 . the solvophobic wall has no structure and is obtained by imposing the repulsive potential @xmath35 where @xmath36 , and @xmath37 @xcite . in the following we consider the parameter @xmath38 . we adopt periodic boundary conditions in the @xmath39 and @xmath40 directions . in order to compute the effective density @xmath41 , we need to compute the effective volume @xmath42 accessible to particles @xmath43 , where @xmath44 is the section of the simulation box and @xmath45 is the effective distance between the plates . by considering the quenched particles forming the philic wall and the repulsive strength of the potential of the phobic wall , we obtain @xmath46 @xcite . the effective density is @xmath47 . to explore different densities for the confined system in the @xmath16 ensemble , we change @xmath48 and @xmath49 . we keep @xmath32 and @xmath1 constant to exclude finite - size effects when we compare results for different densities . for each density , we equilibrate the system by annealing from @xmath50 . for each temperature , the system is equilibrated during @xmath51 time steps . we observe equilibrium after @xmath52 time steps for the range of @xmath53 and @xmath18 considered here . all our results are averaged over @xmath54 independent samples . in order to check the stability of the system , we verify that the energy and the pressure are equilibrated . to compute the pressure we follow the approach of refs.@xcite . due to the inhomogeneous nature of the system , the pressure is a tensor @xmath55 where @xmath56 is the kinetic contribution as in an ideal gas , @xmath57 is the unit tensor , @xmath58 is the potential contribution due to the interparticle interaction , and @xmath59 is the contribution of the walls . at equilibrium the system is mechanically stable if @xmath60 . considering the planar symmetry of the system , the normal ( or orthogonal ) and tangential ( or parallel ) component of the pressure can be written respectively as : @xmath61 and @xmath62 , while mixed components are zero , i.e. , no shear forces are present @xcite . to verify the stability of the system , we computed the normal component of the pressure as a function of @xmath31 @xmath63 the kinetic part is @xmath64 . for the potential part we used the todd , evans and daivis formulation of the pressure @xcite @xmath65\right\rangle\ ] ] where @xmath66 is the z - component of the interaction force between particles @xmath67 and @xmath68 . the products of heaviside step functions , @xmath69 , select couple of particles that lie in different semispace respect to the plane parallel to the walls with coordinate @xmath31 . the walls contribution to the pressure can be computed from eq . [ equ : pressure_pot ] , with @xmath70 for @xmath71 , as @xcite @xmath72\right\rangle\ ] ] where @xmath73 and @xmath74 are the interaction forces of the particle @xmath67 with the philic wall and the phobic wall , respectively . normal pressure profile @xmath75 , for densities @xmath76 and temperatures @xmath77 . the effective densities are : @xmath78 , @xmath79 , @xmath80 , @xmath81 . each component of the normal pressure is displayed . in particular , @xmath82 is the contribution of the philic wall , @xmath83 is the contribution of the phobic wall , @xmath84 is the kinetic contribution as in an ideal gas , @xmath85 is the potential contribution due to the interparticle interaction.,width=604 ] the different components compensate each other in order to keep the normal pressure constant ( fig.[fig : pressures ] ) . in particular , the philic wall contributes with a positive term @xmath86 from @xmath87 to @xmath88 and with a negative term from @xmath88 to @xmath89 , due to the interaction force between the particles of the fluid and those of the philic wall . the phobic wall contributes with a positive term @xmath90 from @xmath91 to @xmath92 due to the repulsive potential . the kinetic pressure @xmath93 is a signature of the density profile , while the potential pressure @xmath94 can contribute with a positive or negative term depending which part of the interparticle interaction , repulsive or attractive , respectively , dominate . after having verified that the system is stable , i.e. , @xmath95 , we compute the normal pressure as @xmath96 where @xmath97 and @xmath98 are z - coordinates sufficiently near the philic wall and the phobic wall , respectively , for which the kinetic @xmath99 and the potential @xmath100 part of the pressure are zero . as pointed out in the introduction ( sec.[sec : introduction ] ) , the confinement can modify the structure of a fluid resulting in an inhomogeneous density profile . in a slit pore geometry , near the confining walls , particles form layers parallel to the walls as the temperature is decreased or the density is increased , as shown by our calculations for the density profile @xmath101 ( fig . [ fig : density ] ) . to establish the aggregation state for each layer , we compute , layer by layer , the lateral radial distribution function @xmath102 ; the 2d voronoi tessellation of each layer ; the mean square displacement ( msd ) and the survival probability ( sp ) of molecules in each layer . all these quantities together , as we discusse in the following , allow us to identify the presence and coesistence of the solid , heterogeneous fluid and homogeneous fluid . our analysis shows that the system organizes forming layers parallel to the solvophilic wall at any temperature @xmath18 and average density @xmath53 . at high @xmath18 and low @xmath53 we find only one well defined layer and no layering near the solvophobic wall , while the whole system is in the fluid state . by decreasing @xmath18 and increasing @xmath53 , the number of layers increases up to eleven . furthermore , at higher value of @xmath53 , the layers appear also near the solvophobic wall . however , for high enough @xmath18 and @xmath53 we observe that away from the walls the system is in the fluid state . nonetheless , the walls affect the density profile , changing system density and aggregation state , over an extension that in our case can be up to @xmath103 layers that depends on temperature and average density of the system . the in - layer radial distribution function @xmath104 for the @xmath105-th layer ( with @xmath106 ) is computed as @xmath107\ ] ] where @xmath108 and @xmath109 are the density of particles and the z - coordinate of the layer @xmath105 , respectively , @xmath110 is the transverse distance between two particles in the same layer ( and in internal units is @xmath111 ) . the heaviside step functions , @xmath69 , select couple of particles that lie in the layer @xmath105 of width @xmath112 . the @xmath104 is proportional to the probability of finding a molecule in the layer @xmath105 at a distance @xmath113 from a randomly chosen molecule of the same layer @xmath105 . the definition of the layer @xmath105 in which lies a particle @xmath67 is once for all established according to the value of the z - coordinate of the particle @xmath67 as : for the first layer @xmath114 , with @xmath115 , and @xmath116 for the others . this is a natural choice because particles at low temperatures or high densities tend to stratify in layers whose interdistance is approximately equal to @xmath117 . at low density ( @xmath118 , fig.[fig : gr_rho011 ] ) , we observe that the system is in a fluid state for high temperature ( @xmath119 ) in any layer . the @xmath104 of the layer @xmath120 shows a first peak around the shoulder radius ( @xmath121 ) and a second peak around the attractive well radius ( @xmath122 ) , while for the other layers only the second peak is present . we interpret this difference between the first and the other layers as the consequence of a `` templating '' effect of the solvophilic wall that at high @xmath123 is observed only on the first layer . at @xmath124 the system shows the same behavior observed for higher temperatures , except that the layer near the solvophilic wall develops patterns . this behavior is reminiscent of what has been observed in monolayers with an interparticle potential composed by a hard core and a soft repulsive shoulder @xcite . at @xmath125 the layer near the solvophilic wall is still showing patterns , while the layer @xmath126 is forming crystal patches . this is evident from the analysis of the @xmath127 that goes to zero for @xmath128 and @xmath129 at @xmath125 , consistent with an incipient triangular crystal with lattice step given by the interaction potential attractive distance @xmath130 . the other layers are in a fluid state . at @xmath131 the first layer is forming a hexagonal crystal . although the hexagonal crystal in @xmath120 does not overlap exactly with the triangular wall structure , the comparison of the @xmath132 and @xmath133 of the wall shows a strong correlation between the two structures , suggesting a `` templating '' effect . this effect due to the attraction to the solvophilic wall is so strong at @xmath131 that is forcing particles to be at their repulsive distance . the high - energy cost of the resulting honeycomb lattice forming in the layer @xmath120 is compensated by the large number of attractive interactions between the particles at @xmath120 and those of the wall ( @xmath134 ) from one hand , and between the particles themeselves at @xmath120 from the other hand . this free energy minimization process is analyzed in sec.[sec : structural ] in the discreet potential approximation to understand the stripe phase formation . the triangular structure that was incipient for @xmath126 at high @xmath123 , for @xmath131 is well defined for @xmath126 and @xmath135 , with defects in the layer @xmath135 . this triangular structure is the dual lattice of the @xmath120 hexagonal layer and its formation is the consequence of a `` molding '' effect of the layer @xmath120 onto the layer @xmath126 . note that while the wall ( @xmath134 ) layer has a templating effect on the @xmath120 layer , the @xmath120 layer has a molding effect on the @xmath126 layer . the difference between the two cases is due to the smaller density of the @xmath120 layer with respect to that of the wall . the smaller density does not allow to compensate the high energy cost of the propagation of the hexagonal crystal to the layer @xmath126 . on the other hand , the triangular crystal of the @xmath126 layer is energetically favorable , because the particles are all at the attractive distance , and at this temperature can propagate to the @xmath135 layer again with a `` templating '' effect . the layer @xmath136 is made of a few triangular crystallites immersed in the fluid , while the other layers are in a fluid state . at @xmath137 both the templating and the molding effect are stronger . in particular the template of the @xmath126 layer propagates over all the six layers that are formed at this density and temperature . at intermediate density ( @xmath138 , fig.[fig : gr_rho022 ] ) , for @xmath119 and @xmath124 we observe the same qualitative behavior as for the low density case . for @xmath125 the layer @xmath120 has less tendency to form patterns respect to the low density case , and the layer @xmath126 to order in a crystal structure . therefore , the confined system is more fluid at this density than at lower density . we understand this result as a consequence of the larger hydration at higer density . at @xmath131 the first layer has partially crystallized in the hexagonal and partially in the triangular structure following the template of the wall . therefore , the templating effect is now stronger then the corresponding case at lower density . the hexagonal crystal shows now a preferred direction of symmetry . this direction propagates to the layer @xmath126 , where we observe stripes along the preferred direction . the stripes propagate up to @xmath136 layer , while the other layers are in a fluid state . the peak of @xmath104 at @xmath139 ( that corresponds to the average second nearest neighbor distance ) is a signature of the stripe phase formation . at @xmath137 the preferred direction in the deformation of the hexagonal crystal for @xmath120 is more evident and we observe a clear stripe phase for the layers from @xmath126 to @xmath136 , with a peak of @xmath104 at @xmath139 more pronounced than the case at @xmath131 . the other layers form a triangular crystal at the attractive distance . at high density ( @xmath140 , fig.[fig : gr_rho030 ] ) , for @xmath119 we observe the same qualitative behavior as for the lower density cases . at @xmath124 we found that the only difference with the lower density case is that the layer @xmath120 is forming crystallites following the template of the wall . at @xmath125 the layer @xmath120 has a different and incipient crystal structure ( kagome lattice ) with defects that is better defined at lower @xmath18 . this is evident from the analysis of the @xmath132 that goes to zero for @xmath121 and @xmath141 at @xmath125 . the layer @xmath126 shows patterns very close to the stripe configuration . the corresponding @xmath127 goes to zero for @xmath142 and shows a peak for @xmath139 . these characteristics of the @xmath143 are typical of a stripe phase . the layer @xmath135 and @xmath136 still show patterns close to the stripe phase , but in a less pronounced way . from the layer @xmath144 to the @xmath145 the pattern is vanishing . the layer @xmath146 is showing an incipient triangular crystal with lattice step given by the interaction potential attractive distance @xmath130 . this is evident from the analysis of the @xmath147 that approaches zero for @xmath128 and @xmath129 at @xmath125 . at @xmath131 the layer @xmath120 is forming a kagome crystal with defects . the layers from @xmath126 to @xmath145 show a stripe phase and the layer @xmath146 is forming a triangular crystal with defects . at @xmath137 the kagome crystal of layer @xmath120 has no defects . the layers from @xmath126 to @xmath145 are in a stripe phase and the layer @xmath146 is forming a well defined triangular crystal . as discussed above , the layer @xmath120 close to the solvophilic wall ( @xmath134 ) is subjects to the templating effect for all densities at low temperatures . in fig.[fig : layers_gr_snap ] we compare the @xmath148 and the snapshots of the first layer for @xmath137 at several densities . in order to compare layers , that correspond to different densities , between them , we considered a portion of each layer of the same size ( @xmath48x@xmath49 of the system at @xmath140 ) . for densities @xmath149 the first layer is forming a distorted hexagonal lattice characterized by a @xmath148 with a first peak at @xmath150 and vanishing for @xmath121 and @xmath141 . by increasing @xmath53 we observe a progressive shift to higher values of @xmath113 of all the peaks of @xmath148 , but the first that , instead , is becoming more pronounced as a consequence of a better local order . for densities between @xmath151 and @xmath152 the first layer shows a polycrystal phase with coexistence of triangular and square lattices . this corresponds to an intermediate stage toward the well defined kagome lattice that is formed for @xmath153 , as showed by the splitting of the second peak of @xmath148 into two close peaks at @xmath154 and @xmath155 . in order to characterize space - dependent diffusion properties of our system , we compute the mean square displacement ( msd ) associated to each layer of the slit . we observe that , except for low temperatures , a particle can visit different layers in which the aggregation state can change from homogeneous to heterogeneous liquid and vice versa . for this reason we calculate the msd only for those particles that remain in a layer over the entire time interval under consideration and we average over all possible time interval . therefore , the msd associated to each layer @xmath105 is defined as @xmath156 where @xmath157 is the time spent in the layer @xmath105 by a particle that entered in the layer at time @xmath158 . in according to the standard definition of the msd , @xmath159 where @xmath160 is the lateral , or parallel , diffusion coefficient and @xmath161 the diffusion exponent . the value @xmath162 means that the system is arrested , as in a solid state where particles can only vibrate around theirs equilibrium positions ; for @xmath163 the system is subdiffusive corresponding in general to particles diffusing in complex structures ( with non trivial microscopic disorder ) ; @xmath164 is the standard diffusive behavior as in a normal fluid state . for @xmath165 the system is superdiffusive . on the other hand , for early times free diffusion we expect the ballistic regime with @xmath166 . our analysis ( fig.[fig : diff_layers ] ) shows that the in - layer msd has always a ballistic regime for @xmath167 . the corresponding mean displacement is approximately half particle diameter at high @xmath18 and low @xmath53 and weakly decreases for increasing @xmath53 and decreasing @xmath18 corresponding to the expected decrease of the mean free path of the particles . for @xmath168 all the layers reach the diffusive ( @xmath164 ) behavior for long times . by decreasing the temperature the behavior of the layers becomes more heterogeneous . in particular , we observe that the layer @xmath120 near to the solvophilic wall slows down in a sensible way with respect to the layers at @xmath169 and becomes arrested for @xmath170 . at these temperatures the other layers , including the one near the solvophobic wall , are diffusive at low densities . however , at @xmath140 and @xmath131 all the layers develop the plateau in the msd typical of glassy dynamics . this behavior is reminiscent of the caging effect in glasses where the plateau in the msd is followed by a diffusive regime . here , instead , at @xmath131 and @xmath140 we observe that for all the layers but the one near the solvophobic wall ( @xmath146 ) , after the plateau , the dynamics enters in a superdiffusive regime with @xmath171 . this effect is related to the presence of defects and of a nonuniform stress field , as discussed in sec.[sec : structural ] . for @xmath137 we observe that all the layers are arrested at @xmath118 . at this low density the slit is only partially filled ( fig.[fig : diff_layers ] ) . at @xmath137 and @xmath172 and @xmath138 also the layer @xmath146 near the solvophobic wall is present and it is characterized by a larger msd with respect to the other layers and by a diffusive regime at long times . the other layers have an arrested dynamics . at @xmath137 and @xmath140 all the layers from @xmath120 to @xmath145 have a superdiffusive regime at long times , while the layer @xmath146 reaches the diffusive log - time regime . however , its msd is smaller than that of the other layer for very long times ( @xmath173 ) . as the time proceeds , the average in eq . [ equ : msd ] for the msd is performed on a decreasing number of particles because some of them can leave the layer . _ a priori _ this reduction of the statistics is not homogeneous , that means that in general there can be a correlation between particles that leave the layer and theirs properties , as theirs velocity components . therefore , for @xmath174 , low enough @xmath53 , and for the most diffusive layers there is a time , @xmath175 , after which the in - layer msd ( fig.[fig : diff_layers ] ) is not well defined . to estimate @xmath175 as function of @xmath18 and @xmath53 for different layers we analyse the population relaxation of particles in each layer . in particular , we compute the survival probability ( sp ) function , @xmath176 , which is the probability that a given particle stay in the layer @xmath67 for a time interval @xmath177 . the sp can be calculated as @xmath178 where @xmath179 is the number of particle in the layer @xmath67 at time @xmath180 and @xmath181 is the number of particle that do not leave the layer @xmath67 during the time interval @xmath182 $ ] . the sp give an indication of the time interval @xmath175 over which the msd is well defined . we observe that @xmath183 has an exponential decay in our simulations ( fig.[fig : stau ] ) . we , therefore , define @xmath175 as the characteristic decay time @xmath184 . this choice is consistent with the observation that the msd in fig.[fig : diff_layers ] is well defined when @xmath185 . we observe that for @xmath186 and all the densities and for @xmath131 and @xmath187 , the sp decay is slower for the layer @xmath120 near the solvophilic wall and becomes faster for the layers away from the two walls . when the layer @xmath146 near the solvophobic wall is present , we observe that it has a decay in sp slower than those layers that are farther away from the wall . at @xmath137 for all densities , and at @xmath131 for @xmath140 , there is no decay in sp , consistent with the crystallization of the layers . the non monotonic behavior of @xmath175 is reported in fig.[fig : tau_layers ] as a function of layers for different densities and temperatures . by comparing fig.[fig : tau_layers ] and fig.[fig : density ] we observe that @xmath175 increases when the layers are more structured in the @xmath31 direction . hence , both walls facilitate the stratification of the fluid , although the structureless phobic wall does it in a less strong way with respect to the structured solvophilic wall . characteristic time decay @xmath175 as a function of fluid layer for density @xmath188 and temperature @xmath77 . for clarity , the points for @xmath124 are shifted up by 100 units.,width=377 ] in sec.[sec : msd ] , analysing the msd layer by layer , we have seen ( fig.[fig : diff_layers ] ) that for high densities and low temperatures , after a plateau , the dynamics can enter in a superdiffusive regime with @xmath171 . in this section we show how this behavior is due to the formation of liquid `` veins '' in such layers . the formation of liquid veins is of particular interest in ice during the freezing of water . recently , experiments and simulations showed the presence of liquid water between nanometer - sized ice crystal @xcite . in polycrystalline systems , the liquid is found along intergranular junctions , as grain boundaries ( see @xcite and references therein for the case of water ) . residual stress in these polycrystal structures can be localized along integranular junctions , and can results in an effective force that acts on fluid particles present in these junctions . the origin of the residual stress in our system is due to the fact that when the fluid solidifies as the temperature is decreased , the minimization process of the free energy take place locally , instead of globally . in glass forming liquids , this effect is caused by a fast cooling , while in our system it is due to the layering of the fluid caused by the confinement . in fig.[fig : snap_z1 - 2 - 3_rand1 - 2 - 3 ] we show the spatial configuration of the first three layers of the system close to the philic wall , for three different runs ( i.e. for three different realization of initial conditions ) at @xmath137 and @xmath140 . we observe that particles in the first layer ( @xmath120 ) are characterized by the same msd , while particles in other layers ( @xmath189 ) can have different msd . in particular in some configurations , we observe veins with mobility higher than the rest of the system ( fig.[fig : config_t_z2-z3 ] ) . we analyzed the trajectories of the particles of these specific realizations of the system ( fig.[fig : config_t_z2-z3 ] ) . for these cases we find that these particles with a msd higher than the majority belong to the same stripe and diffuse along the stripe itself . we observe that the majority of particles in the layer @xmath126 ( and in a less evident way for the layer @xmath135 ( fig.[fig : config_t_z2-z3]b ) , remain spatially localized during the entire simulation , except those belonging to two stripes moving in the same direction as along stripe veins ( fig.[fig : config_t_z2-z3]a ) . we observe a similar situation for the layer @xmath135 , but here all the particles are more mobile and the particles in the veins move in opposite directions . further analysis , that goes beyond the goals of the present work , is necessary to understand the effect of the vicinity of the solvophilic wall and if the veins are related to point - like defects as seems to be suggested by fig.[fig : config_t_z2-z3 ] . spatial configuration of the first three layers of the system close to the solvophilic wall , for three different runs ( i.e. for three different realization of the initial conditions ) at @xmath137 and @xmath140 . the size of circles , representing particle positions , are chosen to be equal to the particles hard core diameter @xmath3.,width=415 ] single - particle msd for layers and runs that are in fig.[fig : snap_z1 - 2 - 3_rand1 - 2 - 3 ] . for run # 1 , red and green colors are used for those particles belonging to veins performing a dynamics different from the rest of the particles in the layer.,width=226 ] in - layer trajectories for particles in the second ( a ) and third ( b ) layer of run # 1 in fig.[fig : snap_z1 - 2 - 3_rand1 - 2 - 3 ] . in red and green we show the trajectories of representative particles belonging to stripes veins . the msd of these particles is represented with the same color code in fig.[fig : diff_z1 - 2 - 3_rand1 - 2 - 3 ] . we apply periodic boundary conditions for @xmath190 where @xmath191 ( dashed lines ) and for @xmath192 where @xmath193.,title="fig:",width=94 ] in - layer trajectories for particles in the second ( a ) and third ( b ) layer of run # 1 in fig.[fig : snap_z1 - 2 - 3_rand1 - 2 - 3 ] . in red and green we show the trajectories of representative particles belonging to stripes veins . the msd of these particles is represented with the same color code in fig.[fig : diff_z1 - 2 - 3_rand1 - 2 - 3 ] . we apply periodic boundary conditions for @xmath190 where @xmath191 ( dashed lines ) and for @xmath192 where @xmath193.,title="fig:",width=94 ] our msd and sp analysis show that at low @xmath18 and high @xmath53 there are layers taht behave as a solid . however , by looking only at the msd and sp is not possible to establish if a solid layer is in an amorphous , crystal or polycrystal state @xcite . in order to better understand the structure of solid layers , we computed the standard 2d voronoi tessellation ( useful to identify defects present in the crystal structures , as vacancies , frenkel - like , dislocations and grain boundaries ) , and a modified version of it ( suitable to identify distorted crystal structures ) . with this analysis we can also disentagle the role that the three relevant length scales ( the diameter of the particles @xmath3 , the repulsive radius @xmath5 , and the attractive minimum @xmath4 ) , giving rise to two competing length scale @xmath194 and @xmath195 , play in the determination of layer s structure . indeed , the interdistance between two adjacent layers is @xmath196 , while when stripes form in a specific layer for intermediate densities , just to consider a specific case , particles within a stripe are compressed at a distance @xmath197 , while the distance between stripes depends on @xmath196 and @xmath198 , as discussed in the last part of this section . in the standard voronoi tessellation we construct polygons centered around particles forming a lattice whose edges are crossed in their middle point by the edges of the voronoi cells . this procedure garantees that each voronoi cell represent the proper volume of each particle . to better visualize the result , we represent voronoi cells having a different number of edges with different colors . to reduce the noise in our analysis we adopt also a modified version of the voronoi tessellation in which we associate a color to a polygon in according to the number of edges of the polygon that have a length @xmath199 larger than that of the average edge lenght calculated over the specific polygon itself . this procedure allows us to better visualize polycrystal structures despite the presence of small lattice deformations . we compute the voronoi tessellation for low density ( @xmath118 ) and high density ( @xmath140 ) at low temperature ( @xmath137 ) and very low temperature ( @xmath200 ) , for three different realizations of initial conditions ( figs.[fig : voronoi_rho0.11]a , b , [ fig : voronoi_rho0.30]a , b , and figs.[fig : voronoi_rho0.11_mod2]a , b , [ fig : voronoi_rho0.30_mod2]a , b in supplementary material ) . the configurations at temperature @xmath200 are obtained by annealing configurations equilibrated at @xmath137 with an annealing rate of @xmath201 . at low density ( @xmath118 ) , the first layer at @xmath137 ( fig.[fig : voronoi_rho0.11_mod2 ] ) is in a frustrated solid state that by annealing toward @xmath200 ( fig.[fig : voronoi_rho0.30_mod2 ] ) becomes a frustrated polycrystal . the very low-@xmath18 polycrystal has point and line defects as grain boundaries dividing a deformed honeycomb lattice ( the deformed green triangles ) from a stripe phase ( the stretched hexagonal cyan polygons ) . for both considered @xmath123 the other layers are organized in a triangular lattice ( where each particle is surrounded by a hexagonal cyan polygon ) . we only observe defects , such as dislocations ( run # 2 in fig.[fig : voronoi_rho0.11_mod2]a and fig.[fig : voronoi_rho0.11]a ) that are not eliminated by annealing ( run # 2 in fig.[fig : voronoi_rho0.11_mod2]b ) . at high density ( @xmath140 ) , the first layer at @xmath137 ( fig.[fig : voronoi_rho0.30_mod2]a ) is in a polycrystal state with defects . at @xmath200 ( fig.[fig : voronoi_rho0.30_mod2]b ) we observe two principal crystal grains : a triangular lattice ( cyan polygons ) and a kagome lattice with defects ( blue rhombouses ) . at @xmath137 , the layers @xmath189 present a zigzagging stripe structure with orientation and angles that can change from run to run . at @xmath200 the stripe structure of these layers becomes more regular . these observations emphasize that the increase of density induces an increase of disorder in the solid layers , propagating from the layer @xmath120 to the other layers and up to the layer @xmath146 . the formation of crystal defects during the annealing , and the fact that they are different for different initial conditions , indicate that the system can not reach easely the global minimum of the free energy landscape , corresponding to the crystal configuration , but is trapped in local minima due to the slowing down of the dynamics and the templating effect of the solvophilic wall . in particular , the mismatch of the wall structure with the bulk crystal structure induces a frustrating effect that is more evident near the wall ( in layers @xmath120 and @xmath126 ) for increasing density . to understand the formation of stripes , we follow the same approach as in refs @xcite to show that , if the principal contribution to the minimization of the free energy comes from the energetic therm , in the discreet potential approximation , under suitable conditions of density and temperature , particles organize in straight or zigzagging stripes . in the discreet potential approximation the energetic terms can be reduced to the soft core ( @xmath7 ) and the attractive well ( @xmath6 ) . as a consequence the energetic contribution coming from the interaction with particles of the adjacent layers is approximately constant when the layer density is fixed . therefore , the energetic cost of stripes formation is determined only by the contribution of the in - layer particle interactions . in particular , if a layer has a triangular structure , as the stable configuration of layers @xmath2022 , ... , 11 at low density ( fig.[fig : voronoi_rho0.11_mod2]a , b ) , then for sufficiently high density the layer will prefer to form stripes . 2d schematic representation of particles composed by an hard core ( in dark blue ) of size @xmath3 , a soft corona ( in cyan ) of size @xmath5 and an attractive external corona ( in green ) that extends uo to the potential cutoff @xmath203 . the equilateral and isosceles triangles represent the unitary cell of the triangular and straight - stripes lattice , respectively . an example of ( maximal ) zigzagging - stripes of the same density of straight - stripes is also shown.,width=264 ] consider our fluid made of particles with a hard core surrounded by a soft corona and an external attractive corona ( fig.[fig : stripes ] ) . if @xmath204 is the lattice constant of the triangular structure at the soft - corona distance , then the density of the layer is @xmath205 , where @xmath206 is the number of particles present in this layer . if we allow the triangular lattice to deform in order to minimize the energy of the layer , the new unit cell will be composed by the isosceles triangle in which one side is equal to @xmath39 and the other two are equals to @xmath40 with @xmath207 the fact that the density does nt change implies that @xmath208 . the energy per particle of the layer is @xmath209u_x+2[(\sqrt{n_l}-1)/\sqrt{n_l}]u_y$ ] . for @xmath210 it becomes @xmath211 , where @xmath212 and @xmath213 are the energy associated to the interaction between the particle along the @xmath39 and @xmath40 side of the triangle respectively . in the discreet potential approximation it is @xmath214 for @xmath215 , @xmath216 for @xmath217 and @xmath218 for @xmath219 ( note that we obtain the same result if instead of @xmath203 we consider any value between @xmath5 and @xmath203 . indeed , the present approach has been applyed to show the stability of stripes cnofiguration for a pure repulsive potential model @xcite ) . the same holds for @xmath213 substituting @xmath39 with @xmath40 . for sufficiently high density , i.e. for @xmath220 or @xmath221 , the energies per particle associated to a layer formed by equilateral or isosceles triangles are @xmath222 or @xmath223 , respectively . therefore , under these conditions , the layer will prefer to form stripes . furthermore , from geometric consideration it is possible to conclude that the zigzagging - stripe lattice can be obtained as a deformation of the straight - stripe lattice without changing the density and keeping the energies per particle @xmath224 @xcite . in general , many different zigzagging stripe lattices are possible all with comparable energy per particle ( fig.[fig : stripes ] ) . considering the stripes that form in the layers @xmath225 for system density @xmath140 and temperature @xmath137 , the resulting average in - layer density is @xmath226 , and @xmath227 . hence , from the previous equation for @xmath228 , we find @xmath229 . in view of all the approximation made , we consider this value consistent with @xmath230 of the distance between the closest particles belonging to two adjacent stripes in the same layer . by considering many layers of a confined anomalous fluid @xcite we show that the effect of the structured solvophilic wall can extend up to the entire slit pore . in particular , we study structural and dynamical properties of a monocomponent anomalous liquid under confinement . the fluid has two characteristic distances and can be considered as a coarse - grained model for globular proteins @xcite , colloidal systems @xcite or , to some extent , liquid metals with water - like anomalies @xcite . we perform molecular dynamics simulations of the fluid in a slit pore with a solvophilic wall and a solvophobic wall . the solvophilic wall has structure while the solvophobic one has no structure . we observe that the molecules organize in an inhomogeneous way , forming layers that are parallel to the surfaces , with higher density near the solvophilic surface with respect to the center of the slit pore . for sufficiently high densities , for which the fluid occupy entirely the pore , we observe an increase of density also close to the solvophobic surface , but in a less prominent way . these results are consistent with experimental and theoretical works for nanoconfined fluids . at low temperature we observe coexistence between the homogeneous liquid , heterogeneous liquid and solid phase of the fluid . the influence of the structured solvophilic surface on the solid layers can extend as far as the sixth hydration layer at low @xmath18 and high @xmath53 . in particular , we find a strong correlation between the structure of the solvophilic surface and that of the first layer suggesting a `` templating '' effect . indeed , the large density of solvophilic surface particles allows the formation of a first layer at high density . the high energy cost of this first layer is compensated by the large number of attractive interactions between the particles of the first layer and those of the surface . further energy gain comes from an extra energy term due to the first in - layer particle interactions . moving further from the wall , we find that the first layer has a `` molding '' effect on the second layer . this is because the density of the first layer is smaller than that of the surface and is not high - enough to propagate its template . nevertheless , the low - density second layer is in condition to template the third layer replicating its structure and inducing a long - range effect that can , eventually , involve the whole system . from the calculation of the mean square displacement and the voronoi tessellation we conclude that at low temperature the first layer close to the solvophilic surface is a polycrystal with two competing phases that generate low - energy states with high degeneracy and very slow dynamics . at low densities the two competing phases are stripes and honeycomb lattice , while at high densities are triangular and kagome lattice . we understand this result as a consequence of the high density ( triangular ) structure of the solvophilic wall with a lattice step that corresponds to the hard repulsive distance of the solvent , and the strong wall - solvent attractive interaction . these properties of the wall generate in the first solvent layer local regions with density and energy that are higher than the average of the layer . as a consequence , other regions within the layer have density and energy below the average , giving rise to a competing crystal structure . our results remind us of some recent experiments and simulations for a thin - film of water on @xmath0 surface for which the authors found a very high density first interfacial layer for all temperatures , while they would expect , from thermodynamic arguments , a lower density liquid at supercooled conditions @xcite . in other recent experiments and simulations @xcite , the authors pointed out that it is necessary to revise the theory of heterogeneous nucleation when the crystalization induces a non - zero entropy at zero temperature and the system initially is far from equilibrium . apart from the layers close to the two surfaces , we observe that the structure of each layer mainly depends on its density . in the case of the stripe phase , using simple geometrical and energetic considerations , we find that straight and zigzagging stripes are the stable configurations for intermediate densities . furtheremore , analysing the mean square displacement layer by layer , we observe layers at high densities and low temperatures with a caging - like behavior characterized by a ballistic dynamics followed by an arrested state ( plateau ) and a superdiffusive regime with a diffusion exponent @xmath171 . our analysis shows that this behavior is due to the formation of liquid veins within the stripe phase . in particular we observe that each vein can behave differently from the others diffusing in one of the two possible directions along the stripes and having a different diffusion exponent @xmath171 . we rationalize the different possible values of @xmath161 as a consequence of the presence of residual stress that could introduce an effective force acting on the fluid . under suitable conditions , e.g. a constant effective force along the stripe , the particles in the vein could perform a biased one - dimensional random walk characterized by an exponent of the msd that approaches the ballistic value ( @xmath166 ) . the behavior of these veins can be analyzed in a more quantitative way by computing , for example , the temporal autocorrelation function , the intermediate scattering function @xcite , the relative displacement of nearest neighbors or the particles displacements following the lindemann criterion @xcite . we will present this analysis in future works . our results show that the dynamical slowing down of the anomalous solvent near the solvophilic wall does not imply by necessity the complete freezing of the first hydration layers , because at low @xmath18 and high @xmath53 we observe largely heterogeneous dynamics in three layers with the formation of liquid veins within a frozen matrix of solvent . therefore , under these considerations the partial freezing of the first hydration layer does not correspond necessarily to an effective reduction of the channel section in terms of transport properties , at variance with the conclusions of ref.@xcite . we thank y. shokef for useful discussion on stripes formation . 12 m .- c . bellissent - funel , s . h. chen , and j .- zanotti , physical review e * 51 * , 4558 ( 1995 ) . s. martin and d. dennis j. , j. chem . phys . * 109 * , 5596 ( 1998 ) . m. r. jay , j. sony , n. r. aluru , and e. jakobsson , nano letters * 3 * , 589 ( 2003 ) . j. mittal , t. m. truskett , j. r. errington , and g. hummer , phys . lett . * 100 * , 145901 ( 2008 ) . g. cicero , j. c. grossman , e. schwegler , f. gygi , and g. galli , journal of the american chemical society * 130 * , 1871 ( 2008 ) . a. de virgiliis , r . l. c. vink , j. horbach , and k. binder , phys . e * 78 * , 041604 ( 2008 ) . f. mallamace , c. corsaro , m. broccio , c. branca , n. gonzlez - segredo , j. spooren , s .- h . chen , and h. e. stanley , proc . usa * 105 * , 12725 ( 2008 ) . n. giovambattista , p. j. rossky , and p. g. debenedetti , j. phys . b * 113 * , 13723 ( 2009 ) . r. mancinelli , s. imberti , a. k. soper , k. h. liu , c. y. mou , f. bruni , and m. a. ricci , j. phys . b * 113 * , 16169 ( 2009 ) . p. gallo , m. rovere , and s .- h . chen , journal of physics : condensed matter * 22 * , 284102 ( 2010 ) . w. rzysko , a. patrykiejew , s. sokolowski , and o. pizio , j. chem . phys . * 132 * , 164702 ( 2010 ) . s. han , m. y. choi , p. kumar , and h. e. stanley , nat . phys . * 6 * , 685 ( 2010 ) . t. m. truskett , p. g. debenedetti , and s. torquato , j. chem . phys . * 114 * , 2401 ( 2001 ) . f. de los santos and g. franzese , j. phys . b * 115 * , 14311 ( 2011 ) . s. k. schnell , t . j. h. vlugt , j .- simon , d. bedeaux , and s. kjelstrup , chem . * 504 * , 199 ( 2011 ) . s. k. schnell , t. j. h. vlugt , j .- m . simon , d. bedeaux , and s. kjelstrup , molecular physics * 110 * , 1069 ( 2011 ) . r. r. nair , h. a. wu , p. n. jayaram , i. v. grigorieva , and a. k. geim , science * 335 * , 442 ( 2012 ) . d. r. paul , science * 335 * , 413 ( 2012 ) . a. l. ferguson , n. giovambattista , p. j. rossky , a. z. panagiotopoulos , and p. g. debenedetti , j. chem . phys . * 137 * , 144501 ( 2012 ) . m. c. stewart , and r. evans , phys . e * 86 * , 031601 ( 2012 ) . l. b. krott , and m. c. barbosa , j. chem . phys . * 138 * , 084505 ( 2013 ) . g. schir , a. cupane , e. vitrano , and f. bruni , j. phys . b * 113 * , 9606 ( 2009 ) . s. karan , s. samitsu , x. peng , k. kurashima , and i. ichinose , science * 335 * , 444 ( 2012 ) . f. biedermann , m. vendruscolo , o. a. scherman , a. de simone , and w. m. nau , j. am . soc . * 135 ( 39 ) * , 14879 ( 2013 ) . g. franzese , and v. bianco , food biophysics * 8 * , 153 ( 2013 ) . n. giovambattista , and p. j. rossky , and p. g. debenedetti , annual review of physical chemistry * 63 * , 179 ( 2012 ) . k. b. jinesh , and j. w. m frenken , phys . * 101 * , 036101 ( 2008 ) . k. koga , x. c. zeng , and h. tanaka , phys . * 79 * , 5262 ( 1997 ) . r. zangi , and a. e. mark , phys . lett . * 91 * , 025502 ( 2003 ) . j. slovk , k. koga , h. tanaka , and x. c. zeng , phys . e * 60 * , 5833 ( 1999 ) . v. w. a. de villeneuve , r. p. a. dullens , d. g. a. l. aarts , e. groeneveld , j. h. scherff , w. k. kegel , and h. n. w. lekkerkerker , science * 309 * , 1231 ( 2005 ) . d. t. limmer , and d. chandler , j. chem . phys . * 137 * , 044509 ( 2012 ) . s. kaya , d. schlesinger , s. yamamoto , j. t. newberg , h. bluhm , h. ogasawara , t. kendelewicz , g. e. brown , l. g. m. pettersson , and a. nilsson , sci . rep . * 3 * , 1074 ( 2013 ) . v. sadtchenko , p. conrad , and g. e. ewing , j. chem . phys . * 116 * , 4293 ( 2002 ) . m. c. foster , and g. e. ewing , j. chem . phys . * 112 * , 6817 ( 2000 ) . s. j. cox , s. m. kathmann , j. a. purton , m. j. gillan , and a. michaelides , phys . * 14 * , 7944 ( 2012 ) . q. chen , s. c. bae , and s. granick , nature * 469 * , 381 ( 2011 ) . g. malescio , and g. pellicane , nature mater . * 2 * , 97 ( 2003 ) . m. a. glaser , g. m. grason , r. d. kamien , a. komrlj , c. d. santangelo , and p. ziherl , eur . . lett . * 78 * , 46004 ( 2007 ) . y. von hansen , s. gekle , and r. r. netz , phys . lett . * 111 * , 118103 ( 2013 ) . y. zhang , a. faraone , w. a. kamitakahara , k .- h . liu , c .- y . mou , j. b. leo , b. juscelino s. chang , and s .- h . chen , proc . usa * 108 * , 12206 ( 2011 ) . a. k. soper , proc . usa * 108 * , e1192 ( 2011 ) . y. zhang , a. faraone , w. a. kamitakahara , k .- h . liu , c .- y . mou , j. b. leo , b. juscelino s. chang , and s .- h . chen , proc . usa * 108 * , e1193 ( 2011 ) . j. bai , and x. c. zeng , proc . usa * 109 * , 21240 ( 2012 ) . e. a. jagla , phys . e * 58 * , 1478 ( 1998 ) . g. franzese , g. malescio , a. skibinsky , s. v. buldyrev , and h. e. stanley , nature * 409 * , 692 ( 2001 ) . g. franzese , g. malescio , a. skibinsky , s. v. buldyrev , and h. e. stanley , phys . e * 66 * , 051206 ( 2002 ) . a. skibinsky , s. v. buldyrev , g. franzese , g. malescio , and h. e. stanley , phys . e * 69 * , 061206 ( 2004 ) . y. zhenyu , v. s. buldyrev , n. giovambattista and h. e. stanley , phys . lett . * 95 * , 130604 ( 2005 ) . l. xu , p. kumar , s. v. buldyrev , s .- h . chen , p. h. poole , f. sciortino , and h. e. stanley , proc . usa * 102 * , 16558 ( 2005 ) . a. b. de oliveira , p. a. netz , t. colla , and m. c. barbosa , j. chem * 124 * , 084505 ( 2006 ) . a. b. de oliveira , a. n. paulo , t. colla , and m. c. barbosa , j. chem . phys . * 125 * , 124503 ( 2006 ) . g. franzese , and h. e. stanley , j. phys . : condens . matter * 19 * , 205126 ( 2007 ) . b. jr m. ney , e. salcedo , and m. c. barbosa , j chem phys * 131 * , 094504 ( 2009 ) . b. guillot , j. mol . liq . * 101 * , 219 ( 2002 ) . t. dotera , t. oshiro , and p. ziherl , nature * 506 * , 208 ( 2014 ) . z. ronen , journal of physics : condensed matter * 16 * , s5371 ( 2004 ) . d. g. lev , k. e. gubbins , r. radhakrishnan , and m. sliwinska - bartkowiak , reports on progress in physics * 62 * , 1573 ( 1999 ) . g. franzese , journal of molecular liquids * 136 * , 267 ( 2007 ) . a. b. de oliveira , g. franzese , p. a. netz , and m. c. barbosa , j. chem . phys . * 128 * , 064901 ( 2008 ) . p. vilaseca , k. a. dawson , and g. franzese , soft matter * 9 * , 6978 ( 2013 ) . h. lwen , physica a : statistical mechanics and its applications * 235 * , 129 ( 1997 ) . a. h. marcus , and s. a. rice , phys . e * 55 * , 637 ( 1997 ) . p. bladon , and d. frenkel , phys . * 74 * , 2519 ( 1995 ) . t. bryk , s. de panfilis , f. a. gorelli , e. gregoryanz , m. krisch , g. ruocco , m. santoro , t. scopigno , and a. p. seitsonen , phys . lett . * 111 * , 077801 ( 2013 ) . g. malescio , g. franzese , a. skibinsky , s. v. buldyrev , and h. e. stanley , phys . e * 71 * , 061504 ( 2005 ) . p. vilaseca and g. franzese , j. chem . phys . * 133 * , 084507 ( 2010 ) . m. p. allen , and d. j. tildesley , _ computer simulation of liquids _ ( oxford university press , 1989 ) . f. varnik , j. baschnagel , and k. binder , j. chem . phys . * 113 * , 4444 ( 2000 ) . j. s. rowlinson , and b. widom , _ molecular theory of capillarity _ ( clarendon oxford , 1982 ) . b. d. todd , d. j. evans , and p. j. daivis , phys . e * 52 * , 1627 ( 1995 ) . d. banerjee , s. v. bhat , and d. leporini , liquid polymorphism * 152 * , 1 ( 2013 ) . s. j. gerbode , u. agarwal , d. c. ong , c. m. liddell , f. escobedo , and i. cohen , phys . lett . * 105 * , 078301 ( 2010 ) . y. han , y. shokef , a. m. alsayed , p. yunker , t. c. lubensky , and a. g. yodh , nature * 456 * , 7224 ( 2008 ) . y. shokef , a. souslov , and t. c. lubensky , proc . usa * 108 * , 11804 ( 2011 ) . s. t. cui , p. t. cummings , and h. d. cochran , j. chem . phys . * 114 * , 7189 ( 2001 ) . a. jabbarzadeh , p. harrowell , and r. i. tanner , j. chem . phys . * 125 * , 034703 ( 2006 ) . j. r. bordin , j. s. andrade jr . , a. diehl , and m. c. barbosa , j. chem . phys . * 140 * , 194504 ( 2014 ) . a. a. milischuk , v. krewald , and b. m. ladanyi , j. chem . phys . * 136 * , 224704 ( 2012 ) . t. s. ingebrigtsen , and j. r. errington , t. m. truskett , and j. c. dyre , phys . lett . * 111 * , 235901 ( 2013 ) . k. zahn , r. lenke , and g. maret , phys . . lett . * 82 * , 2721 ( 1999 ) . e. g. solveyra , e. llave , v. molinero , g. j. a. a. soler - illia , and d. scherlis , j. phys . chem c * 117 * , 3330 ( 2013 ) .

principal component analysis ( pca ) is a key statistical technique for multivariate data analysis . for large data sets the common approach to pca computation is based on the standard nipals - pca algorithm , which unfortunately suffers from loss of orthogonality , and therefore it s applicability is usually limited to the estimation of the first few components . here we present an algorithm based on gram - schmidt orthogonalization ( called gs - pca ) , which eliminates this shortcoming of nipals - pca . also , we discuss the gpu ( graphics processing unit ) parallel implementation of both nipals - pca and gs - pca algorithms . the numerical results show that the gpu parallel optimized versions , based on cublas ( nvidia ) are substantially faster ( up to 12 times ) than the cpu optimized versions based on cblas ( gnu scientific library ) . institute for biocomplexity and informatics university of calgary 2500 university drive nw , calgary alberta , t2n 1n4 , canada

0811.1081

principal component analysis ( pca ) is one of the most valuable results from applied linear algebra , and probably the most popular method used for compacting higher dimensional data sets into lower dimensional ones for data analysis , visualization , feature extraction , or data compression @xcite . pca provides a statistically optimal way of dimensionality reduction by projecting the data onto a lower - dimensional orthogonal subspace that captures as much of the variation of the data as possible . unfortunately , pca quickly becomes quite expensive to compute for high - dimensional data sets , where both the number of variables and samples is high . therefore , there is a real need in many applications to accelerate the computation speed of pca algorithms . for large data sets , the standard approach is to use an iterative algorithm which computes the components sequentially , and to avoid the global methods which calculate all the components simultaneously . nipals - pca @xcite is the most frequently used iterative algorithm , and often considered as the standard pca algorithm . however , for large data matrices , or matrices that have a high degree of column collinearity , nipals - pca suffers from loss of orthogonality , due to the errors accumulated in each iteration step @xcite . therefore , in practice it is only used to estimate the first few components . here , we address both the speed and orthogonality problems , and we offer new solutions which eliminate these shortcomings of the iterative pca algorithms . we formulate an iterative pca algorithm based on the gram - schmidt re - orthogonalization , which we called gs - pca . this algorithm is stable from the orthogonality point of view , and if necessary , it can be used to calculate the full set of principal components . the speed up issue is tackled with a parallel implementation for graphics processing units ( gpus ) . here , we present the gpu parallel implementation of both nipals - pca and gs - pca algorithms . the numerical results show that the gpu parallel optimized versions , based on cublas ( nvidia ) @xcite , are substantially faster ( up to 12 times ) than the cpu optimized versions based on cblas ( gnu scientific library ) @xcite . in the following description , the dataset to be analyzed is represented by the @xmath0 matrix @xmath1 . each column , @xmath2 , @xmath3 , contains all the observations of one attribute . also , we assume that each column is mean centered , i.e. if @xmath4 are the original vectors then @xmath5 pca transforms the set of input column vectors @xmath6 $ ] into another set of column vectors @xmath7 $ ] , called principal component scores . this transformation has the property that most of the original data s information content ( or most of its variance ) is stored in the first few component scores . this allows reduction of the data to a smaller number of dimensions , with low information loss , simply by discarding the last component scores . each component is a linear combination of the original inputs and each component is orthogonal . this linear transformation of the matrix @xmath1 is specified by a @xmath8 matrix @xmath9 so that the matrix @xmath1 is factorized as : @xmath10 where @xmath9 is known as the loadings matrix . there are several pca algorithms in the literature , namely svd ( singular value decomposition ) and nipals ( nonlinear iterative partial least squares ) , which use the data matrix , and power and evd ( eigenvalue decomposition ) which use the covariance of the data matrix @xcite . svd and evd extract all the principal components simultaneously , while nipals and power calculate them sequentially . unfortunately , the traditional implementation of pca through svd or evd quickly becomes prohibitive for very large data sets . in this case , an approximate solution can be more efficiently obtained using the iterative approach based on the nipals algorithm . the nipals - pca algorithm can be described as following @xcite . in the first step , the initial data @xmath1 is copied into the residual matrix @xmath11 . then , in the next steps the algorithm extracts iteratively one component at a time ( @xmath12 ) by repeated regressions of @xmath13 on scores @xmath14 to obtain improved loads @xmath15 , and of @xmath1 on these @xmath15 to obtain improved scores @xmath14 . after the convergence is achieved , this process is followed by a deflation of the data matrix : @xmath16 the convergence test consists in comparing two successive estimates of the eigenvalue @xmath17 and @xmath18 . if the absolute difference @xmath19 is smaller than some small error @xmath20 then the convergence is achieved and the algorithm proceeds to the deflation step . using the nipals - pca algorithm approach , the decomposition of the data matrix @xmath1 takes the form : @xmath21 where @xmath22 $ ] is the matrix formed using the first @xmath23 scores , @xmath24 $ ] is the matrix of the first @xmath23 loadings , and @xmath11 is the residual matrix . the pseudo - code of the nipals - pca algorithm is given below : @xmath25 for(@xmath26 ) do \ { @xmath27 @xmath28 for(@xmath29 ) do \ { @xmath30 @xmath31 @xmath32 @xmath33 if(@xmath34 ) then break @xmath35 } @xmath36 } return @xmath37 , @xmath9 , @xmath11 a well known shortcoming of the nipals - pca algorithm is the loss of orthogonality @xcite . both , the computed scores @xmath14 and the loadings @xmath15 , are supposed to be orthogonal . however , because of the errors accumulated at each iteration step ( which involves large matrix - vector operations ) this orthogonality is quickly lost and in practice one can compute accurately only the first few components . in order to stabilize the iterative pca computation , from the orthogonality point of view , we propose an algorithm which is based on the gram - schmidt ( gs ) re - orthogonalization process . the classical gs algorithm ( cgs ) recursively constructs a set of orthonormal basis vectors for the subspace spanned by a given set of linearly independent normalized vectors @xcite . it is well known that the cgs algorithm is also numerically unstable due to rounding errors . however , the cgs can be easily stabilized by a small modification obtaining the so called modified gram - schmidt ( mgs ) algorithm @xcite . unfortunately , the mgs algorithm can not be expressed by level-2 blas functions ( matrix - vector operations ) and therefore it requires a substantial amount of global communications , when implemented on a parallel computer @xcite . in contrast , the cgs algorithm can be easily expressed using matrix - vector operations and therefore it is more suitable for parallel implementation . also , the numerical stability of cgs can be achieved by applying it iteratively @xcite . in the proposed gs - pca algorithm the re - orthogonalization correction is applied to both the scores and the loadings at each iteration step . for the pseudo - code formulation of the gs - pca algorithm we prefer to use the truncated svd description , since for @xmath38 the algorithm also returns the full svd decomposition of the input matrix : @xmath39 where @xmath40 and @xmath41 are the first @xmath23 left and respectively rigth eigenvectors , @xmath42 are the corresponding eigenvalues , and @xmath11 is the residual . one can easily show that @xmath43 and @xmath44 . the pseudo - code of the gs - pca algorithm can be formulated as following : @xmath25 for(@xmath26 ) do \ { @xmath45 @xmath46 for(@xmath29 ) do \ { @xmath47 if(@xmath48 ) then \ { @xmath49 @xmath50 } @xmath51 @xmath52 if(@xmath48 ) then \ { @xmath53 @xmath54 } @xmath55 @xmath56 if(@xmath57 ) then break @xmath58 } @xmath59 } @xmath60 @xmath61 return @xmath37 , @xmath9 , @xmath11 ( for pca ) or @xmath62 , @xmath63 , @xmath64 ( for svd ) one can see that in every iteration step , if @xmath48 then both the right ( loads ) and the left ( scores ) eigenvectors are re - orthonormalized . this procedure stabilizes the algorithm but it also increases a little bit the computational effort . however , this effort will be compensated by the efficiency of the parallel implementation . the gs - pca algorithm assures the perfect orthogonality of both the loads and the scores . the errors accumulated in gs - pca are only due to to the desired precision @xmath65 in the estimation of the eigenvalues @xmath66 . also , for @xmath38 the gs - pca algorithm returns a full svd decomposition of the original matrix @xmath1 , with a maximum error @xmath65 for eigenvalues and perfectly orthogonal left / right eigenvectors . the newly developed gpus now include fully programmable processing units that follow a stream programming model and support vectorized single and double precision floating - point operations . for example , the cuda computing environment provides a standard c like language interface to the nvidia gpus @xcite . the computation is distributed into sequential grids , which are organized as a set of thread blocks . the thread blocks are batches of threads that execute together , sharing local memories and synchronizing at specified barriers . an enormous number of blocks , each containing maximum 512 threads , can be launched in parallel in the grid . in our implementation of nipals - pca and gs - pca algorithms we use cublas , a recent parallel implementation of blas , developed by nvidia on top of the cuda programming environment @xcite . cublas library provides functions for : * creating and destroying matrix and vector objects in gpu memory ; * transferring data from cpu mainmemory to gpu memory ; * executing blas on the gpu ; * transferring data from gpu memory back to the cpu mainmemory . blas defines a set of low - level fundamental operations on vectors and matrices which can be used to create optimized higher - level linear algebra functionality . highly efficient implementations of blas exist for most current computer architectures and the specification of blas is widely adopted in the development of high quality linear algebra software , such as the gnu scientific library ( gsl ) @xcite . we have selected gsl cblas , for our host ( cpu ) implementation , due to its portability on various platforms ( windows / linux / osx , intel / amd ) and because it is free and easy to use in combination with gcc ( gnu compiler ) . the gsl library provides a low - level layer which corresponds directly to the c - language blas standard , referred here as cblas , and a higher - level interface for operations on gsl vectors and matrices . the cblas ( gnu scientific library ) and respectively cublas ( nvidia cuda ) implementations of the nipals - pca and gs - pca algorithms require the following level 1 , 2 , 3 blas functions ( see the cblas / cublas programming manuals for definition details ) : cblas ( level 2 ) : gsl_blas_dgemv cublas ( level 2 ) : cublasdgemv * computes in double precision the matrix - vector product and sum : @xmath67 @xmath68 and @xmath69 are double precision scalars , and @xmath70 and @xmath71 are double precision vectors . @xmath72 is a matrix consisting of double precision elements . matrix @xmath72 is stored in column major format . cblas ( level 1 ) : gsl_blas_daxpy cublas ( level 1 ) : cublasdaxpy * computes the double precision sum : @xmath73 multiplies double precision vector @xmath70 by double precision scalar @xmath74 and adds the result to double precision vector @xmath75 . cblas ( level 1 ) : gsl_blas_dnrm2 cublas ( level 1 ) : cublasdnrm2 * computes the euclidean norm of a double precision vector @xmath76 : @xmath77 cblas ( level 3 ) : gsl_blas_dger cublas ( level 3 ) : cublasdger * computes in double precision the matrix - matrix sum : @xmath78 where @xmath68 is a double precision scalar , @xmath70 is an @xmath79 element double precision vector , @xmath75 is an @xmath80 element double precision vector , and @xmath72 is an @xmath0 matrix consisting of double precision elements . matrix @xmath72 is stored in column major format . these are the critical functions / kernels which are efficiently exploited in the parallel cublas implementation . the other involved functions are for vector / matrix memory allocation and vector / matrix accessing , device ( gpu ) initialization , host - device data transfer and error handling . in the cublas implementation , the data space is allocated both on host ( cpu ) mainmemory and on device ( gpu ) memory . after the data is initialized on host it is transferred on device , where the main parallel computation occurs . the results are then transferred back on host memory . the double precision code for cblas and cublas implementations are given in the appendices 1 , 2 , 3 and 4 . these implementations can be easily modified in order to meet the end user s specifications . for example the single precision blas data allocation and vector / matrix accessing functions have the prefix gsl_vector_float , gsl_matrix_float etc . ( see cblas / cublas programming manuals for details ) . for convenience , we have included a timer which measures the time of all main operations involved by the algorithm . the cblas ( gsl ) versions can be compiled using the following commands : .... g++ -o2 nipals_pca.cpp -lgsl -lgslcblas -lm .... .... g++ -o2 gs_pca.cpp -lgsl -lgslcblas -lm .... an obvious requirement for the cublas ( nvidia ) version is the presence of an nvidia gpu installed . the gpu must be minimum gtx260 , gtx280 or tesla c1060 , otherwise there is no support for double precision . also , the nvidia driver , toolkit and sdk must be correctly instaled . in order to compile the cublas ( nvidia ) version we recommend to use the following simple make file : .... # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # build script for the cublas ( nvidia ) implementation # # of the nipals - pca and gs - pca project # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # add source files here # ( comment / uncomment your choice ) executable : = nipals_pca # for nipals - pca # executable : = gs_pca # for gs - pca # c / c++ source files ( compiled with gcc / c++ ) # ( comment / uncomment your choice ) cfiles : = nipals_pca.c # for nipals - pca # cfiles : = gs_pca.c # for gs - pca # additional libraries needed by the project usecublas : = 1 # rules and targets include .. / .. /common / common.mk # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # .... the numerical tests have been carried out on the following system : amd phenom 9950 cpu ( 2.6ghz ) ; xfx gtx280 gpu ; nvidia linux 64-bit driver ( 177.67 ) ; cuda toolkit and sdk 2.0 ; ubuntu linux 64-bit 8.04.1 , gnu scientific library v.1.11 ; compilers : gcc ( gnu ) , nvcc ( nvidia ) . the gpu used is a high end graphics card solution with 240 stream processors and 1 gb ddr3 ram , which supports both single and double precision and it is theoretically capable of 1 tflop computational power . in figure 1 we give the cpu vs gpu execution time as a function of the size of the randomly generated input matrix @xmath1 ( @xmath81 , n = m/2 , k=10 , \varepsilon = 10^{-7}$ ] ) . the time gap between cpu and gpu increases very fast by increasing the size of the input matrix , and the cpu time versus the gpu time reaches a maximum for @xmath82 , where the gpu is about 12 times faster than the cpu . the gs - pca algorithm is only about 5 - 7% slower than the standard niplals - pca algorithm , in both cpu and gpu implementation . these results also show that the gpu performance is dependent on the scale of the problem . thus , in order to exploit efficiently the massive parallelism of gpus and to effectively use the hardware capabilities , the problem itself needs to scale accordingly , such that thousands of threads are defined and used in computation . ( @xmath83 , red = nipals - pca , blue = gs - pca ) . ] in conclusion , we have presented an iterative gs - pca algorithm based on gram - schmidt re - orthogonalization . the gs - pca algorithm assures the perfect orthogonality of both the loads and the scores , and thus totally eliminates the loss of orthogonality present in the standard nipals - pca algorithm . also , we have discussed the gpu parallel implementation of both nipals - pca and gs - pca algorithms . we have shown that the gpu parallel optimized versions , based on cublas ( nvidia ) are substantially faster ( up to 12 times ) than the cpu optimized versions based on cblas ( gnu scientific library ) . the author acknowledge the financial support from ibi and the university of calgary . 99 bjorck , a . , paige , c . , ( 1992 ) loss and recapture of orthogonality in the modified gram - schmidt algorithm , _ siam j. matrix anal . _ , * 13*(1 ) , 176 - 190 . galassi , m . _ et al _ , ( 2006 ) gnu scientific library reference manual - revised second edition ( network theory limited ) . golub , g.h . , ( 1996 ) matrix computations , 3rd ed . ( john hopkins university press , baltimore ) . jackson , j.e . , ( 1991 ) a user s guide to principal components ( john wiley and sons , new york ) . jolliffe , i.t . , ( 2002 ) principal component analysis ( springer , new york , 2002 ) . kramer , r.,(1998 ) chemometric techniques for quantitative analysis ( crc press , new york ) . lingen , f.j . , ( 2000 ) efficient gram - schmidt orthonormalisation on parallel computers , _ commun . numer . meth . _ , * 16 * , 57 - 66 . nvidia , ( 2008 ) cuda compute unified device architecture ( programming guide , cublas library ) . wold , s . , esbensen , k . , geladi , p . , ( 1987 ) principal component analysis , _ chemometrics and intelligent laboratory systems _ , * 2 * , 37 - 52 . c / c++ example for the cblas ( gnu scientific library ) // implementation of the nipals - pca algorithm // m. andrecut ( c ) 2008 // // compile with : g++ -o3 nipals.cpp -lgsl -lgslcblas -lm // includes , system # include < math.h > # include < time.h > // includes , gsl & cblas # include < gsl / gsl_vector.h > # include < gsl / gsl_matrix.h > # include < gsl / gsl_blas.h > // declarations int nipals_gsl(int , int , int , gsl_matrix * , gsl_matrix * , gsl_matrix * ) ; int print_results(int , int , int , gsl_matrix * , gsl_matrix * , gsl_matrix * , gsl_matrix * ) ; // main int main(int argc , char * * argv ) { // pca model : x = tp ' + r // input : x , mxn matrix ( data ) // input : m = number of rows in x // input : n = number of columns in x // input : k = number of components ( k<=n ) // output : t , mxk scores matrix // output : p , nxn loads matrix // output : r , mxn residual matrix int m = 1000 , m ; int n = m/2 , n ; int k = 25 ; printf("\nproblem dimensions : mxn=%dx%d , k=%d\n " , m , n , k ) ; // initialize srand and clock srand(time(null ) ) ; clock_t start = clock ( ) ; double htime ; // initiallize some random test data x gsl_matrix * x = gsl_matrix_alloc(m , n ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { gsl_matrix_set(x , m , n , rand()/(double)rand_max ) ; } } // allocate memory for t , p , r gsl_matrix * t = gsl_matrix_alloc(m , k ) ; gsl_matrix * p = gsl_matrix_alloc(n , k ) ; gsl_matrix * r = gsl_matrix_alloc(m , n ) ; htime = ( ( double)clock()-start)/clocks_per_sec ; printf("\ntime for data allocation : % f\n " , htime ) ; // call the nipals_gsl ( ) function start = clock ( ) ; gsl_matrix_memcpy(r , x ) ; nipals_gsl(m , n , k , t , p , r ) ; htime = ( ( double)clock()-start)/clocks_per_sec ; printf("\n\ntime for nipals - pca computation on host : % f\n " , htime ) ; // the results are in t , p , r print_results(m , n , k , x , t , p , r ) ; // memory clean up and shutdown gsl_matrix_free(r ) ; gsl_matrix_free(p ) ; gsl_matrix_free(t ) ; gsl_matrix_free(x ) ; printf("\npress enter to exit ... \n " ) ; getchar ( ) ; return exit_success ; } int nipals_gsl(int m , int n , int k , gsl_matrix * t , gsl_matrix * p , gsl_matrix * r ) { // pca model : x = tp ' + r // input : x , mxn matrix ( data ) // input : m = number of rows in x // input : n = number of columns in x // input : k = number of components ( k<=n ) // output : t , mxk scores matrix // output : p , nxk loads matrix // output : r , mxn residual matrix ( x is initially copied in r ) // maximum number of iterations int j = 10000 ; // max error double er = 1.0e-7 ; // some useful pointers double * a = ( double*)calloc(1 , sizeof(a ) ) ; double * b = ( double*)calloc(1 , sizeof(b ) ) ; int k , n , j ; // mean center the data gsl_vector * u = gsl_vector_calloc(m ) ; for(n=0 ; n < n ; n++ ) { gsl_blas_daxpy(1.0 , & gsl_matrix_column(r , n).vector , u ) ; } for(n=0 ; n < n ; n++ ) { gsl_blas_daxpy(-1.0/n , u , & gsl_matrix_column(r , n).vector ) ; } for(k=0 ; k < k ; k++ ) { gsl_blas_dcopy(&gsl_matrix_column(r , k).vector , & gsl_matrix_column(t , k).vector ) ; * a = 0.0 ; for(j=0 ; j < j ; j++ ) { gsl_blas_dgemv(cblastrans , 1.0 , r , & gsl_matrix_column(t , k).vector , 0.0 , & gsl_matrix_column(p , k).vector ) ; gsl_blas_dscal(1.0/gsl_blas_dnrm2(&gsl_matrix_column(p , k).vector ) , & gsl_matrix_column(p , k).vector ) ; gsl_blas_dgemv(cblasnotrans , 1.0 , r , & gsl_matrix_column(p , k).vector , 0.0 , & gsl_matrix_column(t , k).vector ) ; * b = gsl_blas_dnrm2(&gsl_matrix_column(t , k).vector ) ; if(fabs(*a - * b ) < er*(*b ) ) break ; * a = * b ; } gsl_blas_dger(-1.0 , & gsl_matrix_column(t , k).vector , & gsl_matrix_column(p , k).vector , r ) ; } // clean up memory free(a ) ; free(b ) ; gsl_vector_free(u ) ; return exit_success ; } int print_results(int m , int n , int k , gsl_matrix * x , gsl_matrix * t , gsl_matrix * p , gsl_matrix * r ) { int m , n ; // if m < 13 print the results on screen if(m > 12 ) return exit_success ; printf("\nx\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , gsl_matrix_get(x , m , n ) ) ; } printf("\n " ) ; } printf("\nt\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(t , m , n ) ) ; } printf("\n " ) ; } gsl_matrix * f = gsl_matrix_alloc(k , k ) ; gsl_blas_dgemm ( cblastrans , cblasnotrans , 1.0 , t , t , 0.0 , f ) ; printf("\nt ' * t\n " ) ; for(m=0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(f , m , n ) ) ; } printf("\n " ) ; } gsl_matrix_free(f ) ; printf("\np\n " ) ; for(m=0 ; m < n ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(p , m , n ) ) ; } printf("\n " ) ; } gsl_matrix * g = gsl_matrix_alloc(k , k ) ; gsl_blas_dgemm ( cblastrans , cblasnotrans , 1.0 , p , p , 0.0 , g ) ; printf("\np ' * p\n " ) ; for(m=0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(g , m , n ) ) ; } printf("\n " ) ; } gsl_matrix_free(g ) ; printf("\nr\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , gsl_matrix_get(r , m , n ) ) ; } printf("\n " ) ; } return exit_success ; } .... .... // c / c++ example for the cublas ( nvidia ) // implementation of nipals - pca algorithm // // m. andrecut ( c ) 2008 // includes , system # include < stdio.h > # include < stdlib.h > # include < string.h > # include < time.h > // includes , cuda # include < cublas.h > // matrix indexing convention # define id(m , n , ld ) ( ( ( n ) * ( ld ) + ( m ) ) ) // declarations int nipals_cublas(int , int , int , double * , double * , double * ) ; int print_results(int , int , int , double * , double * , double * , double * ) ; // main int main(int argc , char * * argv ) { // pca model : x = t * p ' + r // input : x , mxn matrix ( data ) // input : m = number of rows in x // input : n = number of columns in x // input : k = number of components ( k<=n ) // output : t , mxk scores matrix // output : p , nxn loads matrix // output : r , mxn residual matrix int m = 1000 , m ; int n = m/2 , n ; int k = 25 ; printf("\nproblem dimensions : mxn=%dx%d , k=%d\n " , m , n , k ) ; // initialize srand and clock srand(time(null ) ) ; clock_t start = clock ( ) ; double dtime ; // initialize cublas cublasstatus status ; status = cublasinit ( ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! cublas initialization error\n " ) ; return exit_failure ; } // initiallize some random test data x double * x ; x = ( double*)malloc(m*n * sizeof(x[0 ] ) ) ; if(x = = 0 ) { fprintf(stderr , " ! host memory allocation error : x\n " ) ; return exit_failure ; } for(m = 0 ; m < m ; m++ ) { for(n = 0 ; n < n ; n++ ) { x[id(m , n , m ) ] = rand ( ) / ( double)rand_max ; } } // allocate host memory for t , p , r double * t ; t = ( double*)malloc(m*k * sizeof(t[0 ] ) ) ; ; if(t = = 0 ) { fprintf(stderr , " ! host memory allocation error : t\n " ) ; return exit_failure ; } double * p ; p = ( double*)malloc(n*k * sizeof(p[0 ] ) ) ; ; if(p = = 0 ) { fprintf(stderr , " ! host memory allocation error : p\n " ) ; return exit_failure ; } double * r ; r = ( double*)malloc(m*n * sizeof(r[0 ] ) ) ; ; if(r = = 0 ) { fprintf(stderr , " ! host memory allocation error : r\n " ) ; return exit_failure ; } dtime = ( ( double)clock ( ) - start)/clocks_per_sec ; printf("\ntime for data allocation : % f\n " , dtime ) ; // call nipals_cublas ( ) start = clock ( ) ; memcpy(r , x , m*n * sizeof(x[0 ] ) ) ; nipals_cublas(m , n , k , t , p , r ) ; dtime = ( ( double)clock ( ) - start)/clocks_per_sec ; printf("\ntime for nipals - pca computation on device : % f\n " , dtime ) ; print_results(m , n , k , x , t , p , r ) ; // memory clean up free(r ) ; free(p ) ; free(t ) ; free(x ) ; // shutdown status = cublasshutdown ( ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! cublas shutdown error\n " ) ; return exit_failure ; } if(argc < = 1 || strcmp(argv[1 ] , " -noprompt " ) ) { printf("\npress enter to exit ... \n " ) ; getchar ( ) ; } return exit_success ; } int nipals_cublas(int m , int n , int k , double * t , double * p , double * r ) { // pca model : x = t * p ' + r // input : x , mxn matrix ( data ) // input : m = number of rows in x // input : n = number of columns in x ( n<=m ) // input : k = number of components ( k < n ) // output : t , mxk scores matrix // output : p , nxk loads matrix // output : r , mxn residual matrix // cublas error handling cublasstatus status ; // maximum number of iterations int j = 10000 ; // max error double er = 1.0e-7 ; int k , n , j ; // transfer the host matrix x to device matrix dr double * dr = 0 ; status = cublasalloc(m*n , sizeof(dr[0 ] ) , ( void**)&dr ) ; if(status ! = cublas_status_success ) { fprintf ( stderr , " ! device memory allocation error ( dr)\n " ) ; return exit_failure ; } status = cublassetmatrix(m , n , sizeof(r[0 ] ) , r , m , dr , m ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! device access error ( write dr)\n " ) ; return exit_failure ; } // allocate device memory for t , p double * dt = 0 ; status = cublasalloc(m*k , sizeof(dt[0 ] ) , ( void**)&dt ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! device memory allocation error ( dt)\n " ) ; return exit_failure ; } double * dp = 0 ; status = cublasalloc(n*k , sizeof(dp[0 ] ) , ( void**)&dp ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! device memory allocation error ( dp)\n " ) ; return exit_failure ; } // mean center the data double * du = 0 ; status = cublasalloc(m , sizeof(du[0 ] ) , ( void**)&du ) ; if(status ! = cublas_status_success ) { fprintf(stderr , " ! device memory allocation error ( du)\n " ) ; return exit_failure ; } cublasdcopy(m , & dr[0 ] , 1 , du , 1 ) ; for(n=1 ; n < n ; n++ ) { cublasdaxpy(m , 1.0 , & dr[n*m ] , 1 , du , 1 ) ; } for(n=0 ; n < n ; n++ ) { cublasdaxpy(m , -1.0/n , du , 1 , & dr[n*m ] , 1 ) ; } double a , b ; for(k=0 ; k < k ; k++ ) { cublasdcopy(m , & dr[k*m ] , 1 , & dt[k*m ] , 1 ) ; a = 0.0 ; for(j=0 ; j < j ; j++ ) { cublasdgemv('t ' , m , n , 1.0 , dr , m , & dt[k*m ] , 1 , 0.0 , & dp[k*n ] , 1 ) ; cublasdscal(n , 1.0/cublasdnrm2(n , & dp[k*n ] , 1 ) , & dp[k*n ] , 1 ) ; cublasdgemv('n ' , m , n , 1.0 , dr , m , & dp[k*n ] , 1 , 0.0 , & dt[k*m ] , 1 ) ; b = cublasdnrm2(m , & dt[k*m ] , 1 ) ; if(fabs(a - b ) < er*b ) break ; a = b ; } cublasdger(m , n , -1.0 , & dt[k*m ] , 1 , & dp[k*n ] , 1 , dr , m ) ; } // transfer device dt to host t cublasgetmatrix(m , k , sizeof(dt[0 ] ) , dt , m , t , m ) ; // transfer device dp to host p cublasgetmatrix(n , k , sizeof(dp[0 ] ) , dp , n , p , n ) ; // transfer device dr to host r cublasgetmatrix(m , n , sizeof(dr[0 ] ) , dr , m , r , m ) ; // clean up memory status = cublasfree(dp ) ; status = cublasfree(dt ) ; status = cublasfree(dr ) ; return exit_success ; } int print_results(int m , int n , int k , double * x , double * t , double * p , double * r ) { int m , n , k ; // if m < 13 print the results on screen if(m > 12 ) return exit_success ; printf("\nx\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , x[id ( m , n , m ) ] ) ; } printf("\n " ) ; } printf("\nt\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , t[id(m , n , m ) ] ) ; } printf("\n " ) ; } double a ; printf("\nt ' * t\n " ) ; for(m = 0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { a=0 ; for(k=0 ; k < m ; k++ ) { a = a + t[id(k , m , m ) ] * t[id(k , n , m ) ] ; } printf("%+f " , a ) ; } printf("\n " ) ; } printf("\np\n " ) ; for(m=0 ; m < n ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , p[id(m , n , n ) ] ) ; } printf("\n " ) ; } printf("\np ' * p\n " ) ; for(m = 0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { a=0 ; for(k=0 ; k < n ; k++ ) { a = a + p[id(k , m , n ) ] * p[id(k , n , n ) ] ; } printf("%+f " , a ) ; } printf("\n " ) ; } printf("\nr\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , r[id ( m , n , m ) ] ) ; } printf("\n " ) ; } return exit_success ; } .... c / c++ example for the cblas ( gnu scientific library ) // implementation of pca - gs algorithm // // m. andrecut ( c ) 2008 // includes , system # include < math.h > # include < time.h > // includes , gsl & cblas # include < gsl / gsl_vector.h > # include < gsl / gsl_matrix.h > # include < gsl / gsl_blas.h > // declarations int gs_pca_gsl(int , int , int , gsl_matrix * , gsl_matrix * , gsl_matrix * ) ; int print_results(int , int , int , gsl_matrix * , gsl_matrix * , gsl_matrix * , gsl_matrix * ) ; // main int main(int argc , char * * argv ) { // pca model : x = tp ' + r // input : x , mxn matrix ( data ) // input : m = number of rows in x // input : n = number of columns in x // input : k = number of components ( k<=n ) // output : t , mxk scores matrix // output : p , nxk loads matrix // output : r , mxn residual matrix int m = 1000 , m ; int n = m/2 , n ; int k = 10 ; printf("\nproblem dimensions : mxn=%dx%d , k=%d " , m , n , k ) ; // initialize srand and clock srand(time(null ) ) ; clock_t start = clock ( ) ; double htime ; // initiallize some random test data x gsl_matrix * x = gsl_matrix_alloc ( m , n ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { gsl_matrix_set(x , m , n , rand()/(double)rand_max ) ; } } // allocate memory for t , p , r gsl_matrix * t = gsl_matrix_alloc ( m , k ) ; gsl_matrix * p = gsl_matrix_alloc ( n , k ) ; gsl_matrix * r = gsl_matrix_alloc ( m , n ) ; htime = ( ( double)clock()-start)/clocks_per_sec ; printf("\ntime for data allocation : % f\n " , htime ) ; // call gs_pca_gsl start = clock ( ) ; gsl_matrix_memcpy ( r , x ) ; gs_pca_gsl(m , n , k , t , p , r ) ; htime = ( ( double)clock()-start)/clocks_per_sec ; printf("\n\ntime for gs - pca host computation : % f\n " , htime ) ; // the results are in t , p , r print_results(m , n , k , x , t , p , r ) ; // memory clean up and shutdown gsl_matrix_free(r ) ; gsl_matrix_free(p ) ; gsl_matrix_free(t ) ; gsl_matrix_free(x ) ; printf("\npress enter to exit ... \n " ) ; getchar ( ) ; return exit_success ; } int gs_pca_gsl(int m , int n , int k , gsl_matrix * t , gsl_matrix * p , gsl_matrix * r ) { // pca model : x = tlp ' + r // input : x , mxn matrix ( data , copied in r ) // input : m = number of rows in x // input : n = number of columns in x // input : k = number of components ( k<=n ) // output : t , mxk left eigenvectors // output : p , nxk right eigenvectors // output : l , kx1 eigenvalues // output : r , mxn residual // maximum number of iterations int j = 10000 ; // max error double er = 1.0e-7 ; double a ; int n , k , j ; // mean center the data gsl_vector * u = gsl_vector_calloc ( m ) ; for(n=0 ; n < n ; n++ ) { gsl_blas_daxpy(1.0 , & gsl_matrix_column(r , n).vector , u ) ; } for(n=0 ; n < n ; n++ ) { gsl_blas_daxpy(-1.0/n , u , & gsl_matrix_column(r , n).vector ) ; } // allocate memory fo eigenvalues gsl_vector * l = gsl_vector_alloc(k ) ; // gs_pca for(k=0 ; k < k ; k++ ) { gsl_blas_dcopy(&gsl_matrix_column(r , k).vector , & gsl_matrix_column(t , k).vector ) ; a = 0.0 ; for(j=0 ; j < j ; j++ ) { gsl_blas_dgemv(cblastrans , 1.0 , r , & gsl_matrix_column(t , k).vector , 0.0 , & gsl_matrix_column(p , k).vector ) ; if(k>0 ) { gsl_blas_dgemv(cblastrans , 1.0 , & gsl_matrix_submatrix ( p , 0 , 0 , n , k).matrix , & gsl_matrix_column(p , k).vector , 0.0 , & gsl_vector_subvector ( u , 0 , k).vector ) ; gsl_blas_dgemv(cblasnotrans , -1.0 , & gsl_matrix_submatrix ( p , 0 , 0 , n , k).matrix , & gsl_vector_subvector ( u , 0 , k).vector , 1.0 , & gsl_matrix_column(p , k).vector ) ; } gsl_blas_dscal(1.0/gsl_blas_dnrm2(&gsl_matrix_column(p , k).vector ) , & gsl_matrix_column(p , k).vector ) ; gsl_blas_dgemv(cblasnotrans , 1.0 , r , & gsl_matrix_column(p , k).vector , 0.0 , & gsl_matrix_column(t , k).vector ) ; if(k>0 ) { gsl_blas_dgemv(cblastrans , 1.0 , & gsl_matrix_submatrix ( t , 0 , 0 , m , k).matrix , & gsl_matrix_column(t , k).vector , 0.0 , & gsl_vector_subvector ( u , 0 , k).vector ) ; gsl_blas_dgemv(cblasnotrans , -1.0 , & gsl_matrix_submatrix ( t , 0 , 0 , m , k).matrix , & gsl_vector_subvector ( u , 0 , k).vector , 1.0 , & gsl_matrix_column(t , k).vector ) ; } gsl_vector_set(l , k , gsl_blas_dnrm2(&gsl_matrix_column(t , k).vector ) ) ; gsl_blas_dscal(1.0/gsl_vector_get(l , k ) , & gsl_matrix_column(t , k).vector ) ; if(fabs(a - gsl_vector_get(l , k ) ) < er*gsl_vector_get(l , k ) ) break ; a = gsl_vector_get(l , k ) ; } gsl_blas_dger ( -gsl_vector_get(l , k ) , & gsl_matrix_column(t , k).vector , & gsl_matrix_column(p , k).vector , r ) ; } for(k=0 ; k < k ; k++ ) { gsl_blas_dscal ( gsl_vector_get(l , k ) , & gsl_matrix_column(t , k).vector ) ; } // memory clean up gsl_vector_free(l ) ; gsl_vector_free(u ) ; return exit_success ; } int print_results(int m , int n , int k , gsl_matrix * x , gsl_matrix * t , gsl_matrix * p , gsl_matrix * r ) { int m , n ; / * if m < 13 print the results on screen * / if(m > 12 ) return exit_success ; printf("\nx\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , gsl_matrix_get(x , m , n ) ) ; } printf("\n " ) ; } printf("\nt\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(t , m , n ) ) ; } printf("\n " ) ; } gsl_matrix * f = gsl_matrix_alloc ( k , k ) ; gsl_blas_dgemm(cblastrans , cblasnotrans , 1.0 , t , t , 0.0 , f ) ; printf("\nt ' * t\n " ) ; for(m=0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(f , m , n ) ) ; } printf("\n " ) ; } gsl_matrix_free(f ) ; printf("\np\n " ) ; for(m=0 ; m < n ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(p , m , n ) ) ; } printf("\n " ) ; } gsl_matrix * g = gsl_matrix_alloc ( k , k ) ; gsl_blas_dgemm ( cblastrans , cblasnotrans , 1.0 , p , p , 0.0 , g ) ; printf("\np ' * p\n " ) ; for(m=0 ; m < k ; m++ ) { for(n=0 ; n < k ; n++ ) { printf("%+f " , gsl_matrix_get(g , m , n ) ) ; } printf("\n " ) ; } gsl_matrix_free(g ) ; printf("\nr\n " ) ; for(m=0 ; m < m ; m++ ) { for(n=0 ; n < n ; n++ ) { printf("%+f " , gsl_matrix_get(r , m , n ) ) ; } printf("\n " ) ; } return exit_success ; } .... .... status = cublasshutdown ( ) ; if(status ! = cublas_status_success ) { fprintf ( stderr , " ! cublas shutdown error\n " ) ; return exit_failure ; } if(argc < = 1 || strcmp(argv[1 ] , " -noprompt " ) ) { printf("\npress enter to exit ... \n " ) ; getchar ( ) ; } return exit_success ; } double * du = 0 ; status = cublasalloc(m , sizeof(du[0 ] ) , ( void**)&du ) ; if(status ! = cublas_status_success ) { fprintf ( stderr , " ! device memory allocation error ( du)\n " ) ; return exit_failure ; } for(j=0 ; j < j ; j++ ) { cublasdgemv ( ' t ' , m , n , 1.0 , dr , m , & dt[k*m ] , 1 , 0.0 , & dp[k*n ] , 1 ) ; if(k>0 ) { cublasdgemv ( ' t ' , n , k , 1.0 , dp , n , & dp[k*n ] , 1 , 0.0 , du , 1 ) ; cublasdgemv ( ' n ' , n , k , -1.0 , dp , n , du , 1 , 1.0 , & dp[k*n ] , 1 ) ; } cublasdscal ( n , 1.0/cublasdnrm2(n , & dp[k*n ] , 1 ) , & dp[k*n ] , 1 ) ;

two ubiquitous aspects of large - scale data analysis are that the data often have heavy - tailed properties and that diffusion - based or spectral - based methods are often used to identify and extract structure of interest . perhaps surprisingly , popular distribution - independent methods such as those based on the vc dimension fail to provide nontrivial results for even simple learning problems such as binary classification in these two settings . in this paper , we develop distribution - dependent learning methods that can be used to provide dimension - independent sample complexity bounds for the binary classification problem in these two popular settings . in particular , we provide bounds on the sample complexity of maximum margin classifiers when the magnitude of the entries in the feature vector decays according to a power law and also when learning is performed with the so - called diffusion maps kernel . both of these results rely on bounding the annealed entropy of gap - tolerant classifiers in a hilbert space . we provide such a bound , and we demonstrate that our proof technique generalizes to the case when the margin is measured with respect to more general banach space norms . the latter result is of potential interest in cases where modeling the relationship between data elements as a dot product in a hilbert space is too restrictive .

0906.4539

two ubiquitous aspects of large - scale data analysis are that the data often have heavy - tailed properties and that diffusion - based or spectral - based methods are often used to identify and extract structure of interest . in the absence of strong assumptions on the data , popular distribution - independent methods such as those based on the vc dimension fail to provide nontrivial results for even simple learning problems such as binary classification in these two settings . at root , the reason is that in both of these situations the data are formally very high dimensional and that ( without additional regularity assumptions on the data ) there may be a small number of `` very outlying '' data points . in this paper , we develop distribution - dependent learning methods that can be used to provide dimension - independent sample complexity bounds for the maximum margin version of the binary classification problem in these two popular settings . in both cases , we are able to obtain nearly optimal linear classification hyperplanes since the distribution - dependent tools we employ are able to control the aggregate effect of the `` outlying '' data points . in particular , our results will hold even though the data may be infinite - dimensional and unbounded . spectral - based kernels have received a great deal of attention recently in machine learning for data classification , regression , and exploratory data analysis via dimensionality reduction @xcite . consider , for example , laplacian eigenmaps @xcite and the related diffusion maps @xcite . given a graph @xmath0 ( where this graph could be constructed from the data represented as feature vectors , as is common in machine learning , or it could simply be a natural representation of a large social or information network , as is more common in other areas of data analysis ) , let @xmath1 be the eigenfunctions of the normalized laplacian of @xmath2 and let @xmath3 be the corresponding eigenvalues . then , the diffusion map is the following feature map @xmath4 and laplacian eigenmaps is the special case when @xmath5 . in this case , the support of the data distribution is unbounded as the size of the graph increases ; the vc dimension of hyperplane classifiers is @xmath6 ; and thus existing results do not give dimension - independent sample complexity bounds for classification by empirical risk minimization ( erm ) . moreover , it is possible ( and indeed quite common in certain applications ) that on some vertices @xmath7 the eigenfunctions fluctuate wildly even on special classes of graphs , such as random graphs @xmath8 , a non - trivial uniform upper bound stronger than @xmath9 on @xmath10 over all vertices @xmath7 does not appear to be known . even for maximum margin or so - called `` gap - tolerant '' classifiers , defined precisely in section [ sxn : background ] and which are easier to learn than ordinary linear hyperplane classifiers , the existing bounds of vapnik are not independent of the number @xmath11 of nodes . on the vc dimension of gap - tolerant classifiers applied to the diffusion map feature space corresponding to a graph with @xmath11 nodes . ( recall that by lemma [ lem : vap1 ] below , the vc dimension of the space of gap - tolerant classifiers corresponding to a margin @xmath12 , applied to a ball of radius @xmath13 is @xmath14 . ) of course , although this bound is quadratic in the number of nodes , vc theory for ordinary linear classifiers gives an @xmath9 bound . ] a similar problem arises in the seemingly very - different situation that the data exhibit heavy - tailed or power - law behavior . heavy - tailed distributions are probability distributions with tails that are not exponentially bounded @xcite . such distributions can arise via several mechanisms , and they are ubiquitous in applications @xcite . for example , graphs in which the degree sequence decays according to a power law have received a great deal of attention recently . relatedly , such diverse phenomenon as the distribution of packet transmission rates over the internet , the frequency of word use in common text , the populations of cities , the intensities of earthquakes , and the sizes of power outages all have heavy - tailed behavior . although it is common to normalize or preprocess the data to remove the extreme variability in order to apply common data analysis and machine learning algorithms , such extreme variability is a fundamental property of the data in many of these application domains . there are a number of ways to formalize the notion of heavy - tailed behavior for the classification problems we will consider , and in this paper we will consider the case where the _ magnitude _ of the entries decays according to a power law . ( note , though , that in appendix [ sxn : learnht ] , we will , for completeness , consider the case in which the probability that an entry is nonzero decays in a heavy - tailed manner . ) that is , if @xmath15 represents the feature map , then @xmath16 for some absolute constant @xmath17 , with @xmath18 . as in the case with spectral kernels , in this heavy - tailed situation , the support of the data distribution is unbounded as the size of the graph increases , and the vc dimension of hyperplane classifiers is @xmath6 . moreover , although there are a small number of `` most important '' features , they do not `` capture '' most of the `` information '' of the data . thus , when calculating the sample complexity for a classification task for data in which the feature vector has heavy - tailed properties , bounds that do not take into account the distribution are likely to be very weak . in this paper , we develop distribution - dependent bounds for problems in these two settings . clearly , these results are of interest since vc - based arguments fail to provide nontrivial bounds in these two settings , in spite of ubiquity of data with heavy - tailed properties and the widespread use of spectral - based kernels in many applications . more generally , however , these results are of interest since the distribution - dependent bounds underlying them provide insight into how better to deal with heterogeneous data with more realistic noise properties . our first main result provides bounds on classifying data whose magnitude decays in a heavy - tailed manner . in particular , in the following theorem we show that if the weight of the @xmath19 coordinate of random data point is less than @xmath20 for some @xmath21 , then the number of samples needed before a maximum - margin classifier is approximately optimal with high probability is independent of the number of features . [ heavy - tailed data ] [ thm : learn_heavytail ] let the data be heavy - tailed in that the feature vector : @xmath22 satisfy @xmath23 for some absolute constant @xmath17 , with @xmath18 . let @xmath24 denote the riemann zeta function . then , for any @xmath25 , if a maximum margin classifier has a margin @xmath26 , with probability more than @xmath27 , its risk is less than @xmath28 where @xmath29 hides multiplicative polylogarithmic factors . this result follows from a bound on the annealed entropy of gap - tolerant classifiers in a hilbert space that is of independent interest . in addition , it makes important use of the fact that although individual elements of the heavy - tailed feature vector may be large , the vector has bounded moments . our second main result provides bounds on classifying data with spectral kernels . in particular , in the following theorem we give dimension - independent upper bounds on the sample complexity of learning a nearly - optimal maximum margin classifier in the feature space of the diffusion maps . [ spectral kernels ] [ thm : learn_spectral ] let the following diffusion map be given : @xmath30 where @xmath31 are normalized eigenfunctions ( whose @xmath32 ) norm is @xmath33 , @xmath34 being the uniform distribution ) , @xmath35 are the eigenvalues of the corresponding markov chain and @xmath36 . then , for any @xmath25 , if a maximum margin classifier has a margin @xmath26 , with probability more than @xmath27 , its risk is less than @xmath37 where @xmath29 hides multiplicative polylogarithmic factors . as with the proof of our main heavy - tailed learning result , the proof of our main spectral learning result makes essential use of an upper bound on the annealed entropy of gap - tolerant classifiers . in applying it , we make important use of the fact that although individual elements of the feature vector may fluctuate wildly , the norm of the diffusion map feature vector is bounded . as a side remark , note that we are not viewing the feature map in theorem [ thm : learn_spectral ] as necessarily being either a random variable or requiring knowledge of some marginal distribution as might be the case if one is generating points in some space according to some distribution ; then constructing a graph based on nearest neighbors ; and then doing diffusions to construct a feature map . instead , we are thinking of a data graph in which the data are adversarially presented , _ e.g. _ , a given social network is presented , and diffusions and/or a feature map is then constructed . these two theorems provides a dimension - independent ( _ i.e. _ , independent of the size @xmath11 of the graph and the dimension of the feature space ) upper bound on the number of samples needed to learn a maximum margin classifier , under the assumption that a heavy - tailed feature map or the diffusion map kernel of some scale is used as the feature map . as mentioned , both proofs ( described below in sections [ sxn : gaphs_htapp ] and [ sxn : gaphs_skapp ] ) proceed by providing a dimension - independent upper bound on the annealed entropy of gap - tolerant classifiers in the relevant feature space , and then appealing to theorem [ thm : vapnik ] ( in section [ sxn : background ] ) relating the annealed entropy to the generalization error . for this bound on the annealed entropy of these gap - tolerant classifiers , we crucially use the fact that @xmath38 is bounded , even if @xmath39 is unbounded as @xmath40 . that is , although bounds on the individual entries of the feature map do not appear to be known , we crucially use that there exist nontrivial bounds on the magnitude of the feature vectors . since this bound is of more general interest , we describe it separately . the distribution - dependent ideas that underlie our two main results ( in theorems [ thm : learn_heavytail ] and [ thm : learn_spectral ] ) can also be used to bound the sample complexity of a classification task more generally under the assumption that the expected value of a norm of the data is bounded , _ i.e. _ , when the magnitude of the feature vector of the data in some norm has a finite moment . in more detail : * let @xmath41 be a probability measure on a hilbert space @xmath42 , and let @xmath43 . in theorem [ thm : ddb_hs ] ( in section [ sxn : gaphs_annent ] ) , we prove that if @xmath44 , then the annealed entropy of gap - tolerant classifiers ( defined in section [ sxn : background ] ) in @xmath42 can be upper bounded in terms of a function of @xmath45 , @xmath12 , and ( the number of samples ) @xmath25 , independent of the ( possibly infinite ) dimension of @xmath42 . it should be emphasized that the assumption that the expectation of some moment of the norm of the feature vector is bounded is a _ much _ weaker condition than the more common assumption that the largest element is bounded , and thus this result is likely of more general interest in dealing with heterogeneous and noisy data . for example , similar ideas have been applied recently to the problem of bounding the sample complexity of learning smooth cuts on a low - dimensional manifold @xcite . to establish this result , we use a result ( see lemma [ lem : vap1 ] in section [ sxn : gaphs_vcdim ] . ) that the vc dimension of gap - tolerant classifiers in a hilbert space when the margin is @xmath12 over a bounded domain such as a ball of radius @xmath13 is bounded above by @xmath46 . such bounds on the vc dimension of gap - tolerant classifiers have been stated previously by vapnik @xcite . however , in the course of his proof bounding the vc dimension of a gap - tolerant classifier whose margin is @xmath12 over a ball of radius @xmath13 ( see @xcite , page 353 . ) , vapnik states , without further justification , that due to symmetry the set of points in a ball that is extremal in the sense of being the hardest to shatter with gap - tolerant classifiers is the regular simplex . attention has been drawn to this fact by burges ( see @xcite , footnote 20 . ) , who mentions that a rigorous proof of this fact seems to be absent . here , we provide a new proof of the upper bound on the vc dimension of such classifiers without making this assumption . ( see lemma [ lem : vap1 ] in section [ sxn : gaphs_vcdim ] and its proof . ) hush and scovel @xcite provide an alternate proof of vapnik s claim ; it is somewhat different than ours , and they do not extend their proof to banach spaces . the idea underlying our new proof of this result generalizes to the case when the data need not have compact support and where the margin may be measured with respect to more general norms . in particular , we show that the vc dimension of gap - tolerant classifiers with margin @xmath12 in a ball of radius @xmath13 in a banach space of rademacher type @xmath47 $ ] and type constant @xmath48 is bounded above by @xmath49 , and that there exists a banach space of type @xmath50 ( in fact @xmath51 ) for which the vc dimension is bounded below by @xmath52 . ( see lemmas [ lem : banach_ub ] and [ lem : banach_lb ] in section [ sxn : gapbs_vcdim ] . ) using this result , we can also prove bounds for the annealed entropy of gap - tolerant classifiers in a banach space . ( see theorem [ thm : ddb_bs ] in section [ sxn : gapbs_annent ] . ) in addition to being of interest from a theoretical perspective , this result is of potential interest in cases where modeling the relationship between data elements as a dot product in a hilbert space is too restrictive , and thus this may be of interest , _ e.g. _ , when the data are extremely sparse and heavy - tailed . gap - tolerant classifiers see section [ sxn : background ] for more details are useful , at least theoretically , as a means of implementing structural risk minimization ( see , _ e.g. _ , appendix a.2 of @xcite ) . with gap - tolerant classifiers , the margin @xmath53 is fixed before hand , and does not depend on the data . e.g. _ , @xcite . with maximum margin classifiers , on the other hand , the margin is a function of the data . in spite of this difference , the issues that arise in the analysis of these two classifiers are similar . for example , through the fat - shattering dimension , bounds can be obtained for the maximum margin classifier , as shown by shawe - taylor _ et al . _ @xcite . here , we briefly sketch how this is achieved . let @xmath54 be a set of real valued functions . we say that a set of points @xmath55 is @xmath56shattered by @xmath54 if there are real numbers @xmath57 such that for all binary vectors @xmath58 and each @xmath59=\{1,\ldots , s\}$ ] , there is a function @xmath60 satisfying , f_(x_i ) = \ { ll > t_i + , & + < t_i - , & . for each @xmath61 , the _ fat shattering dimension _ @xmath62 of the set @xmath54 is defined to be the size of the largest @xmath56shattered set if this is finite ; otherwise it is declared to be infinity . note that , in this definition , @xmath63 can be different for different @xmath64 , which is not the case in gap - tolerant classifiers . however , one can incorporate this shift into the feature space by a simple construction . we start with the following definition of a banach space of type @xmath50 with type constant @xmath48 . [ banach space , type , and type constant ] [ def : rad ] a _ banach space _ is a complete normed vector space . a banach space @xmath65 is said to have ( rademacher ) type @xmath50 if there exists @xmath66 such that for all @xmath11 and @xmath67 _ t^p _ i=1^n x_i_^p . the smallest @xmath48 for which the above holds with @xmath50 equal to the type , is called the type constant of @xmath65 . given a banach space @xmath68 of type @xmath50 and type constant @xmath48 , let @xmath69 consist of all tuples @xmath70 for @xmath71 and @xmath72 , with the norm @xmath73 noting that if @xmath68 is a banach space of type @xmath50 and type constant @xmath48 ( see sections [ sxn : gapbs_prelim ] and [ sxn : gapbs_vcdim ] ) , one can easily check that @xmath69 is a banach space of type @xmath50 and type constant @xmath74 . in our distribution - specific setting , we can not control the fat - shattering dimension , but we can control the logarithm of the expected value of @xmath75 for any constant @xmath76 by applying theorem [ thm : ddb_bs ] to @xmath69 . as seen from lemma 3.7 and corollary 3.8 of the journal version of @xcite , this is all that is required for obtaining generalization error bounds for maximum margin classification . in the present context , the logarithm of the expected value of the exponential of the fat shattering dimension of linear @xmath33-lipschitz functions on a random data set of size @xmath25 taken i.i.d from @xmath41 on @xmath68 is bounded by the annealed entropy of gap - tolerant classifiers on @xmath69 with respect to the push - forward @xmath77 of the measure @xmath41 under the inclusion @xmath78 . this allows us to state the following theorem , which is an analogue of theorem 4.17 of the journal version of @xcite , adapted using theorem [ thm : ddb_bs ] of this paper . [ thm : margin_bs ] let @xmath79 . suppose inputs are drawn independently according to a distribution @xmath41 be a probability measure on a banach space @xmath68 of type @xmath50 and type constant @xmath48 , and @xmath80 . if we succeed in correctly classifying @xmath25 such inputs by a maximum margin hyperplane of margin @xmath53 , then with confidence @xmath27 the generalization error will be bounded from above by @xmath81 where @xmath29 hides multiplicative polylogarithmic factors involving @xmath82 and @xmath53 . specializing this theorem to a hilbert space , we have the following theorem as a corollary . [ thm : margin_hs ] let @xmath79 . suppose inputs are drawn independently according to a distribution @xmath41 be a probability measure on a hilbert space @xmath42 , and @xmath83 . if we succeed in correctly classifying @xmath25 such inputs by a maximum margin hyperplane with margin @xmath53 , then with confidence @xmath27 the generalization error will be bounded from above by @xmath84 where @xmath29 hides multiplicative polylogarithmic factors involving @xmath85 and @xmath53 . note that theorem [ thm : margin_hs ] is an analogue of theorem 4.17 of the journal version of @xcite , but adapted using theorem [ thm : ddb_hs ] of this paper . in particular , note that this theorem does not assume that the distribution is contained in a ball of some radium @xmath13 , but instead it assumes only that some moment of the distribution is bounded . in the next section , section [ sxn : background ] , we review some technical preliminaries that we will use in our subsequent analysis . then , in section [ sxn : gaphs ] , we state and prove our main result for gap - tolerant learning in a hilbert space , and we show how this result can be used to prove our two main theorems in maximum margin learning . then , in section [ sxn : gapbs ] , we state and prove an extension of our gap - tolerant learning result to the case when the gap is measured with respect to more general banach space norms ; and then , in sections [ sxn : discussion ] and [ sxn : conclusion ] we provide a brief discussion and conclusion . finally , for completeness , in appendix [ sxn : learnht ] , we will provide a bound for exact ( as opposed to maximum margin ) learning in the case in which the probability that an entry is nonzero ( as opposed to the value of that entry ) decays in a heavy - tailed manner . in this paper , we consider the supervised learning problem of binary classification , _ i.e. _ , we consider an input space @xmath86 ( _ e.g. _ , a euclidean space or a hilbert space ) and an output space @xmath87 , where @xmath88 , and where the data consist of pairs @xmath89 that are random variables distributed according to an unknown distribution . we shall assume that for any @xmath90 , there is at most one pair @xmath91 that is observed . we observe @xmath25 i.i.d . pairs @xmath92 sampled according to this unknown distribution , and the goal is to construct a classification function @xmath93 which predicts @xmath87 from @xmath86 with low probability of error . whereas an ordinary linear hyperplane classifier consists of an oriented hyperplane , and points are labeled @xmath94 , depending on which side of the hyperplane they lie , a _ gap - tolerant classifier _ consists of an oriented hyperplane and a margin of thickness @xmath12 in some norm . any point outside the margin is labeled @xmath95 , depending on which side of the hyperplane it falls on , and all points within the margin are declared `` correct , '' without receiving a @xmath95 label . this latter setting has been considered in @xcite ( as a way of implementing structural risk minimization apply empirical risk minimization to a succession of problems , and choose where the gap @xmath12 that gives the minimum risk bound ) . the _ risk _ @xmath96 of a linear hyperplane classifier @xmath97 is the probability that @xmath98 misclassifies a random data point @xmath99 drawn from @xmath41 ; more formally , @xmath100 $ ] . given a set of @xmath25 labeled data points @xmath101 , the _ empirical risk _ @xmath102 of a linear hyperplane classifier @xmath97 is the frequency of misclassification on the empirical data ; more formally , @xmath103 $ ] , where @xmath104 $ ] denotes the indicator of the respective event . the risk and empirical risk for gap - tolerant classifiers are defined in the same manner . note , in particular , that data points labeled as `` correct '' do not contribute to the risk for a gap - tolerant classifier , _ i.e. _ , data points that are on the `` wrong '' side of the hyperplane but that are within the @xmath12 margin are not considered as incorrect and do not contribute to the risk . in classification , the ultimate goal is to find a classifier that minimizes the true risk , _ i.e. _ , @xmath105 , but since the true risk @xmath96 of a classifier @xmath98 is unknown , an empirical surrogate is often used . in particular , _ empirical risk minimization ( erm ) _ is the procedure of choosing a classifier @xmath98 from a set of classifiers @xmath106 by minimizing the empirical risk @xmath107 . the consistency and rate of convergence of erm see @xcite for precise definitions can be related to uniform bounds on the difference between the empirical risk and the true risk over all @xmath108 . there is a large body of literature on sufficient conditions for this kind of uniform convergence . for instance , the vc dimension is commonly - used toward this end . note that , when considering gap - tolerant classifiers , there is an additional caveat , as one obtains uniform bounds only over those gap - tolerant classifiers that do not contain any data points in the margin the appendix a.2 of @xcite addresses this issue . in this paper , our main emphasis is on the annealed entropy : [ annealed entropy ] [ def : an_ent ] let @xmath41 be a probability measure supported on a vector space @xmath42 . given a set @xmath106 of decision rules and a set of points @xmath109 , let @xmath110 be the number of ways of labeling @xmath111 into positive and negative samples such that there exists a gap - tolerant classifier that predicts _ incorrectly _ the label of each @xmath112 . given the above notation , h_ann^(k ) : = _ p^k ( z_1 , , z_k ) is the annealed entropy of the classifier @xmath106 with respect to @xmath41 . note that although we have defined the annealed entropy for general decision rules , below we will consider the case that @xmath106 consists of linear decision rules . as the following theorem states , the annealed entropy of a classifier can be used to get an upper bound on the generalization error . this follows from theorem @xmath113 in @xcite and a remark on page 198 of @xcite . note that the class @xmath114 is itself random , and consequently , @xmath115 is the supremum over a random class . [ thm : vapnik ] let @xmath114 be the family of all gap - tolerant classifiers such that no data point lies inside the margin . then , @xmath116 < 8 \exp{\left(\left({h_{ann}^\lambda(\ell)}\right ) - \frac{\epsilon^2\ell}{32}\right)}\ ] ] holds true , for any number of samples @xmath25 and for any error parameter @xmath117 . the key property of the function class that leads to uniform bounds is the sublinearity of the logarithm of the expected value of the `` growth function , '' which measures the number of distinct ways in which a data set of a particular size can be split by the function class . a finite vc bound guarantees this in a distribution - free setting . the annealed entropy is a distribution - specific measure , _ i.e. _ , the same family of classifiers can have different annealed entropies when measured with respect to different distributions . for a more detailed exposition of uniform bounds in the context of gap - tolerant classifiers , we refer the reader to ( @xcite , appendix a.2 ) . note also that normed vector spaces ( such as hilbert spaces and banach spaces ) are relevant to learning theory for the following reason . data are often accompanied with an underlying metric which carries information about how likely it is that two data points have the same label . this makes concrete the intuition that points with the same class label are clustered together . many algorithms can not be implemented over an arbitrary metric space , but require a linear structure . if the original metric space does not have such a structure , as is the case when classifying for example , biological data or decision trees , it is customary to construct a feature space representation , which embeds data into a vector space . we will be interested in the commonly - used hilbert spaces , in which distances in the feature space are measure with respect to the @xmath118 distance ( as well as more general banach spaces , in section [ sxn : gapbs ] ) . finally , note that our results where the margin is measured in @xmath118 can be transferred to a setting with kernels . given a kernel @xmath119 , it is well known that linear classification using a kernel @xmath119 is equivalent to mapping @xmath120 onto the functional @xmath121 and then finding a separating halfspace in the reproducing kernel hilbert space ( rkhs ) which is the hilbert space generated by the functionals of the form @xmath121 . since the span of any finite set of points in a hilbert space can be isometrically embedded in @xmath118 , our results hold in the setting of kernel - based learning as well , when one first uses the feature map @xmath122 and works in the rkhs . in this section , we state and prove theorem [ thm : ddb_hs ] , our main result regarding an upper bound for the annealed entropy of gap - tolerant classifiers in @xmath118 . this result is of independent interest , and it was used in a crucial way in the proof of theorems [ thm : learn_heavytail ] and [ thm : learn_spectral ] . we start in section [ sxn : gaphs_annent ] with the statement and proof of theorem [ thm : ddb_hs ] , and then in section [ sxn : gaphs_vcdim ] we bound the vc dimension of gap - tolerant classifiers over a ball of radius @xmath13 . then , in section [ sxn : gaphs_htapp ] , we apply these results to prove our main theorem on learning with heavy - tailed data , and finally in section [ sxn : gaphs_skapp ] , we apply these results to prove our main theorem on learning with spectral kernels . the following theorem is our main result regarding an upper bound for the annealed entropy of gap - tolerant classifiers . the result holds for gap - tolerant classification in a hilbert space , _ i.e. _ , when the distances in the feature space are measured with respect to the @xmath118 norm . analogous results hold when distances are measured more generally , as we will describe in section [ sxn : gapbs ] . [ annealed entropy ; upper bound ; hilbert space ] [ thm : ddb_hs ] let @xmath41 be a probability measure on a hilbert space @xmath42 , and let @xmath43 . if @xmath44 , then then the annealed entropy of gap - tolerant classifiers in @xmath42 , where the gap is @xmath12 , is @xmath123 let @xmath25 independent , identically distributed ( i.i.d ) samples @xmath124 be chosen from @xmath125 . we partition them into two classes : @xmath126 and @xmath127 our objective is to bound from above the annealed entropy @xmath128 $ ] . by lemma [ lem : submul ] , @xmath129 is sub - multiplicative . therefore , @xmath130 taking an expectation over @xmath25 i.i.d samples from @xmath125 , @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell - k})n^{\lambda}(y_1 , \dots , y_k)].\ ] ] now applying lemma [ lem : vap1 ] , we see that @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell -k})(k+1)^{r^2/\delta^2 + 1 } ] .\ ] ] moving @xmath132 outside this expression , @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell -k})](k+1)^{r^2/\delta^2 + 1 } .\ ] ] note that @xmath133 is always bounded above by @xmath134 and that the random variables @xmath135 $ ] are i.i.d . let @xmath136 $ ] , and note that @xmath137 is the sum of @xmath25 independent bernoulli variables . moreover , by markov s inequality , @xmath138 \,\ , \leq \,\ , \frac{{\mathbb{e}}[\|z_i\|^2]}{r^2 } , \ ] ] and therefore @xmath139 . in addition , @xmath140 \leq { \mathbb{e}}[2^{\ell - k}].\ ] ] let @xmath141 $ ] denote an indicator variable . @xmath142 $ ] can be written as @xmath143 } ] = ( 1+\rho)^\ell \leq e^{\rho \ell}.\ ] ] putting everything together , we see that [ eqn : jensen1 ] [ n^(z_1 , , z _ ) ] ( ( ) ^2 + ( + 1 ) ( + 1 ) ) . if we substitute @xmath144 , it follows that @xmath145 \\ & \leq & \left(\ell^{\frac{1}{2}}\left(\frac{r}{\delta}\right ) + 1\right ) ( 1+\ln(\ell+1 ) ) .\end{aligned}\ ] ] for ease of reference , we note the following easily established fact about @xmath146 . this lemma is used in the proof of theorem [ thm : ddb_hs ] above and theorem [ thm : ddb_bs ] below . [ lem : submul ] let @xmath147 be a partition of the data @xmath148 into two parts . then , @xmath146 is submultiplicative in the following sense : @xmath149 this holds because any partition of @xmath150 into two parts by an element @xmath151 induces such a partition for the sets @xmath152 and @xmath153 , and for any pair of partitions of @xmath152 and @xmath154 , there is at most one partition of @xmath148 that induces them . as an intermediate step in the proof of theorem [ thm : ddb_hs ] , we needed a bound on the vc dimension of a gap - tolerant classifier within a ball of fixed radius . lemma [ lem : vap1 ] below provides such a bound and is due to vapnik @xcite . note , though , that in the course of his proof ( see @xcite , page 353 . ) , vapnik states , without further justification , that due to symmetry the set of points that is extremal in the sense of being the hardest to shatter with gap - tolerant classifiers is the regular simplex . attention has also been drawn to this fact by burges ( @xcite , footnote 20 ) , who mentions that a rigorous proof of this fact seems to be absent . vapnik s claim has since been proved by hush and scovel @xcite . here , we provide a new proof of lemma [ lem : vap1 ] . it is simpler than previous proofs , and in section [ sxn : gapbs ] we will see that it generalizes to cases when the margin is measured with norms other than @xmath118 . [ vc dimension ; upper bound ; hilbert space ] [ lem : vap1 ] in a hilbert - space , the vc dimension of a gap - tolerant classifier whose margin is @xmath12 over a ball of radius @xmath13 can by bounded above by @xmath155 . suppose the vc dimension is @xmath11 . then there exists a set of @xmath11 points @xmath156 in @xmath157 that can be completely shattered using gap - tolerant classifiers . we will consider two cases , first that @xmath11 is even , and then that @xmath11 is odd . first , assume that @xmath11 is even , i.e. , that @xmath158 for some positive integer @xmath159 . we apply the probabilistic method to obtain a upper bound on @xmath11 . note that for every set @xmath160 $ ] , the set @xmath161 can be separated from @xmath162 using a gap - tolerant classifier . therefore the distance between the centroids ( respective centers of mass ) of these two sets is greater or equal to @xmath163 . in particular , for each @xmath164 having @xmath165 elements , @xmath166 suppose now that @xmath164 is chosen uniformly at random from the @xmath167 sets of size @xmath159 . then , @xmath168 \\ & = & k^{-2}\left\{\frac{2k+1}{2k}\sum_{i=1}^n \|x_i\|^2 - \frac{\|\sum_1^n x_i\|^2}{2k } \right\ } \\ & \leq & \frac{4(n+1)}{n^2 } r^2 .\end{aligned}\ ] ] therefore , @xmath169 and so @xmath170 next , assume that @xmath11 is odd . we perform a similar calculation for @xmath171 . as before , we average over all sets @xmath164 of cardinality @xmath159 the squared distance between the centroid of @xmath172 and the centroid ( center of mass ) of @xmath173 . proceeding as before , @xmath174 \\ & = & \frac{\sum_{i=1}^n \|x_i\|^2 ( 1 + \frac{1}{2n } ) - \frac{1}{2n}\|\sum_{1 \leq i \leq n } x_i\|^2}{k(k+1 ) } \\ & \leq & \frac{\sum_{i=1}^n \|x_i\|^2 ( 1 + \frac{1}{2n})}{k(k+1 ) } \\ & = & \frac{4k+3}{2k(2k+1)(k+1)}\{(2k+1)r^2\ } \\ & < & \frac{4r^2}{n-1 } .\end{aligned}\ ] ] therefore , @xmath175 . for a random data sample @xmath120 , ^2 & & _ i=1^n ( c i^-)^2 + & & c^2 ( 2 ) , where @xmath176 is the riemann zeta function . the theorem then follows from theorem [ thm : margin_hs ] . a diffusion map for the graph @xmath177 is the feature map that associates with a vertex @xmath120 , the feature vector @xmath178 , when the eigenfunctions corresponding to the top @xmath179 eigenvalues are chosen . let @xmath34 be the uniform distribution on @xmath180 and @xmath181 . we note that if the @xmath182 are normalized eigenfunctions , _ i.e. _ , @xmath183 ^2 & = & 1 . the above inequality holds because the eigenvalues have magnitudes that are less or equal to @xmath33 : @xmath184 the theorem then follows from theorem [ thm : margin_hs ] . in this section , we state and prove theorem [ thm : ddb_bs ] , our main result regarding an upper bound for the annealed entropy of gap - tolerant classifiers in a banach space . we start in section [ sxn : gapbs_prelim ] with some technical preliminaries ; then in section [ sxn : gapbs_vcdim ] we bound the vc dimension of gap - tolerant classifiers in banach spaces over a ball of radius @xmath13 ; and finally in section [ sxn : gapbs_annent ] we prove theorem [ thm : ddb_bs ] . we include this result for completeness since it is of theoretical interest ; since it follows using similar methods to the analogous results for hilbert spaces that we presented in section [ sxn : gaphs ] ; and since this result is of potential practical interest in cases where modeling the relationship between data elements as a dot product in a hilbert space is too restrictive , _ e.g. _ , when the data are extremely sparse and heavy - tailed . for recent work in machine learning on banach spaces , see @xcite . recall the definition of a banach space from definition [ def : rad ] above . we next state the following form of the chernoff bound , which we will use in the proof of lemma [ lem : banach_ub ] below . [ chernoff bound ] [ lem : chernoff ] let @xmath185 be discrete independent random variables such that @xmath186=0 $ ] for all @xmath64 and @xmath187 for all @xmath64 . let @xmath188 for all @xmath64 and @xmath189 be the variance of @xmath90 . then @xmath190 \leq 2e^{-\lambda^2/4}\ ] ] for any @xmath191 . the idea underlying our new proof of lemma [ lem : vap1 ] ( of section [ sxn : gaphs_vcdim ] , and that provides an upper bound on the vc dimension of a gap - tolerant classifier in hilbert spaces ) generalizes to the case when the the gap is measured in more general banach spaces . we state the following lemma for a banach space of type @xmath50 with type constant @xmath48 . e.g. _ , that @xmath51 for @xmath192 is a banach space of type @xmath193 and type constant @xmath33 . [ vc dimension ; upper bound ; banach space ] [ lem : banach_ub ] in a banach space of type @xmath50 and type constant @xmath48 , the vc dimension of a gap - tolerant classifier whose margin is @xmath12 over a ball of radius @xmath13 can by bounded above by @xmath194 since a general banach space does not possess an inner product , the proof of lemma [ lem : vap1 ] needs to be modified here . to circumvent this difficulty , we use inequality ( [ ineq : rad ] ) determining the rademacher type of @xmath65 . this , while permitting greater generality , provides weaker bounds than previously obtained in the euclidean case . note that if @xmath195 , then by repeated application of the triangle inequality , @xmath196 this shows that if we start with @xmath197 having norm @xmath198 , @xmath199 for all @xmath64 . the property of being shattered by gap - tolerant classifiers is translation invariant . then , for @xmath200 $ ] , it can be verified that @xmath201 the rademacher inequality states that [ ineq : rad]_t^p _ i=1^n x_i^p.using the version of chernoff s bound in lemma [ lem : chernoff ] [ twelve ] [ |_i=1^n _ i| ] 1- 2 e^-^2/4 . we shall denote the above event by @xmath202 . now , let @xmath197 be @xmath11 points in @xmath65 with a norm less or equal to @xmath13 . let @xmath203 as before . @xmath204\\ & \geq & { \mathbb{e}}_\epsilon [ \|\epsilon_i ( x_i-\mu)\|^p| e_{\lambda}]\,\,{\mathbb{p}}[e_{\lambda}]\\ & \geq & { \mathbb{e}}_\epsilon [ ( n - \lambda^2)^p(2\delta)^p ( 1 - 2 e^{-\lambda^2/4})]\,\,\end{aligned}\ ] ] the last inequality follows from ( [ eleven ] ) and ( [ twelve ] ) . we infer from the preceding sequence of inequalities that @xmath205 the above is true for any @xmath206 , by the conditions in the chernoff bound stated in lemma [ lem : chernoff ] . if @xmath207 , choosing @xmath208 equal to @xmath113 gives us @xmath209 therefore , it is always true that @xmath210 finally , for completeness , we next state a lower bound for vc dimension of gap - tolerant classifiers when the margin is measured in a norm that is associated with a banach space of type @xmath47 $ ] . since we are interested only in a lower bound , we consider the special case of @xmath211 . note that this argument does not immediately generalize to banach spaces of higher type because for @xmath212 , @xmath51 has type @xmath213 . [ vc dimension ; lower bound ; banach space ] [ lem : banach_lb ] for each @xmath47 $ ] , there exists a banach space of type @xmath50 such that the vc dimension of gap - tolerant classifiers with gap @xmath12 over a ball of radius @xmath13 is greater or equal to @xmath214 further , this bound is achieved when the space is @xmath51 . we shall show that the first @xmath11 unit norm basis vectors in the canonical basis can be shattered using gap - tolerant classifiers , where @xmath215 . therefore in this case , the vc dimension is @xmath216 . let @xmath217 be the @xmath218 basis vector . in order to prove that the set @xmath219 is shattered , due to symmetry under permutations , it suffices to prove that for each @xmath159 , @xmath220 can be separated from @xmath221 using a gap - tolerant classifier . points in @xmath51 are infinite sequences @xmath222 of finite @xmath51 norm . consider the hyperplane @xmath223 defined by @xmath224 . clearly , it separates the sets in question . we may assume @xmath217 to be @xmath225 , replacing if necessary , @xmath159 by @xmath226 . let @xmath227 clearly , all coordinates @xmath228 of @xmath120 are @xmath229 . in order to get a lower bound on the @xmath51 distance , we use the power - mean inequality : if @xmath192 , and @xmath230 , @xmath231 this implies that @xmath232 for @xmath233 , the type of @xmath51 is @xmath213 @xcite . since @xmath234 is a decreasing function of @xmath50 in this regime , we do not recover any useful bounds . the following theorem is our main result regarding an upper bound for the annealed entropy of gap - tolerant classifiers in banach spaces . note that the @xmath118 bound provided by this theorem is slightly weaker than that provided by theorem [ thm : ddb_hs ] . note also that it may seem counter - intuitive that in the case of @xmath118 ( _ i.e. _ , when we set @xmath235 ) , the dependence of @xmath12 is @xmath236 , which is weaker than in the vc bound , where it is @xmath237 . the explanation is that the bound on annealed entropy here depends on the number of samples @xmath25 , while the vc dimension does not . therefore , the weaker dependence on @xmath12 is compensated for by a term that in fact tends to @xmath238 as the number of samples @xmath239 . [ annealed entropy ; upper bound ; banach space ] [ thm : ddb_bs ] let @xmath41 be a probability measure on a banach space @xmath68 of type @xmath50 and type constant @xmath48 . let @xmath240 , and let @xmath241 . if @xmath242 , then the annealed entropy of gap - tolerant classifiers in @xmath68 , where the gap is @xmath12 , is @xmath243 the proof of this theorem parallels that of theorem [ thm : ddb_hs ] , except that here we use lemma [ lem : banach_ub ] instead of lemma [ lem : vap1 ] . we include the full proof for completeness . let @xmath25 independent , identically distributed ( i.i.d ) samples @xmath124 be chosen from @xmath125 . we partition them into two classes : @xmath126 and @xmath127 our objective is to bound from above the annealed entropy @xmath128 $ ] . by lemma [ lem : submul ] , @xmath129 is sub - multiplicative . therefore , @xmath130 taking an expectation over @xmath25 i.i.d samples from @xmath125 , @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell - k})n^{\lambda}(y_1 , \dots , y_k)].\ ] ] now applying lemma [ lem : banach_ub ] , we see that @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell -k})(k+1)^{(3tr/\delta)^{\frac{p}{p-1}}+ 64 } ] .\ ] ] moving @xmath244 outside this expression , @xmath131 \leq { \mathbb{e}}[n^{\lambda}(x_1 , \dots , x_{\ell -k})](k+1)^{(3tr/\delta)^{\frac{p}{p-1}}+ 64 } .\ ] ] note that @xmath133 is always bounded above by @xmath134 and that the random variables @xmath135 $ ] are i.i.d . let @xmath136 $ ] , and note that @xmath137 is the sum of @xmath25 independent bernoulli variables . moreover , by markov s inequality , @xmath138 \,\ , \leq \,\ , \frac{{\mathbb{e}}[\|z_i\|^\gamma]}{r^\gamma } , \ ] ] and therefore @xmath245 . in addition , @xmath140 \leq { \mathbb{e}}[2^{\ell - k}].\ ] ] let @xmath141 $ ] denote an indicator variable . @xmath142 $ ] can be written as @xmath143 } ] = ( 1+\rho)^\ell \leq e^{\rho \ell}.\ ] ] putting everything together , we see that [ eqn : jensen2 ] [ n^(z_1 , , z _ ) ] ( ( ) ^ + ( k+1)(64 + ) ^ ) . by setting @xmath246 and adjusting @xmath13 so that @xmath247 we see that @xmath248 thus , it follows that @xmath249 \\ & \leq & \left(\eta^{-\eta}(1-\eta)^{-1+\eta } \left(\frac{\ell}{\ln(\ell+1 ) } \left(\frac{3tr}{\delta}\right)^\gamma\right)^\eta + 64\right ) \ln(\ell+1 ) .\end{aligned}\ ] ] in recent years , there has been a considerable amount of somewhat - related technical work in a variety of settings in machine learning . thus , in this section we will briefly describe some of the more technical components of our results in light of the existing related literature . * techniques based on the use of rademacher inequalities allow one to obtain bounds without any assumption on the input distribution as long as the feature maps are uniformly bounded . e.g. _ , @xcite . viewed from this perspective , our results are interesting because the uniform boundedness assumption is not satisfied in either of the two settings we consider , although those settings are ubiquitous in applications . in the case of heavy - tailed data , the uniform boundedness assumption is not satisfied due to the slow decay of the tail and the large variability of the associated features . in the case of spectral learning , uniform boundedness assumption is not satisfied since for arbitrary graphs one can have localization and thus large variability in the entries of the eigenvectors defining the feature maps . in both case , existing techniques based on rademacher inequalities or vc dimensions fail to give interesting results , but we show that dimension - independent bounds can be achieved by bounding the annealed entropy . * a great deal of work has focused on using diffusion - based and spectral - based methods for nonlinear dimensionality reduction and the learning a nonlinear manifold from which the data are assumed to be drawn @xcite . these results are very different from the type of learning bounds we consider here . for instance , most of those learning results involve convergence to an hypothesized manifold laplacian and not of learning process itself , which is what we consider here . * work by bousquet and elisseeff @xcite has focused on establishing generalization bounds based on stability . it is important to note that their results assume a given algorithm and show how the generalization error changes when the data are changed , so they get generalization results for a given algorithm . our results make no such assumptions about working with a given algorithm . * gurvits @xcite has used rademacher complexities to prove upper bounds for the sample complexity of learning bounded linear functionals on @xmath51 balls . the results in that paper can be used to derive an upper bound on the vc dimension of gap - tolerant classifiers with margin @xmath12 in a ball of radius @xmath13 in a banach space of rademacher type @xmath47 $ ] . constants were not computed in that paper , therefore our results do not follow . moreover , our paper contains results on distribution specific bounds which were not considered there . finally , our paper considers the application of these tools to the practically - important settings of spectral kernels and heavy - tailed data that were not considered there . we have considered two simple machine learning problems motivated by recent work in large - scale data analysis , and we have shown that although traditional distribution - independent methods based on the vc - dimension fail to yield nontrivial sampling complexity bounds , we can use distribution - dependent methods to obtain dimension - independent learning bounds . in both cases , we take advantage of the fact that , although there may be individual data points that are `` outlying , '' in aggregate their effect is not too large . due to the increased popularity of vector space - based methods ( as opposed to more purely combinatorial methods ) in machine learning in recent years , coupled with the continued generation of noisy and poorly - structured data , the tools we have introduced are likely promising more generally for understanding the effect of noise and noisy data on popular machine learning tasks . in this appendix section , we state and prove a second result for dimension - independent learning from data in which the feature map exhibits a heavy - tailed decay . the heavy - tailed model we consider here is different than that considered in theorem [ thm : learn_heavytail ] , and thus we are able to prove bounds for exact ( as opposed to maximum margin ) learning . nevertheless , the techniques are similar , and thus we include this result in this paper for completeness . consider the following toy model for classifying web pages using keywords . one approach to this problem could be to associate with each web page the indicator vector corresponding to all keywords that it contains . the dimension of this feature space is the number of possible keywords , which is typically very large , and empirical evidence indicates that the frequency of words decays in a heavy - tailed manner . thus the vc dimension of the feature space is very large , and in a distribution - free setting it is not possible to classify data in such a feature space unless the number of samples is of the order of the vc dimension . more generally , one might be interested in a bipartite graph , _ e.g. _ , an `` advertiser - keyword '' or `` author - to - paper '' graph , in which the nodes are the stated entities and the edges represent some sort of `` interaction '' between the entities , in which case similar issues arise . here , we show that if the probability that the @xmath19 keyword in the above toy example is present is heavy - tailed as a function of @xmath64 , then the sample complexity of the binary classification problem is dimension - independent . more precisely , the following theorem provides a dimension - independent ( _ i.e. _ , independent of the size @xmath11 of the graph and the dimension of the feature space ) upper bound on the number of samples needed to learn by erm , with a given accuracy and confidence , a linear hyperplane that classifies heavy - tailed data into positive and negative labels , under the assumption that the probability of the @xmath19 coordinate of a random data point being non - zero is less than @xmath20 for some @xmath21 . the proof of this result proceeds by providing providing a dimension - independent upper bound on the annealed entropy of the class of linear classifiers in @xmath250 , and then appealing to theorem [ thm : vapnik ] relating the annealed entropy to the generalization error . * remark : * note that although the generalization bound provided by the following theorem seems to be pessimistic in @xmath97 , the dependence on @xmath97 is tight , at least as @xmath97 tends to @xmath33 . clearly , when @xmath251 , the expected number of @xmath33 s in a random sample becomes asymptotically equal to @xmath252 , where @xmath11 is the dimension , in which case , we do not expect a sample complexity that is dimension - independent . [ bounds for heavy - tailed data ] [ thm : learnht_exact ] let @xmath41 be a probability distribution in @xmath250 . ci^{-\alpha}$ ] for some absolute constant @xmath17 , with @xmath18 . then , the annealed entropy of ordinary linear hyperplane classifiers is h_ann^ ( ) & & ( ^+1)consequently , the minimum number of random samples @xmath254 needed to learn , by erm , a classifier whose risk differs from the minimum risk @xmath96 by @xmath255 with probability @xmath256 is less than or equal to 2 ( ( + ) ) ^ ( ( ( + ) ) ^ ) . let the event that a sample @xmath257 has a non - zero coordinate @xmath258 for some @xmath259 be denoted @xmath260 . the probability of this event can be bounded as follows . if @xmath261 and @xmath262 , then [ e_i ] & = & [ k > ^1/ , z_ik 0 ] + & & c_i = k+1^i^- + & & . we partition the @xmath112 into two classes : @xmath263 and @xmath264 @xmath146 is sub - multiplicative by lemma [ lem : submul ] . taking an expectation over @xmath25 i.i.d samples from @xmath41 , [ ( z_1 , , z _ ) ] & & [ ( x_1 , , x_-m)(y_1 , , y_m)]the dimension of the span of @xmath265 is at most @xmath159 , and by a result from vc theory ( @xcite , page 159 ) we have @xmath266 then , @xmath267 \leq { \mathbb{e}}[\nl(x_1 , \dots , x_{\ell -m})e m^k].\ ] ] moving @xmath268 outside this expression , @xmath267 \leq { \mathbb{e}}[\nl(x_1 , \dots , x_{\ell -k})]e m^k .\ ] ] note that @xmath269 is always bounded above by @xmath134 and that the events @xmath270 are independent identically distributed . let @xmath271 $ ] , and note that @xmath137 is the sum of @xmath25 independent @xmath50-bernoulli variables . in addition , @xmath272 \leq { \mathbb{e}}[2^{\ell - k}],\ ] ] and @xmath142 $ ] can be written as _ i=1^(1 + [ e_i ] ) & = & ( 1+p)^ + & & e^p + & = & e^().putting everything together , we see that @xmath267 \leq e(\ell)^k e^{\frac{c \ell\ , k^{-a + 1}}{\a - 1 } } .\ ] ] since @xmath273 , we see that h_ann^ ( ) & = & [ ( z_1 , , z _ ) ] + & & ( ^+1 ) ( ) . in order to obtain sample complexity bounds , we need to apply theorem [ thm : vapnik ] and substitute the above expression for annealed entropy . for the probability that the error of erm exceeds @xmath274 to be less than @xmath275 ( where @xmath98 is the optimal classifier ) , it is sufficient that @xmath25 satisfy @xmath276 for this to be true , it is enough that @xmath277 a calculation shows that @xmath278 is a value of @xmath25 that satisfies the previous expression . l. gurvits . a note on a scale - sensitive dimension of linear bounded functionals in banach spaces . in _ proceedings of the 8th international conference on algorithmic learning theory _ , pages 352363 , 1997 . v. koltchinskii and d. panchenko . rademacher processes and bounding the risk of function learning . in e. gine , d. mason , and j. wellner , editors , _ high dimensional probability ii _ , pages 443459 . birkhauser , 2000 . j. leskovec , k.j . lang , a. dasgupta , and m.w . statistical properties of community structure in large social and information networks . in _ www 08 : proceedings of the 17th international conference on world wide web _ , pages 695704 , 2008 . j. leskovec , k.j . lang , and m.w . mahoney . empirical comparison of algorithms for network community detection . in _ www 10 : proceedings of the 19th international conference on world wide web _ , pages 631640 , 2010 . l. k. saul , k. q. weinberger , j. h. ham , f. sha , and d. d. lee . spectral methods for dimensionality reduction . in o. chapelle , b. schoelkopf , and a. zien , editors , _ semisupervised learning _ , pages 293308 . mit press , 2006 .

we study the normal and the superconducting properties in noncentrosymmetric heavy fermion superconductors cerhsi and ceirsi . for the normal state , we show that experimentally observed linear temperature dependence of the resistivity is understood through the antiferromagnetic spin fluctuations near the quantum critical point ( qcp ) in three dimensions . for the superconducting state , we derive a general formula to calculate the upper critical field , with which we can treat the pauli and the orbital depairing effect on an equal footing . the strong coupling effect for general electronic structures is also taken into account . we show that the experimentally observed features in , the huge value up to(t ) , the downward curvatures , and the strong pressure dependence , are naturally understood as an interplay of the rashba spin - orbit interaction due to the lack of inversion symmetry and the spin fluctuations near the qcp . the large anisotropy between and is explained in terms of the spin - orbit interaction . furthermore , a possible realization of the fulde - ferrell- larkin - ovchinnikov state for is studied . we also examine effects of spin - flip scattering processes in the pairing interaction and those of the applied magnetic field on the spin fluctuations . we find that the above mentioned results are robust against these effects . the consistency of our results strongly supports the scenario that the superconductivity in cerhsi and ceirsi is mediated by the spin fluctuations near the qcp . , ,

1002.4031

in noncentrosymmetric heavy fermion superconductors , in addition to strong electron correlation , there exists another key property , the anisotropic spin - orbit ( so ) interaction due to the lack of inversion symmetry . the anisotropic so interaction plays important roles both in the normal and the superconducting state , and is expected to lead to many interesting phenomena . @xcite for such phenomena , electron correlation is quite important , because it can largely enhance the effect of the so interaction . the interplay of the anisotropic so interaction and electron correlation is truly an unique nature in noncentrosymmetric heavy fermion compounds . @xcite in particular , such an interplay in the superconducting state has been attracting particular interest . in this context , especially , cerhsi@xmath0 @xcite and ceirsi@xmath0 @xcite are promising candidates for the interplay , because they are considered to be located near the antiferromagnetic ( af ) qcps around which strong correlations through the spin fluctuations are essential . cerhsi@xmath0 and ceirsi@xmath0 are af metals at ambient pressure , and begin to exhibit superconductivity at some critical pressures @xmath6 where the nel temperatures seem to be suppressed to absolute zero . according to the neutron experiments for cerhsi@xmath0 , the af ordering vector is @xmath7 and the nature of the af order is spin density wave - like . @xcite this is different from cept@xmath0si in which the af seems to have localized nature and the superconductivity coexists with it even at zero applied pressure . @xcite in nmr experiments in ceirsi@xmath0 , @xmath8 is observed near the critical pressure , which is a characteristic behavior of the systems with 3-dimensional ( 3d ) af spin fluctuations . @xcite in addition , the resistivity @xmath9 in both cerhsi@xmath0 and ceirsi@xmath0 above the superconducting transition temperatures @xmath10 in some pressure regions near the qcp shows the anomalous @xmath11-linear dependence which is different from @xmath12 in canonical fermi liquids . @xcite the qcp related phenomena are observed also in the superconducting state . the large jump in the heat capacity at @xmath10 in ceirsi@xmath0 can be attributed to the strong coupling effect due to the spin fluctuations . @xcite it has strong pressure dependence and is largely enhanced near @xmath6 . the most striking phenomena which would be related to the quantum criticality in cerhsi@xmath0 and ceirsi@xmath0 appear in the behaviors of the upper critical fields @xmath1 when the applied magnetic field is parallel to @xmath13-axis . @xcite the remarkable features of the experimental results are as follows . ( i ) as the pressure approaches a critical value , @xmath1 exhibits extremely high value which exceeds the orbital limit as well as the pauli limit estimated by the conventional bcs theory . the observed @xmath14(t ) is the highest value among the heavy fermion superconductors ever discovered , although @xmath10 is merely @xmath15(k ) . ( ii ) @xmath1 curves have downward curvatures and the increase is accelerated as the temperature is decreased , making a sharp contrast to any other superconductors in which the increase of @xmath1 becomes slower as @xmath11 is decreased . ( iii ) @xmath1 increases very rapidly as the pressure approaches the critical value , while the pressure dependence of @xmath10 is moderate . these characteristic features strongly suggest that there exists a deep connection between the superconductivity and the magnetic quantum criticality . in the previous study , the present authors have shown that these experimental results are well explained as an interplay of the rashba so interaction due to the lack of inversion symmetry and the spin fluctuations near the qcp . @xcite on the other hand , @xmath1 for in - plane fields differs from @xmath2 in some important features . @xmath4 is merely less than @xmath16(t ) and its pressure dependence is moderate , and the @xmath1 curves exhibit usual upward curvatures . @xcite this anisotropy in @xmath1 would be related to the rashba so interaction , since the fermi surface is asymmetrically distorted by the in - plane field and the pauli depairing effect plays essential roles . by contrast , the renormalization of the quasiparticle velocity by the spin fluctuations is almost isotropic , resulting in the enhanced orbital limiting field in all directions of the applied field . another interesting phenomenon in the noncentrosymmetric superconductors in applied magnetic fields is the helical vortex phase which has been discussed theoretically . @xcite in a helical vortex phase , the superconducting gap function is modulated in real space , @xmath17 with the modulation vector @xmath18 , where @xmath19 and @xmath20 are the strength of the so interaction , the fermi velocity , the fermi energy , and the bohr magneton , respectively . for 3d rashba superconductors for @xmath5 , however , it is pointed out that this phase modulation is just a translational shift of the vortex lattice and has no physical importance . @xcite several authors also have discussed a spatially modulated superconducting state under magnetic fields with a large @xmath21 which is continuously connected from @xmath18 . @xcite this large @xmath22 state corresponds to the fulde - ferrell - ovchinnikov - larkin ( fflo ) state . @xcite the stability of the modulating superconducting state with the large @xmath22 depends on the relative strength of the orbital depairing effect to the pauli depairing effect in the compounds . in this paper , we study the normal and the superconducting properties in noncentrosymmetric superconductors cerhsi@xmath0 and ceirsi@xmath0 . we examine the anomalous @xmath11-linear dependence of the resistivity in the normal state . in the previous studies,@xcite at very low temperatures , @xmath23 is predicted for 3d af spin fluctuations . the temperature dependence of the resistivity for 3d af spin fluctuations have been studied in detail by several authors @xcite we , here , show that @xmath24 in cerhsi@xmath0 and ceirsi@xmath0 is actually due to the 3d af spin fluctuations . for the superconducting state , the upper critical fields both for @xmath25 and @xmath5 are investigated . for the calculation of @xmath1 , we derive a general formula which enables us to treat the pauli and the orbital depairing effects on an equal footing . we can also take into account the strong coupling effect for a given electronic structure . we calculate @xmath1 on the basis of the scenario that the superconductivity in cerhsi@xmath0 and ceirsi@xmath0 is mediated by the spin fluctuations , and show that the experimental features are well explained as an interplay of the spin fluctuations and the rashba so interaction . although the formula is applicable for general models , we use a phenomenological model to calculate @xmath1 and neglect the following two points in the model . one is the scattering processes in the pairing interaction in which spins of quasiparticles are flipped by the rashba so interaction . it is pointed out that such processes can enhance the admixture of the singlet and the triplet superconductivity , @xcite and the strength of the admixture affects @xmath4 . @xcite we show that the admixture is still small in cerhsi@xmath0 and ceirsi@xmath0 even if we include the spin - flip scattering processes . the other point is the applied field dependence of the spin fluctuations . because the applied field is so large in cerhsi@xmath0 and ceirsi@xmath0 especially for @xmath25 that one might think that the spin fluctuations are suppressed and they can not contribute to the enhancement of @xmath1 . we show that the spin fluctuations are robust against the applied field @xmath26 up to the strength of the rashba so interaction , @xmath27 , because the rashba so interaction tends to fix the directions of spins on the fermi surface and it competes with the zeeman effect . the consistency of our results with the experiments strongly supports the scenario that the superconductivity in cerhsi@xmath0 and ceirsi@xmath0 is mediated by the spin fluctuations near the af qcp . this paper is organized as follows . in sec.[sec : normal ] , we study the experimentally observed @xmath11-linear dependence of the resistivity . in sec.[sec : eliashberg ] , a general formula for the calculation of @xmath1 is derived from the eliashberg equation . the characteristic features of @xmath1 in cerhsi@xmath0 and ceirsi@xmath0 are well explained with the use of the formula in sec.[sec : hc2 ] . we discuss , in sec.[sec:2points ] , the spin - flip scattering processes in the pairing interaction and the magnetic field dependence of the spin fluctuations which are not included in the approximation used for the computation of @xmath1 . the summary is given in sec.[sec : summary ] . in this section , we discuss the temperature dependence of the resistivity near the af qcp in cerhsi@xmath0 and ceirsi@xmath0 . in ceirsi@xmath0 , nmr @xmath28 behaves as @xmath29 in some pressure regions , which means that the character of the spin fluctuations is 3d antiferromagnetic . @xcite in noncentrosymmetric systems , however , spin fluctuations are not isotropic due to the anisotropic spin - orbit interaction . the anisotropy in the noninteracting susceptibility @xmath30 is of the order of @xmath31 , where @xmath32 is the strength of the spin - orbit interaction and @xmath33 is the fermi energy . actually , in sec.[subsec : susc ] , we show that the anisotropy among @xmath34 and @xmath35 is very small within the random phase approximation . therefore , we can neglect the anisotropy of the spin fluctuations for the discussion of the resistivity and @xmath1 in cerhsi@xmath0 and ceirsi@xmath0 . for the systems with 3d af spin fluctuations , the resistivities are expected to be @xmath23 according to the previous studies . @xcite in cerhsi@xmath0 and ceirsi@xmath0 , however , the temperature dependence is @xmath24 near the af qcp . @xcite the resistivity due to the 3d af spin fluctuations was discussed by several authors , and @xmath24 behavior was found for the weakly disordered systems @xcite and the clean systems . @xcite here , we show that the @xmath11-linear resistivity in cerhsi@xmath0 and ceirsi@xmath0 is naturally understood in terms of the 3d af spin fluctuations , and this behavior has basically nothing to do with the lack of inversion symmetry . cerhsi@xmath0 and ceirsi@xmath0 are heavy fermion systems with kondo temperature @xmath3650 - 100(k ) which is much higher than the superconducting transition temperature @xmath15(k ) . @xcite therefore , to study the properties at @xmath371 - 10(k ) , we consider the low energy quasiparticles mainly formed by @xmath38-electrons through the hybridizations with the conduction electrons . we use the following single band model for the low energy quasiparticles with the asymmetric spin - orbit interaction @xmath39c_k + \sum_kc^{\dagger}_k \alpha { \mbox{\boldmath $ \mathcal l$}}_0({\mbox{\boldmath $ k$}},{\mbox{\boldmath $ h$ } } ) \cdot { \mbox{\boldmath $ \sigma$}}c_k,\\ s_{\rm sf}&=&-\sum_{kk^{\prime}q } \frac{g^2}{6}\chi(q ) { \mbox{\boldmath $ \sigma$}}_{\alpha \alpha^{\prime } } \cdot { \mbox{\boldmath $ \sigma$}}_{\beta \beta^{\prime } } c^{\dagger}_{k+q\alpha}c_{k\alpha^{\prime } } c^{\dagger}_{k^{\prime}-q\beta}c_{k^{\prime}\beta^{\prime } } , \label{eq : action_sf}\end{aligned}\ ] ] where @xmath40 is the annihilation ( creation ) operator of the kramers doublet of the @xmath41 state . @xmath42 is introduced phenomenologically and represents the interaction between the quasiparticles by the strong spin fluctuations near the af qcp . since cerhsi@xmath0 and ceirsi@xmath0 have body - centered tetragonal lattice structures with lattice spacing @xmath43 , @xcite the dispersion relation @xmath44 and the rashba type so interaction are approximated by @xmath45 where @xmath46 is the lattice constant and @xmath47 is the chemical potential . although @xmath48 in this section , we include the zeeman effect in the action for the later discussion . we fix the parameters as @xmath49 by taking @xmath50 as the energy unit . the fermi surface determined by these parameters is in qualitative agreement with the band calculation and can reproduce the peak structures of the momentum - dependent susceptibility observed by the neutron scattering experiments @xcite . since we consider @xmath38-electron systems , we assume that the above parameters include effects of the mass renormalization due to local spin correlations with typical energy scale @xmath51 50 - 100 ( k ) , the kondo temperature ; i.e. @xmath52 50 - 100 ( k ) . the interactions are phenomenologically introduced through the renormalized susceptibility @xmath53 , @xcite @xmath54 where @xmath55 , @xmath56 and @xmath57 are respectively the susceptibility , the length scale and the energy scale of spin fluctuations without strong correlations . these quantities are renormalized through the coherence length @xmath58 as the system approaches the qcp . the critical exponent of @xmath59 is the mean field value @xmath60 and @xmath61 is considered to decrease monotonically as the applied pressure approaches the critical value for the af order . @xcite the temperature dependence of @xmath59 is also consistent with the recent nmr experiment for ceirsi@xmath0 . @xcite the ordering vectors are @xmath62 according to the neutron scattering experiments for cerhsi@xmath0 . @xcite in this study , we fix the parameters in @xmath63 as @xmath64 , @xmath65 and @xmath66 . the value @xmath65 is of the same order as the fermi energy without the effect of the spin fluctuations . @xmath67 is determined so that the maximum of @xmath59 would be @xmath68 at the lowest temperature in this study , which is a reasonable value for the af spin fluctuations . we note that the coupling constant @xmath69 should be regarded as an effective one renormalized by the vertex corrections . @xcite the green s function is , @xcite @xmath70 where @xmath71 and @xmath72 ^ 2}$ ] . selfenergy is introduced as @xmath73 and @xmath74 up to the first order in @xmath75 , @xmath76 and @xmath77 are expressed as @xmath78 \label{eq : sigma_0},\\ \sigma_x(k)&=&\frac{t}{2n}\sum_{k^{\prime}}\frac{g^2}{3}\chi(k - k^{\prime } ) [ -g^0_{\downarrow \uparrow}(k^{\prime } ) -g^0_{\uparrow \downarrow}(k^{\prime } ) ] , \label{eq : sigma_x}\\ \sigma_y(k)&=&\frac{t}{2in}\sum_{k^{\prime}}\frac{g^2}{3}\chi(k - k^{\prime } ) [ -g^0_{\downarrow \uparrow}(k^{\prime } ) + g^0_{\uparrow \downarrow}(k^{\prime } ) ] , \label{eq : sigma_y}\\ \sigma_z(k)&=&\frac{t}{2n}\sum_{k^{\prime}}\frac{g^2}{3}\chi(k - k^{\prime } ) [ -g^0_{\uparrow \uparrow}(k^{\prime } ) + g^0_{\downarrow \downarrow}(k^{\prime } ) ] \label{eq : sigma_z}\end{aligned}\ ] ] where @xmath79 is the non - interacting green s function . we have neglected the constant terms in @xmath80 . in the above expression of @xmath80 , the most dominant term is @xmath76 , and @xmath81 is smaller than @xmath76 by the factor @xmath31 , where @xmath33 is the fermi energy . for rashba superconductors , @xmath82 without magnetic field . the conductivity is calculated from the kubo formula @xmath83 where the current operator @xmath84 is defined as @xmath85.\end{aligned}\ ] ] after the analytic continuation , @xmath86 has the dominant contribution to the conductivity . among the four terms @xmath87 , for sufficiently large @xmath32 , the terms @xmath88 and @xmath89 have no singularity with respect to the quasiparticle damping rate . therefore , we can neglect them , and the resulting expression for @xmath90 is @xmath91 \left ( -\frac{\partial f}{\partial \varepsilon } ( \varepsilon_{k\tau})\right ) \frac{1}{2\gamma_{k\tau } } \label{eq : sigma_xx},\\ \gamma_{k\tau}&=&-[{\rm im}\sigma_0^{\rm r}(0,{\mbox{\boldmath $ k$}})+\tau { \rm re } \hat{{\mbox{\boldmath $ \mathcal l$}}}^r(0,{\mbox{\boldmath $ k$}})\cdot{\rm im}{\mbox{\boldmath $ \sigma$}}^r(0,{\mbox{\boldmath $ k$ } } ) ] \label{eq : gamma}\end{aligned}\ ] ] where @xmath92 , @xmath93 and @xmath94 . here we have neglected the vertex corrections which are necessary for the current conservation law . @xcite this is because , for the resistivity , the scatterings by the af spin fluctuations have large momentum transfers , and therefore , the back - flow included in the vertex corrections does not affect the temperature dependence of the resistivity . @xcite as mentioned above , @xmath77 is much smaller than @xmath76 in amplitude , and from eqs.([eq : sigma_0 ] ) @xmath95([eq : sigma_z ] ) , the temperature dependence of @xmath76 and that of @xmath77 are the same . therefore , hereafter , we neglect @xmath77 and take into account only @xmath76 in this study . the resistivity in noncentrosymmetric systems is almost the same as that in usual centrosymmetric systems . this is different from the situations for the anomalous hall effect , the magnetoelectric effect and so on for which the rashba so interaction plays important roles . @xcite before moving to the numerical calculation , we show a simple analytical result for @xmath96 which determines the qualitative behavior of @xmath97 . since eq.([eq : sigma_0 ] ) is basically the same as the selfenergy @xmath98 in the usual centrosymmetric systems , we consider @xmath98 for brevity . the selfenergy at the hot spots for a sufficiently clean system with the impurity damping @xmath99 is calculated as @xmath100 { \rm im}\chi^r(\varepsilon,{\mbox{\boldmath $ q$}}){\rm im}g^{0r}_{\rm cen } ( \varepsilon , { \mbox{\boldmath $ k$}}_f-{\mbox{\boldmath $ q$}})\nonumber \\ & \sim&\sum_q\frac{\chi_q\gamma_s(\pi t)^2}{\omega_q(\omega_q+\pi t/2 ) } { \rm im}g^{0r}_{\rm cen}(0,{\mbox{\boldmath $ k$}}_f-{\mbox{\boldmath $ q$}})\nonumber \\ & \sim&\xi^0t\ln \bigl [ 1+\pi t/2\gamma_s\bigr],\end{aligned}\ ] ] where @xmath101 is the retarded green s function including the impurity damping @xmath102 and @xmath103 is the retarded susceptibility obtained from the analytic continuation of eq.([eq : chi ] ) . in the above calculation , the dispersion at the hot spots has been expanded as @xmath104 for @xmath105 , because @xmath106 is satisfied at the hot spots . here , we have used the approximation @xmath107\varepsilon/(\omega_q^2+\varepsilon^2 ) = 1/z-2[\psi(z+1)-\psi(z+1/2)]\simeq ( \pi t)^2/[\omega_q ( \omega_q+\pi t/2)]$ ] , where @xmath108 is digamma function and @xmath109 . @xcite this approximate form becomes exact both for @xmath110 and @xmath111 . @xmath112 and @xmath113 are defined as @xmath114 and @xmath115 $ ] , respectively . therefore , we have @xmath116 when the hot spots are dominant for the conductivity . this is a general behavior for the clean 3d systems with the 3d af spin fluctuations . in the numerical calculation , we neglect the real part of the selfenergy which changes the shape of the fermi surface , because such an effect is non - perturbative . we regard @xmath117 as the dispersion that includes @xmath118 . we show the numerical results for @xmath119 by using eq.([eq : sigma_0 ] ) and eq.([eq : sigma_xx ] ) for clean limit . . from the top to the bottom , @xmath120 . ] as shown in fig.[fig : rho ] , for sufficiently small @xmath61 , the resistivity is proportional to @xmath11 in a wide range of temperature where the hot spots are thermally blurred and dominant for the conductivity . in a very low temperature region where such blurring is suppressed , @xmath9 is dominated by the electrons in the cold spots . for large @xmath61 , the canonical fermi liquid behavior @xmath12 can be seen . the calculated @xmath9 well explains the experimentally observed features of the resistivity in cerhsi@xmath0 and ceirsi@xmath0 . therefore , we conclude that the observed @xmath24 above @xmath10 is due to the af spin fluctuations . we put a remark on the impurity effect . @xcite if the impurity scattering is sufficiently strong , the anisotropic scatterings by the spin fluctuations are smeared , which weakens the singularity . we , here , simply estimate the selfenergy by the spin fluctuations @xmath121 in the presence of the impurities for centrosymmetric systems analytically . for the system with the strong impurity effect which smears the anisotropy by the af spin fluctuations , we evaluate the selfenergy @xmath121 averaged on the fermi surface @xmath122 where we have defined @xmath123 . here , we have assumed that its @xmath124-dependence is moderate and it does not contribute to the selfenergy . we obtain @xmath23 for @xmath125 in dirty systems . note that , in the case of @xmath126 , we again have @xmath127 for sufficiently low temperatures . thus , the resistivity in the dirty systems with 3d af spin fluctuations is @xmath23 both for @xmath128 and @xmath126 in agreement with the previous studies . in this section , we derive a formula for the calculation of @xmath1 from the linearized eliashberg equation in real space . the derivation is based on the semiclassical approximation which is legitimate for the systems with @xmath129 , where @xmath130 is the fermi wave number and @xmath131 is the magnetic length . this condition is satisfied for many superconductors including heavy fermion superconductors , and therefore , the resulting equation for @xmath1 is applicable for a number of compounds . our formula is a generalization of the previous studies , @xcite and can be extended easily to more complicated models although we use a single band model in this section . to derive the formula for the calculation of @xmath1 , we use the linearized eliashberg equation in real space with the vector potential @xmath132 which gives a uniform magnetic field , @xmath133 where @xmath134 represents the summation over all lattice sites , and the spin indices are summed over . @xmath135 and @xmath136 are , respectively , the normal green s function , the gap function and the pairing interaction . note that , if @xmath132 is fully taken into account in the above equation , the resulting equation is gauge invariant under the gauge transformation @xmath137 and @xmath138\psi(\tau,{\mbox{\boldmath $ x$}})$ ] where @xmath108 is the field operator of the electrons . by this transformation , each factor in the equation acquires the additional phases as , @xmath139g(i\omega_n,{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}^{\prime}),\\ \delta(i\omega_n,{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}^{\prime})&\rightarrow & \exp [ ie\left(f({\mbox{\boldmath $ x$}})+f({\mbox{\boldmath $ x$}}^{\prime } ) \right)]\delta(i\omega_n,{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}^{\prime}),\\ v(i\omega_n,{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}^{\prime};i\omega_m,{\mbox{\boldmath $ y$}},{\mbox{\boldmath $ y$}}^{\prime } ) & \rightarrow & \exp [ ie\left ( f({\mbox{\boldmath $ x$}})+f({\mbox{\boldmath $ x$}}^{\prime})-f({\mbox{\boldmath $ y$}})-f({\mbox{\boldmath $ y$}}^{\prime})\right ) ] v(i\omega_n,{\mbox{\boldmath $ x$}},{\mbox{\boldmath $ x$}}^{\prime};i\omega_m,{\mbox{\boldmath $ y$}},{\mbox{\boldmath $ y$}}^{\prime}).\end{aligned}\ ] ] in this study , however , we use the semiclassical approximation in which we do not explicitly include the effect of the vector potential on the pairing interaction @xmath136 , because the vector potential in @xmath136 is not responsible for the landau quantization of the gap function which is the most important phenomenon of the orbital effect in type - ii superconductors . by contrast , the lack of translational invariance in @xmath140 and @xmath141 in the presence of the applied vector potential @xmath132 is related to the landau quantization . within the semiclassical approximation , the normal green s function is @xmath142 we can easily perform the integral along the straight line @xmath143 , using the relation @xmath144 which holds for any @xmath132 giving a uniform magnetic field @xmath145 , and obtain @xmath146 although the linearized eliashberg equation is no longer invariant under the gauge transformation defined above within this approximation , it is still gauge invariant under a gauge transformation which involves only the center of mass coordinate of the cooper pairs . next , we proceed to rewrite eq.([eq : eliashberg_r ] ) in @xmath147-space . the pairing interaction @xmath136 should be decomposed into two parts as , @xmath148 where @xmath149 and @xmath150 are the interactions in the relative coordinate and the center of mass coordinate . here , we take @xmath150 to be dimensionless . it is convenient to introduce the following variables @xmath151 in this coordinate , the phase factor @xmath152 which arises from @xmath153 in eq.([eq : eliashberg_r ] ) becomes @xmath154 in the second equality , the neglected term @xmath155 is much smaller than the first term , since for the dominant scattering processes , @xmath156 are satisfied in the systems with short range pairing interaction . the first term represents the phase which the cooper pair with center of mass @xmath157 acquires . we perform the fourier transformation of eq.([eq : eliashberg_r ] ) and assume @xmath158 , then we obtain @xmath159 the phase factor including @xmath132 is rewritten as @xmath160 where @xmath161 , @xmath162 ^ 2\nabla_{r } { \mbox{\boldmath $ a$}}({\mbox{\boldmath $ r$}})$ ] and @xmath163 . @xmath164 is proportional to @xmath165 , and therefore , negligible . @xmath166 is also small compared with @xmath167 , because @xmath168 while @xmath169 . neglecting @xmath166 and @xmath164 , we end up with the linearized eliashberg equation in @xmath147-space in the presence of the vector potential @xmath132 , @xmath170 where @xmath171 and @xmath172 , and we have written @xmath173 as @xmath136 for simplicity . this is a well - known form of the eliahsberg equation and similar expressions are often used for the discussion of @xmath1 in superconductors . as mentioned before , if we define a semiclassical gauge transformation which involves only @xmath174 as @xmath175\delta(k,{\mbox{\boldmath $ r$}}),\end{aligned}\ ] ] this equation is gauge invariant , because , for @xmath176\equiv g(k+\pi/2)g(-k+\pi/2)$ ] , @xmath177e^{i2ef}={\mathcal o}[{\mbox{\boldmath $ a$}}]$ ] is satisfied . the relative coordinate is not involved in the gauge transformation in the semiclassical approximation , and @xmath178 does not change under the transformation . we , next , proceed to rewrite the above eliashberg equation to perform numerical calculations . in the present study , we denote the coordinate as @xmath179 for the perpendicular field and @xmath180 for the in - plane field . with this notation , the gap function for @xmath181 is expanded by the landau functions , @xmath182 where @xmath183 are the usual landau functions , @xmath184 , and @xmath185 . the parameters @xmath186 and @xmath187 represent , respectively , the modulation of the gap function and the anisotropy of the vortex lattice in the @xmath188 plane , and both of them are optimized to give the largest @xmath1 . we introduce the operator @xmath189 with @xmath190 . the landau functions @xmath191 satisfy the following relations , @xmath192 where @xmath193 . by taking an inner product of eq.([eq : eliashberg_k ] ) , we obtain @xmath194 where the completeness relation @xmath195 is used . here , we have neglected the @xmath196 operators in @xmath197 because they only lead to the terms with positive powers of @xmath198 , i.e. , @xmath199 , while @xmath200 is proportional to @xmath201 describing the non - perturbative effect of the formation of the vortex lattice . within the semiclassical approximation , eqs . ( [ eq : delta_pi/2 ] ) and ( [ eq : g_pi/2 ] ) are exact . for numerical calculations , however , we need a cut off in the summation @xmath202 which should be large enough for the calculated results to be reliable . it is hard to solve eq.([eq : delta_pi/2 ] ) and ( [ eq : g_pi/2 ] ) with such a large cut off . so , we introduce an alternative formula for the numerical calculation of @xmath1 in the next section . as mentioned at the end of the previous section , it is difficult to solve the exact formula eqs.([eq : delta_pi/2 ] ) and ( [ eq : g_pi/2 ] ) numerically . then , we approximate them by an alternative equation . instead of eq.([eq : eliashberg_k ] ) , we introduce a modified eliashberg equation , @xmath203 \delta_{\gamma \gamma^{\prime}}(k^{\prime},{\mbox{\boldmath $ r$ } } ) . \label{eq : eliashberg_kpi}\end{aligned}\ ] ] this equation is rewritten as @xmath204 . \label{eq : g_pi}\end{aligned}\ ] ] this is the alternative formula for numerical calculations , and does not need an infinite summation like @xmath205 in eq.([eq : g_pi/2 ] ) . equation([eq : eliashberg_k ] ) is not exactly equivalent to eq.([eq : eliashberg_kpi ] ) when @xmath136 and @xmath141 are @xmath147-dependent as in unconventional superconductors . however , we have confirmed that the two different formulae give the qualitatively same results for @xmath1 , and the quantitative difference is small . therefore , hereafter , we use eqs.([eq : delta_pi ] ) and ( [ eq : g_pi ] ) . with the use of the relation @xmath206 for re@xmath207 , the matrix elements are calculated as @xmath208 where @xmath209 the variables @xmath210 and @xmath13 are given by , @xmath211 ^ 2\\ b_1&=s_{\tau_1}\left(\omega_{\tau_1}(k;h)+i[\tilde{\varepsilon}_{\tau_1 } ( k;h)+{\mbox{\boldmath $ v$}}_{\tau_1}^{\lambda}(k;h)\cdot { \mbox{\boldmath $ q$}}^{\lambda}]\right),\\ z_1&=b_1/\sqrt{2a_1},\\ c_{1\pm}&=c_{\tau_1\pm}(k;h)= -is_{\tau_1}l_h^{-1}v_{\tau_1\pm}^{\lambda}(k;h),\\ s_{\tau_1}&={\rm sgn}(\omega_{\tau_1}(k;h)),\\ v_{\tau \pm}^{\lambda}(k;h)&=v_{\tau 1}^{\lambda}(k;h)\pm iv_{\tau 2}^{\lambda}(k;h),\\ v_{\tau \perp}^{\lambda}&=\sqrt{v_{\tau + } ^{\lambda}v_{\tau -}^{\lambda } } , \end{split}\end{aligned}\ ] ] and @xmath212\right),\\ z_2&=b_2/\sqrt{2a_2},\\ c_{2\pm}&=c_{\tau_2\pm}(k ;- h),\\ s_{\tau_2}&={\rm sgn}(\omega_{\tau_2}(k ;- h ) ) , \end{split}\end{aligned}\ ] ] where @xmath213 , @xmath214 , and @xmath215 . a convenient expression of @xmath216 is obtained through the recurrence formula which is directly derived from eq.([eq : f_def ] ) , @xmath217 the solution is @xmath218 where @xmath219 and @xmath220 . with the expression eqs.([eq : g_pi+ ] ) and ( [ eq : g_pi- ] ) , the numerical calculation of eq.([eq : delta_pi ] ) is straightforward . similar expression for @xmath221 can be obtained in the same way and we can also solve eq.([eq : delta_pi/2 ] ) numerically . as mentioned above , eq.([eq : delta_pi/2 ] ) and eq.([eq : delta_pi ] ) give the qualitatively same results and the quantitative difference is small . the important point is that eq.([eq : delta_pi ] ) allows us to calculate @xmath1 for general lattice models with arbitrary fermi surfaces , taking into account both the orbital and the pauli depairing effect on an equal footing . in the gintzburg - landau approach , the relative strength of the orbital and the pauli depairing effect is characterized by the maki parameter @xmath222 , where @xmath223 and @xmath224 are the orbital and the pauli limiting field , respectively . in our formulation , however , we do not need such a parameter which is difficult to be determined experimentally . the parameter corresponding to @xmath225 in this study is an effective mass of the quasiparticle for the cyclotron motion @xmath226 where @xmath50 is the energy unit of the lattice model and @xmath46 is the length unit , i.e. , the lattice constant . writing @xmath227 and @xmath228 with dimensionless variables @xmath229 and @xmath230 , we have a simple identity , @xmath231 a large effective mass corresponds to a slow velocity of the quasiparticles for the cyclotron motion leading to a suppression of the orbital depairing effect . @xmath232 can be determined reasonably , while evaluating @xmath225 from experiments is rather difficult because @xmath223 and @xmath224 are not directly observed , especially for the strong coupling superconductors . the lattice constant is determined experimentally , and we fix @xmath233( ) in this study , which is consistent with the experiments . @xcite we can also determine the value of @xmath50 in a reasonable way . by solving the eliashberg equation([eq : eliashberg_kpi ] ) at @xmath234 , we obtain @xmath235 in the unit of @xmath50 . then , comparing it with the experimentally observed transition temperature @xmath236 , we have @xmath237 in this way , the parameters of the model are evaluated . however , the choice of all the parameters is not unique and there remains some ambiguity especially for the strength of the interaction on which @xmath235 largely depends . therefore , we change the value of the strength of the interaction depending on the choice of the magnitude of @xmath50 to make @xmath10 consistent with the observed values . for the calculation of @xmath1 in cerhsi@xmath0 and ceirsi@xmath0 , we use two values of @xmath50 and compare the results . in addition to the treatment of the pauli and the orbital depairing effect , the strong coupling effect can be included naturally in eq.([eq : delta_pi ] ) . once we calculate the pairing interaction @xmath136 and the selfenergy @xmath238 for a given hamiltonian , they are directly incorporated into the eliashberg equation([eq : delta_pi ] ) . this feature is essentially important for the study of @xmath1 in cerhsi@xmath0 and ceirsi@xmath0 , because it is considered that they are located near the af qcps and the quasiparticles interact with each other through the strong spin fluctuations . in this section , we show the numerical results calculated from eq.([eq : delta_pi ] ) with eq.([eq : g_pi ] ) . we solve the eliashberg equation both for @xmath234 and @xmath239 . for the latter , we study the two cases : @xmath240 and @xmath241 . in the case of @xmath25 , the pauli depairing effect is strongly suppressed by the anisotropic spin - orbit interaction , and @xmath1 is determined by the orbital limiting field @xmath223 . @xcite on the other hand , for @xmath5 , the pauli depairing effect is significant because of the anisotropic distortion of the fermi surface due to the rashba so interaction , and @xmath1 is mainly determined by the pauli limiting field @xmath224 . in this section , we use the action eq.([eq : action ] ) to calculate @xmath1 . the selfenergy @xmath76 has the real and the imaginary part which have different effects , respectively . @xmath118 only gives the deformation of the fermi surface , and , as in sec.[sec : normal ] , it is reasonable to consider that @xmath117 already includes the shift due to @xmath118 and to replace @xmath242 . on the other hand , @xmath243 gives two important effects for the quasiparticles around the fermi level . one is the damping factor @xmath244 and the other is the mass enhancement factor @xmath245 $ ] . especially , the former gives rise to the depairing effects of the cooper pair due to the inelastic scattering , which would lower @xmath10 . for @xmath246 , however , such a suppression does not occur because @xmath247 as @xmath248 . this property is a key for the colossal enhancement in @xmath1 for @xmath249-axis . we , next , consider the pairing interaction between the quasiparticles due to the strong spin fluctuations near the qcp . they are evaluated at the lowest order in @xmath75 , @xmath250 and the other components are zero . these are directly derived from eq.([eq : action_sf ] ) . although the applied fields might affect @xmath136 , we neglect such an effect in this section . the @xmath26-dependence of @xmath136 can be included within our approach , if the field dependence of @xmath53 is clarified by some experiments . we also note that , in eqs.([eq : int_1])@xmath95([eq : int_3 ] ) , the spin - flip scattering processes are not included . they are expected to enhance the mixing of the spin singlet and the triplet superconductivity . these two neglected effects are discussed in sec.[sec:2points ] . as noted in sec.[sec : normal ] , the coupling constant @xmath69 should be regarded as an effective one renormalized by the vertex corrections @xcite . in this section , we study the gap function and the transition temperature at @xmath234 by solving eq.([eq : eliashberg_kpi ] ) . in this case , @xmath141 does not depend on the center of mass @xmath174 and @xmath251 is simplified as @xmath252 . among the five irreducible representations of the point group c@xmath253 for cerhsi@xmath0 and ceirsi@xmath0 , the most stable symmetry of the gap functions is a@xmath254 symmetry which is consistent with the previous study . @xcite the @xmath147-dependence of the singlet gap function is @xmath255 , and that of the triplet gap function is @xmath256 , as will be discussed in detail in sec.[subsec : flipping ] . in the previous study for @xmath2 , we have neglected the triplet part of the gap function , because it is much smaller than the singlet one in amplitude . @xcite in the present study , we take it into account and show that the results in the previous study are not changed . in fig.([fig : g - t ] ) , the transition temperatures for this a@xmath254 symmetric superconducting state for several @xmath61 are shown as functions of @xmath75 . as functions of @xmath257 for several @xmath61 at @xmath234 . the curves correspond to @xmath258 and @xmath259 from the top to the bottom . ] @xmath10 saturates for large @xmath75 because the strength of the pairing interaction and that of the depairing effect through the normal selfenergy become comparable . note that the dependence of @xmath10 on @xmath61 is weak . the coupling constant @xmath69 is fixed so that the calculated @xmath10 is of the same order as the experimentally observed @xmath10 . in cerhsi@xmath0 and ceirsi@xmath0 , kondo temperature is @xmath51 50 - 100 ( k ) @xcite and the resistivity saturates around 200@xmath95300 ( k ) , @xcite which implies that the hopping integral @xmath50 in our model is @xmath260 - 100 ( k ) . on the other hand , observed @xmath10 is @xmath15(k ) , that is , @xmath261-@xmath262 . to reproduce this @xmath10 in the calculation , we fix @xmath263 - 15 . for these values , the system is in a strong coupling region . the renormalization factor averaged on the fermi surface is @xmath264 for @xmath263 , and @xmath13 is not sensitive to @xmath61 , which is characteristic of the 3d af spin fluctuations . @xcite below , we mainly study the case of @xmath263 for which @xmath10 for the minimum @xmath265 is @xmath266 . setting @xmath267(k ) , which is an averaged value of @xmath10 for cerhsi@xmath0 @xcite and ceirsi@xmath0 , @xcite we have @xmath268(k ) . we also consider the case of @xmath269 for in - plane fields in sec.[subsec : hab ] . in this case , similarly , we have @xmath270 ( k ) and @xmath271(k ) . in this section , we calculate the upper critical fields for @xmath5 . in this case , the parameter @xmath187 which characterizes the anisotropy in the @xmath272-plane is @xmath273 . the other parameter @xmath186 which should be optimized is @xmath274 because , for @xmath186 to be finite , the interband pairing on the split fermi surface is required . however , such a pairing is energetically unfavorable . to study @xmath1 , we fix the strength of the coupling constant as @xmath275 . in this calculation , the admixture of the singlet and the triplet components of the gap functions is fully taken into account , which is neglected in the previous paper . @xcite the results are almost unchanged from the previous ones even if we include the effect of the admixture . in fig.[fig : h_c ] , @xmath2 curves as functions of @xmath11 for several @xmath61 are shown . at @xmath263 for several @xmath61 . the dotted curve with triangles is pauli limiting field @xmath224 for @xmath276 and the solid curves with circles are orbital limiting fields @xmath223 for @xmath277 . the dotted curves with squares are @xmath1 curves including both the pauli and the orbital depairing effects for @xmath278 from the up to the bottom . the @xmath223 curve for @xmath279 coincides with the @xmath1 curve with squares . ] the pauli limiting field @xmath224 is large because the rashba so interaction is strong , @xmath280 . in such a case , the quasiparticles are easily paired on the same band under the applied field @xmath281 . this holds generally and does not depend on the symmetry and the dominant parity of the gap functions for rashba superconductors . @xcite the upper critical field is , therefore , mainly determined by the orbital limiting field @xmath223 . however , as seen in fig.[fig : h_c ] , the orbital limiting field @xmath223 is different from @xmath1 calculated with both the pauli and the orbital depairing effect being taken into account , especially for large @xmath26 . it would be natural to think that this difference is a numerical artifact due to our choice of parameters . the magnitude of @xmath32 used in the above calculations is not sufficiently large for high @xmath26 regions . if one uses the large value of @xmath282 , where @xmath283 $ ] with the density of states at the fermi level @xmath284 , this difference may disappear . in fact , in the experimental data of @xmath1 both in cerhsi@xmath0 and ceirsi@xmath0 , no clear pauli depairing effect can be seen , which implies that the zeeman effect is effectively negligible in the compounds . however , to carry out the numerical calculations of @xmath1 for the larger @xmath282 , we need the large size of the @xmath147-mesh and a large number of matsubara frequencies . we have also calculated @xmath2 for @xmath269 . although , in this case , the zeeman effect much affects @xmath1 compared with the @xmath263 case , the qualitative behavior of @xmath223 is unchanged . the calculated @xmath1 show ( i ) strong @xmath61 dependence and ( ii ) upward curvatures , and ( iii ) they reach @xmath285(t ) . all these characteristic behaviors well explain the experimental observations in cerhsi@xmath0 and ceirsi@xmath0 discussed in sec.[sec : intro ] . @xcite the physical reason for these characteristic behaviors in @xmath2 is quite simple . because , for @xmath25 , @xmath1 is determined mainly by the orbital depairing effect and the orbital limiting field @xmath223 can be strongly enhanced by the spin fluctuations near the qcp . in the quantum critical regime , as @xmath11 is decreased below @xmath286 , the pairing interaction @xmath287 is increased in magnitude while the inelastic scattering between electrons is suppressed and the quasiparticle damping is decreased , @xmath288 . this contrasting behaviors of the pairing interaction and the depairing effect lead to the large enhancement in @xmath1 for @xmath248 near the qcp . on the other hand , as discussed in the next section , the pauli limiting field @xmath224 is not so strongly enhanced by the spin fluctuations at low temperatures . this is a key to resolve the apparent contradiction that although there are many heavy fermion compounds which are considered to be located near magnetic qcps , they do not show such a huge @xmath1 as in cerhsi@xmath0 and ceirsi@xmath0 . in usual centrosymmetric heavy fermion superconductors , @xmath1 is considered to be mainly determined by the pauli depairing effect . therefore , even if the system is close to the qcp , @xmath1 is not anomalously enhanced . the pressure(@xmath61 ) dependence of @xmath289 shows a remarkable feature as a result of above mentioned mechanism . we define normalized @xmath10 and @xmath1 as functions of @xmath61 , @xmath290 and @xmath291 where @xmath292 and @xmath293 . the normalized orbital limiting field @xmath294 is also defined in the same way . in fig.[fig : th - ht ] , @xmath295 and @xmath294 are shown for @xmath275 . for @xmath296 , the dotted curve with triangles is calculated from @xmath223 , and the dotted curve with squares includes both the pauli and the orbital depairing effect . and @xmath296 as functions of @xmath61 . the dotted curve with triangles is calculated from @xmath223 , and the dotted curve with squares includes both the pauli depairing and the orbital effects . the definitions of @xmath297 and @xmath298 are given in the text . ] the @xmath61 dependence of @xmath299 is moderate , while those of both @xmath296 and @xmath294 are significant . as explained above , these behaviors are understood as a result of the strongly enhanced pairing interaction and the suppression of the depairing effect at low temperatures in the vicinity of the qcp(@xmath300 ) . since , in cerhsi@xmath0 and ceirsi@xmath0 , the so interaction makes the superconductivity orbital limited , the huge @xmath1 is a result of the interplay of the rashba so interaction and the electron correlations . generally , such strong enhancement in the pairing interaction and the suppression of the quasiparticle damping at low temperatures are crucial for orbital limited superconductors , because @xmath223 is largely affected by the electron correlations compared with @xmath224 . therefore , the enhanced upper critical field can be considered as a universal property of the orbital limited superconductors near qcps . this would be related to the recent experiments of @xmath301-axis in ucoge in which the relation between the superconductivity and the ferromagnetism has been discussed . @xcite the observed @xmath302 is huge @xmath303 ( t ) while @xmath15 ( k ) . @xcite this issue is now under investigation . we also study @xmath1 for the case of @xmath5 within the same framework . since in this case , the fermi surface is distorted asymmetrically by the in - plane filed through the rashba so interaction , the pauli depairing effect is significant , which implies that the higher landau levels become important . furthermore , the optimization parameter @xmath186 and @xmath187 are nontrivial for @xmath5 . first , at a fixed @xmath26 , we optimize @xmath187 which corresponds to the anisotropy in the quasiparticle velocity of the two directions perpendicular to the applied field , or the anisotropy in the superconducting coherence length . since @xmath187 is characterized by the shape of the fermi surface , the field dependence of @xmath187 is very weak . we fix the optimized @xmath187 , and then , optimize @xmath186 to have the maximum @xmath1 for given temperatures . the optimal @xmath187 is @xmath305 . in fig.[fig : h_ab_10 ] , we show @xmath1 at @xmath275 for two values of @xmath61 , @xmath306 and @xmath307 . each @xmath1 curve is calculated with a single landau function for @xmath308 , respectively . true @xmath1 curve should be calculated by a superposition of the landau functions . we have computed a @xmath1 curve by using a superposition of @xmath309 and @xmath310 landau functions , and found that it almost coincides with the @xmath1 curve calculated by @xmath309 landau function only . at @xmath263 for @xmath311 ( square symbols ) and @xmath312 ( circle symbols ) . for each @xmath61 , three curves correspond to @xmath308 landau levels from the top to the bottom . ] therefore , @xmath1 is mainly determined by the @xmath309 landau level , and the shapes of @xmath309 @xmath1 curves for @xmath313 and @xmath314 are similar . this pressure insensitivity is due to the weak dependence of the pauli limiting field @xmath224 on the electron correlations compared with @xmath223 . the ratio of the calculated value of @xmath315 to that of @xmath316 in the previous section is @xmath317 for @xmath313 . these behaviors in @xmath1 are consistent with the experiments . @xcite we turn to the discussion of the modulation vector @xmath186 . under the field @xmath281 , the dispersion is changed as @xmath318 . in this situation , the momentum pair @xmath319 is energetically degenerate on the one band , where @xmath320 is the fermi momentum for @xmath321-band and @xmath322 satisfies @xmath323 . note that the center of mass momenta of the pairs on each band satisfy @xmath324 , and the electrons on each band favor the center of mass momenta with opposite directions . therefore , it is expected that for sufficiently strong @xmath26 , each band favors each @xmath186 and the resulting superconducting state would be the fulde - ferrell - ovchinnikov - larkin ( fflo ) state @xcite with @xmath325 + \eta_{-}\exp[i{\mbox{\boldmath $ q$}}_-\cdot { \mbox{\boldmath $ r$}}]$ ] . for small @xmath26 , however , the helical vortex state with @xmath326 $ ] is considered to be stabilized in general noncentrosymmetric superconductors . so , the situation is different between the case of small @xmath26 and that of large @xmath26 . we discuss the @xmath26-dependence of the modulation of the gap function qualitatively . when the so split inter - band pairing which is small for @xmath327 is neglected , the eliashberg equation for the diagonal element of the gap @xmath328 is of the form , @xmath329 where @xmath330 is the density of states at the fermi level for the @xmath321-band and @xmath331 is a function depending on @xmath332 . the first term in eq.([eq : gapeq ] ) is proportional to just the sum @xmath333 and therefore , the electrons on each band contribute independently . in contrast , in the second term , the difference between the two bands plays important roles . the term is related to the magnetoelectric effect in the superconducting state due to the anisotropic so interaction , which depends on the difference in the densities of states of the two bands . this effect is incorporated into the ginzburg - landau free energy as @xmath334 $ ] where @xmath335 and @xmath336 is the coefficient of the magnetoelectric effect . @xcite in general noncentrosymmetric superconductors , @xmath337 leads to a spatially modulated gap function with the modulation vector @xmath338 . in this state , the cooper pair is formed by the states with @xmath339 and @xmath340 momenta , not the @xmath341 momenta . this effect arises even under very weak @xmath26 . however , it is pointed out that in 3d rashba superconductors in which only @xmath342 are nonzero , the phase @xmath343 $ ] is absorbed into the landau function as a spatial shift @xmath344 with a @xmath26-dependent vector @xmath345 . @xcite therefore , @xmath346 does not appear in physical observables like @xmath1 , although @xmath347 itself is nonzero . on the other hand , under a high field , the first term in eq.([eq : gapeq ] ) plays important roles for the optimization of @xmath186 . this situation is similar to that of the fflo state in usual centrosymmetric superconductors . since , as noted above , the first term of eq.([eq : gapeq ] ) is a sum of the independent contributions from two bands , it merely favors @xmath348 or @xmath349 . therefore , in the high field region , the candidate for the modulation vector is @xmath348 and @xmath349 . if @xmath26 is applied from zero to some large value for general noncentrosymmetric superconductors , we would see a continuous change of @xmath186 , from @xmath346 to @xmath350 . the threshold value @xmath351 at which @xmath186 changes from @xmath346 to @xmath350 depends on the details of the system . it is pointed out that @xmath352 becomes large as the orbital depairing effect increases . in our study , @xmath186 is determined so that @xmath1 becomes maximum for a given parameter . we find that the optimized @xmath186 vector is parallel to @xmath46-axis , @xmath353 for @xmath354 . in fig.[fig : h_q_g10 ] , the optimized @xmath22 along the @xmath1 curve for @xmath313 is shown . along the @xmath1 curve for @xmath355 and @xmath313 . in the low field region , the optimization of @xmath22 would be due to a numerical artifact and @xmath22 is shown by the dotted curve in this region . @xmath352 is the threshold value . ] we have two regions ; the region where @xmath22 is quite small and the other with large @xmath22 . for the small @xmath22 region , although we have a systematic change of @xmath1 with respect to @xmath22 and can optimize it , this dependence of @xmath1 on @xmath22 would be a numerical artifact because the change in @xmath1 due to nonzero @xmath22 is infinitesimally small . actually , the small @xmath22 region corresponding to the helical vortex state is spurious in a 3d rashba superconductor because of the reason mentioned above . it is expected that the character of the stable vortex state is nothing but the character of the conventional vortex state with @xmath356 in the region @xmath357 . in contrast , for large @xmath358 , we have a finite @xmath22 . this large @xmath22 state would not be a direct result of the lack of the inversion center . rather , it is stabilized by the pairing of the momentum @xmath341 electrons on each band . however , the contribution to the gap function from the second term in eq.([eq : gapeq ] ) is not negligible , resulting in the shift of the degeneracy between @xmath359 and @xmath360 . therefore , we expect that , in this high field region , the fflo state with the gap function @xmath361 + \eta_{-}\exp [ i{\mbox{\boldmath $ q$}}_-\cdot { \mbox{\boldmath $ r$}}]$ ] can be realized . we have performed the calculations for other parameters , and confirmed that the threshold @xmath351 for the two region depends on the effective mass @xmath232 . as @xmath232 becomes larger , @xmath352 decreases , and vice versa , which means that the orbital depairing effect plays important roles for the determination of @xmath352 . to discuss the stability of such a state , we need to compute the free energy in the superconducting state , and it is beyond the linearized calculation of @xmath1 performed in the present study . as mentioned in sec.[sec : eliashberg ] , @xmath75 can not be determined uniquely in our theory , and we can change the value of @xmath75 within the range for which the value of @xmath10 is consistent with the experiments . in the following , we study the case of @xmath362 which gives @xmath363 . in this case , the value of @xmath50 is @xmath271(k ) and the effective mass of the cyclotron motion @xmath232 is large compared with the @xmath268(k ) case . we show @xmath1 for @xmath364 in fig.[fig : h_ab_15 ] . [ cols="^,^ " , ] these behaviors are basically the same as those for @xmath365 , and assert the robustness of the spin fluctuations against the in - plane field . from the above results , we see that neglecting the field dependence in the pairing interaction for any field direction is legitimate provided @xmath366 . although @xmath1 is huge in cerhsi@xmath0 and ceirsi@xmath0 especially for @xmath367-axis , the condition @xmath366 is expected to be satisfied . therefore , the discussion of @xmath1 in sec.[sec : hc2 ] is not changed even if we consider the field dependence of the spin fluctuations . we have discussed the normal and the superconducting properties in noncentrosymmetric heavy fermion superconductors cerhsi@xmath0 and ceirsi@xmath0 . we have shown that the @xmath11-linear dependence of the resistivity above @xmath10 observed experimentally is naturally understood within the 3d spin fluctuations near the af qcp . for the superconducting state , we have derived a formula from the eliashberg equation in real space . the formula enables us to treat the pauli and the orbital depairing effects on an equal footing . furthermore , by using it , we can calculate @xmath1 for strong coupling superconductors with general fermi surfaces . we have calculated @xmath1 with the formula and have well explained the observed features of @xmath1 in cerhsi@xmath0 and ceirsi@xmath0 . for @xmath25 , @xmath224 is infinitely large due to the rashba so interaction , and @xmath1 is determined by @xmath223 . as temperature is lowered and the system approaches the qcp , the pairing interaction becomes larger while the quasiparticle life time becomes longer , which results in the huge @xmath368 with the strong pressure dependence . the enhancement of the orbital limiting field near qcps by this mechanism would be universal . we have also discussed the case for @xmath5 . in this case , the pauli depairing effect is significant because of the asymmetric distortion of the fermi surface , and resulting @xmath1 is moderate against the pressure . the fflo state can be stabilized for a large @xmath26 region , although such a region is very small . the features of the calculated @xmath1 for both @xmath25 and @xmath5 are in good agreement with the experiments . this consistency supports the scenario that the superconductivity in cerhsi@xmath0 and ceirsi@xmath0 is mediated by the spin fluctuations near the af qcp . in the last section , we have checked the legitimacy of our approximation used for the calculation of @xmath1 . in cerhsi@xmath0 and ceirsi@xmath0 , the admixture of the singlet and the triplet gap functions are small even if we take into account the spin - flip scattering processes in the pairing interaction . in noncentrosymmetric systems , the spin susceptibility is robust against the applied magnetic fields @xmath27 . for this reason , the spin fluctuations near the af qcp in cerhsi@xmath0 and ceirsi@xmath0 remain strong even under a large magnetic field @xmath285 ( t ) . therefore , the above mentioned results for @xmath1 is not changed if we refine our approximation used in the calculation of @xmath1 . we thank n. kimura , r. settai and y. nuki for valuable discussions . numerical calculations were partially performed at the yukawa institute . this work is partly supported by the grant - in - aids for scientific research from mext of japan ( grant no.18540347 , grant no.19052003 , grant no.20029013 , grant no.20102008 , grant no.21102510 , and grant no.21540359 ) and the grant - in - aid for the global coe program `` the next generation of physics , spun from universality and emergence '' . y. tada is supported by jsps research fellowships for young scientists . y. okuda , y. miyauchi , y. ida , y. takeda , c. tonohiro , y. oduchi , t. yamada , n. d. dung , t. d. matsuda , y. haga , t. takeuchi , m. hagiwara , k. kindo , h. harima , k. sugiyama , r. settai , and y. nuki , j. phys . . jpn . * 76 * , 044708 ( 2007 ) .

we revisit merton s portfolio optimization problem under boun - ded state - dependent utility functions , in a market driven by a lvy process extending results by karatzas et . al . and kunita . the problem is solved using a _ dual variational problem _ as it is customarily done for non - markovian models . one of the main features here is that the domain of the _ dual problem _ enjoys an explicit `` parametrization '' , built on a _ multiplicative optional decomposition _ for nonnegative supermartingales due to fllmer and kramkov . as a key step in obtaining the representation result we prove a _ closure property _ for integrals with respect to poisson random measures , a result of interest on its own that extends the analog property for integrals with respect to a fixed semimartingale due to mmin . in the case that ( i ) the lvy measure of is atomic with a finite number of atoms or that ( ii ) for a process and a deterministic function , we explicitly characterize the admissible trading strategies and show that the dual solution is a risk - neutral local martingale .

0901.2070

the task of determining good trading strategies is a fundamental problem in mathematical finance . a typical approach to this problem aims at finding the trading strategy that maximizes , for example , the final expected utility , which is defined as a deterministic , concave , and increasing function @xmath5 of the final wealth . there are , however , many applications where a utility function has to change with the underlying securities , or more generally , with the source of randomness ( say a brownian motion ) . for example , in the so - called _ optimal partial replication _ of a contingent claim , introduced by fllmer and leukert @xcite , one tries to find the trading strategy that best replicates the claim @xmath6 under a budget constraint . in particular , when the market is incomplete , it is often more beneficial to allow certain degree of shortfall " in order to reduce the super hedging cost " , a threshold for the minimum initial wealth so that super - hedging is feasible ( see , e.g. , @xcite and @xcite for more details ) . mathematically , such a shortfall risk could be measured by the expected loss @xmath7,\ ] ] where @xmath8 is the loss function " , a convex increasing function that incorporates the investor s attitude towards the shortfall @xmath9 , and the value process @xmath10 is subject to the constraint @xmath11 . such a problem can then be formulated as a utility maximization problem with a bounded _ state - dependent utility _ , in which the utility function is defined by ( cf . @xcite ) : @xmath12 in general , we can define a _ state - dependent utility _ as a function @xmath13 such that @xmath14 is a utility function for each @xmath15 . the utility maximization problem is then defined as @xmath16 : \ ; v\;{\rm is}\;{\rm admissible}\;{\rm and } \ ; v_{0}\leq{}z \right\},\ ] ] where the supreme is taken over all wealth processes @xmath17 generated by admissible trading strategies ( see section [ finmodelsect ] for a precise definition ) . the existence and essential uniqueness of the solution to the problem ( [ primalproblemreal ] ) was proved in @xcite for a general semimartingale price model using a convex duality method , built on a celebrated bipolar theorem by kramkov and schachermayer @xcite . however , this approach does not seem to shed any light on how to compute , in a feasible manner , the optimal trading strategy . this is partly due to the generality of the problem considered there . in this paper we shall consider the market model in which the price is driven by a lvy process , and we propose a more manageable dual problem with a specific domain . we should note that our method can be extended to handle more general jump - diffusion models driven by even additive processes . the problem of utility maximization can be traced back to merton @xcite-@xcite . in a brownian - driven market model , karatzas et . @xcite developed a program , known as the _ convex duality method _ , that has become one of the most powerful methods , yet relatively explicit and simple , to analyze optimal portfolio problems in non - markovian markets . they prove that the marginal utility of the optimum final wealth is proportional to the risk - neutral local martingale that minimizes a dual " problem , defined as another optimization problem with the objective function being the legendre - fenchel - type transformation of the original utility function . to be more precise , consider a minimization problem @xmath18 , \qquad y>0,\ ] ] where @xmath19 , the so - called _ dual domain or class _ , consists of ( risk - neutral ) exponential local martingales , and @xmath20 stands for the _ convex dual function _ of @xmath21 . the idea is first to find , for any @xmath22 , a minimizer @xmath23 of ( [ dualprob0 ] ) , which in turn induces a potential " optimal terminal wealth @xmath24 in the sense that the so - called weak duality relation @xmath25\ ] ] holds . if one can further show that for some @xmath26 , there exists an admissible portfolio @xmath27 such that @xmath28 , then clearly equality holds in ( [ wd ] ) ( a property typically called _ strong duality _ ) , and @xmath27 solves the original problem . customarily , finding the optimal portfolio @xmath27 relies on a variational problem for the dual value function and the existence of the minimizer @xmath29 utilizes the particular form of the market model and some general properties of the utility function . more recently , the convex duality method was further extended to a general `` jump - diffusion '' market by kunita @xcite building on an exponential representation for _ positive _ local supermartingales as well as a variational equality for the dual problem . to ensure the attainability of the dual problem , it is required that the utility function satisfies the same conditions as @xcite ( one of which is unboundedness ) , and that the dual domain @xmath19 contain all positive `` risk - neutral '' local supermartingales . the main purpose of this paper is to further extend the seminal approach of @xcite to the case of _ state - dependent , bounded _ utility functions . for simplicity , we will be contented with a market with only one stock , whose jumps are driven by a lvy process @xmath30 , but our analysis can be readily extended to more general jump - diffusion multidimensional models such as the one considered in @xcite . we should emphasize that the boundedness and potential non - differentiability of the utility function causes some technical subtleties . for example , the dual optimal process can be @xmath31 with positive probability , thus the representation theorem of kunita does not apply anymore . to get around these difficulties we shall reconsider the dual problem over an arbitrary subclass . using an exponential representation for nonnegative supermartingales due to fllmer and kramkov @xcite , we show how to construct suitable explicit dual classes associated with certain classes of semimartingales that are closed under mery s topology . to work with this last condition , we prove a closure property for integrals with respect to poisson random measures , a result of interest on its own that extends the analog property for integrals with respect to a fixed semimartingale due to mmin @xcite . it is also worth mentioning that part of our approach relies on the fundamental characterization of contingent claims that are super - replicable ( see @xcite and @xcite ) , while that of fllmer and leukert @xcite ( see also xu @xcite ) was based on the bipolar theorem of @xcite . we feel that the convex duality approach of @xcite that we develop in this paper offers several advantages . the proofs are more direct and the dual problem might be more suitable for computational purposes since the dual class enjoys an `` explicit '' description and `` parametrization '' . in the case that ( i ) the jumps of the price process @xmath32 are driven by the superposition of finitely - many shot - noise poisson processes or that ( ii ) @xmath2 , we are even able to show that the dual solution is a risk - neutral local martingale . we would like to remark that some of our results in section [ convdualsect ] below may look similar to those in ( * ? ? * chapter 3 ) , but there are essential differences . for example , the model in @xcite exhibits only finite - jump activity , while our model allows general jumps . also , @xcite allows only downward price jumps , an assumption that seems to be crucial for the approach there , which was based on the existence of the solutions to certain stochastic differential equations ( see , e.g. , ( * ? ? ? * lemma 3.3 , proposition 3.4 ) ) . we should point out that our approach is also valid for general additive processes , including the time - inhomogeneous cases considered in @xcite . we present an argument in ( ii ) of section [ fnlrmrkssect ] to justify this point . the paper is organized as follows . in section [ finmodelsect ] we introduce the financial model , along with some basic terminology that will be used throughout the paper . the convex duality method is revised in section [ convdualsect ] , where a potential optimal final wealth satisfying ( [ wd ] ) is constructed . an explicit description of a dual class for which equality in ( [ wd ] ) holds is presented in section [ specdualclsssect ] , along with some interesting simple characterizations of the dual optimum . in particular , as it was mentioned earlier , we prove that under certain conditions in the structure of the jumps , the dual optimum is actually a local martingale and we also provide an explicit characterization of the admissible trading strategies . in section [ rplcbltysect ] we show that the potential optimal final wealth is attained by an admissible trading strategy , as the last step for proving the existence of an optimal portfolio . finally , we give some concluding remarks in section [ fnlrmrkssect ] . some necessary fundamental theoretical results behind our approach are collected in appendix [ covclsssect ] , such as the exponential representation for nonnegative supermartingales of fllmer and kramkov @xcite and the closure property for integrals with respect to poisson random measures that was previously mentioned . throughout this paper we assume that all the randomness comes from a complete probability space @xmath33 , on which there is defined a lvy process @xmath0 with lvy triplet @xmath34 ( see sato @xcite for the terminology ) . by the lvy - it decomposition , there exist a standard brownian motion @xmath35 and an independent poisson random measure @xmath36 on @xmath37 with mean measure @xmath38 , such that @xmath39 where @xmath40 . let @xmath41 be the natural filtration generated by @xmath35 and @xmath36 , augmented by all the null sets in @xmath42 so that it satisfies the _ usual conditions _ ( see e.g. @xcite ) . * the market model * we assume that there are two assets in the market : a risk free bond ( or money market account ) , and a risky asset , say , a stock . the case of multiple stocks , such as the one studied in @xcite , can be treated in a similar way without substantial difficulties ( see section [ fnlrmrkssect ] for more details ) . as it is customary all the processes are taken to be discounted to the present value so that the value @xmath43 of the risk - free asset can be assumed to be identically equal to @xmath44 . the ( discounted ) price of the stock follows the stochastic differential equation @xmath45 where @xmath46 , @xmath47 , @xmath48 , and @xmath49 ( see @xcite for the terminology ) . more precisely , @xmath50 , @xmath51 , and @xmath52 are predictable processes such that @xmath53 a.s . ( hence , @xmath54 a.s . ) , and that @xmath55 are locally integrable . finally , we assume that the market is free of arbitrage so that there exists a risk - neutral probability measures @xmath56 such that the ( discounted ) process @xmath57 , @xmath58 , is an @xmath59-local martingale under @xmath56 . throughout , @xmath60 will stand for the class of all equivalent risk neutral measures @xmath56 . * admissible trading strategies and the utility maximization problem . * a trading strategy is determined by a predictable locally bounded process @xmath61 representing the _ proportion of total wealth invested in the stock_. then , the resulting wealth process is governed by the stochastic differential equation : @xmath62 where @xmath63 stands for the initial endowment . for future reference , we give a precise definition of _ admissible strategies _ " . the process @xmath64 solving ( [ eqdiscvalprocreal ] ) is called the value process corresponding to the self - financing portfolio with initial endowment @xmath63 and trading strategy @xmath61 . we say that a value process @xmath65 is admissible " or that the process @xmath61 is admissible " for @xmath63 if @xmath66 . $ ] for a given initial endowment @xmath63 , we denote the set all admissible strategies for @xmath63 by @xmath67 , and the set of all admissible value processes by @xmath68 . in light of the dolans - dade stochastic exponential of semimartingales ( see e.g. section i.4f in @xcite ) , one can easily obtain necessary and sufficient conditions for admissibility . [ cndadmissibility ] a predictable locally bounded process @xmath61 is admissible if and only if @xmath69=1.\ ] ] to define our utility maximization problem , we begin by introducing the _ bounded state - dependent utility function_. [ condonutility ] a random function @xmath70 is called a bounded and state - dependent utility function " if * @xmath71 is nonnegative , non - decreasing , and continuous on @xmath72 ; * for each fixed @xmath63 , the mapping @xmath73 is @xmath74-measurable ; * there is an @xmath74-measurable , positive random variable @xmath6 such that for all @xmath15 , @xmath71 is a strictly concave differentiable function on @xmath75 , and it holds that @xmath76<\infty ; \end{aligned}\ ] ] notice that the @xmath74-measurability of the random variable @xmath77 is automatic because @xmath78 is @xmath79-measurable in light of the above conditions 1 and 2 . we remark that while the assumption ( [ eufin ] ) is merely technical , the assumption ( [ ubdd ] ) is motivated by the shortfall risk measure ( [ u ] ) . our utility optimization problem is thus defined as @xmath80 : \ ; v\in \calv^w_{ad } ~~{\rm with } ~~ w\leq{}z \right\}.\ ] ] for any @xmath81 . we should note that the above problem is relevant only for those initial wealths @xmath82 that are smaller than @xmath83 , the super - hedging cost of @xmath6 . indeed , if @xmath84 , then there exists an admissible trading strategy @xmath85 for @xmath82 such that @xmath86 almost surely , and consequently , @xmath87 $ ] ( see @xcite and @xcite for this _ super - hedging _ result ) . our main objectives in the rest of the paper are the following : ( 1 ) define the dual problem and identify the relation between the value functions of the primal and the dual problems ; ( 2 ) by suitably defining the dual domain , prove the attainability of the associated dual problem ; and ( 3 ) show that the potential optimum final wealth induced by the minimizer of the dual problem can be realized by an admissible portfolio . we shall carry out these tasks in the remaining sections . in this section we introduce the _ dual problems _ corresponding to the primal problem ( [ primalproblemreal ] ) . we begin by defining the so - called _ convex dual function _ of @xmath14 : @xmath88 we note that the function @xmath89 is closely related to the legendre - fenchel transformation of the convex function @xmath90 . it can be easily checked that @xmath91 is convex and differentiable everywhere , for each @xmath92 . furthermore , if we denote the _ generalized inverse _ function of @xmath93 by @xmath94 with the convention that @xmath95 , then it holds that @xmath96 and the function @xmath97 has the following representation @xmath98 [ measurability2 ] we point out that the random fields defined in ( [ convdual])-([inverse ] ) are @xmath79-measurable . for instance , in the case of @xmath89 , we can write @xmath99 and we will only need to check that @xmath100 is jointly measurable for each fixed @xmath82 . this last fact follows because the random field in question is continuous in the spatial variable @xmath101 for each @xmath92 , and is @xmath74-measurable for each y. in light of ( [ derconvexdual ] ) , it transpires that the random field @xmath102 is jointly measurable . given that the subsequent dual problems and corresponding solutions are given in terms of the fields @xmath103 and @xmath102 ( see definition [ defdualproblem ] and theorem [ mainthrm1 ] below ) , the measurability of several key random variable below is guaranteed . next , we introduce the so - called dual class " . [ dualclass ] let @xmath104 be the class of nonnegative supermartingales @xmath105 such that 1 . @xmath106 , and 2 . for each locally bounded admissible trading strategy @xmath61 , @xmath107 is a supermartingale . to motivate the construction of the dual problems below we note that if @xmath108 and @xmath10 is the value process of a self - financing admissible portfolio with initial endowment @xmath11 , then @xmath109\leq z$ ] , and it follows that @xmath110&\leq & \bbe\left[u\left(v_{t}\wedge h,\cdot\right)\right ] -y\left(\bbe\left[\xi(t ) \left(v_{t}\wedge h\right)\right]-z\right)\nonumber\\ & \leq&\bbe\{\sup_{0\leq{z'}\leq h(\cdot)}\left\{u\left(z',\cdot\right ) -y\xi(t ) z'\right\}\}+zy\\ & = & \bbe\{\widetilde{u}(y\xi(t ) , \cdot)\}+zy . \nonumber\end{aligned}\ ] ] for any @xmath111 . the dual problem is defined as follows . [ defdualprobl ] given a subclass @xmath112 , the minimization problem @xmath113 , \quad y>0,\ ] ] is called the dual problem induced by @xmath19 " . the class @xmath19 is referred to as a dual domain ( or class ) and @xmath114 is called its dual value function . notice that , by ( [ dualrel ] ) and ( [ defdualproblem ] ) , we have the following _ weak duality _ relation between the primal and dual value functions : @xmath115 valid for all @xmath116 . the effectiveness of the dual problem depends on the attainability of the lower bound in ( [ weakdualityeq ] ) for some @xmath117 ( in which case , we say that _ strong duality _ holds ) , and the attainability of its corresponding dual problem ( [ defdualproblem ] ) . the following important properties will be needed for future reference . [ proptydual ] the dual value function @xmath118 corresponding to a subclass @xmath19 of @xmath104 satisfies the following properties : 1 . @xmath118 is non - increasing on @xmath119 and @xmath120\leq v_{_{\gamma}}(y)\leq \bbe\left[u(h;\cdot)\right ] . if @xmath121<\infty,\ ] ] then @xmath118 is uniformly continuous on @xmath119 , and @xmath122-v_{_{\gamma}}(y)}{y}= \sup_{\xi\in\gamma}\bbe\left[\xi(t ) h\right].\ ] ] 3 . there exists a process @xmath123 such that @xmath124 \leq v_{_{\gamma}}(y ) . if @xmath19 is a convex set , then ( i ) @xmath118 is convex , and ( ii ) there exists a @xmath125 attaining the minimum @xmath126 . furthermore , the optimum @xmath127 can be `` approximated '' by elements of @xmath19 in the sense that there exists a sequence @xmath128 for which @xmath129 , a.s . for simplicity , we write @xmath130 . the monotonicity and range of values of @xmath52 are straightforward . to prove ( 2 ) , notice that since @xmath91 is convex , non - increasing , and @xmath131 , we have @xmath132- \inf_{\xi}\bbe\left[\widetilde{u}(y\xi(t ) ) \right]}{y}\leq \sup_{\xi\in\gamma}\bbe\left[\xi(t ) h\right].\ ] ] on the other hand , by the mean value theorem , dominated convergence theorem , ( [ derconvexdual ] ) , and assumption [ condonutility ] , @xmath133 \leq { } \liminf_{y\downarrow{}0}\frac{\bbe\left[u(h;\cdot)\right]-v(y)}{y } \leq \sup_{\xi\in\gamma}\bbe\left[\xi(t ) h\right],\ ] ] for every @xmath134 . then , ( 2 ) is evident . uniform continuity is straightforward since for any @xmath135 small enough it holds that @xmath136 the part ( i ) of ( 4 ) is well - known . let us turn out to prove ( 3 ) and part ( ii ) in ( 4 ) . let @xmath137 be such that @xmath138 = v_{_{\gamma}}(y).\ ] ] without loss of generality , one can assume that each process @xmath139 is constant on @xmath140 . by lemma 5.2 in @xcite , there exist @xmath141 , @xmath142 , and a nonnegative supermartingale @xmath143 with @xmath144 such that @xmath145 is _ fatou convergent to @xmath146 on the rational numbers @xmath147 _ ; namely , @xmath148 for all @xmath149 . by fatou s lemma , it is not hard to check that @xmath150 is a supermartingale for every admissible portfolio with value process @xmath10 , and hence , @xmath123 . next , since the @xmath139 s are constant on @xmath140 and @xmath91 is convex , fatou s lemma implies that @xmath151 & \leq v_{_{\gamma}}(y),\end{aligned}\ ] ] finally , we need to verify that , when @xmath19 is convex , equality above is attained and and that @xmath152 can be approximated by elements of @xmath19 . both facts are clear since @xmath153 and @xmath154 a.s . then , by the continuity and boundedness of @xmath89 , @xmath155 = \bbe\left[\widetilde{u}(y\widetilde\xi_{_{t}}b_{t}^{-1},\omega)\right].\ ] ] we now give a result that is crucial for proving the strong duality in ( [ weakdualityeq ] ) . [ mainthrm1 ] suppose that ( [ condonclassofsprmart1 ] ) is satisfied and @xmath19 is convex . then , for any @xmath156 , there exist @xmath157 and @xmath158 such that 1 . @xmath159 \leq \bbe\left[\widetilde{u}\left ( y(z)\xi(t ) , \omega\right)\right ] , \quad { \forall \quad \xi\in\gamma}$ ] ; 2 . @xmath160=z$ ] , where @xmath161 3 . @xmath162 $ ] . we borrow the idea of @xcite . for simplicity let us write @xmath163 instead of @xmath126 . recall that @xmath164 $ ] and define @xmath165 $ ] . in light of lemma [ proptydual ] , the continuous function @xmath166 satisfies @xmath167 for all @xmath168 . thus , @xmath169 attains its minimum at some @xmath170 . by proposition [ proptydual ] , we can find a @xmath171 such that @xmath172,\ ] ] proving the ( i ) above . now , consider the function @xmath173 , \quad u>0.\ ] ] since @xmath174 can be approximated by elements in @xmath19 , for each @xmath175 there exists a @xmath176 such that @xmath177 > \bbe\left[\widetilde{u}\left(y\xi^{y,\varepsilon}_{y(z)}(t ) \right)\right]-\varepsilon.\ ] ] it follows that for each @xmath175 , @xmath178\right\ } -\varepsilon\\ & \geq\inf_{y>0 } \left\ { yz+v(y)\right\}-\varepsilon\\ & = y(z)z+ \bbe\left[\widetilde{u}\left(y(z)\xi_{y(z)}(t ) \right)\right ] -\varepsilon . \end{aligned}\ ] ] since @xmath175 is arbitrary , the function @xmath179 attains its minimum at @xmath180 . on the other hand , @xmath181 equals @xmath182 , \end{aligned}\ ] ] which converges to @xmath183\ ] ] as @xmath184 . here , we use ( [ derconvexdual ] ) and the dominated convergence theorem . then , @xmath185=z.\ ] ] this proves ( ii ) of the theorem , and also , ( iii ) in light of ( [ utilde ] ) and ( [ dualrel ] ) . we note that theorem [ mainthrm1 ] essentially provides an upper bound for the optimal final utility of the form @xmath186 $ ] , for certain reduced " contingent claim @xmath187 . by suitably choosing the dual class @xmath19 , we shall prove in the next two sections that this reduced contingent claim is ( super- ) replicable with an initial endowment @xmath82 . we now give a full description of a dual class @xmath19 for which _ strong duality _ , i.e. , @xmath188 , holds . denote @xmath189 to be the class of all real - valued cdlg , non - decreasing , adapted processes @xmath190 null at zero . we will call such a process increasing " . in what follows we let @xmath191 be the dolans - dade stochastic exponential of the semimartingale @xmath192 ( see e.g. @xcite for their properties ) . let @xmath193 and consider the associated class of exponential local supermartingales : @xmath194 in ( [ bscclass3 ] ) , we assume that @xmath195 , @xmath196 , and that @xmath197 , for all @xmath198 . the following result shows not only that the class @xmath199 is convex , but also that the dual optimum , whose existence is deduced from theorem [ mainthrm1 ] , remains in @xmath19 . the proof of this result is based on a powerful representation for nonnegative supermartingales due to fllmer and kramkov @xcite ( see theorem [ thrmfollmerkramkov ] in the appendix ) , and a technical result about the closedness of the class of integrals with respect to poisson random measures , under mery s topology . we shall differ the presentation of these two fundamental results to appendix [ covclsssect ] in order to continue with our discussion of the dual problem . [ existencedual ] the class @xmath19 is convex , and if ( [ condonclassofsprmart1 ] ) is satisfied , the dual optimum @xmath200 of theorem [ mainthrm1 ] belongs to @xmath19 , for any @xmath201 . let us check that @xmath202 meets with the conditions in theorem [ thrmfollmerkramkov ] . indeed , each @xmath192 in @xmath202 is locally bounded from below since , defining @xmath203 @xmath204 where @xmath205 . condition ( i ) of theorem [ thrmfollmerkramkov ] is straightforward , while condition ( ii ) follows from theorem [ closenessresult ] . finally , condition ( iii ) holds because the processes in @xmath202 are already local martingales with respect to @xmath206 and hence @xmath207 with @xmath208 . by the corollary [ closednessclasses ] , we conclude that @xmath209 is convex and closed under fatou convergence on dense countable sets . on the other hand , @xmath104 is also convex and closed under fatou convergence , and thus so is the class @xmath210 . to check the second statement , recall that the existence of the dual minimizer @xmath200 in theorem [ mainthrm1 ] is guaranteed from proposition [ proptydual ] , where it is seen that @xmath200 is the fatou limit of a sequence in @xmath19 ( see the proof of proposition [ proptydual ] ) . this suffices to conclude that @xmath211 since @xmath19 is closed under under fatou convergence . in the rest of this section , we present some properties of the elements in @xmath19 and of the dual optimum @xmath212 . in particular , conditions on the `` parameters '' @xmath213 so that @xmath214 is in @xmath104 are established . first , we note that without lose of generality , @xmath190 can be assumed predictable . [ lmma1sectdualclass ] let @xmath215 then , there exist a predictable process @xmath216 and a process @xmath217 such that @xmath218 . let @xmath219 . since @xmath196 , there are stopping times @xmath220 such that @xmath221 cf . theorem ii.1.33 in @xcite . now , define @xmath222 and @xmath223 . then , @xmath224 + \bbe \left[\delta a_{\tau_{n}}\right]\leq n+1 + \bbe\left[|f(\tau_{n},z_{\tau_{n}})|\right]\\ & \leq n+2+\bbe\int_{0}^{\tau_{n}}\int_{\bbr}|f(s , z)|{\bf 1}_{|f|>1}\nu(dz)ds<\infty,\end{aligned}\ ] ] where we used that @xmath225 . therefore , @xmath190 is locally integrable , increasing , and thus , its predictable compensator @xmath226 exists . now , by the representation theorem for local martingales ( see theorem iii.4.34 @xcite ) , the local martingale @xmath227 admit the representation @xmath228 finally , @xmath229 the conclusion of the proposition follows since @xmath230 is necessarily in @xmath202 . the following result gives necessary conditions for a process @xmath214 to belong to @xmath104 . recall that a predictable increasing process @xmath190 can be uniquely decomposed as the sum of three predictable increasing processes , @xmath231 where @xmath232 is the _ absolutely continuous _ part , @xmath233 is the _ singular continuous _ part , and @xmath234 is the jump part ( cf . theorem 19.61 in @xcite ) . [ budgetconstraintresult ] let @xmath235 , where @xmath236 , @xmath237 and @xmath190 is an increasing predictable process . let @xmath238 be the `` sinking time '' of the supermartingale @xmath105 : @xmath239 also , let @xmath240 . then , @xmath241 is a supermartingale if and only if the following two conditions are satisfied : 1 . there exist stopping times @xmath242 such that @xmath243 2 . for @xmath206-a.e . @xmath15 , @xmath244 for almost every @xmath245 $ ] , where @xmath246 recall that @xmath105 and @xmath32 satisfies the sde s @xmath247 integration by parts and the predictability of @xmath190 yield that @xmath248 suppose that @xmath249 is a nonnegative supermartingale . then , the integral @xmath250 must have locally integrable variation in light of the doob - meyer decomposition for supermartingale ( see e.g. theorem iii.13 in @xcite ) . therefore , there exist stopping times @xmath251 such that @xmath252 then , ( i ) is satisfied with @xmath253 , where @xmath254 and @xmath255 . next , we can write ( [ intgrbypart ] ) as @xmath256 by the doob - meyer representation for supermartingales and the uniqueness of the canonical decomposition for special semimartingales , the last integral must be increasing . then , @xmath257 for @xmath258 since @xmath259 and @xmath260 for @xmath261 ( see i.4.61 in @xcite ) . we now turn to the sufficiency of conditions ( i)-(ii ) . since @xmath262 is locally bounded , @xmath263 is locally integrable . then , from ( [ intgrbypart ] ) , we can write @xmath264 condition ( ii ) implies that @xmath265 is a supermartingale , and by fatou , @xmath266 will be a supermartingale . this concludes the prove since @xmath260 for @xmath261 , and thus , @xmath267 , for all @xmath149 . the following result gives sufficient and necessary conditions for @xmath214 to belong to @xmath268 . its proof is similar to that of proposition [ budgetconstraintresult ] . [ conditionslocalsuper ] under the setting and notation of proposition [ budgetconstraintresult ] , @xmath214 belongs to @xmath268 if and only if condition ( i ) in proposition [ budgetconstraintresult ] holds and , for any locally bounded admissible trading strategies @xmath61 , @xmath269\right\}\right]=1.\ ] ] the previous result can actually be made more explicit under additional information on the structure of the jumps exhibited by the stock price process . we consider two cases : when the jumps come from the superposition of shot - noise poisson processes , and when the random field @xmath52 exhibit a multiplicative structure . let us first extend proposition [ cndadmissibility ] in these two cases . [ twocases ] ( i ) suppose @xmath1 is atomic with finitely many atoms @xmath270 . then , a predictable locally bounded strategy @xmath61 is admissible if and only if @xmath271-a.e . @xmath272 ( ii ) suppose that @xmath273 for a predictable locally bounded process @xmath3 such that @xmath271-a.e . @xmath274 and @xmath275 is locally bounded , and a deterministic function @xmath4 such that @xmath276 . then , a predictable locally bounded strategy @xmath61 is admissible if and only if @xmath271-a.e . @xmath277 where @xmath278 and @xmath279 . from proposition [ cndadmissibility ] , recall that @xmath206-a.s . @xmath280 for a.e . @xmath281 . then , for any closed set @xmath282 , @xmath283 , and @xmath284 , @xmath285 taking expectation , we get @xmath286 since such processes @xmath287}(\omega , u)$ ] generate the class of predictable processes , we conclude that @xmath271-a.e . @xmath288 let us prove ( ii ) ( the proof of ( i ) is similar ) . notice that @xmath289 where @xmath290 is the support of @xmath1 . suppose that @xmath291 . then , by considering closed sets @xmath292 such that @xmath293 as @xmath294 , we can prove the necessity . the other two cases ( namely , @xmath295 or @xmath296 ) are proved in a similar way . sufficiency follows since , @xmath206-a.s . , @xmath297 [ exmpl ] it is worth pointing out some consequences : 1 . in the time homogeneous case , where @xmath298 , the extreme points of the support of @xmath1 ( or what accounts to the same , the infimum and supremum of all possible jump sizes ) determine completely the admissible strategies . for instance , if the lvy process can exhibit arbitrarily large or arbitrarily close to @xmath299 jump sizes , then @xmath300 a constraint that can be interpreted as absence of shortselling and bank borrowing ( this fact was already pointed out by hurd @xcite ) . 2 . in the case that @xmath301 , the admissibility condition takes the form @xmath302 if in addition @xmath303 ( such that the stock prices exhibits only downward sudden movements ) , then @xmath304 and @xmath305 , with @xmath306 arbitrary , is admissible . in particular , from proposition [ conditionslocalsuper ] , if @xmath214 belongs to @xmath268 , then a.s . @xmath307 this means that @xmath308 if and only if condition ( i ) in proposition [ budgetconstraintresult ] holds and @xmath309a.s . @xmath310 , for a.e . @xmath258 . for a general @xmath3 and still assuming that @xmath301 , it follows that @xmath61 admissible and @xmath311 satisfy that @xmath206-a.s . @xmath312 for a.e . @xmath149 . we now extend proposition [ conditionslocalsuper ] in the two cases introduced in proposition [ twocases ] . its proof follows from propositions [ conditionslocalsuper ] and [ twocases ] . [ twocasesb ] suppose that either ( i ) or ( ii ) in proposition [ twocases ] are satisfied , in which case , define : @xmath313 then , a process @xmath214 belongs to @xmath268 if and only if condition ( i ) in proposition [ budgetconstraintresult ] holds , and for @xmath314-a.e . @xmath92 , @xmath315 , for a.e . @xmath149 . we remark that the cases @xmath316 and @xmath317 do not lead to any absurd in the definition of @xmath318 above as we are using the convention that @xmath319 . indeed , for instance , if @xmath316 , it was seeing that @xmath320 , for a.e . @xmath321 , and thus , we set the second term in the definition of @xmath318 to be zero . now we can give a more explicit characterization of the dual solution @xmath322 to the problem ( [ defdualproblem ] ) , which existence was established in proposition [ existencedual ] . for instance , we will see that @xmath323 is absolutely continuous up to a predictable stopping time . below , we refer to proposition [ budgetconstraintresult ] for the notation . [ abscontdual ] let @xmath324 , @xmath325 and @xmath326 . the followings two statements hold true : 1 . @xmath327 . furthermore , @xmath108 if and only if @xmath328 . 2 . suppose that either of the two conditions in proposition [ twocasesb ] are satisfied and denote @xmath329 where @xmath330 is defined accordingly to the assumed case . then , @xmath331 , and furthermore , the process @xmath332 belongs to @xmath268 if @xmath108 . let @xmath333 denote the increasing predictable processes in the decomposition ( [ decmpincr ] ) of @xmath190 . since @xmath190 is predictable , there is no common jump times between @xmath192 and @xmath190 . then , @xmath334 where we used that @xmath335 , and @xmath336 . since both processes @xmath105 and @xmath146 enjoy the same absolutely continuous part , and the same sinking time , the second statement in ( 1 ) is straightforward from proposition [ conditionslocalsuper ] . part ( 2 ) follows from proposition [ twocasesb ] since the process @xmath337 is nonnegative , predictable ( since @xmath135 is predictable ) , and locally integrable ( since @xmath338 ) . we remark that part ( 2 ) in proposition [ abscontdual ] remains true if we take @xmath339 . the following result is similar to proposition 3.4 in xu @xcite and implies , in particular , that the optimum dual @xmath127 can be taken to be a local martingale . [ abscontdual2 ] suppose that either condition ( i ) or ( ii ) of proposition [ twocases ] is satisfied . moreover , in the case of condition ( ii ) , assume additionally that @xmath340 for @xmath341 if @xmath342 , and for @xmath343 if @xmath344 . let @xmath345 . then , there exists @xmath346 such that @xmath347 and @xmath348 . furthermore , @xmath349 is a local martingale for all locally bounded admissible trading strategies @xmath61 . let us prove the case when condition ( i ) in proposition [ twocases ] is in force . in light of proposition [ abscontdual ] , we assume without loss of generality that @xmath350 with @xmath351 assume that @xmath352 . otherwise if , for instance , @xmath353 , then it can be shown that @xmath310 , a.s . ( similarly to case ( b ) in example [ exmpl ] ) , and the first term of @xmath318 is @xmath31 under our convention that @xmath354 . notice that , in any case , one can find a predictable process @xmath82 taking values on @xmath355 , such that @xmath356 write @xmath357 for a @xmath358 to be determined in the sequel . for @xmath359 it suffices to prove the existence of a field @xmath360 satisfying both conditions below : @xmath361 ( then , @xmath362 is defined as @xmath363 ) . similarly , for @xmath364 to belong to @xmath268 it suffices that @xmath365 taking @xmath366 clearly non - negative , ( b ) and ( c ) hold with equality . moreover , the fact that inequalities ( c ) hold with equality implies that @xmath367 is a local martingale for all locally bounded admissible trading strategy @xmath61 ( this can be proved using the same arguments as in the sufficiency part of proposition [ budgetconstraintresult ] ) . now suppose that condition ( ii ) in proposition [ twocases ] holds . for simplicity , let us assume that @xmath368 ( the other cases can be analyzed following arguments similar to example [ exmpl ] ) . notice that ( [ cndlcdo ] ) implies the existence of a borel @xmath369 ( resp . @xmath370 ) such that @xmath371 on @xmath369 ( resp . @xmath372 on @xmath370 ) and @xmath373 . taking @xmath374 ( b ) and ( c ) above will hold with equality . we now show that the tentative optimum final wealth @xmath375 , suggested by the inequality ( iii ) in proposition [ existencedual ] , is ( super- ) replicable . we will combine the dual optimality of @xmath127 with the _ super - hedging theorem _ , which states that given a contingent claim @xmath376 satisfying @xmath377 , one can find for any fixed @xmath378 an admissible trading strategy @xmath85 ( depending on @xmath82 ) such that @xmath379 almost surely ( see kramkov @xcite , and also delbaen and schachermayer @xcite ) . recall that @xmath60 denotes the class of all equivalent risk neutral probability measures . [ attainresult ] under the setting and conditions of proposition [ existencedual ] , for any @xmath201 , there is an admissible trading strategy @xmath85 for @xmath82 such that @xmath380 and thus , the optimum of @xmath381 is reached at the strategy @xmath85 . in particular , @xmath382 when @xmath383 . for simplicity , we write @xmath384 , @xmath385 , and @xmath386 fix an equivalent risk neutral probability measure @xmath387 , and let @xmath388 be its corresponding density processes . here , @xmath389 ( resp . @xmath390 ) is the restriction of the measure @xmath56 ( resp . @xmath206 ) to the filtration @xmath391 . under @xmath56 , @xmath392 is a local martingale , and then , for any locally bounded @xmath61 , @xmath393 is a @xmath56-local martingale . by iii.3.8.c in @xcite , @xmath394 is a @xmath206-local martingale ( necessarily nonnegative by admissibility ) , and thus , @xmath395 is in @xmath268 . on the other hand , @xmath395 belongs to @xmath209 due to the exponential representation for _ positive _ local martingales in kunita @xcite ( alternatively , by invoking theorems iii.8.3 , i.4.34c , and iii.4.34 in @xcite , @xmath396 even if @xmath0 were just an additive process @xmath0 ) . by the convexity of the dual class @xmath397 , @xmath398 belongs to @xmath19 , for any @xmath399 . moreover , since @xmath89 is convex and @xmath400 , @xmath401 the random variable @xmath402 is integrable since by assumption @xmath403 . we can then apply dominated convergence theorem to get @xmath404-\bbe\left[\widetilde{u}\left(y\xi^{*}_{_{t } } \right)\right]\right\ } = -y\bbe\left [ v^ { * } \left(\xi'_{_{t}}-\xi^{*}_{_{t}}\right ) \right],\ ] ] which is nonnegative by condition ( i ) in proposition [ existencedual ] . then , using condition ( ii ) in proposition [ existencedual ] , @xmath405 = \bbe\left [ v^{*}\xi'_{_{t}}\right ] \leq\bbe\left [ v^{*}\xi^{*}_{_{t } } \right]=z.\ ] ] since @xmath387 is arbitrary , @xmath406 \leq z. $ ] by _ the super - heading theorem _ , there is an admissible trading strategy @xmath85 for @xmath82 such that @xmath407 the second statement of the theorem is straightforward since @xmath408 is strictly increasing on @xmath409 . we conclude the paper with the following remarks . the dual domain of the dual problem can be taken to be the more familiar class of equivalent risk - neutral probability measures @xmath60 . to be more precise , define @xmath410 since @xmath411 is obviously a convex subclass of @xmath268 , theorem [ mainthrm1 ] implies that , as far as @xmath412 < \infty,\ ] ] for each @xmath413 , there exist @xmath414 and @xmath415 ( not necessarily belonging to @xmath411 ) such that ( i)-(iii ) in proposition [ existencedual ] hold with @xmath416 . finally , one can slightly modify the proof of proposition [ attainresult ] , to conclude the replicability of @xmath417 indeed , in the notation of the proof of the proposition [ attainresult ] , the only step which needs to be justified in more detail is that @xmath418\leq \bbe\left[\widetilde{u } \left(y\xi^{(\varepsilon)}_{_{t } } \right)\right],\ ] ] for all @xmath419 , where @xmath420 ( here , @xmath395 is a fixed element in @xmath411 ) . the last inequality follows from the fact that , by proposition [ attainresult ] ( c ) , @xmath127 can be approximated by elements @xmath421 in @xmath422 in the sense that @xmath423 a.s . thus , @xmath424 can be approximated by the elements @xmath425 in @xmath411 , for which we know that @xmath426\leq \bbe\left[\widetilde{u } \left(y\xi^{(\varepsilon , n)}_{_{t } } \right)\right].\ ] ] passing to the limit as @xmath294 , we obtain ( [ neededrel ] ) . in particular we conclude that condition ( [ sfcondexit ] ) is sufficient for both the existence of the solution to the primal problem and its characterization in terms of the dual solution @xmath427 of the dual problem induced by @xmath416 . we now further know that @xmath127 belongs to the class @xmath428 defined in ( [ maindualclass ] ) , and hence , enjoys an explicit parametrization of the form @xmath429 for some triple @xmath430 . * ( ii ) market driven by general additive models . * our analysis can be extended to more general multidimensional models driven by additive processes ( that is , processes with independent , possibly non - stationary increments ; cf . sato @xcite and kallenberg @xcite ) . for instance , let @xmath431 be a complete probability space on which is defined a @xmath432-dimensional additive process @xmath0 with lvy - it decomposition : @xmath433 where @xmath35 is a standard @xmath434dimensional brownian motion , @xmath435 is an independent poisson random measure on @xmath436 , and @xmath437 . consider a market model consisting of @xmath438 securities : one risk free bond with price @xmath439 and @xmath440 risky assets with prices determined by the following stochastic differential equations with jumps : @xmath441 where the processes @xmath442 , @xmath50 , @xmath51 , and @xmath52 are predictable satisfying usual integrability conditions ( cf . kunita @xcite ) . we assume that @xmath443 , where @xmath41 is the natural filtration generated by @xmath35 and @xmath36 ; namely , @xmath444\times a ) : s\leq{}t , a\in\calb(\bbr^{d}))$ ] . the crucial property , particular to this market model , that makes our analysis valid is the representation theorem for local martingales relative to @xmath0 ( see theorem iii.4.34 in @xcite ) . the definition of the dual class @xmath19 given in section [ specdualclsssect ] will remain unchanged , and only very minor details will change in the proof of theorem [ closenessresult ] . some of the properties of the results in section [ specdualclsssect ] regarding the properties of @xmath19 will also change slightly . we remark that , by taking a real ( nonhomogeneous ) poisson process , the model and results of chapter 3 in xu @xcite will be greatly extended . we do not pursue the details here due to the limitation of the length of this paper . another classical portfolio optimization in the literature is that of optimal wealth - consumption strategies under a budget constraint . namely , we allow the agent to spend money outside the market , while maintaining `` solvency '' throughout @xmath445 $ ] . in that case the agent aims to maximize the cost functional that contains a _ running cost _ " : @xmath446,\ ] ] where @xmath447 is the instantaneous rate of consumption . to be more precise , the cumulative consumption at time @xmath448 is given by @xmath449 and the ( discounted ) wealth at time @xmath448 is given by @xmath450 here , @xmath451 is a ( state - dependent ) utility function and @xmath452 is a utility function for each @xmath448 . the dual problem can now be defined as follows : @xmath453,\ ] ] over a suitable class of supermartingales @xmath19 . for instance , if the support of @xmath1 is @xmath454 , then @xmath19 can be all supermartingales @xmath105 such that @xmath455 and @xmath456 is a supermartingale . the dual theorem [ mainthrm1 ] can be extended for this problem . however , the existence of a wealth - consumption strategy pair @xmath457 that attains the potential final wealth induced by the optimal dual solution ( as in section [ rplcbltysect ] ) requires further work . we hope to address this problem in a future publication . the goal of this part is to establish the theoretical foundations behind theorem [ existencedual ] . we begin by recalling an important optional decomposition theorem due to fllmer and kramkov @xcite . given a family of supermartingales @xmath202 satisfying suitable conditions , the result characterizes the nonnegative exponential local supermartingales @xmath458 where @xmath459 and @xmath460 , in terms of the so - called _ upper variation process _ for @xmath202 . concretely , let @xmath461 be the class of probability measures @xmath462 for which there is an increasing predictable process @xmath463 ( depending on @xmath56 and @xmath202 ) such that @xmath464 is a local supermartingale under @xmath56 , for all @xmath459 . _ smallest _ _ satisfies such a property then @xmath465 is increasing . ] of such processes @xmath190 is denoted by @xmath466 and is called the upper variation process for @xmath202 corresponding to @xmath56 . for easy reference , we state fllmer and kramkov s result ( see @xcite for a proof ) . the next result is a direct consequence of the previous representation . recall that a sequence of processes @xmath472 is said to be fatou convergent on @xmath147 " to a process @xmath105 if @xmath472 is uniformly bounded from below and it holds that @xmath473 almost surely for all @xmath149 . [ closednessclasses ] if @xmath202 is a class of semimartingales satisfying the conditions in theorem [ thrmfollmerkramkov ] , then @xmath474 is convex and closed under fatou convergence on any fix dense countable set @xmath147 of @xmath475 ; that is , if @xmath476 is a sequence in @xmath477 that is fatou convergent on @xmath147 to a process @xmath105 , then @xmath478 . the convexity of @xmath477 is a direct consequence of theorem [ thrmfollmerkramkov ] , since the convex combination of supermartingales remains a supermartingale . let us prove the closure property . fix a @xmath471 and denote @xmath479 . notice that @xmath480 because @xmath481 is increasing and hence , its jumps are nonnegative . since @xmath482 , @xmath483 is a supermartingale under @xmath56 . then , for @xmath484 , @xmath485 \leq{}c^{-1}_{s'}\xi^{n}_{s'}.\ ] ] by fatou s lemma and the right - continuity of process @xmath486 , @xmath487= \bbe^{\bbq}\left [ \liminf_{t'\downarrow{}t : t'\in\pi } \liminf_{n\rightarrow\infty } c^{-1}_{t'}\xi^{n}_{t'}|\calf_{s'}\right ] \leq{}c^{-1}_{s'}\xi^{n}_{s'}.\ ] ] finally , using the right - continuity of the filtration , @xmath488\leq \liminf_{s'\downarrow{}s : s'\in\pi}\liminf_{n\rightarrow\infty } c^{-1}_{s'}\xi^{n}_{s'}= c_{s}^{-1}\xi_{s},\ ] ] where @xmath489 . since @xmath56 is arbitrary , the characterization of theorem [ thrmfollmerkramkov ] implies that @xmath478 . the most technical condition in theorem [ thrmfollmerkramkov ] is the closure property under mery distance . the following result is useful to deal with this condition . it shows that the class of integrals with respect to a poisson random measure is closed with respect to mery distance , thus extending the analog property for integrals with respect to a fixed semimartingale due to mmin @xcite . [ closenessresult ] let @xmath490 be a closed convex subset of @xmath491 containing the origin . let @xmath492 be the set of all predictable processes @xmath493 , @xmath196 and @xmath494 , such that @xmath197 , for all @xmath198 , and @xmath495 , for @xmath496-a.e . then , the class @xmath498 is closed under convergence with respect to mery s topology . consider a sequence of semimartingales @xmath499 in the class @xmath202 . let @xmath192 be a semimartingale such that @xmath500 under mery topology . to prove the result , we will borrow some results in @xcite . for some @xmath462 , we denote @xmath501 to be the banach space of all @xmath56-square integrable martingales on @xmath445 $ ] , endowed with the norm @xmath502_{t}\right)^{1/2}$ ] , and @xmath503 to be the banach space of all predictable processes on @xmath445 $ ] that have @xmath56-integrable total variations , endowed with the norm @xmath504 . below , @xmath505 stands for the localized class of increasing process in @xmath503 . by theorem ii.3 in @xcite , one can extract a subsequence from @xmath506 , still denote it by @xmath506 , for which one can construct a probability measure @xmath56 , defined on @xmath74 and equivalent to @xmath507 ( the restriction of @xmath206 on @xmath74 ) , such that the following assertions hold : let us extend @xmath513 and @xmath514 to @xmath72 by setting @xmath515 and @xmath516 for all @xmath517 . also , we extend @xmath56 for @xmath518 by setting @xmath519 , so that @xmath462 ( on @xmath42 ) . in that case , it can be proved that @xmath520 this follows essentially from proposition iii.3.5 in @xcite and doob s theorem . now , let @xmath521 , $ ] denote the density process . since @xmath105 is bounded , both @xmath522 and @xmath523 are bounded . by lemma iii.3.14 and theorem iii.3.11 in @xcite , the @xmath309quadratic covariation @xmath524 $ ] has @xmath309locally integrable variation and the unique canonical decomposition @xmath525 of @xmath526 relative to @xmath56 is given by @xmath527 also , the @xmath206-quadratic variation of the continuous part @xmath528 of @xmath526 ( relative to @xmath206 ) , given by @xmath529 is also a version of the @xmath56-quadratic variation of the continuous part of @xmath526 ( relative to @xmath56 ) . by the representation theorem for local martingales relative to @xmath0 ( see e.g. theorem iii.4.34 in @xcite or theorem 2.1 in @xcite ) , @xmath105 has the representation @xmath530 for predictable @xmath360 and @xmath531 necessarily satisfying that @xmath532 , @xmath533 then , @xmath534 we conclude that @xmath535 . hence , @xmath536 , where @xmath537 is the purely discontinuous local martingale ( relative to @xmath56 ) defined by @xmath538 where @xmath539 is the compensator of @xmath36 relative to @xmath56 ( see theorem iii.3.17 in @xcite ) . it can be shown that @xmath540 . notice that @xmath537 is well - defined since @xmath541 and the definition iii.1.27 in @xcite . then , the purely discontinuous part of the local martingale @xmath513 ( relative to @xmath56 ) is given by @xmath537 ( see i.4.19 in @xcite ) , and since @xmath542 , @xmath543_{t } & \bbe^{\bbq}\int_{0}^{t } \left(f^{n}(s , z)\right)^{2 } y(s , z ) \nu(dz)ds + \bbe^{\bbq}\int_{0}^{t } \left(g^{n}(s)\right)^{2}ds<\infty.\end{aligned}\ ] ] similarly , since @xmath510 is a cauchy sequences under the norm @xmath544_{t}$ ] , @xmath545_{t } & = \bbe^{\bbq}\int_{0}^{t } \left(f^{n}(s , z)-f^{m}(s , z)\right)^{2 } y(s , z ) \nu(dz)ds\\ \quad&+\bbe^{\bbq}\int_{0}^{t } \left(g^{n}(s)-g^{m}(s)\right)^{2}ds\rightarrow{}0,\end{aligned}\ ] ] as @xmath546 . using the notation @xmath547 and @xmath548 , where @xmath549 is the predictable @xmath550 field , we conclude that @xmath551 is a cauchy sequence in the banach space @xmath552 and thus , there is @xmath553 such that @xmath554 , as @xmath294 . similarly , there exists a @xmath555 in the banach space @xmath556 such that @xmath557 , as @xmath294 . in particular , @xmath493 satisfies condition ( iv ) since @xmath540 is strictly positive , and each @xmath558 satisfies ( iv ) . also , @xmath196 relative to @xmath56 in light of @xmath541 . similarly , @xmath559 belongs to @xmath560 , and hence , belongs to @xmath561 . it follows that the process @xmath562 is a well - defined local martingale relative to @xmath206 . applying girsanov s theorem to @xmath563 relative to @xmath56 and following the same argument as above , the purely discontinuous local martingale and bounded variation parts of @xmath563 are respectively @xmath564 the continuous part of @xmath563 has quadratic variation @xmath565 . we conclude that @xmath566 and @xmath567 on @xmath568 . then , @xmath526 converges under mery s topology to @xmath563 and hence , @xmath569 .

we study the use of entanglement purification for quantum communication over long distances . for distances much longer than the coherence length of a corresponding noisy quantum channel , the fidelity of transmission is usually so low that standard purification methods are not applicable . it is however possible to divide the channel into shorter segments that are purified separately and then connected by the method of entanglement swapping . this method can be much more efficient than schemes based on quantum error correction , as it makes explicit use of two - way classical communication . an important question is how the noise , introduced by imperfect local operations ( that constitute the protocols of purification and the entanglement swapping ) , accumulates in such a compound channel , and how it can be kept below a certain noise level . to treat this problem , we first study the applicability and the efficiency of entanglement purification protocols in the situation of imperfect local operations . we then present a scheme that allows entanglement purification over arbitrary long channels and tolerates errors on the per - cent level . it requires a polynomial overhead in time , and an overhead in local resources that grows only logarithmically with the length of the channel . # 1

quant-ph9808065

in quantum communication , qubits @xcite are sent across quantum channels that connect distant nodes of a quantum network . in general , these quantum channels are noisy and therefore limit the fidelity of the transmission . a bottleneck for quantum communication is the scaling of the probability of transmission errors with the length of the channels connecting the nodes . different from classical ( digital ) communication , the signals " in quantum communication consist of single qubits which may be entangled . owing to fundamental principles , these qubits can not be cloned @xcite nor amplified @xcite without destroying the essential quantum feature . this fact limits the maximum distance between the nodes to a few multiples of the absorption length ( or the coherence length ) of the channel , and poses a severe restriction on any practical application . for example , in recent experiments on quantum cryptography @xcite , single photons are sent through optical fibers with a given absorption and depolarization length . this has two effects : @xmath0 to transmit a photon without absorption , the number of trials scales exponentially with @xmath1 ; @xmath2 even when a photon arrives , the fidelity of the transmitted state decreases exponentially with @xmath1 . in the experiment , the distance between the nodes is therefore presently limited by a few multiples of the absorption length of the fiber @xcite . in the context of fault - tolerant quantum computing @xcite , knill and laflamme @xcite have discussed an important scheme that allows , in principle , to transmit a qubit over arbitrarily long distances with a polynomial overhead in the resources . their method requires to encode a single qubit into a concatenated quantum code ( i.e. an entangled state of a large number of qubits ) and to operate on this code repeatedly while it is propagated through the channel . the tolerable error probabilities for transmission are less than @xmath3 , whereas for local operations they are less than @xmath4 . this seems to be outside the range of any practical implementation in the near future . a crucial figure for any experiment will be the number of particles that can be manipulated locally in a coherent fashion , together with the precision with which such local manipulations can be realized . for instance , even when the number of particles scales only polynomially with the length of the channel , a polynomially large number may still be far too large for any practical implementation . a central tool in the theory of quantum information , specifically for quantum communication is entanglement purification @xcite . it allows , in principle , to create maximally entangled states of particles at different locations even if the channel that connects those locations is noisy @xcite . these entangled particles can then be used for faithful teleportation @xcite or secure quantum cryptography @xcite . it seems therefore natural to use this method as an ingredient for a quantum repeater . furthermore , it allows highly efficient two - way protocols which can not be realized with quantum error correction procedures . if a noisy quantum channel is much longer than its coherence length , one can not directly employ standard purification schemes . these schemes first create an ensemble of low fidelity epr pairs across the channel , and then purify / distill a few perfect epr pairs out of the ensemble . this , however , requires a minimum fidelity @xmath5 of the initial pairs to operate with , which can not be achieved as the length @xmath1 of the channel increases . in this paper , we treat the problem of a quantum repeater based on the method of entanglement purification . the idea of such a repeater is to divide a long quantum channel into shorter segments , which are purified separately , before they are connected . connecting two segments of a channel means here to build up quantum correlations across the compound channel from correlations that exist across the individual segments . this can be done by entanglement swapping , or teleportation of entanglement . a quantum repeater must therefore combine the methods of entanglement purfication and teleportation . although the combination of these methods should , in principle , allow to create entanglement over arbitrary distances , it is another question how much this `` costs '' in terms of resources needed for purification . resources means here the number of low fidelity epr pairs that have to be provided for purification of each channel segment . this quantity is related to the number of particles that have to be manipulated locally ( at the connection points between the segments ) in a coherent fashion . if the resources grow too fast with the length of the channel , not much will be gained by the whole procedure . a further important quantity is the error tolerance for the local operations . in every real situation , the local operations applied to one or more particles , will bear some imperfections . since such operations are the building blocks for any entanglement purification protocol , their imperfections will limit the maximum attainable fidelity for an epr pair and the efficiency of the protocol . in the context of the quantum repeater , a maximum fidelity @xmath6 corresponds to a residual amount of noise for each segment . when the segments are connected , this noise accumulates . to give a realistic treatment of the quantum repeater , we will therefore first study the applicability and the efficiency of entanglement purification protocols in the situation of imperfect local operations . the general conditions under which standard purification protocols can be used in the presence of errors have been studied by giedke _ this includes , in particular , thresholds and lower bounds for the attainable fidelities . for a generic class of stochastic errors , one can give explicit representations of the imperfect operations in terms of completely positive maps and/or povms @xcite . in terms of these maps , we derive recursion formulas for the fidelities , which generalize the results given by bennett _ _ @xcite and by deutsch _ _ @xcite . from these results , we estimate that , for a generic class of errors , standard entanglement purifcation protocols work even for error probabilities of the order of a few per cent . in the context of long distance communication , we develop a purification procedure that combines the standard protocols with quantum teleportation in a certain way to be specified . this purification procedure allows quantum communication via noisy channels of arbitrary length . since it explicitly exploits two - way classical communication @xcite , our procedure is much more efficient for quantum communication than protocols based on quantum error correction @xcite . specifically , this solution of the quantum repeater tolerates errors on the per - cent level ; it requires a polynomial overhead in time and an overhead in local resources that grows only logarithmically with the length of the channel . the paper is organized as follows : in sec . [ error ] we introduce a generic error model to describe imperfect local operations such as noisy 1 bit and 2 bit operations and imperfect measurements . with the aid of these models , we revisit in sec . [ secelements ] the basic elements of quantum communication purfication and teleportation . this discussion includes results for the maximum attainable fidelities and the working conditions of the standard protocols under the condition of noisy local operations . after this discussion , in sec . [ secqrep ] , we are ready to attack the quantum repeater problem itself . our solution combines the methods of entanglement purification and entanglement swapping into a single ( meta- ) protocol , which we call nested entanglement purification . this protocol allows to distribute entanglement with a given fidelity across arbitrary long distances , even when local operations are imperfect . it requires an overhead in local resources that grows polynomially with the length of the channel . a modification of this protocol works with resources that grow only logarithmically with the length of the channel ( while the build - up time grows polynomially ) . imperfections of local operations can have various origins . an example for an imperfect one - qubit operation is a unitary ` overrotation ' @xcite , as it would be realized by a laser pulse that effects a bloch rotation around an angle that slightly deviates from the ideal ( intended ) one . if the laser is well stabilized but the pulse area tuned incorrectly , say , the deviation of the real from the intended ( unitary ) rotation will be the same when we repeat the operation , representing an example of a systematic error . if , on the other hand , the angle of the rotations fluctuates randomly around a certain mean value , every pulse introduces a certain amount of noise into the system , and the resulting operation would be described by a non - unitary map . in the present paper , we concentrate on this latter situation , that is we assume only stochastic errors and neglect systematic errors . to give a precise description of a noisy 1-bit or 2-bit operation , one needs to know the exact form of the error mechanisms , which in turn depend on the specific physical implementation . in general , we have only limited knowledge of these details . a generic model for a noisy channel is the so - called ( symmetric ) depolarizing channel . it transforms a qubit with initial state @xmath7 in the way @xmath8 where @xmath9 is the identity operator and @xmath10 depends on the time for which the qubit is subjected to the channel . the action of the channel thus results in admixing a completely depolarized state @xmath11 to the initial density operator . the limit @xmath12 corresponds to a very noisy channel , while @xmath13 describes a channel with very little noise ( ideal storage " ) . one can easily generalize ( [ depol ] ) to the case where the ideal operation is some arbitrary unitary ( 1-bit ) operation @xmath14 , @xmath15 in this situation , the ideal operation is accompanied by the action of the depolarizing channel . the whole map becomes non - unitary . it describes an imperfect 1-bit operation , which for @xmath13 becomes identical to the ideal unitary map . we call @xmath16 the _ reliability _ of the imperfect operation , which represents a lower bound to the probability that its result corresponds to the ideal one . for a parametrization @xmath17 an imperfect ( noisy ) rotation may explicitly be defined by the map @xmath18 where the noise function @xmath19 is normalized to unity and describes the fluctuations @xmath20 of the ( generalized ) bloch vector @xmath21 . one can easily show that for isotropic noise , " @xmath22 , the integral reduces to @xmath23 wherein @xmath24 describes the ideal rotation . motivated by these observations , we generalize this model to arbitrary one - qubit and two - qubit gates @xcite acting on an entangled state @xmath25 of several qubits . we thereby require that errors introduced by local operations may only affect the local qubits on which the ideal operation acts ( locality condition ) . let @xmath25 be a state of @xmath26 qubits , labeled by @xmath27 . an imperfect one - qubit operation acting on the first qubit , say , is then described by the map @xmath28 whereas an imperfect two - qubit operation acting on qubits 1 and 2 is described by @xmath29 in these expressions , @xmath30 is the ideal ( perfect ) operation and @xmath31 and @xmath32 denote unit operators on the subspace where the ideal operation acts , corresponding to totally depolarized one and two - qubit states . the state of the other particles are described by the partial traces @xmath33 and @xmath34 , respectively , of the initial density operator over theses subspaces . the quantities @xmath35 and @xmath36 measure the _ reliability _ of the operations , where perfect operations correspond to @xmath37 . technically speaking , an imperfect operation is modeled by the corresponding ideal operation accompanied / followed by a depolarizing channel that acts ( only ) on the same subspace as the ideal operation . note that the maps ( [ one_qubit_op ] ) and ( [ two_qubit_op ] ) are linear and trace conserving , albeit non - unitary . for any state @xmath7 which is diagonal in the bell basis defined by any two of the particles all states we deal with in this paper are of this form , the model is self consistent in the sense that @xmath38 ( two single - qubit operations , each described by an error parameter @xmath35 ) can be written as a certain two - qubit operation @xmath39 described by an error parameter @xmath40 . generally , any sequence of local operations can be written as a single joint operation within this model . the resulting error parameter ( i.e. reliability ) describing the joint operation is obtained by multiplying the error parameters of the single operations . we will use these maps to estimate the role of imperfect operations in standard protocols for entanglement purification and for teleportation . although the model is simple , it is generic at the same time . we just mention here that similar maps may be derived from a jaynes - type principle , insofar as their resulting states maximize the entropy of the resulting state with respect to variations that are trace conserving and keep the average fidelity of the operation constant . we finally describe an _ imperfect measurement _ on a single qubit ( in the computational basis ) by a povm @xcite corresponding to the ` imperfect projectors ' @xmath41 the parameter @xmath42 is a measure for the quality of the projection onto the basis states . assume the qubit to be measured is in the state @xmath43 and we are trying to measure its state with the aid of an measurement apparatus described by ( [ povm ] ) . the expectation values @xmath44 and @xmath45 simply mean that the apparatus will give us the wrong result ( `` 1 '' ) with probability @xmath46 . that is , the result is not completely reliable . an ideal measurement , in contrast , is described by @xmath47 . it is clear that the effect of the measurement on the measured qubit is not fully specified by the povm ( [ povm ] ) . in the present context , however , this description is sufficient as all measured particles are removed from the system i.e. we trace over their degrees of freedom after the measurement . the operations ( [ one_qubit_op ] ) and ( [ two_qubit_op ] ) together with single - qubit measurements ( [ povm ] ) are sufficient to describe all operations occurring in the context of teleportation and entanglement purification . for example , a bell measurement ( the measurement of a projector in the bell basis ) on two particles , say 1 and 2 , can be realized by a two - qubit operation @xmath48 ( controlled not ) followed by a hadamard rotation of particle 1 and two single - qubit measurements @xmath49 , @xmath50 on particles 1 and 2 @xcite . instead of performing the hadamard transformation one can also measure particle 1 in a rotated basis . in summary , an imperfect bell measurement is described by an imperfect two - qubit operation followed by 2 imperfect single - qubit measurements , effecting e.g. @xmath51 . in this section we reconsider the basic elements of quantum communication , which are teleportation and purification , in the presence of local errors . purification is the distillation of few perfect epr pairs out of many imperfect pairs . in the following , we will generalize two different purification protocols that have been treated in the literature , by introducing imperfect gate and measurement operations . in subsection [ purificationc ] , we then discuss a modified version of these schemes which , generally speaking , has less favorable convergence properties but works with a smaller and constant number of _ physical resources_. this purification scheme was introduced by bennett _ _ @xcite . in short , the scheme takes two pairs ( 1 - 2 and 3 - 4 ) as in fig . [ purify1 ] , both in a werner state = f|^+^+|+()(|^-^-|+|^+^+|+|^-^-| ) . [ werner ] with fidelity @xmath52 . then it performs local ( 1 @xmath53 2-bit ) operations on the particles at the same ends of the pairs . writing @xmath54 , this essentially involves two cnot operations , cnot@xmath55 and cnot@xmath56 acting on the state @xmath57 , followed by a simple measurement @xmath58 , @xmath59 of the particles @xmath60 and @xmath61 . if the particles are found in the same state ( 00 or 11 ) , the remaining pair , described by the state @xmath62 is kept , otherwise it is discarded . to obtain a recursive mapping between werner states , a non - unitary depolarization operation ( twirl ) is applied to the resulting state before it is used for the next purification step . the original treatment of this protocol @xcite assumes that all operations and measurements of the protocol are perfect . to estimate the effect of local errors , we evaluate the protocol with the aid of the imperfect operations ( [ one_qubit_op][povm ] ) . the calculation of the fidelity @xmath63 of the new pair @xmath62 is somewhat lengthy but straightforward , and the result can be given in analytic form . the result of this calculation is summarized in the formula @xmath64[\eta^2+(1-\eta)^2 ] + [ f(\frac{1-f}{3})+(\frac{1-f}{3})^2)][2\eta(1-\eta ) ] + ( \frac{1-p_2 ^ 2}{8p_2 ^ 2})}{[f^2+\frac{2}{3}f(1-f ) + \frac{5}{9}(1-f)^2][\eta^2+(1-\eta)^2 ] + [ f(\frac{1-f}{3})+(\frac{1-f}{3})^2)][8\eta(1-\eta ) ] + 4(\frac{1-p_2 ^ 2}{8p_2 ^ 2 } ) } \label{modified_bennett}\ ] ] which reduces to the formula given in ref . @xcite in the limiting case @xmath47 and @xmath65 ( perfect operations ) . although @xmath62 is no longer a werner state , it can again be brought to werner form using depolarization , which is considered to be noiseless here @xcite . if one starts with an ensemble of pairs with fidelity @xmath66 , @xmath63 gives the fidelity of the remaining pairs that are left after one purification step . this defines a fraction of @xmath67 of the initial ensemble of pairs , where @xmath68 denotes the probability for finding the particles @xmath60 and @xmath61 in the state 00 or 11 , and is given by the denominator in ( [ modified_bennett ] ) . the factor 2 arises since the pair originally composed of particles @xmath60 and @xmath61 is lost due to the measurement . each ( iterated ) application of ( [ modified_bennett ] ) corresponds to a purification step . for each purification step , 2 identically pairs ( which both result from previous successful purification steps ) are used . purification resources _ are defined by the average number @xmath69 of pairs needed to perform @xmath70 successful purification steps , and are given by @xmath71 where the probabilities @xmath72 depend on the purification step @xmath73 . the _ physical resources _ , defined by the total number of particles at a or b that are used to store the pairs needed for the purification process , are for this scheme equal to the purification resources , since all pairs are created at the beginning and have to be stored as can be seen in fig.[tree ] . when calculating the fixpoints of ( [ modified_bennett ] ) , one finds that one fixpoint is always @xmath74 , independent of the error parameters @xmath36 and @xmath75 . the other two fixpoints @xmath5 and @xmath76 ( see fig . [ figpuriloop ] ) are given by the expression @xmath77 they depend on the error parameters and give the borders of the interval within which purification is possible . if @xmath78 , then @xmath79 . since @xmath76 is an attractor , iterative application of ( [ modified_bennett ] ) leads to a resulting pair with fidelity @xmath80 . the value @xmath5 thus gives the threshold for f where this purification scheme can be successfully applied , while @xmath76 gives the maximal reachable fidelity . for perfect local operations , @xmath81 and @xmath82 , meaning that all pairs with @xmath83 can be purified to @xmath84 . for imperfect local operations , @xmath85 and @xmath86 , i.e. no perfect epr pairs can be created . if the error parameters become larger , these two fixpoints approach each other and the region where purification is possible shrinks to zero . the limiting situation @xmath87 defines the threshold for the applicability of the purification protocol . for all pairs ( @xmath88 ) for which there is only one real fixpoint ( at @xmath89 ) , the imperfections of the local operations introduce more noise than one gains from the purification , so the scheme breaks down . for example for @xmath47 , ( perfect measurements ) , eq . ( [ fixpoints ] ) simplifies to @xmath90 with the threshold at @xmath91 . that is , the cnot gate must work with a reliability of at least @xmath92 . the threshold gets tighter when the measurements are imperfect as well , i.e. for @xmath93 . in general , the threshold values for the error parameters @xmath36 and @xmath75 are of the order of some percent as can be seen in fig.[fixb3 ] . the fixpoints as a function of the error parameters are plotted in fig.[benoxf3 ] , where we set @xmath94 to get a 2 dimensional plot . a purification protocol that converges faster and involves less resources was proposed by deutsch _ et al . _ @xcite . generally speaking , this scheme can be described as a mapping on states that are diagonal in the bell basis , but need not necessarily be werner states , @xmath95 where @xmath96 and @xmath97 are the bell states between particles 1 and 2 in the usual notation . in the notation of ( [ bell_diag ] ) , the protocol corresponds to a mapping @xmath98 between the diagonal elements ( a , b , c , d ) . in particular , between two successive purification steps , the states are not depolarized . this feature , together with additional @xmath99 rotations that are applied to the qubits before the cnot operations , makes scheme b a much faster converging protocol . in the following analysis , we evaluate scheme b using the imperfect operations ( [ one_qubit_op])([povm ] ) instead of perfect ones . a @xmath99 rotation followed by a cnot operation is thereby treated as a joint 2-qubit operation with a single error parameter @xmath36 . the resulting map between the diagonal elements @xmath98 is described in @xcite ; it completely characterizes the action of this purification protocol . successive purification steps are now described by iterated applications of this map . for each purification step , two identical pairs ( which result from the previous step ) are used . the resources can be calculated in a similar way as for scheme a. the fixpoints of this map are no longer described by a single parameter f but by a set of 4 numbers @xmath100 . the fidelity of such a diagonal bell state is given by @xmath101 , the @xmath102 component and , for simplicity , we shall continue to call this component fixpoint `` of scheme b , although the fixpoint is not sufficiently described by this single parameter . for example , it may happen that a werner state with a certain fidelity @xmath103 can not be purified , while a binary state with the same fidelity @xmath103 that has only two nonzero components @xmath102 and @xmath104 , say , ( in short : ' ' binary state ( a , c ) " ) can be purified . we have numerically compared schemes a and b , and found that scheme b converges much faster towards the upper fixpoint @xmath76 if @xmath105 as can be seen in fig.[benoxf1 ] . surprisingly , the upper fixpoint for scheme b is above the one for scheme a for given error parameters and does not depend on the shape " of the initial state . the lower fixpoint @xmath5 is smaller for scheme b and therefore the interval within which purification is possible is _ larger _ for scheme b , and so is the maximal reachable fidelity . the fixpoints as a function of the error parameters @xmath94 are plotted in fig.[benoxf3 ] . one can see that scheme b is significantly less sensitive to noisy local operations than scheme a. both schemes ( a and b ) need many elementary pairs to start with , which have to be stored somewhere before the protocols are applied . the physical resources are thus quite large and grow with the number of necessary purification steps . in the following , we modify scheme b in a certain way , and call this scheme c. different from scheme b , we do not use two identical pairs at each purification step ( which are left over from previous purification steps ) but always purify one and the same pair with the help of an auxiliary pair @xmath106 with constant fidelity @xmath107 . apart from that , we employ the same protocol ( that is the sequence of local operations and measurements ) as in scheme b. the resulting map is thus the same , the only difference being that @xmath108 is a diagonal bell state which is constant throughout the whole purification procedure and is given by @xmath109 . this auxiliary pair is repeatedly created before each purification step between two of the four particles as visualized in fig.[ibkresfig ] . if a purification process is not successful , one has to start from the beginning ( when both pairs 12 and 34 are put in the state @xmath106 ) . using this procedure , the fidelity converges towards a given fixpoint @xmath110 which is _ not _ equal to one @xcite . the fixpoint highly depends on both the initial fidelity _ and _ the shape ( other diagonal elements in the bell basis ) of the state @xmath106 . the properties of the fixpoint will be discussed in more detail later on . for ideal operations , this method is generally less favorable as far as its convergence properties are concerned . however , for imperfect operations , the situation changes and the drawback that the reachable fixpoint @xmath111 is smaller than unity becomes less important . as we shall see later , for imperfect local operations the fixpoint of scheme c may lie even _ above _ the fixpoints of schemes a and b ( see fig . [ ibkoxf1 ] ) . the main advantage of this scheme is , however , that the _ physical resources _ , that is the number of particles needed at a or b to store the pairs , are _ constant _ ( namely 2 ) and independent of the number of necessary purification steps . this fact makes scheme c particulary interesting for the repeater problem , where the accumulation of local particles at the connection points of a compound long channel plays an important role . the _ purification resources _ , on the other hand , that is the ( average ) number of how often an auxiliary pair has to be created in order to perform @xmath70 purification steps , have now to be calculated in a different way . let @xmath112 be the probability to succeed at the @xmath113 purification step . the total resources can be calculated using the iteration formula m_k+1=(m_k+1 ) . to close this formula , note that @xmath114=1 . this can be understood as follows : starting at the elementary level , one needs two pairs for purification and has to repeat the procedure on average @xmath115 times . thus @xmath116 . for the next purification step , one needs one additional elementary pair to perform the next purification step . this step has to be performed on average @xmath117 times . the resources to perform the second purification step are thus given by @xmath118 and so on . to perform @xmath70 purification steps , one needs to create on average m_k_max=_i=1^k_max(_k = i^k_max ( ) ) [ iteration ] times an auxiliary pair . a related quantity is the average number of purification steps ( including also those which fail ) . this is given by the iteration formula s_k+1=(s_k+1 ) where @xmath119 and k denotes the number of _ successful _ purification steps . if no purification is performed , the number of steps @xmath119 . this is the only difference to the iteration formula ( [ iteration ] ) for the resources , since every time an elementary pair is created , one performs a purification step . thus the average total number of purification steps performed ( assuming that @xmath70 steps were successful ) is given by s_k_max=_i=1^k_max(_k = i^k_max ( ) ) .[iteration1 ] the fixpoint discussion for this purification protocol is more involved due to the many parameters , namely the fidelity @xmath107 and the shape ( other diagonal elements @xmath120 ) of the elementary pair , and the error parameters @xmath36 and @xmath75 describing the noisy local operations . for this protocol , purification is possible if the reachable fidelity ( fixpoint ) for a certain auxiliary pair lies above its initial fidelity . to illustrate first the shape dependence of the fixpoint , we introduce the following parameterization of diagonal bell states with constant fidelity @xmath103 . in the notation of ( [ bell_diag ] ) we write a&=&f_0 + b = c&=&(1-f_0)(1-)/2 + d&=&(1-f_0 ) where @xmath121 . a werner state thereby described by @xmath122 and a binary state ( a , d ) is obtained for @xmath123 . figure [ ibk1 ] shows the dependence of the fixpoint as a function of @xmath124 for different error parameters . for the manifold of states covered by this parametrization one sees that the optimum shape of the elementary pair is a binary state ( a , d ) , meaning that the reachable fixpoint for a given initial fidelity is the largest for this shape . figure [ ibkoxf1 ] demonstrates two different things . first , it shows that the fixpoint ( maximally reachable fidelity ) depends on the initial fidelity of the elementary pair even if the other components are kept constant . second , it demonstrates that the reachable fidelity using this purification scheme may lie above the reachable fidelity using schemes a or b. ( the upper fixpoint for scheme a is not plotted here , but as we already know ( see fig.[benoxf3 ] ) it always lies below the fixpoint for scheme b ) . one should however mention that this is only the case if the shape of the elementary pair is close to the optimal configuration ( a , d ) . this can be understood as follows : scheme b always picks 2 pairs that result from a previous purification step and the shape of such a pair converges towards a working state " which no longer depends on the initial shape but only on the error parameters . scheme c , on the other hand , always uses the same auxiliary pair for each purification step which has _ not _ been influenced by noisy local operations in previous steps . as we have pointed out earlier , in general one can only increase the fidelity by a certain amount . in the context of the quantum repeater , teleportation comes into play when two ( purified ) segments of a channel are connected . the channel segments are represented by epr pairs between particles 1 - 2 and 3 - 4 that have been created across the corresponding segments , see fig . [ teleport_2 ] . in general , to teleport the state of particle 2 to particle 4 , a bell measurement has to be made on particles 2 and 3 ( and the result of this measurement communicated particle 4 ) . if particle 2 is itself entangled to some other particle , say particle 1 as in fig . [ teleport_2 ] , the effect of the teleportation is to transfer this entanglement to an entanglement between particles 1 and 4 . this process has also been termed entanglement swapping " @xcite . as a result , one obtains an epr pair between particles 1 and 4 which can be used as a single channel ( again with the help of teleportation ) . in this sense , the segments of the compound channel have been connected . put in different terms , the connection process creates quantum correlations across the compound channel from correlations that exist across the individual segments . we will study this process first for the particular case where the involved pairs ( 1 - 2 and 3 - 4 ) are not maximally entangled but are both in a werner state ( [ werner ] ) with fidelity f. to connect the pairs ( 1 - 2 ) and ( 3 - 4 ) , we make a bell measurement on particles 2 and 3 , which is realized by a ( imperfect ) cnot operation followed by two ( imperfect ) single - particle measurements , as described in the last paragraph of section [ error ] . depending on the outcome of this measurement , a one - particle operation @xmath125 is performed on particle 4 , as in the teleportation scheme @xcite . the measurement outcome needs thereby to be sent to particle 4 using classical communication . the result of this sequence of operations , followed by a subsequent depolarization , is a werner state between particle 1 and 4 with a smaller fidelity @xmath126 , which is independent of the outcome of the measurement . more generally , imagine that we have not only 2 pairs but a whole string of @xmath127 pairs as visualized in fig . [ figmanypairs ] . each of the pairs ( @xmath128),(@xmath129), .... (@xmath130 ) is assumed to be in a werner state with fidelity f. connecting these @xmath127 pairs using the procedure described above leads to a pair ( @xmath131 ) with fidelity @xmath132 given by the formula @xmath133 the parameters @xmath134 and @xmath36 appearing in this formula quantify the amount of noise that is introduced by the connection processes . the connection therefore leads to an exponential decrease of the resulting fidelity , unless both the elementary pairs and all the operations involved have unity fidelity . the connection can be performed in two different ways , which both lead to the same resulting state with fidelity @xmath132 but involve a different temporal ordering of the operations . the first way is to connect the pairs _ sequentially _ , i.e. first connect at @xmath135 , then at @xmath136 , and so on , each time only connecting one additional pair . this involves a sequence of @xmath137 connection procedures . the second way is to connect the pairs in _ parallel_. to achieve this , first connect simultaneously the neighboring pairs at @xmath135 , @xmath138 , ... @xmath139 . this leaves us with longer pairs @xmath140 , @xmath141 , ... , @xmath142 . then connect simultaneously these longer pairs at @xmath136,@xmath143, ... ,@xmath144 , and so on , until we get a final pair between a and b. to have at each step pairs of equal length and fidelity , @xmath127 should be some power of 2 , @xmath145 , although this is not an essential requirement . this method is much faster as it requires _ fewer _ iteration steps , namely @xmath146 instead of @xmath137 , although the number of local connections to be made is , of course , the same . the elementary pairs that are connected need not be in a werner state . for example , for bell diagonal states one can derive a map @xmath147 that relates the elements @xmath148 of the corresponding states before and after the connection . that is , bell diagonal states are mapped to bell diagonal states under connection . all other remarks about the connection procedure for a string of pairs apply here as well . the parallel way of connecting the pairs is preferable not only because can be done faster , but also because , after each connection step , one deals with an ensemble of _ identical _ pairs . further details are given in @xcite . we have now all necessary tools available to introduce the concept of the quantum repeater . our goal is to create an epr pair of high fidelity between two distant locations . since nonlocal entanglement between distant particles can not be created using only local operations , this involves the usage of a quantum channel , which is noisy in general . the bottleneck for communication over large distances is the scaling of the error probability with the length of the channel . when using , for example , optical fibers and single photons as a quantum channel , both the absorption losses and the depolarization errors scale _ exponentially _ with the length of the channel . the state of the photon or the photon itself will therefore be destroyed with almost certainty if the channel is longer than a few half - lengths of the fiber . to overcome this limitation , we divide the long channel into @xmath127 smaller segments and create less distant epr pairs across each segment as visualized in fig . [ figmanypairs ] . the number of segments @xmath127 is thereby chosen in such a way that it is possible to create epr pairs with sufficiently high initial fidelity @xmath149 over the distance of such a segment . in a next step , we connect these elementary " pairs as described in section [ connection ] . this leaves us with a pair between a and b with reduced fidelity @xmath132 as given in ( [ connect ] ) . in principle , one could now create many pairs between a and b in a similar way and then use this ensemble of pairs for purification . but purification is only possible if the fidelity of the initial pairs is above a certain threshold value @xmath5 as we have seen in section [ purification ] . this limits the number of of pairs one can connect before purification becomes impossible . we therefore connect a smaller number @xmath150 of pairs so that the resulting fidelity @xmath151 stays above the threshold value for purification ( @xmath152 ) and purification is possible . the general strategy will be to design an alternating sequence connection and ( re-)purification procedures in such a way that the number of resources needed remains as small as possible , and in particular does not grow exponentially with @xmath127 and thus with @xmath1 . in the remainder of this section we describe a _ nested purification protocol _ which consists of connecting and purifying certain groups of pairs simultaneously in the following sense ( see fig . [ fignesting ] ) . for simplicity , assume that @xmath153 for some integer @xmath154 . on the first level , we simultaneously connect the pairs ( initial fidelity @xmath155 ) at all the checkpoints except at @xmath156 . as a result , we have @xmath157 pairs of length @xmath158 ( and fidelity @xmath151 ) between @xmath101 & @xmath159 , @xmath159 & @xmath160 and so on . to purify these pairs , we need a certain number @xmath69 of copies that we construct in parallel fashion . [ for keeping track of the resources , it is convenient to arrange them in form of an array of elementary pairs as is done in fig . [ fignesting ] for @xmath161 and @xmath162 . we then use these copies on the segments @xmath101 & @xmath159 , @xmath159 & @xmath160 etc . , to purify and obtain one pair of fidelity @xmath163 on each segment . this last condition determines the ( average ) number of copies @xmath69 that we need , which will depend on the initial fidelity , the degradation of the fidelity under connections , and the efficiency of the purification protocol . the total number of elementary pairs we used up to this point is @xmath164 . on the second level , we connect @xmath158 of these larger pairs at every checkpoint @xmath165 ( @xmath166 ) except at @xmath167 . as a result , we have @xmath168 pairs of length @xmath169 between @xmath101 & @xmath170 , @xmath170 & @xmath171 , and so on of fidelity @xmath172 . again , we need @xmath69 parallel copies of these long pairs to re - purify up to a fidelity @xmath163 . the total number of elementary pairs involved up to this point is @xmath173 . we iterate the procedure to higher and higher levels , until we reach the @xmath154th level . as a result , we have obtained a final pair between @xmath101 & @xmath174 of length @xmath127 and fidelity @xmath163 . in this way , the total number @xmath175 of elementary pairs will be @xmath176 [ where @xmath177 alone gives the number of required ` parallel channels ' in fig . [ fignesting ] ] . we can re - express this result in the form @xmath178 which shows that the resources grow _ polynomially _ with the distance @xmath127 . the central feature of this nested purification protocol is the two - step process connection - purification on each nesting level . this purification loop is visualized in fig . [ figpuriloop ] . the curves shown in this figure are described by the formulas ( [ modified_bennett ] ) for purification ( upper curve ) , and ( [ connect ] ) for connection ( lower curve , with @xmath127 replaced by @xmath158 ) , respectively . let us consider a given nesting level k , where we have @xmath179 pairs of fidelity f each . starting from f , the fidelity @xmath151 after connecting @xmath158 pairs can be read off from the curve below the diagonal . reflecting this value back to to the diagonal line , as indicated by the arrows in fig . [ figpuriloop ] , sets the starting value for the purification curve . if @xmath151 lies within the purification interval @xmath180 , then the purification curve lies above the diagonal , @xmath181 , and iterative application of ( [ modified_bennett ] ) leads back to a fidelity larger or equal to the initial value @xmath66 . each step in the ` staircase ' of fig.[figpuriloop ] corresponds to a successful purification step , and with help of the total number of necessary steps @xmath70 needed to recover one can calculate the needed resources m as described in section [ purification ] . once the initial value @xmath66 of the fidelity is reobtained , we have @xmath182 pairs and we can start with the next level k+1 . in summary , each level in the protocol corresponds to one cycle in fig . [ figpuriloop ] . there are two conditions that have to be fulfilled to realize a closed loop . first , the fidelity after the connection has to be larger than the minimum value for purification ( lower fixpoint ) , @xmath183 . if @xmath184 , then repurification is impossible . second , the fidelity @xmath66 one starts with and wants to reach again has to be smaller than the maximal reachable fidelity ( upper fixpoint ) , @xmath185 . these two conditions determine both a threshold value for the error tolerances of the local operations and they limit the value @xmath158 ( the number of pairs which can be connected before purification becomes necessary ) . please note that , being polynomial in @xmath158 , the lower curve gets steeper and steeper near @xmath186 for higher and higher values of @xmath158 . from this one can see that , for a given initial fidelity @xmath66 , there is a maximum number of pairs one can connect before repurification becomes impossible . the threshold value for the nested purification scheme , i.e. the repeater , is tighter than the threshold for simple purification on a single segment . consider , as an example , the situation in fig . [ figpuriloop ] . the chosen value @xmath187 of the error parameter ( with @xmath47 ) lies above the purification threshold of @xmath188 that we have found after eq . ( [ simple_fixpoints ] ) . on the other hand , for @xmath189 , the purification interval in fig . [ figpuriloop ] would shrink essentially to a point at @xmath190 . clearly , in this situation no loop can be realized ; even if we start initially at @xmath191 , the purification interval " is left after the first connection process and the fidelity would subsequently converge towards the trivial fixpoint @xmath192 . when we employ the protocols a or b in the purification part of the nested scheme , we find error thresholds that are typically in the percent region . for scheme c , the situation is similar , _ as long as _ we do not use depolarization after each purification step ( i.e. keep all involved pairs during the purification process in the werner form ) . if we do use depolarization , thus realizing a modified version of protocol a , the loop can not be closed . differently speaking , for werner states , the fidelity that is lost by the connection process can never be regained by purification with this scheme . for this to be true , the repurification condition @xmath193 must be fulfilled , where the upper fixpoint @xmath76 is a function of the fidelity of the auxiliary @xmath158-pair . it can be shown analytically that this is not possible for scheme c with werner states . this underlines the fact that the purification scheme c is not a trivial variant of either of the schemes a and b. note that both connection and purification are not smooth processes . this means that one can only perform whole steps and not parts of a step . due to this fact , one will not get an exactly closed purification loop , in general , but the final fidelity after the purification ( this is the value at which the ladder in fig . [ figpuriloop ] ends ) may be slightly larger than the initial fidelity ; secondly , the final fidelity may be different at each nesting level , but it is always larger or equal to the fidelity of the elementary pairs . the length of a segment is limited by the transmission errors , and the number of segments @xmath127 one splits up the total channel should be varied in order to find out the optimal configuration . the quantum repeater can be characterized by two quantities , which depend on the used connection and purification protocol * the total _ physical resources per segment _ needed to build up the epr pair - this gives the number of necessary parallel channels between two checkpoints . * the _ total time _ which is needed to build up the epr pair . the way to calculate these quantities is similar for schemes a and b , while the properties of scheme c are totally different . the physical resources per segment @xmath194 gives us the number of required connection lines between two checkpoints , and also half of the particles which should be stored at each checkpoint . on the borders ( at a and b ) , it is equal to the number of particles to be stored , while at each checkpoint @xmath195 , one has to take care about the incoming and outgoing string and therefore needs the double number of particles . @xmath194 ( vertical axis in fig.[fignesting ] ) is completely determined by the resources m needed for each purification loop and is given by @xmath196 , where @xmath197 is the number of nesting levels . the resources @xmath69 needed for purification can be calculated as described in section [ purification ] . for schemes a and b , the physical resources are equal to the resources , since all pairs have to be created and therefore stored at the beginning of the process . figure [ figbennett ] shows the resources @xmath69 needed for a single purification loop , where it is assumed that @xmath198 , i.e. only two pairs are connected before re - purification takes place . @xmath69 is plotted against a working fidelity , which is the fidelity one starts with before connection and one ends up after re - purification . although the difference between schemes a and b does not look very dramatic , note that @xmath69 has to be taken to the @xmath199 power to calculate the resources per segment , where for long range communication @xmath200 ( see sec . [ compare ] ) . one can see that there exists an optimal working region , which can be understood as follows : for small working fidelities f , two things happen : firstly due to the polynomial law for the connection , the fidelity decreases strongly . secondly , purification becomes less effective near the lower fixpoint @xmath5 ( the gain per purification step gets smaller ) . for both reasons , many purification steps are needed to recover and thus the needed resources are large . for working fidelities f close to the maximal reachable fidelity ( upper fixpoint @xmath76 ) , one does not loose much due to the connection , but purification is less effective - the gain per purification step is smaller . therefore the the number of necessary steps is again quite large and so are the needed resources . in between there exists a region which is optimal in the sense that the needed resources are minimized . here a working fidelity @xmath201 turns out to be optimal for the chosen error parameters . the error dependence of the resources is shown in fig.[figoxford ] . for smaller error probabilities ( larger error parameters ) , two things happen : the optimal working fidelity gets larger and the needed resources @xmath69 decrease . for imperfect local operations , the needed resources increase to infinity if one really wants to reach the maximal achievable fidelity @xmath76 , only for perfect operations and @xmath186 , this is not true because one starts with perfect pairs and does not loose due to connection . therefore no purification is necessary and @xmath202 . using scheme c for purification , the situation is different . the vertical axis of fig.[fignesting ] ( which corresponds to the resources ) is translated into a `` temporal axis '' . instead of creating all needed pairs at the beginning and operating parallelly on ensembles of pairs at each checkpoint ( as in scheme a and b ) , one creates the pairs needed for purification every time one needs one and operates sequentially on the pairs . to perform the purification process , every time only two pairs are involved . one is needed to store the purified state and the other one is repeatedly created to purify the first pair as visualized in fig.[ibkresfig ] . the number of _ physical resources _ required for purification is thus only 2 , but every time the purification was not successful one has to start from the very beginning . using scheme a or b one obtains a purified pair after @xmath70 successful purification steps , which is in this case also the number of really performed purification steps . looking at fig.[tree ] , one sees that a purification step which was not successful eliminates one branch of the tree , but after 4 steps ( in this example ) one ends up with a purified pair . in scheme c , on the other hand , one does not follow all branches parallel but follows the branches sequentially ( and using always the same elementary pair for each purification step ) . this limits the number of involved particles , but increases the required time . looking now at the nested algorithm , it turns out that the required number of physical resources for scheme c grows linearly with the number of nesting levels and thus logarithmically with the number of segments ( and the distance ) . this is because one additional pair is needed for storage purposes at each nesting level . one can understand this by inspecting fig.[ibkrep ] . first , three elementary pairs are connected ( line 4 ) and used to purify the pair at line 3 , which now is the elementary pair for the next nesting level . second , three of these re - purified pairs ( line 3 ) are connected and used to purify the pair at line 2 . since one has to create such a pair repeatedly and all particles at lines 3 and 4 are involved , it is necessary to store the pair to be purified ( here in line 2 ) . therefore one needs one additional particle to store the pair at each new nesting level . this particle is not needed in all checkpoints , but only in those which lie on the _ borders _ of the corresponding nesting levels . that is why the maximum number of additional particles ( physical resources ) is required at the outermost places ( a and b ) , while e.g. at checkpoint @xmath135 , no additional particle is needed . the _ maximal _ number of additional particles ( physical resources ) grows only linearly with the number of nesting levels and therefore logarithmically with the number of segments in contrast to the polynomially growing resources when using scheme a or b. the maximal resources per segment ( in this case particles needed to store the pair ) can be calculated by r_segment=1+log_ln=1+n here we will discuss the total time needed to create an epr pair . this involves at least three parameters : * the typical time @xmath203 needed to perform local operations ( single qubit , two qubit and measurements ) . * the typical time @xmath204 to create an elementary pair . using optical fibers and the model of the absorption free channel ( afc ) @xcite , this can be expressed in terms of the other 2 parameters and is given by _ pair=_afc=(5_op+2_class)e^ ( ) where @xmath205 is the half length of the fiber and @xmath206 is the length of a segment . this just reflects the fact that for a single use of the afc , all together 5 operations and 2 classical communications ( in this case the transmission of a photon ) are necessary . the exponential function gives the average number of repetitions which are necessary due to absorption losses . * the time @xmath207 needed for classical communication to broadcast the result of the measurement . this time is determined by the length of the segment and the speed of light c and is given by _ class= to simplify the discussion , we will assume that the number of pairs @xmath158 which are connected at each nesting level is some power of 2 , @xmath208 , because in this case the connection process can be performed in parallel . for each connection process , 3 elementary operations ( cnot , measurement , operation depending on the outcome of the measurement ) and classical communication over the distance of one segment are needed . for purification , also three elementary operations ( e.g. for scheme a depolarization , bilateral cnot and bilateral measurement ) and classical communication of the result are necessary . using scheme a or b , the time needed to perform a _ purification loop _ ( the connection of @xmath208 pairs and re - purification , where @xmath70 successful purification steps are necessary ) at nesting level m is given by t_loop(m)= 3l_op+f(m)(2^k-1)_class+k_max(3_op+f(m)2^l_class ) where @xmath209 gives the length dependence of the elementary pair on the nesting level . the first two terms give the time required to perform the connection process in parallel fashion , where @xmath1 connections and classical communication over the distance of @xmath210 times the length of an elementary pair is needed . the purification process involves @xmath70 purification steps and for each purification step classical communication over the distance of @xmath211 times the length of an elementary pair ( since @xmath208 pairs were connected before ) . one only has to take into account the successfull purification steps , since one runs through the whole purification tree ( see fig.[tree ] ) in a parallel fashion . to calculate the total time needed to perform the nested algorithm , one can use the iteration formula t_tot(m)=t_tot(m-1)+t_loop(m ) [ iteratetime]where m corresponds to the @xmath212 nesting level and @xmath213 gives the total time needed for the creation of an epr pair between a and b. to complete this iteration formula , note that @xmath214 which is just the time needed to create an elementary pair at the beginning . the total time if one has @xmath153 segments and @xmath208 is thus t_tot^a , b= n(3l+3k_max)_op+(2^l-1+k_max2^l())_class + _ pair the main contribution is given by the classical communication if @xmath203 is not too large and is just @xmath215 . using scheme c for purification , the time needed to perform a purification loop on nesting level m is given by t_loop(m)=(\{t_loop(m-1)}+3l_op+f(m)(2^l-1)_class)m + s(3_op+f(m)2^l_class ) where the first term is the time needed to create a pair for purification ( by building up @xmath211 pairs at nesting level ( m-1 ) and connecting them ) , which must be done _ sequentially _ m times . the second term gives the time needed for purification ( operations and classical communication over the distance of @xmath211 times the length of an elementary pair ) , where s purification steps have to be performed . the function @xmath209 reflects the length dependence of the elementary pair on the nesting level and the time needed to create an elementary pair at the lowest nesting level @xmath216 , which completes this iteration formula . the maximum appearing in this expression is taken over all @xmath211 segments on nesting level @xmath217 . this is necessary because all @xmath211 new elementary pairs have to be built up before the next connection process can start . therefore one has to wait for the slowest one . note that the time for purification is fixed for schemes a and b ( obtaining bad results in this case corresponds to the elimination of a branch in fig.[tree ] ) , while for scheme c the purification process has to be started from the very beginning when obtaining a bad result . this results in the fact that the performed purification steps s and the used purification resources m ( and therefore the needed time ) might vary from case to case . this formula is sufficient to calculate the total time needed to perform the nested algorithm up to nesting level m and @xmath218 gives the total time needed to build up an epr pair between a and b. a rough estimation of the total time can be obtained when using ( [ iteration ] ) and ( [ iteration1 ] ) to calculate the _ average _ number of purification resources m and purification steps s. in this case , one can write down a closed expression for the total time , which is polynomially growing with the number of segments @xmath127 and thus with the distance . comparing this time with the results obtained from a simulation of the whole process , it turns out to be smaller by a factor of @xmath219 . in some sense , the vertical axis ( physical resources ) in fig.[fignesting ] is translated into a temporal axis . we will give some typical numbers to quantify the properties of the nested purification protocol if the purification loop is performed with help of scheme a , b or c. the following tables are based on error parameters of @xmath220 and a working fidelity f=0.96 , the fidelity one starts with at the lowest nesting level and one ends up after each purification loop . the length of a segment is assumed to be in the order of half length of an optical fiber , @xmath221 . in this case , the number of segments is proportional to the distance between two locations . if the number of segments is in the order of @xmath222 , this will be called continental scale ( distance of @xmath223 ) , while @xmath224 is called intercontinental scale ( distance of @xmath225 ) , since one can create an epr pair between two international cities . it is assumed that the connection is performed in parallel . the columns from left to right have the following meaning : * resources : this gives the _ physical _ resources per segment . * time : gives the total time in seconds needed to create an epr pair . the number is based on the application of the afc to create elementary pairs ( @xmath226 ) . the half length of the fiber is considered to be 10 km , which is also the length of the elementary segment ( @xmath227 ) . all operations ( single qubit , 2 qubit , measurement ) are considered to be performed in @xmath228 . the time is calculated using a simulation of the nested entanglement purification process , averaging over a few hundred runs . [ cols="<,^,^,^,^ " , ] one sees that the physical resources needed when using scheme c are some orders of magnitude smaller , while the time needed is larger . for a practical implementation , the achievable bit rates are not very impressive , but note that all numbers get much better if the local operations can be performed better ( error parameters closer to 1 ) as can be seen in fig.[figoxford ] . so the reachable fidelity is larger and at the same time , the required resources are smaller when using better local operations . note that the total time needed is in the order of the time for classical communication over this distance , which is just @xmath229 . we have shown that it is possible to create epr pairs via a noisy channel , with an overhead in physical resources that grows only logarithmically with the length of the channel , while the time needed for the creation grows polynomially . the central idea is to use a nested purification protocol for connecting a sequence of epr pairs at certain connection points " within the channel , whose roles are reminiscent of to classical repeaters . different from the classical situation , the concept of the quantum repeater is not a local amplifier , but it involves both the local checkpoints and global purification protocol . our scheme tolerates errors for local operations and measurements that are in the percent region . permanent address : institut fr theoretische physik , universitt mnchen , theresienstrasse 37 , d80333 mnchen . b. schumacher , phys . rev . a , * 54 * , 2614 ( 1996 ) ; b. schumacher and m. d. westmoreland , _ ibid . _ * 56 * , 131 ( 1997 ) . w. k. wootters and w. h. zurek , _ nature _ * 299 * , 802 ( 1982 ) . d. dieks , phys . a * 92 * 271 ( 1982 ) . r. j. glauber , in _ frontiers in quantum optics _ , e. r. pike & s. sarkar ) , pp . 534 - 582 . bristol : adam hilger . w. tittel , j. brendel , b. gisin , t. herzog , h. zbinden , and n. gisin , _ los alamos report _ quant - ph/9707042 . for optical fibers , this length is typically 10 km ; see e.g. @xcite . p. shor , quant - ph/9605011 ; a. m. steane _ phys . lett . _ * 78 * 2252 ( 1997 ) . d. gottesman , quant - ph/9702029 ; for a review see e.g. j. preskill , quant - ph/9712048 . e. knill and r. laflamme , quant / ph-9608012 . see also d. aharonov and m. ben - or , quant - ph/9611025 ; c. zalka , quant - ph/ 9612028 . c. h. bennett , g. brassard , s. popescu , b. schumacher , j. a. smolin , and w. k. wootters , phys . lett . * 76 * , 722 ( 1996 ) . c. h. bennett , d. v. divincenzo , j. a. smolin , and w. k. wootters _ phys . a _ * 54 * , 3824 ( 1996 ) . n. gisin , phys . a * 210 * 151 ( 1996 ) . d. deutsch , a. ekert , c. macchiavello , s. popescu , and a. sanpera , phys . lett . * 77 * , 2818 ( 1996 ) . c. h. bennett , c. h. brassard , c. crepeau , r. josza , a. peres , and w. k. wootters , phys . lett . * 70 * , 1895 ( 1993 ) . a. k. ekert , _ phys . lett . _ * 70 * , 661 - 663 ( 1991 ) . g. giedke _ _ , unpublished . j. briegel , w. dr , j. i. cirac , and p. zoller , quant - ph-9803056 ( 1998 ) . c. miquel , j. p. paz and w.h . zurek , phys . * 78 * , 3971 ( 1997 ) . a similar form as in ( [ result_bloch_av ] ) may be obtained for two - bit rotations with a corresponding isotropic noise function , although the calculation is more complicated . in particular , we have not yet been successful in giving a proof in terms of a manifest _ parametrization - independent average _ over the invariant group manifold of su(4 ) . isotropic noise in this case means that the depolarizing channel does not distinguish between the two qubits and acts on them in a symmetric way . this assumption is not essential ; it is motivated by the fact that the depolarization is a itself an inessential although simplifying step in the bennett et al . protocol . if the reliability @xmath35 is the same for all one - bit operations involved in the realization of the twirl , an _ imperfect depolarization _ can effectively be described by @xmath230 where @xmath231 summarizes reliability of the total twirl operation @xmath232 and @xmath62 is the bilocal state before the twirl is applied . this operation still depolarizes @xmath62 and brings it to the werner form . however , the fidelity @xmath233 of the resulting state is reduced with respect to the fidelity @xmath63 of the state @xmath62 in ( [ modified_bennett ] ) by the relation @xmath234 c. w. helstrom , _ quantum detection and estimation theory _ , associate press , london 1976 . any sequence of operations at the same node , e.g. two single qubit operations @xmath38 followed by a cnot operation will be described by a single two - qubit operation @xmath39 and consequently with only a single error parameter w. dr , master thesis , available under http://www.uibk.ac.at/c/c7/c705/staff/wd.html see also n. ltkenhaus , unpublished . a. zeilinger _ _ , _ phys . _ 78 _ , 3031 ( 1997 ) in their scheme , however , a three - particle entangled state is obtained rather than a distant epr pair . this procedure is not necessary , but to have a consistent model for both connection and purification , we want to describe all operations in terms of ( [ two_qubit_op][povm ] ) . s. j. van enk , j. i. cirac , and p. zoller , _ science _ , * 279 * , 205 ( 1998 ) . s. j. van enk , j. i. cirac , and p. zoller , _ phys . _ , * 78 * , 4293 ( 1997 ) .

we have thoroughly characterized the ultraviolet to near - infrared ( 0.15 - 2.2 ) spectral energy distribution ( sed ) of the central parsec of the m33 nucleus through new infrared photometry and optical / near - infrared spectroscopy , combined with ultraviolet / optical observations from the literature and the hst archive . the sed shows evidence for a significant level of attenuation , which we model through a monte carlo radiative transfer code as a shell of clumpy milky way - type dust ( ) . the discovery of milky way - type dust ( with a strong 2175 bump ) internal to the m33 nucleus is different from previous work which has found smc - like dust ( no bump ) near starburst regions . the amount by which dust can be processed may be related to the mass and age of the starburst as well as the extent to which the dust can shield itself . our starburst models include the effects of this dust and can fit the sed if the nucleus was the site of a moderate ( at 10 myrs ) episode of coeval star formation about 70 myrs ago . this result is quite different from previous studies which resorted to multiple stellar populations ( between 2 and 7 ) attenuated by either no or very little internal dust . the m33 nuclear starburst is remarkably similar to an older version ( 70 myr versus 10 myr ) of the ultra - compact starburst in the center of the milky way .

astro-ph9902043

the nucleus of m33 ( ngc 598 , triangulum galaxy ) has been studied extensively , but its true nature has remained elusive . it is bright and compact , has a fwhm of @xmath4 ( @xmath5 pc ) in hst f555w observations , and lacks a substantial black hole ( @xcite ; @xcite ; @xcite ) . van den bergh ( 1991 ) reviews studies on the stellar populations in m33 , including the nucleus . previous work has found that the nucleus is composed of roughly two stellar populations , one with an age of @xmath6 years and one with an age of @xmath7 years ( @xcite ; @xcite ; @xcite ) . previous attempts at modeling the stellar population of the m33 nucleus have either ignored the effects of reddening internal to m33 or modeled it assuming a simple screen of dust . the presence of ultraviolet ( uv ) ( especially far - uv [ @xcite ] ) flux from the m33 nucleus , and its small size , imply a coeval , young stellar population . the flat uv / optical spectrum of the m33 nucleus [ figure 1c of mcquade , calzetti , & kinney ( 1995 ) ] , when coupled with the assumption of a single young stellar population , can only be explained by significant reddening by dust internal to m33 ( @xcite ) . the work of witt , thronson , & capuano ( 1992 ) demonstrates the importance of accurately treating the radiative transfer through the dust in galaxies . the m33 nucleus is ideal for investigating the role of dust in starburst galaxies . previous work on starburst galaxies found that the extinction curve lacks a 2175 bump ( @xcite ; @xcite ) . the only other dust which does not have a 2175 bump is found in the small magellanic cloud ( smc ) . gordon & clayton ( 1998 ) reinvestigated the uv extinction in the smc and found evidence to support the processing of dust near sites of active star formation . dust with weak 2175 bumps is seen along some galactic and large magellanic cloud ( lmc ) sightlines ( e.g. @xcite ; @xcite ) . it is possible that the lack of a 2175 bump is the result of processing the dust near sites of active star formation . the work of gordon et al . ( 1997 ) was based on an iue selected sample of starburst galaxies and due to the large aperture of iue , the sample was observationally biased toward intrinsically bright starbursts with small amounts of dust . we probe the m33 nucleus with a model including stars , gas , and dust . we find that its dust has a strong 2175 bump , in contrast to the behavior of the starburst dust studied previously ( @xcite ) . this paper represents the beginning of a series of papers devoted to understanding the role of dust on the interpretation of stellar populations in starburst galaxies . in this paper , we find that the m33 sed can be modeled as a modest ( @xmath2@xmath8 ) @xmath270 myr old starburst enshrouded by a shell of mw - like dust ( @xmath9 ) by using a starburst model which includes stars , gas , and dust . the indication of a young , very compact , and moderate luminosity starburst in the central pc of m33 is interesting because of the possible similarity to the very young and massive stars within the central 0.5 pc of the milky way . because of their modest outputs and small sizes , such systems would not stand out clearly in the nuclei of galaxies outside the local group . the presence of two similar events in local group galaxies suggests that such starbursts may be relatively common and hence may play an important role in the evolution of the central parsec - sized regions of galaxies . the observations of the m33 nucleus can be divided into two groups ; photometry and spectroscopy . for photometric measurements , the background level is determined using an annulus surrounding the nucleus . as a result , the photometry of the m33 nucleus does not include a significant contribution from the disk or bulge of m33 . the background levels for spectroscopic observations are usually measured a significant distance from the nucleus . this means they include contributions from regions surrounding the nucleus , which makes studying just the nucleus difficult . thus , to isolate the spectral behavior of the nucleus requires a judicious combination of both photometry and spectroscopy . the ultraviolet imaging telescope ( uit , @xcite ) observed m33 during the astro-1 mission in the far - uv ( b1 filter , @xmath10 ) and the near - uv ( a1 filter , @xmath11 ) . the uit images , along with ground - based ubv images , were analyzed by massey et al . they present uit - b1 , uit - a1 , u , b , and v measurements of the nucleus of m33 ( see table [ tab_obs_sed ] ) and note the nucleus is very red compared to the other sources in their images . additional ground - based measurements at b and r are given by kormendy & mcclure ( 1993 ) . the zero points for the u , b , v , and r magnitudes were taken from bessell , castelli , & plez ( 1998 ) . lcccc uit - b1 & 1534 & 355 & @xmath12 & 1 f160bw & 1554 & 575 & @xmath13 & 2 f170w & 1695 & 520 & @xmath14 & 2 uit - a1 & 2607 & 1147 & @xmath15 & 1 f300w & 3045 & 879 & @xmath16 & 2 f336w & 3349 & 495 & @xmath17 & 2 u & 3641 & 641 & @xmath18 & 1 b & 4421 & 960 & @xmath19 & 1 b & 4421 & 960 & @xmath20 & 3 f555w & 5314 & 1578 & @xmath21 & 2 v & 5507 & 984 & @xmath22 & 1 r & 6563 & 1591 & @xmath23 & 3 f814w & 8365 & 2504 & @xmath24 & 2 f1042 m & 10444 & 854 & @xmath25 & 2 j & 12302 & 2036 & @xmath26 & 2 h & 16379 & 2863 & @xmath27 & 2 k & 21201 & 2882 & @xmath28 & 2 numerous wfpc2 images of the central region of m33 exist in the hst archive . we chose post - costar images taken with at least two images in each filter to facilitate the removal of cosmic rays . the names , filters , and exposure times of the images used are listed in table [ tab_wfpc2_obs ] . all but the f170w and f336w images were taken with the nucleus centered on the pc chip . for each filter , we combined the available images to create a cosmic ray free image without saturated pixels . the total counts from the nucleus in each filter were calculated by summing the counts out to a radius @xmath29 after subtracting the background counts , which were determined using an annulus with radii between @xmath30 and @xmath31 . it was necessary to use an aperture at least @xmath32 in radius due to the broad wings on the hst point spread function ( @xcite ) . these counts were converted to the fluxes presented in table [ tab_obs_sed ] using the algorithms presented by voit et al . ( 1997 ) and whitmore , heyer , & baggett ( 1996 ) . the uncertainties were assumed to be 15% for f160bw , f170w , and f1042 m images . all other images were assumed to have an uncertainty of 10% . these values reflect the uncertainties in both the absolute calibration and determination of the background level . ccccc u3mr0101 m & f160bw & 1200 u3mr0102 m & f160bw & 1200 u2ye0207 t & f170w & 900 u2ye0209 t & f170w & 900 u3mr0103 m & f300w & 800 u3mr0104 m & f300w & 800 u2ye020bt & f336w & 900 u2ye020dt & f336w & 900 u2e20507 t & f555w & 400 u2e20508 t & f555w & 400 u2e20509 t & f555w & 400 u2e2050at & f555w & 400 u2e2050bt & f555w & 40 u2e2050ct & f555w & 40 u2e2050dt & f814w & 400 u2e2050et & f814w & 400 u2e2050 ft & f814w & 400 u2e2050gt & f814w & 40 u2e2050ht & f814w & 40 u2e20501 t & f1042 m & 500 u2e20502 t & f1042 m & 500 u2e20503 t & f1042 m & 700 u2e20504 t & f1042 m & 300 u2e20505 t & f1042 m & 500 u2e20506 t & f1042 m & 500 near infrared images at @xmath33 , @xmath34 , and @xmath35 of the central region of m33 were taken on 1998 march 2 ( ut ) using the nsfcam instrument on nasa s infrared telescope facility ( irtf ) . the detector was a @xmath36 insb array with an image scale of @xmath37/pixel giving a field - of - view of @xmath38 . for each filter , four images were taken alternating with four sky exposures . the sky images were combined using a sigma clipping algorithm giving a final sky image without stars . the flatfield image was calculated by subtracting a dark image from the sky image and normalizing the result . each of the images of m33 was sky subtracted and flat fielded before being combined using the sigma clipping algorithm to remove cosmic rays and bad pixels . the calibration of the images was done using standard stars from hunt et al . ( 1998 ) chosen to bracket @xmath39 colors between -0.15 and 0.63 . the standard stars were observed throughout the night bracketing airmasses between 1.0 and 2.2 . the @xmath40 magnitudes by hunt et al . ( 1998 ) were converted to @xmath35 magnitudes using @xmath41 ( @xcite ) . total exposure times were 80 seconds for each of the three filters . the @xmath33 , @xmath34 , and @xmath35 magnitudes of the nucleus were extracted using a sky annulus with radii between 40 and 60 pixels . the magnitudes were transformed from the cit system to the homogeneous system defined by bessell & brett ( 1988 ) . the zero points for the @xmath33 , @xmath34 , and @xmath35 magnitudes were taken from bessell et al . ( 1998 ) and the resulting fluxes are given in table [ tab_obs_sed ] . these fluxes are consistent with those given by gallagher et al . ( 1982 ) within the uncertainties . international ultraviolet explorer ( iue ) observations provide the only ultraviolet spectra of the m33 nucleus . the observations used the large aperture ( @xmath42 ) . examining the line - by - line ( silo ) spectra of the two iue observations with similar position angles ( swp 52142 and lwp 1584 ) , we found two sources were present . using the mgex and newcalib routines provided in the iueidl package ( @xcite ) , we extracted and calibrated the two spectra . we imposed the iue aperture on the uit images of m33 and confirmed that one of the spectra was from the nucleus and one was from a nearby star ( uit 219 , @xmath43 , @xcite ) . the integrated flux from the uit images ( fuv0496 and nuv0402 ) in the iue aperture matched that of the sum of the two spectra within the uncertainties . optical spectroscopy of the m33 nucleus was obtained using the goldcam ccd spectrometer on the 2.1meter telescope at kitt peak national observatory on 1998 june 29 ( ut ) . grating # 201 was used with a slit width of 240 @xmath44 m ( 3@xmath45 ) resulting in a resolution of 19 fwhm over a spectral range from 33009100 . in the spatial direction , the slit extended @xmath46 with a scale of 0.78@xmath45 pix@xmath47 . the m33 spectrum was obtained in a single 900 sec exposure with a position angle of 100@xmath48 and at an airmass of 1.39 . the flat field used in reducing this spectrum was constructed from exposures of a quartz lamp corrected for the slit illumination function using twilight sky exposures . the spectrum was wavelength calibrated using henear lamp exposures taken throughout the night . the rms deviation of the adopted solution about the comparison line wavelengths was @xmath49 . atmospheric h@xmath50o and o@xmath50 bands were removed using observations of hot stars and the method described by bica & alloin ( 1987 ) . a mean sky frame was constructed from several exposures of galaxies that did not fill the spatial direction on the chip . this frame was subsequently scaled to match the night sky lines in the m33 spectrum and subtracted from the m33 image . a 1dimensional spectrum was extracted using a 10@xmath45 aperture centered on the m33 nuclear spectrum . the non - nuclear contribution to the spectrum was subtracted by fitting a 5th degree polynomial to the galactic background along the spatial direction . the resulting spectrum was flux calibrated using several observations of spectrophotometric standards and the mean extinction coefficients appropriate for kpno . the optical spectrum of the m33 nucleus is shown in figure [ fig_id_opt ] with the strong lines identified ( @xcite , 1987 ) . our spectrum is similar to that given in figure 1 of schmidt et al . ( 1990 ) except ours has a slightly steeper slope and a more pronounced paschen jump . in addition , our spectrum is absolutely calibrated while the schmidt et al . ( 1990 ) spectrum was only relatively calibrated . the near - ir observations were made using the steward observatory infrared spectrometer , fspec ( @xcite ) , on the multiple mirror telescope ( mmt ) on 1996 june 21 , 22 ( ut ) . the spectra were taken using the medium resolution , 300 lines / mm grating . four wavelength settings were used , @xmath33-band from 1.15 to 1.35 @xmath44 m , @xmath34-band from 1.54 to 1.77 @xmath44 m and two settings at @xmath40-band , from 2.015 to 2.23 @xmath44 m and from 2.19 to 2.41 @xmath44 m . the spectral resolution , as measured via oh sky emission lines , or neon krypton calibration lamps in the case of the long wavelength k - band , corresponded to approximately 2.5 resolution pixels ( 21 ) , or r @xmath51 590 at 1.25 @xmath44 m , r @xmath51 780 at 1.65 @xmath44 m , r @xmath51 1000 at 2.12 @xmath44 m and r @xmath51 1100 at 2.3 @xmath44 m . the same observing procedure was used on both nights and with all grating positions . the spectrometer has a slit size of @xmath52 on the mmt . this allowed for four unique positions as the m33 nucleus was stepped @xmath53 along the slit . the integration time at each position was 2 minutes giving a total of 8 minutes per set of observations . a total integration of 24 minutes was obtained at both @xmath33- and @xmath34-band . at 2.12 @xmath44 m ( short @xmath40-band ) and 2.30 @xmath44 m ( long @xmath40-band ) , the total integration times were 40 and 32 minutes respectively . the reductions made use of a dark - current - corrected flat field obtained from a uniformly illuminated screen . after dark current had been subtracted and the flat field divided into the raw two - dimensional images , sky emission ( and additional thermal background in the k - band ) was removed by subtracting one slit position from the next . in many cases , a few percent scaling was required to get the telluric atmospheric oh features to disappear entirely . background renormalization by a few percent was also required to remove fluctuations in thermal background between integrations in the k - band . interspersed between our m33 observations , we obtained spectra at a similar airmass of hr 410 , a late f dwarf , to be used to remove telluric artifacts . intrinsic features in the spectrum of this star were removed using a kurucz model atmosphere , with t@xmath54 = 6500 , log(g ) = 4.5 , solar abundance . dividing the spectrum of m33 by this modified stellar spectrum allowed us to cancel features due to the atmosphere . while our near - infrared spectra are not flux calibrated ( due to slit loss and non - photometric conditions ) we can determine the accurate relative flux distributions for each grating position . the long slit of fspec allowed us to look for evidence of light extended beyond the nuclear region . while there likely is some low level flux outside the nucleus , dividing one slit position by another in the reductions revealed no structure outside of the nuclear point source . measurement of the full - width , half - maximum flux of the nucleus along the slit also reveals the nucleus to be unresolved spatially . typical seeing during the observations was very good , approximately 1 to @xmath55 ; however by the end of the integration , the multiple mirrors of the mmt drifted slightly relative to each other on the sky , reducing the effective seeing to @xmath2 @xmath56 . lccc ? ? & & 1.196 & 1.5 & ( 3 - 5 ) & 1.2818 & 1.5 @xmath57co & ( 3,0 ) & 1.558 & 1.3 ? ? & & 1.565 & 0.8 & ( 4s-4p , tr ) & 1.577 , 1.575 , 1.575 & @xmath58 @xmath57co & ( 4,1 ) & 1.578 & @xmath59 & & 1.5894 & 2.5 & ( 4p-5s ) & 1.596 & 0.3 @xmath57co & ( 5,2 ) & 1.598 & 1.5 & ( 4 - 13 ) & 1.6109 & @xmath601 @xmath57co & ( 6,3 ) & 1.619 & 2.5 & ( 4 - 12 ) & 1.6407 & 1.8 ? ? & & 1.6525 & 1.3 @xmath57co & ( 8,5 ) & 1.662 & 4.0 & ( 4 - 11 ) & 1.6811 & @xmath611 & ( 4s-4p ) & 1.7146 & 4.6 @xmath57co & ( 10,7 ) & 1.7067 & 2.0 & & 1.7307 & 1.2 & & 1.7332 & 2.0 & ( 4 - 10 ) & 1.7362 & @xmath59 ? ? & & 2.0443 & 1.0 & & 2.0704 & @xmath62 & ( 4p-5s ) & 2.1360 & @xmath601 & ( 4 - 7 ) & 2.1661 & 2.0 & & 2.1789 & @xmath20.3 & & 2.1885 & 1.0 & & 2.2062 & @xmath21 & & 2.2090 & @xmath610.5 @xmath57co & ( 2,0 ) & 2.2935 & @xmath210 @xmath57co & ( 3,1 ) & 2.3227 & @xmath28 @xmath63co & ( 2,0 ) & 2.3448 & @xmath23.5 @xmath57co & ( 4,2 ) & 2.3525 & @xmath29 @xmath63co & ( 3,1 ) & 2.3739 & @xmath24 @xmath57co & ( 5,3 ) & 2.3830 & @xmath25.5 @xmath63co & ( 4,2 ) & 2.4037 & @xmath22.5 to model the spectral energy distribution ( sed ) of a starburst region , we need to model the stellar and gas emission accurately , including the modification of this emission by absorption and scattering by dust in the starburst region . other input parameters include the distance to the starburst , the radial velocity of the starburst , and the dust column in our galaxy towards the starburst . for the m33 nucleus , the distance is 795 kpc , the foreground dust column has an @xmath64 and was assumed to be milky way ( mw ) dust with @xmath65 ( @xcite ) , and the radial velocity is -172 km s@xmath47 ( @xcite ; @xcite ) . the starburst model is an improved version of that used by gordon et al . the stars and gas in a starburst are modeled using leitherer et al . ( 1995 , 1998 ; hereafter l98 ) and fioc & rocca - volmerage ( 1997 , 1998 ; hereafter pegase ) stellar evolutionary synthesis ( ses ) models . we used two ses models in an attempt to estimate the uncertainties in the model seds due to uncertainties in the assumed stellar evolutionary tracks , stellar spectra , and computational algorithms . both models were run for burst and constant star formation scenarios , an initial mass function ( imf ) with a salpeter ( 1955 ) slope ( @xmath66 ) , a mass range of 0.1100 m@xmath67 , and solar metallicity . the l98 model used geneva evolutionary tracks and the pegase model used padova evolutionary tracks . the choice of a salpeter imf for stellar masses greater than @xmath68 is supported by the observations in local group clusters and associations ( @xcite ) . furthermore , imf8 , found by rieke et al . ( 1993 ) to be optimum for the m82 starburst , has a net high mass slope similar to the salpeter value . there are no direct constraints on the form of the imf below @xmath69 in starburst - type regions , but the local imf has a slope of @xmath70 1 below masses of @xmath2@xmath71 ( e.g. , @xcite ) . for this work , we assume the imf has the form @xmath72 where @xmath73 is the normalization constant of the imf . the use of a different imf below @xmath69 changes the total mass needed for the burst , but does not affect the computed sed for the ages appropriate for the m33 nucleus . for example , using the imf given in equation [ eq_imf ] would imply 59% of the mass given by the assumption of a salpeter slope for the entire imf . to obtain realistic masses , we have therefore multiplied the model values by 0.59 . the metallicity of the m33 nucleus should be similar to the surrounding region and the central abundances of m33 are approximately solar ( @xcite ; @xcite ) . seds were produced for ages up to @xmath74 years for l98 and @xmath75 years for pegase . both models used the lejeune , cuisinier , & buser ( 1997 ) spectral library which has a resolution of @xmath240 . an additional input to both models is the fraction of lyman continuum photons which contribute to the nebular emission ( the rest are assumed to be absorbed by dust ) . since the nucleus of m33 does not show any emission at h@xmath76 , the contribution to the sed from nebular emission is negligible and the value of this model parameter is unimportant . both ses models produce similar seds for ages between 10 and 100 million years ( myrs ) . for ages @xmath61 10 myrs , there are quite significant differences in the optical / ir between the seds which can be traced to the different evolutionary tracks used ( geneva versus padova ; @xcite ) . after 100 myrs , the pegase model produces consistently higher ir fluxes , which is due to the inclusion of the thermally pulsating asymptotic giant branch ( tp - agb ) evolutionary tracks in this model ( @xcite ) . these tracks are not included in the l98 model , limiting its validity to ages @xmath77 myrs ( c. leitherer 1998 , private communication ) . the effects of dust were modeled using a monte carlo radiative transfer code ( @xcite ; @xcite ; @xcite ; @xcite ) . calculation of the radiative transfer through dust is crucial when any geometry other than the screen geometry is suspected . our radiative transfer model was run for dust global geometries characterized as dusty ( dust , stars , and gas uniformly mixed together ) and shell ( a sphere of stars and gas enshrouded by a shell of dust ) . these two geometries define the extremes one would expect for star formation regions . the local dust geometry was characterized with inter - clump to clump density ( @xmath78 ) ratios between 1.0 and 0.001 and a clump filling factor of 15% . in a given model , the physical properties of the dust grains ( @xmath79 , albedo , and scattering phase function ) were characterized by either mw ( @xmath80 , @xcite ) or small magellanic cloud bar ( @xcite ) type dust ( @xcite ) . the amount of dust is parameterized by the v band optical depth , @xmath81 . the dust models were run for @xmath81 values ranging from 0.25 to 50 . combining the ses and dust radiative transfer models provides a model for the seds of starburst regions which accurately includes stars , gas , _ and _ dust . the parameter space covered by the starburst model is large and , as a result , an algorithm to determine which of the model seds fit the observed sed was needed . following gordon et al . ( 1997 ) , we use the @xmath82 ( reduced @xmath83 ) method . the @xmath82 was computed using @xmath84 ^ 2\ ] ] where @xmath85 is the number of points in the observed sed , @xmath86 is the number of degrees of freedom , @xmath87 is the observed flux at the @xmath88th point in the sed , @xmath89 is the same for the model flux , and @xmath90 is the @xmath88th uncertainty in the observed flux ( @xcite ) . in the paper by gordon ( 1997 ) , a typographical error resulted in the square being left off the term in the sum of equation [ eq_chisqr ] . in an attempt to also account for the observational uncertainties in the ses and radiative transfer model predictions , we have added , in quadrature , a 15% ( 10% ) uncertainty to @xmath90 of the ultraviolet / infrared ( optical ) observations . the uncertainties associated with the ses models are due to their use of observational spectra to calibrate the stellar seds ( @xcite ) . the uncertainties associated with the radiative transfer model are due to the use of observationally determined values of the physical properties of dust grains . the 10 - 15% uncertainty quoted above for the starburst model is a reasonable assessment of these uncertainties . in fitting the model seds to the observed sed , there are 4 parameters which are derived from the observed sed ( age , mass , @xmath81 , and @xmath78 ) . thus , @xmath91 . in our fitting algorithm , all of the model seds were computed for the input distance , reddened by the foreground dust , and then shifted to the rest frame of m33 . the model seds were fitted to the photometric data given in table [ tab_obs_sed ] . the model fluxes in the observed photometric bands , @xmath89 , were calculated by convolving the band transmission curves ( @xcite ; @xcite ; @xcite ; @xcite ; m. bessell 1998 , private communication ) with the full resolution model seds . we examined all model seds which had @xmath92 as this includes all model seds which fit with a @xmath931% significance , i.e. the probability of the correct model getting a higher @xmath82 [ @xmath94 is only 1% . these model seds were grouped into two sets , ( a ) those with @xmath270 myr old burst of star formation and ( b ) those with @xmath21 gyr of constant star formation . for both sets , no models with a dusty geometry fit the observations . the best fit model parameters for set a and b are given in table [ tab_best_fit ] . the uncertainties on these parameters were determine by the 99% confidence level as described above . the observed photometric sed of the m33 nucleus is plotted in figure [ fig_obs_sed ] along with the best fit model seds in from both set a and b. the set a models fit the photometric sed better than the set b models according to the @xmath82 values ( table [ tab_best_fit ] ) . this can be seen in figure [ fig_obs_sed ] where the best fit set b model sed is consistently too high in the far - uv , too low in the optical , and too high in the ir . lccc star formation type & burst & burst & constant geometry & shell & shell & shell dust type & milky way & milky way & milky way age [ years ] & @xmath95 & @xmath96 & @xmath97 mass [ @xmath98 & @xmath99 & @xmath100 & @xmath101 @xmath78 & @xmath102 & @xmath102 & @xmath103 @xmath81 & @xmath104 & @xmath105 & @xmath106 @xmath107 & @xmath108 & @xmath109 & @xmath110 @xmath111 [ @xmath112 & @xmath113 & @xmath114 & @xmath115 @xmath82 & @xmath116 & @xmath117 & @xmath118 the validity of sets a and b can be tested further with the spectroscopic observations . the uv , optical , and ir spectra of the m33 nucleus are plotted in figures [ fig_iue]-[fig_ir_spectra ] . as the spectroscopic data were taken with different apertures , it is not possible to fit simultaneously the uv to near - ir spectroscopy of the m33 nucleus . in addition , the use of relatively large apertures results in the flux from the nucleus being diluted by flux from the surrounding region . therefore , we have used the photometric data along with the spectra for the comparisons with the best fit model seds . figure [ fig_iue ] plots the iue spectrum along with the photometric sed and the best fit model seds from sets a and b. due to the faintness of the m33 nucleus , the s / n in the iue spectrum is quite low . we display the spectrum with 20 bins to increase the s / n and match the resolution of the model sed . the iue spectrum obviously includes flux from both the nucleus and the surrounding region since the photometric sed is always below the iue spectrum . the shape of the iue spectrum is flatter than the photometric and model seds , implying that the uv spectrum of the region surrounding the nucleus has a much smaller depression at 2175 and , therefore , is attenuated by less dust than the nucleus . this behavior arises from mixing flux from regions which have differing amounts of dust : the least attenuated region will contribute disproportionately to the flux , washing out the signatures of dust . the optical spectrum of the m33 nucleus is plotted in figure [ fig_optical ] along with the best fit model seds from sets a and b. unlike the iue spectrum , the optical spectrum is roughly consistent with the photometric sed . while the steeper slope makes the observed spectrum slightly bluer than all of the model seds , the strengths of the features are comparable , with some notable exceptions . for example , the observed strength of the @xmath1195889 ( 1.8 ) can be traced to the presence of a large amount of interstellar material . this is consistent with our detection of a shell of dust with @xmath9 enshrouding the nucleus . both the differing line strengths and the bluer color could arise from inclusion of flux from the region surrounding the nucleus as the observed spectrum includes light from a @xmath120 region . the general agreement with the nuclear photometric levels would still be consistent if some of the nuclear flux spilled out of the slit ( the seeing was @xmath2@xmath56 ) . the optical spectrum shows no significant emission in h@xmath76 or any other line . set b models predict h@xmath76 in emission with an equivalent width of @xmath2100 which would easily have been seen in the observed spectrum . the ses models predict the strength of h@xmath76 , but do not add them to their output seds . thus , no emission lines are seen in figure [ fig_optical ] for the set b model sed even through they would be very prominent . in comparison , the set a models predict no detectable h@xmath76 emission , in good agreement with the observations . the infrared spectra are plotted in figure [ fig_ir_spectra ] . all the models do a fairly good job of fitting them , as expected since all the model seds fit the photometric sed . most of the observed absorption lines are present in the model seds , but not necessarily at the same strength . the model seds have much weaker brackett lines than observed ( ( 4 - 12 ) and ( 4 - 7 ) are good examples , but we also detect ( 4 - 10 ) ) , and they do not show the @xmath63co lines . furthermore , the @xmath57co bands are somewhat weak in the models , both at the first overtone near @xmath121 and the second near @xmath122 . since the inputs to the infrared models have greater uncertainty and are less complete than those for the optical , it is not surprising that this is also the region where the models are most deviant . there is only one infrared spectral feature that differs significantly between the set a and b models , the broad @xmath122 h@xmath123 peak is seen in stars with @xmath124 ( @xcite ) . this feature is not seen in the observations ( figure [ fig_obs_sed ] and [ fig_ir_spectra]b ) , which suggests that the set a models better describe the m33 nucleus than the set b models . the set b models reproduce the slope of the observed short k band spectrum ( figure [ fig_ir_spectra]c ) , but given the difficulties in determining general slopes in the data , this disagreement is probably not significant . because of its sensitivity to the stellar luminosity , the co band strength has proven useful in determining the age of starbursts . this parameter is commonly expressed as a `` co index '' based on the average depth of the feature in intermediate width photometric bands ( @xcite ) . we have used the formulation of doyon , joseph , & wright ( 1994 ) to compute co indices for our observations of m33 and for the l98 and pegase synthetic spectra . for m33 , the index is approximately 0.21 . the synthetic spectra show obvious shortcomings in both cases , such as discontinuous changes in index strength for the pegase models and an index that never reaches the highest observed strengths for l98 . for the starburst models of @xcite , the co index of the individual stellar types was carefully calibrated empirically against the extensive available photometric observations . the run of co index with starburst age for these models ( @xcite ) does not have the problems we find in the synthetic spectra , and we prefer it for comparison with m33 . there are two cases that represent a relatively short burst , either a star formation rate gaussian in time with a full width at half maximum of 5 myr , or a burst that starts abruptly and decays with an exponential time constant of 20 myr . these cases produce an index of 0.21 after 50 to 70 myr , in excellent agreement with the parameters of the burst we have derived from the spectral synthesis in the optical and ultraviolet . assuming the rotation curve of m33 is flat within @xmath32 , kormendy & mcclure ( 1993 ) calculate an upper limit of @xmath125 on the mass of the nucleus . lauer et al . ( 1998 ) calculate a lower limit by determining a core mass of @xmath126 . a better estimate of the mass in the starburst region should be possible from the measured velocity dispersion ( 21 km s@xmath47 , kormendy & mcclure 1990 ) and the size of the starburst . we measured the fwhm of the pc images of the m33 nucleus by fitting them with 2-dimensional gaussian convolved with the tinytim point spread function ( psf ) appropriate for each filter ( @xcite ) . it was not possible to fit the nucleus in the f160bw filter due to its low signal . the results are given in table [ tab_nuc_size ] . lauer et al . ( 1998 ) found the fwhm was @xmath127 ( 0.27 pc ) in the f555w image using a different analytic form for the nuclear profile and taking better care to correct for the effects of the psf and aliasing . the increasing size of the nucleus as a function of wavelength confirms the color gradient measured by kormendy & mcclure ( 1993 ) and lauer et al . ( 1998 ) . lccc f300w & 2.65 & 0.12 & 0.46 f555w & 4.33 & 0.20 & 0.77 f814w & 4.62 & 0.21 & 0.81 f1042 m & 5.58 & 0.25 & 0.98 because they dominate the light , the observed velocity dispersion and size reflect only properties of the more massive stars . this is confirmed by the composite spectral type of the nucleus , for which a range from late - a to early - f main sequence has been reported ( @xcite ; @xcite ; @xcite ; b. garrison , private communication ) . assuming equipartition in kinetic energies , the true velocity dispersion ( @xmath128 ) of the nucleus is then @xmath129 where @xmath130 is the observed velocity dispersion , @xmath131 is the average mass of the stars in the nucleus , and @xmath132 is the average mass of the stars contributing to @xmath130 . assuming the imf given in equation [ eq_imf ] , @xmath133 . from the spectral type of the nucleus , @xmath134 . thus , the true velocity dispersion of the nucleus is 40 km s@xmath47 . the virial mass is computed from @xmath135 where r is the effective radius of the nucleus in pc , @xmath128 is in km s@xmath47 , and @xmath136 is in @xmath137 . estimating @xmath138 for the nucleus is nontrivial and so we will use a measured @xmath138 value and compute a lower limit . we assume @xmath138 to be the size of the nucleus as measured in the f814w image ( hwhm @xmath139 pc ) since @xmath130 was measured from the infrared triplet ( @xcite ) . the mass of the nucleus is then @xmath140 . this is similar to the @xmath141 of the best fit set a models . in addition to the starburst population , it is likely there is at least one other stellar population in the nuclear region . the presence of an older ( @xmath21 gyr ) underlying stellar population with a large @xmath111 is strongly implied by the growth of surface brightness and stellar counts approaching the nuclear region ( @xcite ) and the mass ( @xmath142 ) of the inner @xmath32 of m33 ( @xcite ) . within 4@xmath45 , 3/4 of the observed v band flux comes from the nucleus ( @xcite ) . this implies that the other stellar population in the nuclear region has a @xmath111 ratio of 5.5 assuming the set a best fits or 1.8 assuming the set b best fit . thus , the mass required by the set a models ( @xmath143 ) is more consistant with the available data on the nuclear region . the photometric and spectroscopic evidence indicate that set a models fit the observations of the m33 nucleus better than the set b models , as does the mass determination . the set a models are consistent with the small size of the nucleus and flat profile of the inner disk of m33 . the flat profile implies there is a lack of a substantial density gradient which could funnel gas to the nucleus ( @xcite ) , which means fueling continuous star formation ( as set b fits would require ) would be difficult . the stellar population of the m33 nucleus is described by a 35 to 180 myr burst of star formation with a mass of 0.4 - 1.1 @xmath144 ( table [ tab_best_fit ] ) . the best fit seds in set a for both ses models give very similar results . they imply the best description of the m33 nucleus is a burst of star formation 70 - 75 myrs old and a mass of @xmath20.7 @xmath144 ( table [ tab_best_fit ] ) . our finding that the m33 nucleus is well described by a single burst of star formation enshrouded by a significant amount of dust is quite different than previous studies ( @xcite ; @xcite ; @xcite ) . in these studies , the role of dust was underestimated and , as a result , luminous young and old stellar populations were both needed to reproduce the relatively red ( flat ) spectrum of the m33 nucleus . for example , ciani et al . ( 1984 ) modeled the iue spectrum plus @xmath145 , @xmath146 , and @xmath147 photometry of the m33 nucleus . they found that the data were best represented by two stellar populations , one extremely young ( @xmath148 years ) and one old ( @xmath149 years ) . they argued that the effects of foreground and internal dust were well represented by a screen of mw - type dust with an @xmath150 . schmidt et al . ( 1990 ) obtained a long - slit optical spectrum of the m33 nucleus with a @xmath151 slit and modeled it using a combination of star clusters of different ages . they found that the m33 nucleus was composed of 7 star clusters with ages ranging from @xmath152 to @xmath153 years . this is similar to other studies where over 50% of the v band light was claimed to be from stars with ages @xmath154 years . in addition , schmidt et al.(1990 ) modeled the effects of dust as due to a galactic foreground screen with @xmath155 and a m33 internal screen with @xmath156 . the mass - to - light ratio ( @xmath111 ) we compute from our model fits , corrected for foreground and internal dust , is @xmath20.1 . this is significantly different from previous studies , which found @xmath157 23 ( @xcite ; @xcite ) which is in excess of the value likely to be present from dynamical considerations . by including dust properly , we have arrived at a simpler model of the m33 nucleus which reproduces its observed sed from the uv to the near - ir . we can estimate the peak luminosity of the m33 starburst , although there is some uncertainty because our models do not constrain the rate of star formation with time uniquely . for example , the starburst mass in m33 is roughly 500 times less than that in m82 , assuming similar imfs ( e.g. , imf8 of @xcite ) and the somewhat extended burst which best fits the m82 data . thus , at an age of @xmath210 million years , the m33 starburst would have luminosity of 0.002 times that of m82 , or @xmath158 , alternately , from our abrupt burst model and the current luminosity of @xmath159 , the luminosity at 10 myrs would have been @xmath160 . the center of the milky way contains a very compact ( @xmath161 pc ) cluster of young ( @xmath162 myrs old ) stars with an integrated luminosity of @xmath2@xmath163 ( @xcite ; @xcite ) . the analogy between these two cases is striking . it appears that in m33 we may well be seeing a later development of an event virtually identical to the starburst that currently dominates the galactic center . the relatively young age of the nucleus that we derive from our models may clear up another mystery . the m33 nucleus has a very strong x - ray source with a @xmath164 ergs s@xmath47 ( @xcite ; @xcite ) . the detection of a 106 day period requires that this object be a single source ( @xcite ) . the only known type of object which could reproduce both observed effects is a stellar mass black hole ( @xmath2@xmath165 ) accreting from a companion star . this type of object was mentioned by dubus et al . ( 1997 ) , but discarded as the solution as previous work on the m33 nucleus argued against a significant population of o b stars . the requirements of such an x - ray binary are a black hole with a companion star with enough mass to withstand the supernova blast that created the black hole without being ejected from the binary . for example , a black hole with an early b giant companion would be transfering a significant amount of matter to the black hole , resulting in the observed x - ray luminosity . the type , geometry , and amount of dust associated with the m33 nucleus is fairly well constrained by the observed photometric sed . the dust is best described by a shell of somewhat clumpy mw - like dust ( @xmath166 ) with a radial v - band optical depth of @xmath22 . to show the effects of both foreground and internal dust on the intrinsic sed of the m33 nucleus , we present figure [ fig_dust_ex ] . even in the galactic foreground dust corrected sed , the flux at 2607 ( uit - a1 ) is significantly lower than the flux in the bands around 1600 ( uit - b1 , f160bw , and f170w ) and around 3300 ( f300w , f336w , u ) . the only known way to get such a depression in the sed of a stellar population is to attenuate its sed by dust with a 2175 feature , i.e. mw - type dust . it is interesting to compare this with the result found by gordon e al . they found that smc - like dust was responsible for the attenuation in the 30 starburst galaxies in their sample . they used iue - selected starburst galaxies , which biased their study towards intrinsically bright , lightly attenuated starburst regions . the interstellar environment near an intense starburst is very harsh , associated with an intense uv radiation density and supernovae shocks . there is evidence from the observed dust extinction curves in the magellanic clouds which suggests that star formation modifies nearby dust . in the smc , the only extinction curve outside the star forming bar is very similar to that found in the milky way . the other three extinction curves in the bar are the most extreme known in terms of a non - existent 2175 bump and strong far - uv rise ( @xcite ) . in the lmc , there are two different extinction curves ( @xcite ) . the extinction curve associated with stars near the lmc 2 supergiant shell ( which is itself near the 30 dor star formation region ) has a weak 2175 bump . the extinction curve associated with stars elsewhere in the lmc is similar to that found in the milky way . however , the 30 dor region in the lmc exhibits much more intense star formation than any region in the smc , but the most extreme extinction curves are found in the smc . this fact implies that the relationship between star formation activity and the modification of nearby dust is not a simple one . recent theoretical models of dust exposed to shocks ( @xcite ; @xcite ) illustrate the complicated behavior expected for dust near active star formation . the number of small particles in shocked dust is increased through shattering of larger grains and results in an increased extinction in the far - uv . however , the models which produce a stronger far - uv rise , also produce normal or strong 2175 bumps . this is not what is seen in the behavior of dust in starburst galaxies or the magellanic clouds and probably reflects our incomplete understanding of the carrier of the 2175 feature . the m33 nucleus was also observed by iue , but is not typical of the iue sample used by gordon et al . m33 is much closer than the majority of the iue sample ( 0.795 versus 60 mpc ) and its faintness implies it is intrinsically fainter and/or suffers a higher attenuation than the rest of the sample . in the case of m33 , the starburst is surrounded by a larger optical depth of dust and , since it lacks emission lines , is older than those seen in the iue sample . thus , the dust in the m33 nucleus has likely undergone less processing since it is near a small starburst and self - shielded by a large amount of dust . the amount by which dust can be processed in starburst regions is likely related to the mass and age of the starburst as well as the extent to which the dust can shield itself . additional work on other fairly reddened starburst regions is needed to test this interpretation of the origin of the mw - type dust we find in the m33 nucleus . the low velocity dispersion and extreme compactness of the m33 nucleus have caused lauer et al . ( 1998 ) to suggest that stellar mergers might play an important role in creating the blue stellar population ( see also @xcite ) . these arguments are based on a calculation that a typical star lying within the present - day environment of the m33 nucleus would have a collision / binary capture timescale of about a hubble time . this model could apply if the previous estimate of an old ( @xmath167 yr ) stellar population in the nucleus of m33 is assumed ( @xcite ; @xcite ) . that is , if conditions in the nucleus have not changed over a hubble time , most stars in the nucleus will have undergone such an event . to assess if collisions and mergers are important , we will assume the age of the nucleus is on the order of the hubble time and see if reasonable arguments lead to the observed characteristics of the m33 nucleus . the optical spectrum of m33 ( figure [ fig_id_opt ] ) demonstrates that the nucleus contains many stars of @xmath168 . the observed nuclear b magnitude is 14.6 . corrected for foreground and internal dust , the intrinsic b magnitude is 12.5 , corresponding to a m@xmath169 ( for a distance of 795 kpc ) . for the purposes of this exercise , we attribute most of the nuclear mass of @xmath170 to an old stellar population that creates more massive stars by mergers . the absolute blue magnitude would require the presence of @xmath171 main sequence stars with masses @xmath23 @xmath172 , if these stars were the most massive in the nucleus . this hypothetical population accounts for an uncomfortably large portion of the total nuclear mass . instead , it seems required that the light from the nucleus is dominated by stars closer to the top of the permitted mass range ( @xmath173 ) . we assume the old stellar population is described by an imf similar to the one taken for the starburst , up to a main sequence turnoff mass of @xmath174 . allowing for the mass lost in mergers , stars at the turnoff must undergo on the order of six to eight mergers to build a @xmath175 star . thus , the calculation by lauer et al . ( 1998 ) that , on average , each star might undergo one merger in a hubble time implies that the dominant stars could not be created in this fashion . a more rigorous argument would account for the dynamical evolution of the nucleus . if the age of the m33 nucleus is on the order of the hubble time , then it is possible that it has gone through core collapse ( @xcite ; @xcite ; @xcite ) . lee ( 1987 ) has studied stellar mergers in dynamical models of systems of @xmath176 stars which , in their post core collapse configuration , resemble the m33 nucleus . the merger rate reaches a sharp peak around core collapse . stars that are the result of eight mergers , as required to build the blue stars in the m33 nucleus , are @xmath177 less common than the original population members shortly after core collapse in lee s ( 1987 ) model . we have roughly corrected this result for the approximately three times higher number of seed stars in m33 compared with the modeled cluster . in addition , lee ( 1987 ) made the optimistic assumptions that all binary captures lead to mergers ( rather than the expected @xmath225% ; @xcite ) and that the merged stars have main sequence lifetimes similar to normal stars ( whereas the lifetimes are shortened considerably by the presence of enriched nuclear material ; e.g. , @xcite ) . we project that mergers would yield at best an order of magnitude less than the numbers of massive stars required . this conclusion agrees with the statement by hernquist et al . ( 1991 ) . the merger hypothesis is attractive primarily because of the very small observed core radius in m33 . however , the observations are dominated by the relatively massive stars ; hence we have to examine the time scale for mass segregation to be sure that the radius measured is representative of the distribution of lower mass stars that hypothetically merge due to their high density in this core . we have estimated the dynamical decay time for stars of @xmath178 in a total mass of @xmath179 distributed over a radius of 0.5 pc , assuming that the density goes as @xmath180 , @xmath181 , or is constant within this radius . the calculations are based on equation ( 7 - 18 ) of binney & tremaine ( 1987 ) . we find that the dynamical decay time from formation at a radius of 0.5 pc is in all three cases substantially less than the lifetimes of the stars ; that is , mass segregation is likely to have occurred . the increasing size of the nucleus with wavelength ( see table [ tab_nuc_size ] ) and corresponding observed color gradients ( @xcite ; @xcite ) confirm that the blue , massive stars have probably sunk to the center and are surrounded by a population of lower mass stars . thus , estimates of the merger rate based on the small observed core radius ( e.g. , @xcite and our discussion in the preceding paragraphs ) are likely to be too high . the merger hypothesis can also be tested by examining the properties of the merger products . these products have huge angular momentum compared with normal stars ( e.g. , @xcite ) . if the interiors become convective , the resulting high rotation rates may be spun down through magnetic interaction with circumstellar disks ( @xcite ) . however , the resulting mixing of metal rich material causes the evolution of the stars to differ significantly from that of normal stars ( e.g. , @xcite ; @xcite ) . the possibility of abnormal stellar evolution is made very unlikely by the success in our spectral synthesis evolutionary models in fitting the properties of the nucleus across the entire spectrum , since these models assume normal evolution . the possibility that the nuclear population is characterized by abnormally high rotational velocities is contradicted by the small net velocity dispersion . if the merger mechanism played an important role in m33 , then the conclusions drawn from starburst models would largely be vitiated . however , a variety of arguments show that mergers play a negligible role and hence that the blue stellar population must arise in a starburst as described in section [ sec_stell_pop ] . by fitting the uv to near - ir sed of the m33 nucleus with a starburst model which includes stars , gas , and dust , we find the m33 nucleus to be composed of a @xmath270 myr burst of star formation with a mass of @xmath2@xmath182 enshrouded by a shell of mw - type dust ( @xmath9 ) . this result is different from previous work which modeled the nucleus using both young ( @xmath6 years ) and old ( @xmath183 years ) stellar populations subject only to modest reddening . by including the effects of dust internal to m33 , we are able to reproduce the relatively red sed of the m33 nucleus without requiring a luminous old stellar population . as a result , this work is more consistent with the small size and the detection of far - uv flux from the nucleus . the dust in m33 has a strong 2175 bump , in contrast to other uv - bright starbursts that do not show this feature . not much can be said with a sample of one galaxy , but it is possible that the different dust characteristics in m33 nucleus are related to the higher optical depth of dust and/or the age of the starburst . thus , there is not a single type of dust found in starburst galaxies but a range of types varying from that found in the mw to the smc . correcting for the effects of dust in starburst galaxies must be done with care , especially in the uv . the m33 nuclear starburst resembles the ultracompact population of very massive and luminous stars in the center of the milky way , both in the peak luminosity deduced for the starbursts responsible and in the size of the starburst region . it seems likely that ultracompact starbursts ( @xmath184 pc ) are common in the nuclei of normal galaxies and may influence the evolution of these regions significantly . confirmation of our interpretation of the m33 nucleus would be greatly helped by additional uv and far - ir observations . a good uv spectrum of just the nucleus would be easily obtained by the stis instrument on hst due to it s high sensitivity and small apertures . far - ir observations at high spatial resolution would help put stringent constraints on the mass of the dust internal to the nucleus . these observations should be possible with sirtf . we thank the anonymous referee for valuable suggestions which improved the presentation of our work . this work benefited from discussions about ses models with claus leitherer , michel fioc , and brigette rocca - volmerange . in addition , we are indebted to claus leitherer and michel fioc for providing us with new grids of seds from the l98 and pegase ses models . we thank juhan frank for helpful discussion on the x - ray properties of binaries and the intricacies of cluster dynamics . support for this work was provided by nasa through grant number [ ar-08002.01 - 96a ] from the space telescope science institute . additional support was provided by the nsf under grant ast95 - 29190 . mmh received support for this work provided by nasa through hubble fellowship grant # hf-1072.01 - 94a awarded by the space telescope science institute . bailyn , c. d. , & pinsonneault , m. h. 1995 , , 439 , 705 bessell , m. s. 1990 , , 102 , 1181 bessell , m. s. & brett , j. m. 1988 , , 100 , 1134 bessell , m. s. , castelli , f. , & plez , b. 1998 , , 333 , 231 bica , e. & alloin d. 1986 , , 162 , 21 bica , e. & alloin d. 1987 , , 186 , 49 binney , j. , & tremaine , s. 1987 , `` galactic dynamics '' ( princeton univ . press : princeton , n. j. ) , 425 biretta et al . 1996 , wfpc2 instrument handbook , version 4.0 ( baltimore : stsci ) ciani , a. , dodorico , s. , & benvenuti , p. 1984 , , 137 , 223 calzetti , d. , kinney , a. l. , & storchi - bergmann , t. 1994 , , 429 , 582 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 cardelli , j. a. & savage , b. d. 1988 , , 325 , 864 doyon , r. , joseph , r. d. , & wright , g. s. 1994 , , 421 , 101 dubus , g. , charles , p. a. , long , k. s. , & hakala , p. j. 1997 , , 490 , l47 fioc , m. & rocca - volmerange , b. 1997 , , 326 , 950 fioc , m. & rocca - volmerange , b. 1998 , in preparation frogel , j. a. , persson , s. e. , aaronson , m. , matthews , k. 1978 , , 220 , 75 garnett , d. r. , shields , g. a. , skillman , e. d. , sagan , s. p. , & dufour , r. j. 1997 , , 489 , 63 gallagher , j. s. , goad , j. w. , & mould , j. 1982 , , 263 , 101 gordon , k. d. , calzetti , d. , & witt , a. n. 1997 , , 487 , 625 gordon , k. d. & clayton , g. c. 1998 , , 500 , 816 gordon , k. d. , witt , a. n. , & friedmann , b. c. 1998 , , 498 , 522 hernquist , l. , hut , p. , & kormendy , j. 1991 , nature , 354 , 376 holtzman , j. , et al . 1995 , , 107 , 156 hunt , l. k. , mannucci , f. , testi , l. , migliorini , s. , stanga , r. m. , baffa , c. , lisi , f. , & vanzi , l. 1998 , , 115 , 2594 hunter , d. a. , light , r. m. , holtzman , j. a. , lynds , r. , oneil , e. j. , jr . , & grillmair , c. j. 1997 , , 478 , 124 jones , a. p. , tielens , a. g. g. m. , & hollenbach , d. j. 1996 , , 469 , 740 krist , j. & hook , r. 1997 , the tiny tim user s guide , version 4.4 ( baltimore : stsci ) kormendy , j. & mcclure , r. d. 1993 , , 105 , 1793 lanon , a. & rocca - volmerange , b. 1992 , , 96 , 593 lauer , t. r. , faber , s. m. , ajhar , e. a. , grillmair , c. j. , & scowen , p. a. 1998 , , 116 , 2263 lee , h. m. 1987 , , 319 , 801 lee , h. m. , & ostriker , j. p. 1986 , , 310 , 176 leitherer , c. & heckman , t. m. 1995 , , 96 , 9 leitherer , c. et al . 1998 , , submitted lejeune , th . , cuisinier , f. , & buser , r. 1997 , , 125 , 229 leonard , p. j. t. , & livio , m. 1995 , , 447 , l121 massey , p. , bianchi , l. , hutchings , j. b. , & stecher , t. p. 1996 , , 469 , 629 mcquade , k. , calzetti , d. , & kinney , a. l. 1995 , , 97 , 331 minniti , d. , olszewski , e. w. , & rieke , m. 1993 , , 410 , l79 misselt , k. a. , clayton , g. c. , & gordon , k. d. 1998 , , in press monteverde , m. i. , herrero , a. , lennon , d. j. , & kurdritzki , r .- 1997 , , l107 morgan , w. w. & mayall , n. u. 1957 , , 69 , 291 nichols , j. s. , garhart , m. p. , de la pea , m. d. , & levay , k. l. 1994 , new spectral image processing system information manual : low dispersion data - version 1.0 , _ iue _ newsletter no . 53 oconnell , r. w. 1983 , , 267 , 80 odonnell , j. e. & mathis , j. s. 1997 , , 479 , 806 rasio , f. a. , & shapiro , s. l. 1995 , , 438 , 887 rieke , g. h. & lebofsky , m. j. 1982 , in `` the galactic center '' ( am . inst . of phys . : new york ) , 194 rieke , g. h. , loken , k. , rieke , m. j. , & tamblyn , p. 1993 , , 412 , 99 salpeter , e. e. 1955 , , 121 , 161 scalo , j. 1998 , proceedings of herstmonceux conference on imf , in press schmidt , a. a. , bica , e. , & alloin , d. 1990 , , 243 , 620 schulman , e. & bergman , j. n. 1995 , , 441 , 568 shier , l. m. , rieke m. j. , & rieke , g. h. 1996 , , 470 , 222 sills , a. , lombardi , j. c. , bailyn , c. d. , demarque , p. , rasio , f. a. , & shapiro , s. l. 1997 , , 487 , 290 stecher , t. p. , et al . 1997 , , 109 , 584 taylor , j. r. 1982 , an introduction to error analysis ( mill valley , ca : univ . science books ) trinchieri , g. , fabbiano , g. , & peres , g. 1988 , , 325 , 531 van den bergh , s. 1991 , , 103 , 609 voit , m. , et al . 1997 , hst data handbook , version 3.0 ( baltimore : stsci ) wainscoat , r. j. & cowie , l. l. 1992 , , 103 , 332 werner , m. w. & davidson , j. a. 1989 , in `` the center of the galaxy '' ed . m. morris ( kluwer : dordrecht ) , 423 whitmore , b. , heyer , i. , baggett , s. 1997 , instrument science report wfpc2 96 - 4 ( baltimore : stsci ) williams , d. m. , thompson , c. l. , rieke , g. h. & montgomery , e. f. 1993 , , 1946 , 482 witt , a. n. 1977 , , 35 , 1 witt , a. n. & gordon , k. d. 1996 , , 463 , 681 witt , a. n. & gordon , k. d. 1998 , in preparation witt , a. n. , thronson , h. a. , & capuano , j. m. 1992 , , 393 , 611 zaritsky , d. , elston , r. , & hill , j. m. 1989 , , 97 , 97

entanglement being a uniquely quantum attribute , to gain a better understanding of the nature of macroscopic quantum phenomena we explore in this paper ways to qualify and quantify the quantum entanglement between two macroscopic objects by way of model studies . knowing that a macroscopic object is a composite , how does one determine in terms of the entanglements between its micro - constituents ? we assert that the notion of ` levels of structure ' , the coupling strength between constituents in different levels , and the use of collective variables to capture the contribution of each level are all pertinent considerations . to understand how the coupling pattern amongst the constituents of the two macro objects enters into the picture , we consider two types of coupling , each constituent particle is coupled to only one other particle ( 1-to-1 ) versus coupled to all particles ( 1-to - all ) . in the 1 - 1 case with pairwise interactions of equal strength , the entanglement is independent of the number of constituent particles in the macroscopic object . in the 1-to - all case the relative coordinates are decoupled and the center of mass ( com ) coupling scales with . here we expect the entanglement between the com variables to increase with increasing size of the macroscopic objects and survive at higher temperatures . for a quantum many body system containing constituents , we provide a proof of the conditions whereby the com variable decouples , a cause for the special role the com variable plays in the entanglement between two such macroscopic objects . similar qualitative behavior is found when the couplings between the constituents of the macroscopic objects are statistically independent gaussian random variables . in another model study we analyzed the entanglement pattern of 4 coupled oscillators in two pairs , representing the two objects a and b ( or two adjacent levels of structure ) , each with two constituents . by assigning different coupling strengths we can investigate the interplay of inter - level entanglement with intra- level interactions representing tight or weak binding of the constituents within one level of structure . from the entanglement dynamics of the 4-oscillator system with varying coupling strength we see the entanglement between constituents meeting sudden death while the com variables may sustain over longer times . this offers another way to determine under what conditions quantum entanglement between macroscopic objects can persist . we explore the implications of our findings for quantifying the entanglement measure for macroscopic objects which is different from that of the entanglement entropy dependent on the area of a partition for a quantum many body system . finally we mention how the model studies here bear on quantifying the entanglement in physical systems such of nucleons in a nucleus , layers of graphene and between dna macromolecules . macroscopic quantum phenomena , quantum entanglement , open quantum systems , many - body theory

1308.4225

macroscopic quantum phenomena ( mqp ) refer to quantum features in objects of ` large ' sizes , systems with many components or degrees of freedom , organized in some ways where they can be identified as macroscopic objects . what makes mqp interesting is because in the traditional view ` quantum ' belongs to the microscopic domain , in the sense that only small objects necessitate a quantum description , while classical mechanics , a limiting case of quantum mechanics , is sufficient for the description of the macro world . in the face of new challenges from macroscopic quantum phenomena , viz , quantum features may show up even at macroscopic scales , this common belief now requires a much closer scrutiny , involving possible reformulations and/or re - interpretations . in addition to the worthy effort invested in the last two decades to gain a better understanding of the quantum - classical relation , as was the aim of quantum decoherence studies , and what could be viewed as characterizing the ` quantumness ' of a system , such as the existence of quantum entanglement in the system , we note that faced with the challenge of mqp , even a naive and seemingly unequivocal notion , like what is meant by ` macroscopic ' , need be re - considered . this emerging field is ushered in by several categories of definitive experiments . a common example of mqp is superconductivity , where the cooper pairs can extend to very large scales compared to interatomic distances , and bose - einstein condensate ( bec ) , where a large number n of atoms occupy the same quantum state , the n - body ground state . other important examples are in nanoelectromechanical devices @xcite , where the center of mass of a macroscopic object , the cantilever , is seen to obey a quantum mechanical equation of motion . experimental proposals to detect the superposition between a mirror and the quantum field , and between two mirrors , have been proposed @xcite while the interference pattern formed when a large object composed of @xmath4 molecules passing through two slits have been observed @xcite . likewise for experiments in quantum optomechanics , see e.g. , @xcite . by contrast , this new field , which is rich in open issues at the foundation of quantum and statistical physics , remains little explored theoretically ( with the important exception of the work of leggett @xcite ) . our attitude on this issue is not to give up quantum mechanics at a certain meso or macro scale , but to examine carefully the conditions whereby quantum features may survive at such scales , or if they do nt , we want to know how and why ? there is also a subtle difference between the term ` phenomena ' in mqp and macroscopic quantum mechanics ( mqm ) @xcite . as we see it , mqp explores the quantum features in macroscopic systems based on quantum mechanics whereas mqm conjures up the possibility that quantum mechanics as a well - tested theory for microscopic objects may either yield different results , or need be replaced by a different theory , at the macroscopic level . in two recent essays @xcite we have suggested two pathways toward understanding mqp . the first concerns what macroscopic means large size ? large number of components ? for this we used the @xmath5 quantum mechanical model to examine what exactly does the leading order and next - to - leading order large n behaviour of this theory say at the large n limit does the system become classical , as often conjured ? we find even at the mean field level when a gaussian system obeys an equation with the same form as the classical equation of motion ( cf ehrenfest theorem ) , the system is not classical but remains quantum in nature . we reckon that there is no a priori good reason why quantum phenomena in macroscopic objects can not exist . the second pathway explores how quantum correlations and fluctuations impact on mqp . examples we used to examine this aspect include the running of coupling constants with energy and scale , the critical behavior of a system near phase transition , with the help of two - particle - irreducible ( 2pi ) effective action and effective dimensional reduction in the infrared regime , in the spirit of kaluza - klein theories . this perspective offers a dynamics - dependent and an interaction - sensitive definition of `` macroscopia '' . this paper explores a third avenue , from the quantum entanglement perspective . since , according to schrdinger @xcite this is an exclusively quantum feature , it should be a good measure of quantumness in a system . quantum entanglement has been applied extensively to qualify quantum phase transition ( at zero temperature ) @xcite . it is also claimed to persist at high temperatures @xcite and large scales @xcite under specific conditions and special domains . traditionally one expects classical behaviour at high temperatures and large scales , not quantum . to explore the entanglement between macroscopic objects , such as had been suggested in terms of mirror superposition @xcite , one needs to first ask two questions : 1 ) what constitutes a macro object ? we mentioned large size , large number . but how they are put together certainly makes a difference . what about the degree of complexity of its constituents ? what if the constituents are non - interacting versus interacting ? weakly interacting versus strongly interacting ? 2 ) which parties are being entangled ? rather , as in the quantum world in principle everything may be entangled with everything else , the essential question is , are there ` representative ' physical variables which capture the essential features and correctly depict the dynamics of a macroscopic object as a distinct entity ? for example , why is it that an ostensibly macroscopic object such as a cantilever should follow a quantum equation of motion ? what is the physical variable of a macroscopic object interacting with an environment which behaves quantum mechanically ? more specifically , why is it that its center of mass variable does and not other variables , at least not ostensibly ? this `` center - of - mass axiom '' is implicitly assumed in many descriptions of mqp but rarely justified . the conditions upon which this can be justified are explored in @xcite with the derivation of a master equation for a system of n coupled quantum oscillators ( nos ) weakly interacting with a finite temperature bath made of n quantum oscillators , in the form of the hpz master equation @xcite , extending earlier work for 2 coupled oscillators @xcite . ( a mathematically more vigorous and complete treatment of nos system is given in @xcite . ) the above dissection immediately reveals the importance of how the basic constituents of the objects interact and are organized in addition to the simple characterization of micro - meso - macro by their scales and total masses . the aim of this paper is to find out a way , amongst presumably many , to quantify quantum entanglement between macroscopic objects . to do so we need to first understand how different levels of structure in a macroscopic object contribute to , or partake of , the entanglement with another object of similar structures . for this purpose in sec . 2 we describe the level of structure concept ( see , e.g. , @xcite in a composite body , the importance of choosing the proper collective variables , of which the center of mass is the most directly representative , to capture its behavior and dynamics . we revisit the center of mass axiom shown in an earlier paper and apply it to address the present issue . specifically , for a system composed of n constituents ( particles or components ) , we want to see what kind of coupling or under what conditions the entanglement would scale with n. in sec . 3 we consider two types of coupling configurations , first ( fig . 1a ) , when any one constituent of object a is coupled to one specific constituent of object b ( called 1 - 1 or pairwise interaction ) . the second type ( fig 1b ) , is when each constituent of object a is coupled to all the constituents of object b ( called 1-to - all interaction ) . we will show that the latter type is when mqp would appear , as this interaction configuration scales with n. next , we provide another proof of the conditions whereby the com variable decouples in a quantum many body system containing @xmath3 constituents , and with this , the special role the com variable plays in the entanglement between the two such macroscopic objects . in sec . 4 we look more closely into the specific case of 4 coupled oscillators in two pairs , representing the two objects a ( constituents 1 , 2 ) and b ( constituents 3 , 4 ) . we assign different coupling strengths : @xmath6 as that between the two ( intra - level ) constituents 1 - 2 or 3 - 4 in each object , @xmath7 as the ( inter - level ) coupling strength between 1 - 3 ( and also 2 - 4 ) ( parallel ) , and @xmath8 between 1 - 4 ( and also 2 - 3 ) ( cross ) . considering @xmath6 to be large compared to @xmath7 or @xmath8 refers to strong intra- level interactions or tight binding of constituents within one object . this model can show the difference between the cases studied in sec . 3 , namely , the pairwise ( 1 - 3 , 2 - 4 ) interaction versus the one to all ( 1 - 3 and 1 - 4 ) interaction . as in sec . 3 we see that it is only in the one - to - all case that the entanglement scales with n , assuming that entanglement increases with coupling strength . we then show the entanglement dynamics of the 4-oscillator system with varying coupling strength , while distinguishing three cases of entanglement : the case between the center of mass variables ( com ) of a and b , the case between the com variable of one object a and one constituent particle of b , and the case between one constituent particle of a and one of b. we see some pariwise entanglement meets sudden death while others sustain over long time ranges . this offers another way to determine under what conditions quantum entanglement between macroscopic objects can persist . finally in sec . 5 we summarize the key findings of this paper , compare with earlier results , and discuss their implications for the existence of mqp . the model studies here can in principle be applied to a number of realistic experiments in cold atom and condensed matter systems . ` macroscopic ' conveys a sense of being ` large ' , but what exactly does ` largeness ' mean ? do all the basic constituents of a large object contribute equally towards its quantum feature ? ( this point is highlighted in footnote 2 of @xcite . ) in some cases we may actually know what the basic constituents are and how they are organized . a @xmath4 molecule is made of carbon atoms , each atom is made of nuclei and electrons , each nucleus contains a certain number of protons and neutrons , each of them in turn is made up of quarks and gluons . are we to simply count the number of quarks /gluons or protons /neutrons when we say an object is macroscopic ? obviously the tight binding of them to form a nucleus enters into our consideration when we treat the nucleus as a unit which maintains its own more or less distinct identity , features and dynamics . thus when one talks about the mesoscopic or macroscopic behavior of an object one needs to specify which level of structure is of special interest , and how important each level contributes to these characteristics . the coupling strength between constituents at each level of structure ( e.g. , inter - atomic ) compared to that structure s coupling with the adjacent and remaining levels ( which can be treated as an environment to this specific level of structure in an effective theory description , and its influence on it represented as some kind of noise @xcite ) will determine the relative weight of each level of structure s partaking of the macroscopic object s overall quantum behavior . often the best description of the behavior and dynamics of a particular level of structure is given by an effective theory for the judiciously chosen collective variables " . same consideration should enter when one looks for the quantumness " of an object , be it of meso or macro scale . one can quantize any linear system of whatever size , even macroscopic objects , such as sound waves from their vibrations . giving it a name which ends with an `` on '' such as phonon and crowning it into a quantum variable is almost frivolous compared to the task of identifying the correct level of structure and finding the underlying constituents the atoms in a lattice in this example , and their interactions . constructing the relevant collective variables which best capture the salient physics of interest should come before one considers their quantum features . thus , viewed in this perspective in terms of collective variables , we see that quantum features need not be restricted to microscopic objects . in fact ` micro ' is ultimately also a relative concept as new elementary " particles are discovered which make up the once regarded ` micro ' objects . we illustrate this idea first with a discussion of the relevance of the center of mass variable in capturing the quantum features of a macroscopic object , then in the following sections , with a description of quantum entanglement between two macroscopic objects . we can ask the question : what are the conditions upon which the mechanical and statistical mechanical properties of a macroscopic object can be described adequately in terms mainly of its center - of - mass ( com ) variable kinematics and dynamics , as captured by a master equation ( for the reduced density matrix , with the environmental variables integrated out ) . this is an implicit assumption made in many mqp investigations , namely , that the quantum mechanical behavior of a macroscopic object , like the nanoelectromechanical oscillator @xcite , a mirror @xcite , or a @xmath4 molecule @xcite , placed in interaction with an environment behavior such as quantum decoherence , fluctuations , dissipation and entanglement can be captured adequately by its com behavior . for convenience we refer to this as the ` com axiom ' . this assertion is intuitively reasonable , as one might expect it to be true from normal- mode decompositions familiar in classical mechanics , but when particles ( modeled by nho ) interact with each other ( such as in a quantum bound state problem ) in addition to interacting with their common environment , all expressed in terms of the reduced density matrix , it is not such a clear - cut result . at least we have not seen a proof of it . with the aim of assessing the validity of the com axiom the authors of @xcite considered a system modeled by @xmath2 harmonic oscillators interacting with an environment consisting of @xmath9 harmonic oscillators and derived an exact non - markovian master equation for such a system in a bath with arbitrary spectral density and temperature . the authors outlined a procedure to find a canonical transformation to transform from the individual coordinates @xmath10 to the collective coordinates @xmath11 where @xmath12 are the center - of - mass coordinate and momentum respectively . in fact they considered a more general type of coupling between the system and the environment in the form @xmath13 ( instead of the ordinarily assumed @xmath14 ) and examined if the com coordinate dynamics separates from the reduced variable dynamics . they noted that if the function @xmath15 has the property @xmath16 , for example @xmath17 or @xmath18 , one can split the coupling between the system and environment into couplings containing the com coordinate and the relative coordinates . tracing out the environmental degrees of freedom @xmath19 , one can easily get the influence action which characterizes the effect of the environment on the system . however , as the authors of @xcite emphasized , the coarse - graining made by tracing out the environmental variables @xmath19 does not necessarily lead to the separation of the com and the relative variables in the effective action . when they are mixed up and can no longer be written as the sum of these two contributions , the form of the master equation will be radically altered as it would contain both the relative variable and the center - of - mass variable dynamics . with these findings they conclude that for the n harmonic oscillators quantum brownian motion model , the coupling between the system and the environment need be bi - linear , in the form @xmath14 , for this axiom to hold . they also proved that the potential @xmath20 is independent of the center - of - mass coordinate . in that case , one can say that the quantum evolution of a macroscopic object in a general environment is completely described by the dynamics of the center - of - mass canonical variables ( @xmath21 ) obeying a master equation of the hpz @xcite type . what is the relevance of this finding to mqp ? within the limitations of the n harmonic oscillator model it conveys at least two points : 1 ) for certain types of coupling the center of mass ( com ) variable of an object composed of a large number of constituents does play a role in capturing the collective behavior of this object 2 ) otherwise , more generally , the environment - induced quantum statistical properties of the system such as decoherence and entanglement could be more complicated . ( for a similar conclusion considering the cross level ( of structure ) coarse - graining , see @xcite . ) we next investigate the quantum entanglement between two macroscopic objects , comparing the entanglement between the micro - variables of their constituents in two types of couplings : one - to - one and one - to - many . the very different natures between these two types serve to illustrate the relevance of how the micro - constituents organize into a macro object and how entanglement between collective variables reveals the quantum features of a macroscopic entity . we first apply the methods developed in @xcite to the study of the entanglement between the com coordinates of two macroscopic objects . each macroscopic object is modeled by @xmath2 identical coupled oscillators ( nos ) . however , unlike @xcite , we do not include an environment in our discussion because our focus is on the entanglement between the two objects induced by various types of direct interactions between their microscopic constituents . we denote the coordinates and the momenta of the microscopic constituents of the two macroscopic objects a and b by @xmath22 and @xmath23 respectively . the interactions between the microscopic constituents of one macroscopic object are assumed to be functions of the difference of variables only and we restrict ourselves to bilinear couplings between the microscopic constituents of the two macroscopic objects . the total hamiltonian is then given by : @xmath24 the canonical transformation described in the appendix a of @xcite can be applied to each object separately to yield a new set of phase space variables @xmath25 and @xmath26 and the associated masses @xmath27 . here @xmath28 and @xmath29 are the com coordinates . the hamiltonians of the macroscopic objects can be written in terms of these variables as : @xmath30 it has been proven in @xcite that the potential @xmath31 is not a function of the com variable . this is a consequence of the form assumed for the potential energy . for a general bilinear coupling characterized by @xmath32 the interaction hamiltonian @xmath33 can take on a complicated form , possibly mixing the com variables with the relative variables in what follows we will focus on two particular choices of @xmath32 , inspired by martins @xcite . the use of the new set of canonical variables which include the com will help interpret the behaviour of macroscopic entanglement . the pairwise interaction pattern is defined by @xmath34 ( see fig . 1(a ) ) . in other words one constituent particle modeled by an oscillator from object a couples to one oscillator from object b , assuming that all pairwise couplings have the same strength . using the canonical transformation of @xcite it can be shown that the interaction hamiltonian takes the form : @xmath35 note that pairwise interactions among the original variables translate into pairwise interactions among the transformed variables . a very important difference is that whereas the pairwise interactions in the original variables were all of equal strength , the strength of the interactions scale with the mass of the variables after the transformation . as a result the relative strength of interactions between variable pairs are the same for all the variables , including the com . to see this explicitly let us consider the case with @xmath36 for simplicity , namely the micro - constituents of each macroscopic object do not interact with each other . then we rescale the coordinates by @xmath37 and @xmath38 , after which the hamiltonian takes the form : @xmath39 in this form it is easy to see that the effective strength of interactions in the com variable is the same as the effective strength of interactions in all the other variables . hence the pairwise interaction pattern will induce the same amount of entanglement between pairs of transformed variables , without distinguishing the com variable . entanglement between non - com variables would be effected if the interactions among the oscillators within the same object , i.e. @xmath40 , are not set to zero . [ fig : pattern ] if we only focus on the effect of the pairwise interactions , it is fair to say that such interactions couple the pairwise transformed variables with equal effective strength independent of the size @xmath2 of the macroscopic objects . as a consequence we expect the behavior of entanglement between the corresponding variables of the objects to be independent of the size of the macroscopic objects , even for the com coordinate . for instance , at a given temperature the amount of entanglement between the two corresponding variables of the objects will not depend on @xmath2 . the critical temperature above which the entanglement ceases to exist also should not depend on @xmath2 . the one - to - all interaction pattern is characterized by @xmath41 ( see fig . then it is easy to see that the interaction hamiltonian in the transformed variables takes the form : @xmath42 note that interaction pattern corresponds to a coupling only between the com variables of the macroscopic objects , the relative variable hamiltonian is unaffected . thus pattern differs from the pairwise pattern in that it distinguishes the com variable . moreover if we perform the same rescaling of the previous section to determine the effective strength of this coupling we get : @xmath43 we see that the effective strength of the coupling increases with increasing @xmath2 for the pattern . thus in this case we expect the entanglement between the com variables to increase with increasing size of the macroscopic objects and survive at higher temperatures . the interaction pattern is crucial for the scaling of the entanglement of com variables with @xmath2 . hence it is important to investigate if this type of interaction pattern can occur in realistic situations and if so how common it is . in the discussion section we will present a different perspective in understanding the one - to - all coupling case , as characterizing different identical constituents at the same level of structure of one macro - object rather than between the two objects . in this section we derive the necessary and sufficient conditions for the com variables of two macroscopic objects to decouple from the rest of the degrees of freedom . the macroscopic objects are modeled by n coupled oscillators and interact via bilinear couplings . eqs.([h1v2 ] , [ h2v2 ] ) show that @xmath44 and @xmath45 obey the com axiom independently . below , we demonstrate under which conditions @xmath46 where @xmath47 is a function of the com coordinates of both macroscopic objects only and @xmath48 does not depend on the com coordinates of either object . specifically , we will derive the conditions under which @xmath49 to this end we follow the strategy adopted in appendix c of ref.@xcite and determine the functional form of @xmath33 by calculating its partial derivatives . @xmath50 where in the second equality we used @xmath51 for all @xmath52 , which can be shown by explicitly constructing a coordinate transformation whereby one coordinate is the com coordinate . for an example consider the construction described in the appendix b of @xcite or the explicit construction described in section [ sec : a ] of this paper , albeit with a different normalization for the com coordinate . since we want @xmath33 to have the form given by eq.([h_iform ] ) we require @xmath53 , which is independent of @xmath54 . repeating this derivation by replacing subscript @xmath55 with @xmath56 we obtain the second condition that @xmath57 , which is independent of @xmath52 . to summarize , the necessary and sufficient conditions for the com variables of both macroscopic objects to decouple from the relative coordinates is : @xmath58 if we do the same rescaling as in eq . we obtain : @xmath59 as a quick check it can be easily verified that both the `` pairwise '' and `` one - to - all '' couplings satisfy this condition with @xmath60 and @xmath61 respectively . these results agree with eq . and eq . . up to this point we have only discussed two different patterns of couplings in detail , the pairwise and one - to - all . by analyzing eq . we can see how the effective coupling strength of the com variables will behave for different patterns . for instance , if the summations converge as @xmath62 we conclude that in the thermodynamic limit the effective coupling strength of the com variables is an intrinsic quantity , independent of the size of the system . the only interaction pattern for which the effective coupling strength is extensive is the one - to - all pattern . any other pattern for which the summation in eq . is divergent corresponds to an effective coupling that increases with the system size . if @xmath63 { } \log(n ) + c_2,\end{aligned}\ ] ] where @xmath64 is a constant that depends on @xmath65 and we used the convention @xmath66 for @xmath67 . the reason we used @xmath68 in the interaction term is in order to satisfy the condition eq . . we see that in the thermodynamical limit the effective interaction strength scales as @xmath69 . in this section we describe a change of coordinates from the original set @xmath71 into a new set @xmath72 . this new canonical transformation is more symmetric than the one used in the previous section and in @xcite , and allows for a general analysis for @xmath70 , where @xmath73 is an arbitrary integer . since we are mainly interested in the behavior of the com coordinate ( and how it differs from the rest of the coordinates ) , we require that the new set includes two coordinates @xmath74 and @xmath75 , which correspond to the com of objects @xmath55 and @xmath56 . , i.e. @xmath76 . see figure [ a ] for the definition of the rest of the coordinates . this rescaling is purely conventional and does not effect the physical conclusions drawn about the com . ] this set of new coordinates allow us to generalize our previous analysis to include randomness in the couplings between the microscopic constituents of objects @xmath55 and @xmath56 which is shown in the next subsection . our results show that the com variable is singled out by i.i.d . ( independent and identically distributed ) random couplings between the micro - variables . moreover , the properties of the canonical transformation provide insights into the reason why the com variable is special . to distinguish this set of coordinates from those defined in section [ sec : inter - intra ] we drop the tilde . the transformation described below is only valid for @xmath70 . however , this is enough for our purpose of addressing mqp for large @xmath2 . let @xmath77 be the matrix associated with the linear transformation from the original coordinates @xmath78 to the new coordinates @xmath79 . in the rest of the paper we denote vectors and matrices with bold characters , whereas individual entries will be indicated by regular characters with subscripts . then @xmath80 explicit form of @xmath77 for @xmath81 is given in figure [ a ] and the procedure for obtaining @xmath70 for arbitrary integer @xmath9 is described . for brevity of notation the subscript @xmath2 will be dropped in the rest of this section . a nice property of the transformation matrix is that @xmath82 . note that @xmath83 has the first column ( row ) of identical entries , which corresponds to the com coordinate . the relative coordinates defined in this section are different from those used in section [ sec : inter - intra ] as defined in @xcite . the coordinates of this section are in a way more symmetric ; for instance all the associated masses are equal to the original mass @xmath84 . demanding this symmetry together with the condition that there are two coordinates proportional to the com of each object forces a specific form on the matrix @xmath83 : all the columns ( rows ) corresponding to the relative coordinates have half of the entries with positive and the other half negative sign . this property , which singles out the com coordinate kinematically , will play an important role in the proceeding analysis . is the transformation matrix from the original coordinates to the new set of coordinates which include the com . a procedure to explicitly construct @xmath77 is illustrated above for @xmath70 . if @xmath85 is known , @xmath86 can be constructed by copying the block matrix @xmath85 to the the off - diagonal blocks and by putting the negative of it to the lower diagonal . note that any transformation matrix for a smaller @xmath2 can be obtained by restricting to the upper left corner of the larger matrix.,title="fig : " ] [ a ] using the transformation we can write the hamiltonian for the new set of canonical variables . in this section we will set @xmath87 since we are interested in the effect of interactions between the constituents of the two macroscopic objects . the hamiltonians @xmath88 and @xmath45 preserve their original form under the transformation since @xmath89 . the interaction hamiltonian becomes : @xmath90 as a quick check of this formalism we calculate @xmath91 for the pairwise and one - to - all interaction patterns studied in sections [ sec : pairwise ] and [ sec : onetoall ] . for pairwise interactions @xmath92 and we get @xmath93 . note that each coordinate is coupled with equal strength in this type of coupling ( there was no need to renormalise the coordinates since the associated masses are already equal in this set of coordinates ) . for one - to - all interactions we have @xmath94 which translates to @xmath95 . note that only the com coordinates are coupled in this type of interaction and the coupling strength scales as @xmath2 , which agrees with previous analysis . in previous sections we treated coupling patterns that are deterministic . it is reasonable to ask whether the conclusions we reach about the significance of the com coordinate and its decoupling from the relative coordinates are stable under perturbations . to investigate this issue , we reconsider the one - to - one and patterns and this time allow for gaussian variations around the non - zero coupling strengths . our analysis shows that the conclusions of previous sections regarding the significance of the com variable and the decoupling of it from the relative coordinates are not altered by the addition of fluctuations . note that in what follows we do not allow for fluctuations in the vanishing coupling strengths , for example the non - pairwise coupling strengths in the one - to - one pattern ( see eq . ) . we motivate this choice by noting that the vanishing couplings can be the result of a constraint based on symmetry or geometry and thus immune to fluctuations . on the other hand , to assume that the values of non - zero coupling constants are fixed without fluctuations would be more difficult to justify . with the canonical transformation of the previous section we can address the case where the coupling constants @xmath96 are sampled from identical independent gaussian distributions characterised by the mean and variances : @xmath97 we now ask the question , how do the coupling constants @xmath98 behave ? we can use the transformation and the properties of the matrix @xmath83 to calculate the statistical properties as : @xmath99 thus if the couplings between the constituents of the macroscopic objects are statistically independent gaussian random variables , the corresponding couplings between the new variables are also independent gaussian random variables , which follows from the fact that the new and old variables are related by a linear transformation . the main difference is that only the com - to - com coupling has a nonvanishing expectation value which is equal to the expectation value of the couplings of the original coordinates multiplied by @xmath2 . as expected , the average behavior is that of the deterministic rule around which we are perturbing . on the other hand the coupling constants of all of the new coordinates , i.e. both com and relative coordinates , have the same variance that is the same as the variance of the couplings of the original coordinates . note that the fluctuations of the com coupling become negligible in the thermodynamic limit , but not those of the relative coordinates . thus the conclusions of previous sections about the com hold with respect to the perturbations considered here and in the thermodynamic limit . here we repeat the analysis of the previous section for the one - to - one coupling pattern . the coupling constants @xmath96 are sampled from identical independent gaussian distributions characterised by the mean and variances : @xmath100 we now ask the question , how do the coupling constants @xmath98 behave ? after some algebra we get : @xmath101 in the last step we used the fact that the matrix elements of @xmath83 are @xmath102 to conclude that @xmath103 . we see that , unlike the one - to - all case , the com coupling behaves the same as those of the relative coordinates and for com as well as relative coordinates the fluctuations become negligible in the thermodynamic limit . in the above section we showed how the scaling with n depends on the type of couplings amongst the constituent particles of two objects and provided formal proofs for the decoupling of the center of mass variables . in this section we use an explicit model calculation to illustrate the ideas presented in prior sections , namely , the level of structure scheme in the consideration of entanglement of composite objects , and the special role of the com variables in the consideration of macroscopic entanglement . the model is made up of 4 coupled oscillators in two pairs representing the two objects a ( constituents 1 , 2 ) and b ( constituents 3 , 4 ) . we assign different coupling strengths : @xmath6 as that between the two ( intra - level ) constituents 1 - 2 or 3 - 4 in each object , @xmath7 as the ( inter - level ) coupling strength between 1 - 3 and also 2 - 4 ( parallel ) , and @xmath8 between 1 - 4 and also 2 - 3 ( cross ) . considering @xmath6 to be large compared to @xmath7 or @xmath8 refers to strong intra- level interactions or tight - binding of constituents within one level of structure . this model can show the difference between the cases studied in sec . 3 , namely , the pairwise ( 1 - 3 , 2 - 4 ) interaction versus the one to all ( 1 - 3 and 1 - 4 ) interaction . as in sec . 3 we see that it is only in the one - to - all case that the entanglement scales with n. after this we calculate the entanglement dynamics of the 4-oscillator system with disparate coupling strength , while distinguishing three cases of entanglement : the case between the center of mass variables ( com ) of a and b , the case between the com variable of one object a and one constituent particle of b , and the case between one constituent particle of a and one of b. we see some pariwise entanglement meets sudden death while others sustain over long time ranges . this offers another way to determine under what conditions quantum entanglement between macroscopic objects can persist . the total hamiltonian of the system of 4 coupled harmonic oscillators of equal masses but disparate coupling strength as described above is given by @xmath104 where @xmath105 and we have set the mass of each oscillator to unity . we use indices @xmath106 to label the objects a , b and @xmath107 to label the constituent particles . the @xmath108 characterizes the interactions among the oscillators and has the matrix form @xmath109 this hamiltonian can be diagonalized into the form @xmath110 where @xmath111 . the @xmath112 could be written in terms of the center - of - mass ( @xmath113 ) and relative coordinates ( @xmath114 ) of object a and similarly for object b. the explicit form of @xmath112 are @xmath115 as a simple illustration we assume that at time ( @xmath116 ) the couplings among the four oscillators were zero and the initial wave function of the system was the direct product of the gaussian ground state wave function of each oscillator , i.e. , @xmath117 , where @xmath118 . at time @xmath119 the interactions among the oscillators were turned on and the total system is then driven by the hamiltonian @xmath120 and the final wave function at time @xmath121 is @xmath122 which can be easily obtained from the evolution operator of the system . there are well defined measures of entanglement for continuous variables . define the two - point correlation matrix @xmath123 with elements @xmath124 where @xmath125 , @xmath126 . note that here @xmath127 is the conjugate momentum operator of @xmath128 and the @xmath129 could be @xmath130 or @xmath131 . the partial transpose of @xmath132 is @xmath133 , where @xmath134 . starting with the gaussian initial state , @xmath135 , the reduced density matrix or wigner function of the two oscillators is always gaussian by virtue of the quadratic structure of the hamiltonian of our model . therefore the oscillator @xmath136 and oscillator @xmath137 is entangled at time @xmath121 if and only if @xcite @xmath138 < 0 \label{entcond}\ ] ] where @xmath139 is a symplectic matrix . note that when @xmath140 the behavior of @xmath141 is quite similar to the behavior of the negative eigenvalues which are connected to the logarithm negativity @xcite . thus the value of @xmath141 itself is a good indicator of the degree of entanglement , at least in this specific model . it is then straightforward ( although somewhat tedious ) to obtain the analytic form of @xmath142 and hence the entanglement dynamics . below we calculate the two - point correlation matrix @xmath143 and @xmath141 for various degrees of freedom in this model and discuss the implications on mqp . we now study the entanglement structure and its dynamics for different pairs of variables , those which represent the objects a , b specifically the com variables @xmath144 and the relative coordinates @xmath145 and those representing the constituents particles @xmath146 . in this way we can see how these variables and how the inter - level and intra - level coupling strengths enter into the entanglement structure . the results ranges from the somewhat obvious , such as that @xmath147 and @xmath148 to the more complicated and subtle , such as the existence of sudden death for certain pairs of variables and others which persists , as illustrated in the figures . we divide into two categories , that when the intra - level coupling strength @xmath6 is zero or nonzero . one can think of the former as depicting systems with very weak coupling amongst the constituent particles ( molecular gas ) and the latter as objects with stronger constituent couplings ( solid body ) . we will present the formal results in the following and attempt to understand their physical relevance in the next section . note that @xmath149 is a special case for sec . 3 . in this category there is no direct interactions between the constituents of each object . we will then concentrate on the interactions between objects a and b there are two special subcases : * @xmath151 : pairwise interaction case . oscillator 1(2 ) in subsystem a couples only to oscillator 3(4 ) in subsystem b. fig . 3(a ) shows the entanglement dynamics of @xmath152 and fig . 3(b ) shows the entanglement dynamics of @xmath153 . note that @xmath154 in this case and hence there is no entanglement among @xmath155 and @xmath156 . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158 , @xmath159(in purple ) , @xmath160(in yellow ) . it can be shown that @xmath147 and @xmath148 irrespective of the couplings . in these figures we set @xmath161 , @xmath162,@xmath163 , @xmath164 . under this special choice of parameter @xmath157 is almost twice @xmath165 and @xmath166 . these can be easily understood from the definition of com and relative coordinates . note that @xmath167 for @xmath168 hence there is always quantum entanglement between @xmath169 and @xmath170 once the coupling between them , @xmath7 , was turned on.,title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158 , @xmath159(in purple ) , @xmath160(in yellow ) . it can be shown that @xmath147 and @xmath148 irrespective of the couplings . in these figures we set @xmath161 , @xmath162,@xmath163 , @xmath164 . under this special choice of parameter @xmath157 is almost twice @xmath165 and @xmath166 . these can be easily understood from the definition of com and relative coordinates . note that @xmath167 for @xmath168 hence there is always quantum entanglement between @xmath169 and @xmath170 once the coupling between them , @xmath7 , was turned on.,title="fig:",width=226 ] * @xmath171 : one - to - all case . oscillator 1(2 ) in subsystem a couples to both oscillator 3 and 4 in subsystem b with equal strength . 4(a ) shows the entanglement dynamics of @xmath152 and fig . 4(b ) shows the entanglement dynamics of @xmath153 . note that @xmath172 in this case @xmath173 . hence there is no entanglement among @xmath174 and @xmath175 . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath176 , @xmath177,@xmath163 , @xmath164 . note that under these parameters @xmath178 and do not appear in figure ( b ) . here one can see for the on - to - all case , @xmath179 hence the com axion applies as explained in sec . 3.0.2.,title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath176 , @xmath177,@xmath163 , @xmath164 . note that under these parameters @xmath178 and do not appear in figure ( b ) . here one can see for the on - to - all case , @xmath179 hence the com axion applies as explained in sec . 3.0.2.,title="fig:",width=226 ] now let s turn on the interactions among the oscillators of each subsystem a , b. in this case @xmath180 . we consider the following subcases . * @xmath151 : pairwise interaction case . 5(a ) shows the entanglement dynamics of @xmath152 and fig . 5(b ) shows the entanglement dynamics of @xmath153 . note that @xmath154 in this case and hence there is no entanglement among @xmath155 and @xmath156 . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) , @xmath159(in purple ) , @xmath160(in yellow ) . here @xmath181 , @xmath182,@xmath163 , @xmath164 . one can see the effect of the coupling parameter @xmath183 makes the peak values of @xmath165 and @xmath157 differ by a sizable amount . @xmath184 ( although very small and can hardly read from the figure ) in this case , this reflects the separability of the states in the hilbert spaces of @xmath155 and @xmath175 . that means @xmath155 and @xmath175 are always separable . it s interesting to note that the entanglement between @xmath174 and @xmath175 is even stronger than the entanglement among @xmath155 and @xmath185 in this case.,title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) , @xmath159(in purple ) , @xmath160(in yellow ) . here @xmath181 , @xmath182,@xmath163 , @xmath164 . one can see the effect of the coupling parameter @xmath183 makes the peak values of @xmath165 and @xmath157 differ by a sizable amount . @xmath184 ( although very small and can hardly read from the figure ) in this case , this reflects the separability of the states in the hilbert spaces of @xmath155 and @xmath175 . that means @xmath155 and @xmath175 are always separable . it s interesting to note that the entanglement between @xmath174 and @xmath175 is even stronger than the entanglement among @xmath155 and @xmath185 in this case.,title="fig:",width=226 ] * @xmath186 : special one - to - all case . 6(a ) shows the entanglement dynamics of @xmath152 and fig . 6(b ) shows the entanglement dynamics of @xmath153 . note that in this special case we have @xmath172 in this case @xmath173 . hence there is no entanglement among @xmath174 and @xmath175 . also it is found that when @xmath187 , @xmath188 . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath189,@xmath163 , @xmath164 . @xmath190 as expected from the com axiom even now the coupling between oscillators in each macroscopic object , @xmath6 , is non - zero . combined with the fact that @xmath191 , one has @xmath192 under this special @xmath187 case . hence the com coordinate does play a very special role in macroscopic quantum phenomena as illustrated in this section . @xmath159= @xmath160=0 and do not appear in figure ( b).,title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath189,@xmath163 , @xmath164 . @xmath190 as expected from the com axiom even now the coupling between oscillators in each macroscopic object , @xmath6 , is non - zero . combined with the fact that @xmath191 , one has @xmath192 under this special @xmath187 case . hence the com coordinate does play a very special role in macroscopic quantum phenomena as illustrated in this section . @xmath159= @xmath160=0 and do not appear in figure ( b).,title="fig:",width=226 ] * @xmath193 : strongly interacting one - to - all case . 7(a ) shows the entanglement dynamics of @xmath152 and fig . 7(b ) shows the entanglement dynamics of @xmath194 . in this strong coupling regime , the entanglement dynamics of @xmath157 shows the sudden death and revival phenomena . it s interesting to note that @xmath195 while @xmath157 can be greater than zero for some finite period of time . this can be easily understood as follows : as the coupling @xmath6 between the constituents among each macro - objects ( a or b ) gets stronger and stronger , the correlations among them become stronger hence the entanglement among the micro - constituents @xmath196 will be strongly affected by the strong coupling between @xmath197 and @xmath198 ( and also @xmath199 and @xmath200 ) . this will make the entanglement between @xmath197 and @xmath199 vanish for some time intervals . however the two macro - objects ( a and b ) will interact mainly through the coupling @xmath7 in the one - to - all case hence the entanglement between @xmath201 and @xmath199 still holds . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath202,@xmath163 , @xmath164 . two major feature differences as compared to @xmath187 case . one is that @xmath203 is larger than twice the amount of @xmath204 . the other feature is the entanglement dynamics of @xmath157 shows the sudden death and revival phenomena while those for @xmath165 does not . @xmath159= @xmath160=0 and do not appear in figure ( b).,title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath202,@xmath163 , @xmath164 . two major feature differences as compared to @xmath187 case . one is that @xmath203 is larger than twice the amount of @xmath204 . the other feature is the entanglement dynamics of @xmath157 shows the sudden death and revival phenomena while those for @xmath165 does not . @xmath159= @xmath160=0 and do not appear in figure ( b).,title="fig:",width=226 ] * @xmath205 : weak interacting one - to - all case . 8(a ) shows the entanglement dynamics of @xmath152 and fig . 8(b ) shows the entanglement dynamics of @xmath153 . this is similar to the @xmath206 case . the effect of the small nonzero coupling @xmath6 will slightly reduce the amount of entanglement . + ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath207,@xmath163 , @xmath164 . note that under these parameters @xmath178 and do not appear in figure ( b ) . here one can see for the on - to - all case , @xmath179 hence the com axiom applies as explained in sec . the effect of the coupling makes the amount of entanglement being slightly reduced as compared to the noninteracting case as those in figure 3(where @xmath150 ) . , title="fig:",width=226 ] ( in blue ) , @xmath157(in purple ) . figure ( b ) shows the entanglement dynamics of @xmath158(in blue ) . here @xmath207,@xmath163 , @xmath164 . note that under these parameters @xmath178 and do not appear in figure ( b ) . here one can see for the on - to - all case , @xmath179 hence the com axiom applies as explained in sec . the effect of the coupling makes the amount of entanglement being slightly reduced as compared to the noninteracting case as those in figure 3(where @xmath150 ) . , title="fig:",width=226 ] in summary , the 4 harmonic oscillators case gives a concrete example supporting the physical arguments and analysis in previous sections . here we provide the entanglement structure and dynamics depending on whether there is interaction ( coupling strength = @xmath6 ) among constituent oscillators of each subsystem . we analyze two special cases : one - to - one(pairwise ) and one - to - all in detail . no quantum entanglement between the center - of - mass @xmath155 and the relative coordinates @xmath174(@xmath175 ) . ( ii ) . for non - interacting @xmath208 one - to - one case @xmath209 as can be easily understood from the definition of the center - of - mass coordinate . ( fig.3 ) ( iii ) . for non - interacting @xmath208 one - to - all case @xmath210 hence supporting the com axiom . ( fig.4 ) ( iv ) . for interacting @xmath211 one - to - one case one can see the effect of the coupling on the entanglement dynamics.(fig.5 ) ( v ) . for special interacting one - to - all case @xmath212 @xmath188 . for this special case , the model processes a larger symmetry and hence the clear - cut behavior of the corresponding entanglement dynamics . this also supports the analysis in the previous section . ( fig.7 ) ( vi ) . in the one - to - all cases , as the coupling @xmath6 changes from weak to strong ( compared with @xmath7 ) , we see how the strong coupling alters the entanglement dynamics and drives the sudden death and revival of entanglement among constituents between different subsystems . ( fig.7,8 ) in short , the 4 harmonic oscillators case gives a concrete example illustrating the physical arguments and analysis in previous sections , lending concrete support to the com axiom . in this paper we try to understand mqp from the quantum entanglement perspective by asking several key questions and suggesting some pathways to understand them . as a background to this question we have in two earlier papers @xcite asked the question , what is meant by macroscopic ? is it the size of an object ? its contents , as measured by the number of particles or components ? what about its composition ? here the coupling strength of its constituents matter . what about its ` appearance ' ? in the sense that the energy scale or precision level one probes the object matters , etc . against this backdrop , we ask these questions : \1 ) one really needs to start with the entanglement between the constituent particles . but immediately one encounters the puzzle as to what would be regarded as constituent , since each constituent particle is made up of sub - constituent particles . here unavoidably a ) _ the level of structure scheme _ of macro or even meso objects enters in an essential way . when we talk about the entanglement between two objects are we referring to the entanglement between the quarks of one object with that of the other , or the molecules of each object ? the obvious difference between them is that quarks are strongly coupled and the interactions among them is short - ranged while interactions among molecules are much weaker but with longer effective interaction range . thus b ) the _ coupling strength of constituents _ should enter . thirdly , c ) assuming that one can find a measure of entanglement between two macro objects of a definite value , call it @xmath0 , and we know that their constituents fall under levels of structures , how does one derive @xmath0 from the entanglements between these micro - constituents , call them @xmath1 . this problem , let s call it _ the sum - rule " question _ , is highly nontrivial . the quarks and gluons are strongly coupled and are expected to have stronger entanglements , but their interaction is short ranged , thus their contributions to e(m ) may not be as high as the van der walls force between molecules . a measure of entanglement at the macro - level could involve some coarse - graining of the contributions of the sub - constituents . but what is the criteria for coarse - graining ? if there is some way to express @xmath0 in terms of @xmath1 would the sum - rule include a weighing factor attributed to the entanglement measure from each level of structure ? is there some kind of hierarchical ordering in systemizing the weighing factors in the contributions from each level of structure for @xmath1 to add up to a finite value ? these are interesting questions worthy of further investigation . we have only some glimpses of an answer to this question : they do nt usually add up , but depend on the coupling pattern of the constituents and what and how different parties are entangled . this last point is what enables one to address questions 2 ) and 3 ) below . \2 ) acknowledging the complexity of what makes up a macroscopic object , constructing or _ identifying a set of collective variables _ from the constituents in each level of structure is therefore an essential step towards organizing their contributions . we showed the conditions under which the center of mass of an object plays a special role in mqp . its distinguished role is highlighted by the fact that the com variables are in general decoupled from the relative variables and the dynamics of the com variables behaves differently from the relative variables . to understand how the _ coupling pattern amongst the constituents _ of the two macro objects enters into the picture , we consider two types of coupling , the 1-to-1 versus the 1-to - all . for the 1 - 1 or pairwise interactions with equal strength , the entanglement is independent of the size @xmath2 of the macroscopic objects . the 1 - 1 interaction pattern is the only one for which the com and relative coordinate couplings are effectively the same . in the 1-to - all case the relative coordinates are decoupled and the com coupling scales with n. in this case we expect the entanglement between the com variables to increase with increasing size of the macroscopic objects and survive at higher temperatures . note that this scaling property is proven here for gaussian systems only . we have also considered gaussian fluctuations around the 1-to-1 and 1-to - all patterns and shown that these conclusions continue to hold in the presence of such perturbations in the thermodynamic limit . our detailed analysis of the specific model , that between two pairs of constituent oscillators [ ( 3 ) below ] also displays further distinct behavior of the com in terms of entanglement dynamics , like the absence of sudden death of entanglement in the 1-to - all case , etc . \3 ) we pause here to compare our findings with prior work on entanglement structure of gaussian systems such as that carried out systematically by adesso and illuminati @xcite . in our analysis we have shown that the one - to - all interaction pattern , when viewed in the transformed coordinates , corresponds to only com - com interactions being non - vanishing , which moreover scale linearly with @xmath2 . this ability to concentrate the interactions between two macro object each consisting of @xmath2 oscillators into interactions between @xmath213 modes only is related to the results of @xcite . in that paper it is shown that if a gaussian state of @xmath214 oscillators is bi - symmetric , then there exists a unitary transformation which acts on each object separately and results in @xmath215 uncorrelated modes and @xmath213 correlated modes , one in each object . we showed through the coupling pattern analysis ( in sec . 3 ) and explicit results of a 4-oscillator system ( in sec . 4 ) that these two correlated modes correspond to the com variables which serve to capture the essence of the interactions between the two macroscopic objects completely . reference @xcite has shown that the @xmath216 entanglement vanishes in the limit @xmath217 , whereas the @xmath218 entanglement diverges . this shows that 1x1 entanglement ( even if summed over all pairs ) is not enough to account for the macroscopic entanglement between two objects . the localized entanglement , in our case between the com variables , quantifies the total entanglement between the two objects , and its divergence with n is consistent with our result that the interaction strength is proportional to n. in contrast to the one - to - all case , it can be shown that in the pairwise coupling case no coordinate transformation can bring forth a `` concentration '' of the interactions between the macro objects . all transformed coordinates are still pairwise coupled with equal coupling strengths . this does not contradict the results of @xcite however , since the pairwise coupling scheme is not bisymmetric . as a result we can not apply the general results about bisymmetric gaussian states to the pairwise coupling case . in this case the com fails to capture the interactions between the two objects completely . \4 ) before drawing implications of these results we should make two points clear here : a ) reference @xcite considered states ( equivalently , covariance matrices ) whereas we analysed hamiltonians , which are more directly related to the dynamics . however , when considering thermal states or dynamics following initially uncorrelated states , the properties of evolved states follow closely the properties of the hamiltonian . because of this our results can be related to theirs . b ) our results in sec . 3 addresses how the coupling strength depends on the coupling pattern and varies with the constituent versus the collective variables . statements made stemming from there pertaining to entanglement strength assume that entanglement is proportional to coupling strength . even though this is intuitively acceptable , the relation between entanglement magnitude and coupling strength remains to be made more explicit . allowing for this connection we may continue our discussions further : the fact that the entanglement between two ( or more ) macroscopic objects coupled in the pattern grows as @xmath2 increases ( in some power or pattern ) is easy to understand : since the coupling is identical for all constituent particles , it does not matter which object the particles are the constituents of , the partition between the two objects can be viewed as a conceptual artifact . it can as well be regarded as the entanglement between two parts of one macro object . one can also turn this around and introduce the entanglement between the ( coms of the ) two halves as providing a measure of the _ entanglement strength of a macroscopic object . _ furthermore , since in reality the coupling strength often depends on the separation between the constituent particles the case studied here is realized only in short - ranged yet strongly - interacting cases , such as nucleons in a nucleus , whose interaction is quite adequately depicted by a square well potential . note that it is the one - to - all pattern , rather than the strength , of nuclear interactions that allows us to use them as an example here . motivated by this , even apart from mqp considerations the way how entanglement varies with coupling strength and pattern can in principle be used to explore the _ entanglement entropy for a quantum many - body system_. e.g. , depending on the coupling pattern the entanglement entropy can scale differently with the number of particles in the system , from which one can see how the proportionality shifts , say , from the volume to the surface or perhaps even somewhere in between . this carries a different meaning from the conventional one where for short - ranged forces the entanglement entropy of a quantum many - body system is related to the surface area of an artificially introduced partition . ( for a review on this subject see , e.g. , @xcite . the original area theorem found by bekenstein for black holes is in a more natural setting because of the existence of event horizons . ) the analysis of the model here brings out a different facet of the problem , namely , the coupling strength and pattern of constituents , in defining a suitable measure of entanglement entropy for quantum many - body systems . \5 ) to quantify these behavior , specifically aiming at how entanglement depends on how the constituents are organized by their coupling strengths and how the com axiom applies to the entanglement of macro - objects , we analyzed the entanglement structure of a four oscillator system in detail , not only in the kinematics as is done in the two interaction patterns and in the proofs , but here also in the dynamics of entanglement . by assigning different strengths to the three pairwise coupling constants we can investigate the entanglement differences between the intra- and inter- level of structures . by setting the coupling strength between the intra - level constituents to be weak , we reproduce the results for the 1-to-1 pattern , while keeping the three coupling strengths the same we obtain the results for the 1-to - all pattern . we verify the dominance in the entanglement of the com variables by showing the existence of sudden death in the entanglement dynamics of inter - level constituents and the persistence of the entanglement between the com variables . this model continues what was studied before in the entanglement dynamics of say two two - level atoms interacting with common field modes @xcite or two oscillators interacting with a common field @xcite , but with a new angle toward issues in mqp . in terms of practical relevance , we commented on the 1-to - all pattern above , as relating to the entanglement entropy of a many - body system . for the 1-to-1 pattern it can be applied to finding the entanglement in materials with layered structures , from two bec slabs , to graphene , even to the two strands of dna sequences where the dna a(c ) couples pairwise with dna t(g ) . our effort here is to search for pathways into analyzing the entanglement structure and dynamics of macro - objects . the field is wide open for various theoretical probings starting even with asking good questions with intellectual rewards , and the results will bear immediate practical applications . * acknowledgment * we wish to thank dr . j. t. hsiang for sharing his calculation results for the entanglement pattern of three coupled oscillator systems ; and r. zhou and k. sinha for discussions . chc thanks the support from national center for theoretical science(south ) of taiwan . this work is supported by the national science council of taiwan under grant nsc 101 - 2112-m-006 - 003 . ys was a visiting graduate fellow at perimeter institute in the spring semester of 2013 . blh thanks professor m. c. chu of the chinese university of hong kong and professors y. s. wu and j. q. you of fudan university for their kind hospitality during his visits in the spring of 2013 when this work was carried out . brian pepper , evan jeffrey , roohollah ghobadi , christoph simon , dirk bouwmeester , macroscopic superpositions via nested interferometry : finite temperature and decoherence considerations " new j. phys . 14 , 115025 ( 2012 ) [ arxiv:1207.1946 ] k. sinha , n.i . cummings , b.l . hu effect of interatomic separation on entanglement dynamics in a two - atom two - mode model " j. phys . b : at . mol . ( 2012 ) 035503 . chris fleming , nicholas cummings , c. anastopoulos and b. l. hu , non - markovian dynamics and entanglement of two - level atoms in a common field " , j. phys . a : math . theor . 45 ( 2012 ) 065301 .

polynomial chaos expansions ( pce ) allow us to propagate uncertainties in the coefficients of differential equations to the statistics of their solutions . their main advantage is that they replace stochastic equations by systems of deterministic equations . their main challenge is that the computational cost becomes prohibitive when the dimension of the parameters modeling the stochasticity is even moderately large . we propose a generalization of the pce framework that allows us to keep this dimension as small as possible in favorable situations . for instance , in the setting of stochastic differential equations ( sdes ) with markov random forcing , we expect the future evolution to depend on the present solution and the future stochastic variables . we present a restart procedure that precisely allows pce to depend only on that information . the computational difficulty then becomes the construction of orthogonal polynomials for dynamically evolving measures . we present theoretical results of convergence for our dynamical generalized polynomial chaos ( dgpc ) method . numerical simulations for linear and nonlinear sdes show that it adequately captures the long - time behavior of their solutions as well as their invariant measures when the latter exist . keywords : polynomial chaos , stochastic differential equations , uncertainty quantification , markov processes

1505.00047

_ polynomial chaos _ ( pc ) is a computational spectral method that is used to propagate and quantify uncertainties arising in the modeling of physical systems . the method originated from the works of wiener @xcite and cameron and martin @xcite on the decomposition of functionals in a basis of hermite polynomials of gaussian random variables . ghanem and spanos @xcite used such an expansion to solve stochastic equations with random data . xiu and karniadakis @xcite introduced _ generalized polynomial chaos _ ( gpc ) involving non - gaussian random parameters . extensions to arbitrary probability measures followed in the works @xcite ; see also @xcite for recent theoretical developments and a convergence result for gpc expansions . _ polynomial chaos expansions _ ( pce ) provide an explicit expression of quantities of interests as functionals of the underlying uncertain parameters and in some situations allow us to perform uncertainty quantifications at a considerably lower computational cost than monte carlo methods . however , they suffer from the curse of dimensionality and thus typically work efficiently for systems involving low dimensional uncertainties . a related major drawback appears in the long - term integration of evolution equations . in the presence of random forcing in time , the number of stochastic variables in the system increases linearly with time and thus quickly becomes overwhelming . moreover , standard pce utilize orthogonal polynomials of the initial distribution and as time evolves , the dynamics deviate from the initial data substantially ( e.g. , due to nonlinearities ) and the solution may become poorly represented in the initial basis . this led the authors in @xcite to legitimately question the usefulness of ( standard ) pce to address long - term evolution properties of solutions such as intermittent instabilities . this paper aims to address the aforementioned drawbacks . we propose a pc - based method that constructs evolving chaos expansions based on polynomials of projections of the time dependent solution and the random forcing . more precisely , chaos expansions at each iteration are constructed based on the knowledge of the moments of underlying distributions at a given time . in the setting of dynamics satisfying a markov property , we crucially exploit this feature to introduce projections of the solution at prescribed time steps that allow us to `` forget '' about the past and as a consequence keep the dimension of the random variables fixed and independent of time . our chaos basis is adapted to the evolving dynamics with the following consequences : ( i ) our expansion retains its optimality for long times ; and ( ii ) the curse of dimensionality is mitigated . notably , we establish , with appropriate modifications , theoretical convergence analysis which sheds light on our numerical findings . further , asymptotic analysis for computational complexity will also be discussed . inspired by examples of @xcite , we will apply our method to a nonlinear coupled system of stochastic differential equations ( sdes ) and its variants . other iterative methodologies have already been proposed in @xcite in various contexts of random differential equations ( rdes ) . the main novelty of our approach is that it allows us to compute long - term solutions of evolution equations with complex stochastic forcing , which here will be a brownian motion but could be generalized to more complex models . white noise driven evolution equations have important roles in modeling the small scale effects and certain uncertainties in many applications such as turbulence , filtering , mathematical finance , and stochastic control @xcite . the plan of the paper is as follows . section [ sec : pc ] introduces background material and establishes the necessary notation for pce . our methodology , called dynamical generalized polynomial chaos ( dgpc ) , is described in detail in section [ sec : method ] . numerical experiments comparing dgpc to hermite pc and monte carlo simulations are presented in section [ sec : numeric ] . some conclusions are offered in section [ sec : conclu ] . we briefly introduce the original pc framework . the notation we follow can be found in @xcite . given a probability space @xmath0 where @xmath1 is a general sample space , @xmath2 is a @xmath3-algebra of subsets and @xmath4 is a probability measure , we consider @xmath5 , the space of real - valued random variables with finite second moments . let @xmath6 be a countable collection of independent and identically distributed ( i.i.d ) standard gaussian random variables belonging to the probability space , and @xmath7 . then , we define the wick polynomials @xmath8 where @xmath9 belongs to set of multi - indices with a finite number of nonzero components @xmath10 @xmath11 is the @xmath12th order normalized one - dimensional hermite polynomial and @xmath13 . note that the wick polynomials are orthogonal to each other with respect to the measure induced by @xmath14 . the cameron and martin theorem @xcite establishes that the wick polynomials form a complete orthonormal basis in @xmath15 . this means that any functional @xmath16 , can be expanded as @xmath17,\end{aligned}\ ] ] and the sum converges in @xmath18 . we define @xmath19 as expectation with respect to @xmath4 . the expansion is called polynomial chaos expansion ( pce ) . we will also use the term hermite pce to emphasize that the expansion utilizes gaussian random variables . one of the notable features of is that it separates the randomness in @xmath20 such that the coefficients @xmath21 are deterministic and all statistical information is contained in the coefficients . in particular , the first two moments are given by @xmath22 = u_0 $ ] and @xmath23 = \sum_{\alpha \in \mathcal{j } } |u_{\alpha}|^2 . $ ] higher order moments may then be computed using either triple products of wick polynomials @xcite or the hermite product formula @xcite ; see section [ sec : moments ] . in numerical computations , the doubly infinite expansion is truncated so that it becomes a finite expansion @xmath24 where we simply used polynomials up to degree @xmath25 in the variables @xmath26 . throughout the paper , we use the graded lexicographic ordering for multi - indices ; see ( * ? ? ? * table 5.2 ) . in the context of white noise driven sdes , the random variables @xmath27 can be obtained by the projection @xmath28 where @xmath29 is a brownian motion on @xmath30 $ ] and @xmath31 is a complete orthonormal system in @xmath32 $ ] . moreover , @xmath33 is comprised of i.i.d . standard gaussian random variables , and the expansion @xmath34 converges in @xmath18 for all @xmath35 and has the error bound @xmath36 ^ 2 = o(t k^{-1 } ) , \end{aligned}\ ] ] provided the @xmath31 s are trigonometric polynomials ; see and @xcite . here , brownian motion @xmath37 is projected onto @xmath32 $ ] for a fixed time @xmath38 so that the corresponding wick polynomials @xmath39 depend implicitly on @xmath40 . the simple truncation leads to @xmath41 terms in the approximation . thus , the computational cost increases rapidly with high dimensionality , which in turn decreases the efficiency of pce . this is the `` curse of dimensionality '' . another related major problem of pc is that expansions may converge slowly and even fail to converge for long time evolutions @xcite . these problems led to numerous extensions of the hermite pce , which we now briefly discuss . the paper @xcite proposed a method called generalized polynomial chaos ( gpc ) , in which the above random variables @xmath42 have specific , non - gaussian , distributions in the askey family @xcite . further generalizations to arbitrary probability measures beyond the askey family were proposed in @xcite , where the probability space is decomposed into multiple elements and chaos expansions are employed in each subelement . the approach taken in @xcite is to use a restart procedure , where the chaos expansion is restarted at different time - steps in order to mitigate the long - term integration issues of chaos expansions . another generalization toward arbitrary distributions is presented in @xcite , using only moment information of the involved distribution with a data - driven approach . our key idea is to adapt the pce to the dynamics of the system . consider for concreteness the following sde : @xmath43,\end{aligned}\ ] ] where @xmath44 is a general function of @xmath20 ( and possibly other deterministic or stochastic parameters ) , @xmath29 is a brownian motion , @xmath45 is a constant , and @xmath46 is an initial condition with a prescribed probability density . we assume that a solution exists and is unique on @xmath30 $ ] . as the system evolves , a fixed polynomial chaos basis adapted to @xmath46 and @xmath47 may not be optimal to represent the solution @xmath48 for long times . moreover , the dimension of the representation of @xmath47 increases linearly with time @xmath40 , which renders the pce method computationally intractable even for moderate values of @xmath40 . we will therefore introduce an increasing sequence of restart times @xmath49 and construct a new pce basis at each @xmath50 based on the solution @xmath51 and all additional random variables that need to be accounted for . a very similar methodology with @xmath52 was considered earlier in @xcite . when random forcing is present and satisfies the markov property as in the above example , the restarting strategy allows us to `` forget '' past random variables that are no longer necessary and focus on a significantly smaller subset of random variables that influence the future evolution . as an example of application , we will show that the restarting strategy allows us to capture the invariant measure of the above sde , when such a measure exists . to simplify notation , we present our algorithm on the above scalar sde with @xmath44 a function of @xmath20 , knowing that all results also apply to systems of sdes with minor modifications ; see sections [ sec : higherdim ] and [ sec : numeric ] . first , we notice that the solution @xmath48 of is a random process depending on the initial condition and the paths of brownian motion up to time @xmath40 : @xmath53 therefore , recalling the expansion for @xmath54 , the solution at time @xmath40 can be seen as a nonlinear functional of @xmath46 and the countably infinite variables @xmath14 . as previously noticed in @xcite , the solution @xmath48 can be represented as a linear chaos expansion in terms of the polynomials in itself and therefore , for sufficiently small later times @xmath55 , @xmath56 , the solution @xmath57 can be efficiently captured by low order chaos expansions in @xmath48 everything else being constant . moreover , the solution @xmath57 on the interval @xmath58 depends on @xmath59}$ ] and not on values of @xmath60 outside of this interval . this significantly reduces the number of random variables in @xmath14 that need to be accounted for . this crucial observation clearly hinges upon the markovian property of the dynamics . hence , dividing the time horizon into small pieces and iteratively employing pce offer a possible way to alleviate both curse of dimensionality and long - term integration problems . we decompose the time horizon @xmath30 $ ] into @xmath12 subintervals ; namely @xmath61,[t_1,t_2],\ldots,[t_{n-1},t]$ ] where @xmath62 . the idea is then to employ polynomial chaos expansion in each subinterval and restart the approximation depending on the distributions of both @xmath63 and @xmath51 at each @xmath50 , @xmath64 ; see figure [ fig : prop_chaos ] . here , @xmath63 denotes the gaussian random variables required for brownian forcing on the interval @xmath65 $ ] . throughout the paper , we utilize the term @xmath66 to represent orthonormal chaos basis involving its arguments . in order to establish evolution equations for chaos basis @xmath67 , triple products @xmath68 $ ] are needed . this procedure basically corresponds to computing the coefficients and indices in the hermite product formula . the probability distribution of @xmath51 does not belong to any classical family such as the askey family in general . the construction of the orthogonal polynomials in @xmath51 is therefore fairly involved computationally and is based on the general pce results mentioned earlier in the section . at each time step , we compute the moments of @xmath51 using its previously obtained chaos representation and incorporate them in a modified gram schmidt method . this is a computationally expensive step . armed with the orthogonal basis , we then compute the triple products in @xmath51 to perform the necessary galerkin projections onto spaces spanned by this orthogonal basis and thus obtain evolution equations for the deterministic expansion coefficients . letting @xmath69 , equations for the coefficients @xmath70 of @xmath71 are given in the general form @xmath72 ds + \sigma \ , { \mathbb{e}}\left[t_{\alpha}({\boldsymbol{\xi}}_j , u_j ) \ , d{w_s}\right ] , \end{aligned}\ ] ] where @xmath73 . before integration of these expansion coefficients in time , @xmath74 is represented in terms of its own orthogonal polynomials , which provides a description of the initial conditions on that interval . we then perform a high order time integration . cumulants of the resulting solution are then computed to obtain relevant statistical information about the underlying distribution . the proposed methodology does not require the computation of any probability density function ( pdf ) and rather depends only on its moments at each restart time . the statistical information contained in such moments provides useful data for decision making in modeling and , for determinate distributions , moments characterize the distribution uniquely . this aspect will be essential in the proof of convergence in section [ sec : conv ] . comparing our method with probability distribution function methods , e.g. , the fokker planck equation , we note that it essentially evolves the coefficients of orthogonal polynomials of the projected solution in time instead of evolving the pdf . let @xmath75 represents the nonlinear chaos expansion mapping @xmath76 to @xmath77 . our scheme may be demonstrated by figure [ fig : prop_chaos ] . mathematically , we have @xmath78 in simulations , @xmath63 is truncated with finite dimension @xmath79 . let also @xmath80 denote the maximum degree of the polynomials in the variables @xmath81 . here , to simplify , @xmath82 and @xmath25 are chosen independently of @xmath83 . the multi - indices having @xmath82 dimensions and degree @xmath25 belong to the set @xmath84 we now present our algorithm , called _ dynamical generalized polynomial chaos _ ( dgpc ) , in one dimension for concreteness ; see also section [ sec : higherdim ] . + * algorithm 1d * : dgpc + - decompose time domain @xmath30 = [ t_0,t_1 ] \cup \ldots.\cup [ t_{n-1},t]$ ] + - initialize degrees of freedom @xmath85 + - compute coefficients / indices in hermite product formula for @xmath86 + - integration in time : + - time step @xmath87 : obtain random variable @xmath74 + - calculate moments @xmath88 $ ] + - construct orthogonal polynomials @xmath89 + - compute triple products @xmath90 $ ] + - perform galerkin projection onto @xmath91 + - set up initial conditions for the coefficients @xmath92 + - evolve the expansion coefficients @xmath92 + - calculate cumulants + - postprocessing + several remarks are in order . first , since our stochastic forcing has identically distributed independent increments , we observe that at each subinterval the distribution of @xmath63 is the same . computing and storing ( in sparse format ) the coefficients for the product formula only once drastically reduces the computational time . at each iteration , the random variable @xmath93 is projected onto a finite dimensional chaos space and the next step is initialized with the projected random variable . for a @xmath94-dimensional sde system , this projection leads to @xmath95 terms in the basis for the subinterval @xmath96 $ ] . therefore , the total number of degrees of freedom used in @xmath30 $ ] becomes @xmath97 ; see also section [ sec : cost ] . we emphasize that small values of @xmath85 are utilized in each subinterval such that computations can be carried out quickly . the idea of iteratively constructing chaos expansions for arbitrary measures was considered earlier in @xcite . in @xcite , an iterative method is proposed to solve coupled multiphysics problems , where a dimension reduction technique is used to exchange information between iterations while allowing the construction of pc in terms of arbitrary ( compactly supported ) distributions at each iteration . a similar iterative procedure for hermite pc was suggested for stochastic partial differential equations ( spdes ) with brownian forcing in the conclusion section of @xcite without any mathematical construction or numerical example . we also stress an important difference between our approach and @xcite : our scheme solely depends on the statistical information , i.e. , moments , which are available through chaos expansion , whereas @xcite either requires the pdf of @xmath74 at @xmath50 or uses mappings to transform @xmath74 back to the original input random variables , which in our problem is high dimensional , including the @xmath42 variables . we now describe the implementation of our algorithm . provided that the distribution of the initial condition @xmath46 is known , several methods allow us to compute moments : analytic formulas , quadrature methods , or monte carlo sampling . also , in the case of limited data , moments can be generated from raw data sets . throughout the paper , we assume that the moments of the initial condition can be computed to some large finite order so that the algorithm can be initialized . we first provide the hermite product formula , which establishes the multiplication of two pce for random variables @xmath20 and @xmath98 . the chaos expansion for the product becomes @xmath99 with @xmath100^{1/2 } , $ ] @xcite . as our applications include nonlinear terms , this formula will be used repeatedly to multiply hermite chaos expansions . to compute moments of the random variable @xmath74 ( recall ) at time step @xmath50 , @xmath101 , we use its chaos representation and previously computed triple products recursively as follows . due to the markov property of the solution , the normalized chaos basis in @xmath102 and @xmath103 becomes a tensor product , and thus we can write @xmath104 where @xmath105 and @xmath106 . then for an integer @xmath107 , @xmath108 here , @xmath109 and @xmath110 . hence , using , the coefficients @xmath111 , where @xmath112 and @xmath113 , can be calculated using the expression @xmath114 \ , ( u_j^{m-1})_{\gamma-\theta+\alpha , l}\,(u_j)_{\theta+\alpha , k},\end{aligned}\ ] ] which allows us to compute moments by simply taking the first coefficient , i.e. @xmath115 = ( u_j^m)_{\boldsymbol{0},0}$ ] . even though the coefficients @xmath116 can be computed offline , the triple products @xmath117 $ ] must be computed at every step as the distribution of @xmath77 evolves . given the real random variable @xmath74 obtained by the orthogonal projection of the solution @xmath20 onto homogeneous chaos at @xmath50 , the gram schmidt orthogonalization procedure can be used to construct the associated orthogonal polynomials of its continuous distribution . for theoretical reasons , we assume that the moment problem for @xmath74 is uniquely solvable ; i.e. , the distribution is nondegenerate and uniquely determined by its moments @xcite . to construct orthogonal polynomials , the classical gram schmidt procedure uses the following recursive relation @xmath118 where the coefficients are given by @xmath119 \left({\mathbb{e}}[t^2_l(u_j)]\right)^{-1}$ ] . note that @xmath19 denotes the lebesgue stieltjes integral with respect to distribution of @xmath74 . we observe that the coefficient @xmath120 requires moments up to degree @xmath121 as @xmath122 is at most of degree @xmath123 . thus , normalization would need first @xmath124 moments for orthonormal polynomials up to degree @xmath125 . after the construction of orthonormal polynomials of @xmath74 , the basis @xmath126 can be generated by tensor products since @xmath127 and @xmath74 are independent . moreover , triple products @xmath128 $ ] , where @xmath129 can be found by simply noting that this expectation is a triple sum involving moments up to order @xmath130 . hence , the knowledge of moments yields not only the orthonormal set of polynomials but also the triple products required at each iterative step . in our numerical computations , we prefer to utilize a modified gram schmidt algorithm as the classical approach is overly sensitive to roundoff errors in cases of high dimensionality @xcite . we refer the reader to @xcite for other algorithms , e.g. , stieltjes and modified chebyshev methods . when the algorithm is reinitialized at the restart point @xmath50 , it needs to construct the initial condition in terms of chaos variables for the next iteration on @xmath65 $ ] . in other words , we need to find the coefficients @xmath131 in terms of @xmath74 and @xmath127 . to this end , we observe that the first two polynomials in @xmath74 are given by @xmath132 and @xmath133 ) , $ ] where @xmath134 is the standard deviation of the random variable @xmath74 . notice that @xmath135 , because we assumed in the preceding section that the distribution was nondegenerate . hence , the initial condition can be easily found by the relation @xmath136 t_0(u_j ) + \sigma_{u_j } t_1(u_j ) = \sum_{\alpha \in \mathcal{j } , \ , k \in \mathbb{n}_0 } ( u_j)_{\alpha , k } \ , t_{\alpha}({\boldsymbol{\xi}}_{j } ) \,t_k(u_{j}).\end{aligned}\ ] ] we now extend our algorithm to higher dimensions , and for concreteness , we consider two dimensions . let @xmath137 be a two - dimensional random variable obtained by projection of the solution @xmath20 onto homogeneous chaos space at time @xmath138 . the two - dimensional case requires the calculation of the mixed moments @xmath139 , l , m \in \mathbb{n } \cup \{0 \}$ ] . in the case of independent components @xmath140 , the moments become products of marginal moments and the basis @xmath141 is obtained by tensor products of the one - dimensional bases . for a nonlinear system of equations , the components of the solution typically do not remain independent as time evolves . therefore , we need to extend the procedure to correlated components . denoting by @xmath142 and @xmath143 the corresponding chaos expansions on @xmath65 $ ] so that @xmath144 and @xmath145 , we compute the mixed moments by the change of variables formula @xmath146= { \mathbb{e}}_{(v_1,v_2)}[v^l_1 \ , v^m_2 ] = { \mathbb{e}}_{({\boldsymbol{\xi}}_{j},u_{j } ) } \left[z_{j+1,1}^l \ , z_{j+1,2}^m \right],\end{aligned}\ ] ] where @xmath147 represents the expectation with respect to the joint distribution induced by the subscript . note that the latter integral in can be computed with the help of previously computed triple products of @xmath63 and @xmath77 . incidentally , similar transformation methods were used previously in @xcite . since the components may not be independent , the tensor product structure is lost but the construction of orthogonal polynomials is still possible . based on the knowledge of marginal moments , we first compute the tensor product @xmath141 . however , this set is not orthogonal with respect to the joint probability measure of @xmath148 . therefore , we further orthogonalize it via a gram schmidt method . note that this procedure requires only mixed moments @xmath149 $ ] , which have already been computed in the previous step of the algorithm . it is worth noting that the resulting set of orthogonal polynomials is not unique as it depends on the ordering of monomials @xcite . in applications , we consider the same ordering used for the set of multi - indices . finally , calculation of triple products and initial conditions can be extended in an obvious way . in this section , we discuss the computational costs of our algorithm using a vector - valued version of the sde in @xmath150 for a fixed @xmath151 . for each scalar sde , we assume that the nonlinearity is proportional to @xmath125th order monomial with @xmath152 . furthermore , we compare costs of dgpc and hermite pc in the case where both methods attain a similar order of error . throughout this section , @xmath85 will denote the dimension and the degree in @xmath153 for hermite pc . here @xmath14 represents the truncation of @xmath94-dimensional brownian motion ; therefore , @xmath154 . the time interval is fixed and given by @xmath155 $ ] . we also let @xmath156 @xmath157 be the degrees of freedom for the simple truncation so that the dimension of truncated multi - indices @xmath158 becomes @xmath159 . for the dgpc method , we divide the interval into @xmath160 identical subintervals so that @xmath161 . we now slightly change the notation of previous sections and denote by @xmath162 the dimension and degree of polynomials of @xmath163 used in dgpc approximation for each @xmath64 . with additional @xmath94 variables at each restart , the dimension of the multi - index set for dgpc becomes @xmath164 due to the tensor product structure . further , let @xmath165 and @xmath166 denote the time step and global order of convergence for the time integration method employed in hermite pc , respectively . with the above notation , we summarize the computational costs in table [ table : compcost ] . all terms should be understood in big @xmath167 notation . r | p6.5cm|p3.9 cm flops & dgpc & hermite pc + offline & @xmath168 & @xmath169 + time evolution & @xmath170 $ ] & @xmath171 $ ] + moments & @xmath172 $ ] & + gram schmidt & @xmath173 & + triple products & @xmath174 & + initials & @xmath175 & + the estimates in table [ table : compcost ] are obtained as follows . triple products for @xmath14 with @xmath176 can be calculated by the equation @xmath177 & = { \mathbb{e}}\left [ \prod_{i=1}^k h_{\alpha_i } ( \xi_i ) \prod_{j=1}^k h_{\beta_j } ( \xi_j ) \prod_{k=1}^k h_{\gamma_k } ( \xi_k ) \right]= \prod_{i=1}^k { \mathbb{e}}[h_{\alpha_i}(\xi_i)h_{\beta_i}(\xi_i)h_{\gamma_i}(\xi_i)].\end{aligned}\ ] ] thus , the offline cost is of order @xmath178 assuming that one - dimensional triple products can be read from a precomputed table . since a time discretization with error @xmath179 is employed for hermite pc , then for dgpc in each subinterval , time steps of order @xmath180 should be used to attain the same global error . without nonlinearity , the total cost in hermite pc for evolution of a @xmath94-dimensional system becomes @xmath181 . due to the functional @xmath182 , the cost should also include the computation of @xmath183 at each time integration step . this requires additional @xmath184 work ( see the computation of moments below ) . hence , the corresponding total cost of evolution becomes the sum of these costs . the coefficients of the @xmath185th power of a single projected variable @xmath74 are given by @xmath186 , \quad 2 \leq k \leq 3n_*,\ ] ] assuming that , after each multiplication , the variable is projected onto the pc basis . here , @xmath187 . thus , the computation of marginal moments of all variables requires @xmath188 calculations at each restart time in dgpc . since we use the same ordering for multidimensional moments as @xmath189 , computing joint moments further needs @xmath190 amount of work . the gram schmidt procedure costs are cubic in the size of the gram matrix , which has dimension @xmath191 in dgpc . each triple product @xmath192 , \alpha,\beta,\gamma \in \mathcal{j}_{d , n_*}$ ] , is a triple sum ; therefore , the total cost is proportional to @xmath193 . also , initial conditions can be computed by orthogonal projection onto the dgpc basis , which costs @xmath164 per dimension . note that although we employed simple truncation for multi - indices , sparse truncation techniques can be utilized to further reduce the costs of both computing moments and evolution ; see @xcite . we now discuss the error terms in both algorithms . we posit that the hermite pce converges algebraically in @xmath25 and @xmath82 and that the error at time @xmath40 satisfies @xmath194 and @xmath195 depending on the sde . here @xmath196 denotes that @xmath197 for @xmath198 with @xmath199 . this assumption enforces an increase in the degrees of freedom @xmath200 if one wants to maintain the accuracy in the long term . algebraic convergence in @xmath82 and the term @xmath201 stems from the error bound . we do not show errors resulting from time discretization in since we already fixed those errors to the same order in both methods . incidentally , even though we assumed algebraic convergence in @xmath25 , exponential numerical convergence , depending on the regularity of the solution in @xmath14 , is usually observed in the literature @xcite . although it is usually hard to quantify the constants @xmath202 , this may be done for simple cases . for instance , for the ornstein uhlenbeck ( ou ) process , using the exact solution and fourier basis of cosines , we can obtain the parameters @xmath203 and @xmath204 for short times . the parameter @xmath205 does not play a role in this case since the sde stays gaussian and hence we take @xmath206 . we also note that for complex dynamics , @xmath205 may depend on @xmath25 ; see @xcite in the case of the stochastic burgers equation and advection equations . now , we fix the error terms in to the same order @xmath207 , @xmath56 , by choosing @xmath208 . since dgpc uses truncated @xmath14 of dimension @xmath209 and polynomials of degree @xmath210 in each subinterval of size @xmath211 , the @xmath18 error at @xmath212 has the form @xmath213 of accuracy , we can choose @xmath214 from table [ table : compcost ] , we observe that to minimize costs for dgpc , we can take @xmath215 and @xmath216 and maximize @xmath12 according to the previous equation as @xmath217 with these choices of parameters , the total computational cost for dgpc is of algebraic order in @xmath82 and in general , is dominated by the computation of moments . for hermite pc with large @xmath82 and @xmath218 ( or @xmath219 in the case of exponential convergence in @xmath25 ) , both offline and evolution stages include the cubic term @xmath220 . both costs increase exponentially in @xmath82 . therefore , the asymptotic analysis suggests that dgpc can be performed with substantially lower computational costs using frequent restarts ( equation ) in nonlinear systems driven by high dimensional random forcing . we now consider the convergence properties of our scheme as degrees of freedom tend to infinity . there is an extensive literature on the moment problem for probability distributions ; see @xcite and their references . we provide the relevant background in the analysis of dgpc . _ ( hamburger moment problem ) _ for a probability measure @xmath221 on @xmath222 , the moment problem is uniquely solvable provided moments of all orders @xmath223 , @xmath224 exist and they uniquely determine the measure . in this case , the distribution @xmath221 is called _ there are several sufficient conditions for a one - dimensional distribution to be determinate in hamburger sense : these are compact support , exponential integrability and carleman s moment condition @xcite . for instance , gaussian and uniform distributions are determinate , whereas the lognormal distribution is not . the moment problem is intrinsically related to the density of associated orthogonal polynomials . indeed , if the cumulative distribution @xmath225 of a random variable @xmath20 is determinate , then the corresponding orthogonal polynomials constitute a dense set in @xmath226 and therefore also in @xmath227 @xcite . additionally , finite dimensional distributions on @xmath150 with compact support are determinate ; see @xcite . now , let @xmath228 denote an independent collection of general random variables , where each @xmath229 ( not necessarily identically distributed ) has finite moments of all orders and its cumulative distribution function is continuous . under these assumptions , @xcite proves the following theorem . [ thm : conv_ernst ] for any random variable @xmath230 , gpc converges to @xmath205 in @xmath18 if and only if the moment problem for each @xmath229 , @xmath231 , is uniquely solvable . this result generalizes the convergence of hermite pce to general random variables whose laws are determinate . notably , to prove @xmath18 convergence of chaos expansions , it is enough to check one of the determinacy conditions mentioned above for each one - dimensional @xmath229 . we now consider the relation between determinacy and distributions of sdes . consider the following diffusion @xmath232,\end{aligned}\ ] ] where @xmath54 is @xmath94-dimensional brownian motion , @xmath233 , and @xmath234 are globally lipschitz and satisfy the usual linear growth bound . these conditions imply that the sde has a unique strong solution with continuous paths @xcite . determinancy of the distribution of @xmath235 is established by the following theorem ; see @xcite for details . [ thm : det_sde ] the law of the solution of is determinate if @xmath236 is finite . consider the solution of the sde written as @xmath237 where @xmath238 is the exact evolution operator mapping the initial condition @xmath46 and @xmath239 to the solution @xmath235 at time @xmath240 . here , with a slight change of notation , @xmath241 represents brownian motion on the interval @xmath242 $ ] . note that future solutions @xmath243 , @xmath244 , are obtained by @xmath245 , where @xmath246 denotes brownian motion on the interval @xmath247 $ ] . even though the distribution of the exact solution @xmath235 is determinate under the hypotheses of thm . [ thm : det_sde ] , it is not clear that this feature still holds for the projected random variables . to address this issue , we introduce , for @xmath248 , a truncation function @xmath249 @xmath250 where @xmath251 \subset { \mathbb{r}}^d$ ] , and such that @xmath252 decays smoothly and sufficiently slowly on @xmath253 such that the lipschitz constant @xmath254 of @xmath255 equals @xmath256 ; i.e. , for @xmath257 , the euclidean distance in @xmath150 , @xmath258 . such a function is seen to exist by appropriate mollification for each coordinate using the continuous , piecewise smooth , and odd function @xmath259 defined by @xmath260 on @xmath261 $ ] , and extended by @xmath262 outside @xmath263 $ ] . this truncation is a theoretical tool that allows us to ensure that all distributions with support properly restricted on compact intervals remain determinate ; see @xcite . we can now introduce the following approximate solution operators for @xmath248 and @xmath264 : @xmath265 where @xmath266 denotes the orthogonal projection onto polynomials in @xmath20 and @xmath241 , and @xmath159 is the total number of degrees of freedom used in the expansion . throughout this section , we use a linear indexing to represent polynomials in both @xmath20 ( restricted on @xmath267 ) and the random variables @xmath14 . the main assumptions we imposed on the solution operator @xmath268 are summarized as follows : a. @xmath269 , where @xmath270 and @xmath271 with @xmath272 . [ item : zero ] b. @xmath273 , where @xmath274 , @xmath275 , and @xmath271 with @xmath272 . [ item : one ] c. for @xmath56 and @xmath276 , there is @xmath277 so that @xmath278 , where @xmath279 . [ item : two ] assumption [ item : zero ] is a stability estimate ensuring that the chain remains bounded in @xmath280th norm starting from a point in @xmath150 ( here @xmath262 without loss of generality owing to [ item : one ] ) . assumption [ item : one ] is a lipschitz growth condition controlled by a constant @xmath281 . these assumptions involve the @xmath282 norm , and hence are slightly stronger than control in the @xmath18 sense . note that @xmath283 also satisfies assumption [ item : one ] with the same constant @xmath281 since @xmath284 . finally , assumption [ item : two ] is justified by the weierstrass approximation theorem for the @xmath20 variable in @xmath267 and by the cameron martin theorem to handle the @xmath14 variable . consider now the following versions of the markov chains : @xmath285 where @xmath286 ( possibly random and independent of the variables @xmath14 ) and @xmath287 is a sequence of i.i.d . gaussian random variables representing brownian motion on the interval @xmath288 $ ] . we also use the notation @xmath289 to denote brownian motion on the interval @xmath290 $ ] . then we have the first result . [ lemma : rapprox ] assume [ item : zero ] , [ item : one ] , and [ item : two ] hold , and let @xmath291 be finite . suppose also that @xmath292 . then , for each @xmath56 , there exists @xmath248 depending on @xmath12 , @xmath293 , and @xmath294 such that @xmath295 for each @xmath296 . let @xmath56 be fixed . using properties [ item : zero ] and [ item : one ] , we observe that @xmath297 where for all @xmath83 , @xmath127 and @xmath241 are identically distributed . then , by induction , we obtain @xmath298 where the constant @xmath299 is bounded . the last inequality indicates that @xmath280th norm of the solution grows with @xmath83 , possibly exponentially , but remains bounded . for @xmath300 , we note that @xmath301 since @xmath74 is independent of future forcing . by markov s inequality and , we get @xmath302 let @xmath303 be the indicator function of the set @xmath304 . using and , we compute the following error bound @xmath305 , \quad 1 \leq j \leq n,\\ \nonumber & \leq 4 \ , { \mathbb{e } } [ \mathbbm{1 } _ { \ { u_{j } \in a_r^c \ } } |u_{j } |^2 ] \leq 4 \ , ( { \mathbb{e}}|u_j|^p ) ^{2/p } \ , { \mathbb{p}}(u_j \in a_r^c)^{1/q } , \quad p=2+\epsilon , q= 1 + 2\epsilon^{-1 } , \\ \label{eq : ujrbound } & \leq c_{s , n , c_0 } \ , r^{-2/q}.\end{aligned}\ ] ] then , using property [ item : one ] gives @xmath306 taking expectation with respect to remaining measures and using yields @xmath307 setting @xmath308 in entails @xmath309 since @xmath310 , we also have @xmath311 . thus , we can choose @xmath312 large enough such that @xmath313 for each @xmath296 . based on the preceding lemma , we prove that the chain converges to the solution of the chain in the setting of a finite number of restarts . [ thm : convfinite ] under the assumptions of lemma [ lemma : rapprox ] , for each @xmath56 , there exists @xmath314 and then @xmath315 such that @xmath316 for each @xmath296 . let @xmath317 be fixed . by the previous lemma , choose @xmath276 such that @xmath318 for @xmath296 . then , using calculations similar to those above , by [ item : one ] we get @xmath319 where @xmath320 and @xmath321 . then by [ item : two ] , we choose @xmath322 sufficiently large so that @xmath323 the last inequality can be rewritten as @xmath324 therefore , @xmath325 for each @xmath326 . consider the setting of the solution operator to the sde given in . as @xmath293 increases to @xmath327 , the random variable @xmath328 may converge in distribution to a limiting random variable @xmath329 , whose distribution is the invariant measure of the evolution equation . although there are many efficient ways to analyze and compute such invariant measures ( see for instance @xcite ) , we wish to show that our iterative algorithm also converges to that invariant measure as degrees of freedom tend to infinity . in other words , our pce - based method allows us to remain accurate for long - time evolutions something that can not be achieved without restarts . now we iterate each discrete chain in , and for an arbitrary @xmath330 . in order to ensure long - time convergence , we need a stricter condition than [ item : one ] and impose instead the following contraction condition 1 . @xmath273 , where @xmath331 , @xmath332 @xmath272 , and @xmath274 . [ item : onemod ] we first prove the following result about the existence and uniqueness of an invariant measure for the original chain . [ lemma : invariant ] under conditions [ item : zero ] and [ item : onemod ] , there exists a unique invariant measure @xmath333 of the chain with bounded @xmath280th moment . moreover , if @xmath310 , then @xmath334 is bounded uniformly in @xmath83 . for @xmath335 ( and using that @xmath336 ) , we compute @xmath337 where @xmath338 is chosen so that @xmath339 and @xmath340 . thus , there exists a lyapunov function @xmath341 with a constant @xmath342 . this in turn implies the existence of an invariant measure @xmath333 with finite @xmath280th moment for the process ; see ( * ? ? ? * corollary 4.23 ) . uniqueness also follows from assumption [ item : onemod ] ; see ( * ? ? ? * theorem 4.25 ) . let @xmath343 be a random variable with law @xmath333 and independent of @xmath46 . consider the chain @xmath344 started from @xmath343 . note @xmath345 for all @xmath330 . then , from the bound @xmath346 we conclude that @xmath347 . the following theorem establishes the exponential convergence of the pc chain to the chain as time @xmath83 increases . assume [ item : zero ] , [ item : onemod ] and [ item : two ] hold , and @xmath310 . then , for each @xmath56 , there exists @xmath276 independent of @xmath83 , and then @xmath348 , such that @xmath316 for all @xmath349 . let @xmath317 be given . by lemma [ lemma : invariant ] , @xmath334 is bounded uniformly in @xmath83 . the inequality holds with a constant independent of @xmath83 . then , choosing @xmath350 so that @xmath351 in implies that there exists a constant @xmath352 independent of @xmath83 such that @xmath353 . as in the proof of theorem [ thm : convfinite ] , we choose @xmath350 so that @xmath354 . then , with @xmath348 large enough , is replaced by @xmath355 therefore , @xmath356 this concludes our proof of convergence of the dgpc method to the invariant measure @xmath333 . let us conclude this section with the following remark . it would be desirable to obtain that the chain ( corresponding to @xmath357 , or equivalently no support truncation ) @xmath358 remains determinate . we were not able to do so and instead based our theoretical results on the assumption that the true distributions of interest were well approximated by compactly supported distributions . in practice , the range of @xmath159 has to remain relatively limited for at least two reasons . first , large @xmath159 rapidly involves very large computational costs ; and second , the determination of measures from moments becomes exponentially ill - posed as the degree of polynomials @xmath25 increases . for these reasons , the support truncation has been neglected in the following numerical section since , heuristically , for large @xmath359 and limited @xmath25 , the computation of ( low order ) moments of distributions with rapidly decaying density at infinity is hardly affected by such a support truncation . in this section , we present several numerical simulations of our algorithm and discuss the implementation details . we consider several of the equations described in @xcite to show that pce - based simulations may perform well in such settings . we mostly consider the following two - dimensional nonlinear system of coupled sdes : @xmath360 where @xmath361 , @xmath362 are damping parameters , @xmath363 are constants , and @xmath364 and @xmath365 are two real independent wiener processes . the system was proposed in @xcite to study filtering of turbulent signals which exhibit intermittent instabilities . the performance of hermite pce is analyzed in various dynamical regimes by the authors in @xcite , who conclude that truncated pce struggle to accurately capture the statistics of the solution in the long term due to both the truncated expansion of the white noise and neglecting higher order terms , which become crucial because of the nonlinearities . for a review of the different dynamical regimes that exhibits , we refer the reader to @xcite . for the rest of the section , @xmath366 stands for the endpoint of the time interval while @xmath367 denotes the time - step after which restarts occur at @xmath368 . moreover , @xmath82 denotes the number of basic random variables @xmath27 used in the truncation of the expansion . in the presence of two wiener processes it will denote the total number of variables . we slightly change the previous notation and let @xmath25 and @xmath369 denote the maximum degree of the polynomials in @xmath14 and @xmath74 , respectively . recall that @xmath74 is the projected pc solution at @xmath50 . furthermore , following @xcite we choose the orthonormal bases for @xmath370 $ ] as @xmath371.\end{aligned}\ ] ] other possible options for @xmath372 include sine functions , a combination of sines and cosines , and wavelets . we refer the reader to @xcite for details on the use of wavelets to deal with discontinuities for random inputs . assuming that the solution of is square integrable , we utilize intrusive galerkin projections in order to establish the following equations for the coefficients of the pce : @xmath373 , \\ \dot{v}_{\alpha } & = -b_v\ , v_{\alpha}+ a_v \ , ( uv)_{\alpha } + \sigma_v \,e[\dot{w}_v t_{\alpha}].\end{aligned}\ ] ] this system of odes is then solved by either a second- or a fourth - order time integration method . finally , we note that other methods are also available to compute the pce coefficients such as nonintrusive projection and collocation methods @xcite . in most of our numerical examples , the dynamics converge to an invariant measure . to demonstrate the convergence behavior of our algorithm , we compare our results to exact second order statistics or monte carlo simulations with sufficiently high sampling rate @xmath374 ( e.g. , euler maruyama or weak runge kutta methods @xcite ) where the exact solution is not available . comparisons involve the following relative pointwise errors : @xmath375 where @xmath376 represents the mean and variance . in the following figures , we plot the evolution of mean , variance , @xmath377 and @xmath378 using the same legend . in addition , we exhibit the evolution of higher order cumulants , which will be denoted by @xmath379 to demonstrate the convergence to steady state as time grows . finally , in our numerical computations , we make use of the c++ library uq toolkit @xcite . as a first example , we consider the one - dimensional ornstein - uhlenbeck ( ou ) process @xmath380 , \quad v(0 ) = v_0,\end{aligned}\ ] ] on @xmath381 $ ] with the parameters @xmath382 . we first present the results of hermite pce . figure [ fig : ex1_1 ] shows that hermite pc captures the mean accurately but approximation for the variance is accurate only at the endpoint . this is a consequence of the expansion in , which violates the markov property of brownian motion and is inaccurate for all @xmath383 , while it becomes ( spectrally ) accurate at the endpoint @xmath40 . using frequent restarts , these oscillations are significantly attenuated as expected ; see figure [ fig : ex1_2b ] . this oscillatory behavior could also be alleviated by expanding the brownian motion as in on subintervals of @xmath384 . note that the global basis functions @xmath385 may also be filtered by taking the conditional expectation @xmath386 $ ] , where @xmath387 is the @xmath3-algebra corresponding to brownian motion up to time @xmath293 @xcite . we do not pursue this issue here but note that the accuracy of the different methods should be compared at the end of the intervals of discretization of brownian motion . next , we demonstrate numerical results for dgpc taking @xmath206 , @xmath388 and a varying @xmath389 . figure [ fig : ex1_2 ] shows that as the exact solution converges to a steady state , our algorithm captures the second order statistics accurately even with a small degrees of freedom utilized in each subinterval . moreover , although we do not show it here , it is possible , by increasing @xmath369 , to approximate the higher order zero cumulants @xmath390 , with an error on the order of machine precision , which implies that the algorithm converges numerically to a gaussian invariant measure . in figure [ fig : ex1_3 ] , we illustrate an algebraic convergence of the variance in terms of the dimension @xmath82 for @xmath391 and @xmath392 . dgpc uses the same size of interval @xmath393 for both cases . for comparison , we also include the convergence behavior of hermite pce . values of @xmath82 are taken as @xmath394 for figures [ fig : ex1_3a ] and [ fig : ex1_3c ] and @xmath395 for figures [ fig : ex1_3b ] and [ fig : ex1_3d ] . we deduce that our algorithm maintains an algebraic convergence of order @xmath396 for both @xmath391 and @xmath392 whereas the convergence behavior of hermite pce drops from @xmath396 to @xmath397 as time increases . this confirms the fact that the degrees of freedom required for standard pce to maintain a desired accuracy should increase with time . we now introduce a nonlinearity in the equation so that the damping term includes a cubic component : @xmath398 figure [ fig : ex2_1 ] displays several numerical simulations for the degree of polynomials in @xmath14 given by @xmath399 and @xmath400 . this is compared with the weak runge kutta method using @xmath401 and @xmath402 . we observe that increasing @xmath25 improves the accuracy of the solution to this nonlinear equation as expected . moreover , table [ table : cubic ] presents a comparison for statistical cumulants between our method with @xmath403 , @xmath404 , @xmath405 and monte carlo ( mc ) at time @xmath406 . the ( stationary ) cumulants of the invariant measure are also estimated by solving a standard fokker planck ( fp ) equation @xcite for the invariant measure . we conclude this example by noting that the level of accuracy of approximations in dgpc for cumulants is similar to that of the mc method . |r | r | r |r|r|r|r| & @xmath407&@xmath408 & @xmath409 & @xmath410&@xmath411 & @xmath412 + dgpc & 3.58e-5 & 7.33e-1 & -2.08e-5 & -3.37e-1 & 6.02e-5 & 9.35e-1 + mc & -2.53e-3 & 7.33e-1 & 7.85e-4 & -3.38e-1 & -3.34e-3 & 9.58e-1 + fp & 0 & 7.33e-1 & 0 & -3.39e-1 & 0 & 9.64e-1 + as a third example we consider an ou process in which the damping parameter is random and uniformly distributed in @xmath413 $ ] , i.e. @xmath414 . this is an example of non - gaussian dynamics that may be seen as a coupled system for @xmath415 with @xmath416 . we consider a time domain @xmath417 $ ] and divide it into @xmath418 subintervals . the initial condition is normally distributed @xmath419{90}{$\models$}}w_v$ ] and @xmath400 . in the next figure , we compare second order statistics obtained by our method to monte carlo method for which we use the euler maruyama scheme with the time step @xmath420 and sampling rate @xmath421 implying @xmath422 samples in total . we stress again that this problem is essentially two - dimensional since the damping is random . as expected , the mean decreases monotonically and is approximated accurately by the algorithm . the estimation of the variance becomes more accurate as @xmath25 increases . furthermore , figures [ fig : ex3_1c ] and [ fig : ex3_1d ] show that the cumulants become stationary for long times indicating that the numerical approximations converge to a measure which is non - gaussian . table [ table : ex3 ] compares the first six cumulants obtained by our algorithm at time @xmath423 with @xmath424 , @xmath425 , @xmath405 to the monte carlo method . for further comparison , we also provide cumulants obtained by averaging fokker planck density with respect to the known , explicit , distribution of the damping . it can be observed from the following table that both our algorithm and monte carlo capture cumulants reasonably well although the accuracy degrades at higher orders . |r | r | r |r|r|r|r| & @xmath407&@xmath408 & @xmath409 & @xmath410&@xmath411 & @xmath412 + dgpc & 1.68e-5 & 1.10 & 3.80e-5 & 3.78e-1 & 8.72e-5 & 5.04e-1 + mc & -1.78e-4 & 1.10 & -3.25e-3 & 3.70e-1 & -6.34e-2 & 4.65e-1 + fp & 0 & 1.10 & 0 & 3.79e-1&0 & 5.29e-1 + the above calculations provide an example of stochasticity with two distinct components : the variables @xmath42 that are markov and can be projected out and the non - markov variable @xmath426 that has long - time effects on the dynamics . the variables @xmath415 are strongly correlated for positive times . pce thus need to involve orthogonal polynomials that reflect these correlations and can not be written as tensorized products of orthogonal polynomials in @xmath98 and @xmath426 . we demonstrate that our method is not limited to equations forced by brownian motion . we apply it to an equation where the forcing is a nonlinear function of brownian motion : @xmath427 with parameters @xmath428 , @xmath429 . to depict the effect of different choices of number of restarts , we take @xmath430 using @xmath403 , @xmath404 , @xmath431 in each expansion . from figure [ fig : ex4 ] , we observe that second order statistics are captured accurately and an increasing number of restarts provides better approximations as anticipated . although the system does not converge to a steady state ( variance of @xmath432 increases linearly ) , this example illustrates that dgpc is able to capture behaviors of solutions where the forcing is a nonlinear function of wiener process . this example concerns the nonlinear system , where the dynamics of @xmath20 exhibit intermittent non - gaussian behavior . a time dependent deterministic periodic forcing is considered with the second equation being an ou process , i.e. @xmath433 in , and the parameters are taken as @xmath434 , @xmath435 , @xmath436 , @xmath437 , @xmath438 , @xmath439 , with initial conditions @xmath440{90}{$\models$}}v_0 = n(0 , { \sigma_v^2/8b_v } ) $ ] ; see also @xcite . the initial variables are independent of each other and of the stochastic forcing . intermittency is introduced by using an ou process , which acts as a multiplicative noise fluctuating in the damping . the second order statistics for this case ( @xmath433 ) can be derived analytically @xcite and we plot them in black in the following figures using quadrature methods with sufficiently high number of quadrature points . we first depict the results for @xmath20 obtained by standard hermite pce in the first row of figure [ fig : ex6_1 ] . it can be observed that truncated pc approximations are accurate for short times . however , they quickly lose their accuracy as time grows . this is consistent with the simulations presented in @xcite . the second row of the same figure shows that the accuracy is vastly improved in dgpc using @xmath441 , and second order statistics are accurate up to three or four digits . moreover , errors in the variance oscillate around @xmath442 for @xmath443 throughout the time length suggesting that the approximation retains its accuracy in the long term , which is consistent with theoretical predictions . our algorithm easily outperforms standard pce in capturing the long - time behavior of the nonlinear coupled system in this scenario . + for this scenario , we consider the system parameters as @xmath444 , @xmath433 , @xmath445 , @xmath446,@xmath447 , @xmath448 without the deterministic forcing @xmath449 and with the initials @xmath450{90}{$\models$}}\ , v_0 = n(0,{\sigma_v^2/8b_v } ) . $ ] in this regime , the dynamics of @xmath328 are characterized by unstable bursts of large amplitude @xcite . second order statistics obtained by hermite pce and our algorithm are presented in figure [ fig : ex7_2 ] , where the first two parts ( figures [ fig : ex7_1a ] and [ fig : ex7_1b ] ) correspond to hermite pce . we observe from figure [ fig : ex7_1b ] that the relative errors for standard hermite expansions are unacceptably high . on the other hand , as the degrees of freedom is increased in dgpc , there is a clear pattern of convergence of second order statistics ; see figures [ fig : ex7_2e ] and [ fig : ex7_2f ] . note , however , that short - time accuracy is better than accuracy for long times . this behavior may be explained by the onset of intermittency as nonlinear effects kick in after short times . further , figure [ fig : ex7_2 g ] shows the error behavior of variance at @xmath451 as @xmath25 and @xmath369 vary , and @xmath82 is fixed . the rate of convergence is consistent with exponential convergence in @xmath25 and @xmath369 as the logarithmic plot is almost linear . for similar convergence behaviors , see @xcite . + + finally , in figure [ fig : ex7_3 ] , we present the evolution of the first few cumulants obtained by dgpc , which shows that the system converges to a steady state distribution as time increases . bivariate cumulants are ordered so that first and second subscripts correspond to @xmath20 and @xmath98 , respectively . figure [ fig : ex7_3f ] shows kurtosis excess for @xmath452 and clearly indicates that the dynamics of @xmath98 stay gaussian , whereas those of @xmath20 converge to a non - gaussian state . as a final example , using the same set of parameters of the previous example , we introduce a small perturbation to the second equation and set @xmath453 in . another choice of perturbation as @xmath454 was considered before in @xcite . comparing figurs [ fig : ex7_3 ] and [ fig : ex8 ] , we see that evolutions of higher order cumulants are perturbed in most cases as expected . in this case , it took a longer transient time to converge to an invariant measure as additional nonlinearity was introduced into the system . polynomial chaos expansions ( pce ) applied to the simulation of evolution equations suffer from two drawbacks : in the presence of stochastic forcing , the dimension of randomness is too large , and the long - time solution may not be sparsely represented in a fixed pce basis @xcite . in the setting of markovian random forcing , we have proposed a restart method that addresses the two aforementioned drawbacks . such restarts of the pce , which we called here dynamical generalized polynomial chaos ( dgpc ) , allow us to both to keep the number of random variables small and to obtain a solution that remains reasonably sparse in the evolving basis of orthogonal polynomials , as was done earlier in @xcite . following @xcite , we applied dgpc to the long - time numerical simulation of various stochastic differential equations ( sdes ) . to demonstrate the ability of the algorithm to reach long - time solutions , we computed invariant measures for sdes that admit one and found a very good agreement between dgpc and other standard methods such as monte carlo - based simulations . we also presented a simple theoretical justification for such a convergence . the main computational difficulty in dgpc , as in most gpc - based methods , is the estimation of the orthogonal polynomials of evolving arbitrary measures . our method , as in @xcite , is based on estimating moments of the multivariate distribution of interest and estimating orthogonal polynomials by a gram schmidt procedure . the bottleneck in such estimations is the cost of computing moments . this is similar to the cost of solving differential equations by pce methods , where moment estimations are also the most costly step . however , in dgpc , such estimations can not be performed offline . from a theoretical point of view , we need to ensure that the evolving measures remain _ determinate _ so that their orthogonal polynomials span square integrable functionals . since distributions with compact support are determinate , our theoretical results have been applied in the setting where sde solutions are well approximated by compactly supported distributions . in this connection , we note that the calculation of orthogonal polynomials from knowledge of moments is an ill - posed problem , which is a serious impediment to gpc methods in general . extension of the method to larger systems than those considered here , including to systems obtained by solving stochastic partial differential equations ( spdes ) is challenging . the method works reasonably well for low dimensional sdes and it needs modifications for larger systems . however , the general idea remains applicable . to keep the number of uncertain variables in check , the method has to be coupled with a compression technique such as karhunen loeve projection at each restart time . moreover , sparse truncation techniques further help to reduce costs . the analysis of extensions of dgpc to larger systems , faster methods to compute orthogonal polynomials , and adaptive restart procedures is the subject of ongoing research . this work was partially funded by afosr grant nsseff- fa9550 - 10 - 1 - 0194 and nsf grant dms-1408867 . height 2pt depth -1.6pt width 23pt , _ reduced chaos expansions with random coefficients in reduced - dimensional stochastic modeling of coupled problems _ , internat . methods engrg . , 97 ( 2014 ) , 352376 . , _ moment problems related to the solutions of stochastic differential equations _ , in stochastic theory and control ( lawrence , ks , 2001 ) , vol . 280 of lecture notes in control and inform . sci . , springer , berlin , 2002 , pp .

before a cell divides into two daughter cells , chromosomes are replicated resulting in two sister chromosomes embracing each other . each sister chromosome is bound to a separate proteinous structure , called kinetochore ( kt ) , that captures the tip of a filamentous protein , called microtubule ( mt ) . two oppositely oriented mts pull the two kts attached to two sister chromosomes thereby pulling the two sisters away from each other . here we theoretically study an even simpler system , namely an isolated kt coupled to a single mt ; this system mimics an _ in - vitro _ experiment where a single kt - mt attachment is reconstituted using purified extracts from budding yeast . our models not only account for the experimentally observed `` catch - bond - like '' behavior of the kt - mt coupling , but also make new predictions on the probability distribution of the lifetimes of the attachments . in principle , our new predictions can be tested by analyzing the data collected in the _ in - vitro _ experiments provided the experiment is repeated sufficiently large number of times . our theory provides a deep insight into the effects of ( a ) size , ( b ) energetics , and ( c ) stochastic kinetics of the kt - mt coupling on the distribution of the lifetimes of these attachments .

1310.1579

chromosomes , the genetic material of a cell , are duplicated and properly segregated before the cell divides into two daughter cells @xcite . each of the sister chromatids , that result from chromosome replication , is bound to a proteinous structure , called kinetochore ( kt ) that , in turn , is coupled to the plus ends of stiff polar filaments called microtubules ( mt ) @xcite . the negative ends of these mts are located at the poles of the fusiform structure , called mitotic spindle @xcite . the process of chromosome segregation , called mitosis @xcite , is carried out in eukaryotic cells by the dynamic mitotic spindle which is self - organized from its components for this purpose . there are strong indications that molecular motors @xcite are located at the kt - mt interface ( though not in all eukaryotes ) ; these motor proteins , which are capable of generating force by hydrolyzing atp , are believed to generate poleward force or anti - poleward force depending on the family to which they belong . however , polymerizing and depolymerizing mts are also found to make significant contributions to the forces that cause chromosomal movements @xcite . the force - generation capability of depolymerizing mts have been demonstrated _ in - vitro _ however , how a mt maintains contact with the kinetochore while depolymerizing from its tip and how it exerts force on the kt remain challenging open questions in spite of the recent progress in identifying the molecular components of a kt and their spatial organization @xcite , moreover , in most of the eukaryotic organisms multiple mts couple to each kt ; how the ( de-)polymerization of mutiple mts are coordinated is still a mystery . _ budding yeast _ is a much simpler system because each kt is coupled to only a single mt . but , even in budding yeast , the two embraced sister chromatids are coupled to two mts approaching them from the two opposite poles of the spindle . in this paper we theoretically study an even simpler system , namely an isolated kt coupled to a single mt ; this system mimics an _ in - vitro _ experiment @xcite where a single kt - mt attachment was reconstituted using purified extracts from budding yeast . in those _ in - vitro _ experiments the average lifetime was found to vary nonmonotonically with the externally applied load tension . in other words , the kt - mt attachment is stabilized ( instead of getting destabilized ) at sufficiently small load tension . similar tension - induced stabilisation of attachment between e - coli fimbriae and target cells have been interpreted in the past in terms of the concept of `` catch - bond '' @xcite . akiyoshi et al . @xcite developed a 2-state kinetic model that involves four phenomenological rate constants . by assuming the rate constants to vary exponentially with the load tension , they obtained a reasonably good fit of the model predictions with their experimental data . here we do not attempt any quantitative comparison with the experimental data reported in ref.@xcite . instead , our approach allows to derive _ analytical _ expressions for the lifetimes of the kt - mt attachments within the framework of the simple theoretical models that we develop here . these models describe the polymerization- depolymerization of the mt explicitly and capture the effects of its interaction with the coupler by a potential energy landscape that gets modified when the mt is subjected to an external tension . therefore , our results for the mean lifetimes provide deep physical insight into the interplay of opposing forces and competing kinetic processes that , together , determine the stability of the kt - mt coupler . we also make theoretical predictions on the probability distribution of the lifetimes which also indicate the probability of survival of an attachment at least for a time interval @xmath0 after the attachment is established . in principle , our new predictions can be tested by analyzing the data collected in the _ in - vitro _ experiments provided the experiment is repeated sufficiently large number of times . this paper is organized as follows . in sections [ sec - minimalcont ] and [ sec - minimaldisc ] we develop a _ minimal theoretical model _ of a device that couples a single mt with a single kt and subject it to an external tension . space is represented by a continuous variable in section [ sec - minimalcont ] and by a discrete variable in section [ sec - minimaldisc ] . our results on the distribution of the lifetimes of the model kt - mt attachments , derived analytically in section [ sec - minimalcont ] , are compared with the corresponding numerical data , obtained by numerical simulations , in section [ sec - minimaldisc ] . next , in section [ sec - realistic ] , we extend the minimal model by representing the kt - mt interaction by a more realistic potential . calculation of the full distribution of the lifetimes in this case is too difficult to be carried out analytically . therefore , in this case we have directly calculated the mean lifetime of the kt - mt attachments and found that the realistic potential leads to quantitative changes without affecting the qualitative features of the kt - mt lifetimes observed in the minimal model . finally , motivated by recent evidence that the kt - mt coupler in mammalian cells might be a hybrid nano - device , composed of spatially separated active and passive components , we have extended our study also to a simple model of a hybrid coupler in appendix [ sec - hybrid ] . in the first version of the minimal model space is represented by a continuous variable and the kinetics of the system is formulated in terms of a fokker - planck equation . we find it more convenient to derive our analytical results using this formulation . for the convenience of numerical simulations , in the next section , we discretize the same model following prescriptions proposed earlier by wang , peskin and elston ( wpe ) @xcite . we also compare the results of the two versions . for the minimal model we do not need to make any assumption about the molecular constituents and their spatial organization in the coupler . the only postulates are as follows : + _ postulate ( a ) _ : the energy of the system is lowered monotonically with the increasing overlap between the inner surface of the cylindrically shaped coupler and the outer surface of the mt , and the corresponding binding energy is proportional to the projected length of this overlap along the mt axis ; + _ postulate ( b ) _ : the rate of depolymerization of a mt is suppressed by external force . to our knowledge , the postulate ( a ) is consistent with the scenario envisaged originally by hill @xcite , except for the fact that the `` roughness '' of the mt - coupler interface in the hill model , which is also an integral part of our realistic model ( described in the next section ) , is not captured by the minimal version of our model . the existence of the sleeve was postulated by hill @xcite long before information on the the molecular structure of mt - kinetochore coupler began to emerge from experiments . the current knowledge on the inventory of the mitotic machinery have identified the plausible candidates that give rise to the effectively sleeve - like coupler . the dam1 ring ( also called dash ) and the ndc80 complex seem to be the strongest candidates for the components of the coupler @xcite . the postulate ( b ) is supported by the observation of franck et al . @xcite in their _ in - vitro _ experiments that the rate of depolymerization of mt is suppressed by externally applied tension . it has been known from quite some time that the tips of the depolymerizing protofilaments are curled radially outward from the central axis of the cylindrically shaped mt . based on this fact , franck et al.@xcite speculated that the dam1 ring complex may transmit the external tension to the curled tips of the protofilaments thereby tending to straighten them . such straightening of the protofilaments is likely to suppress the tendency of the protofilaments to peel away from the depolymerizing tip of the mt . it is worth emphasizing that our results presented here require only the validity of the postulates ( a ) and ( b ) irrespective of the nature of the underlying cause of their validity . the length of the `` coupler''is denoted by @xmath1 . in the continuum formulation of the stochastic kinetics , the time - dependent variable @xmath2 denotes the instantaneous length of overlap between the outer surface of the mt and inner surface of the coaxial cylindrical coupler . thus , @xmath3 and @xmath4 corresponds to minimum and maximum overlap , respectively ; the mt is just on the verge of exiting the coupler when @xmath3 . guided by our postulate ( a ) , in the minimal version of our model we capture the mt - coupler interaction by a potential energy that is proportional to the overlap @xmath5 , i.e. , @xmath6 parametrized by @xmath7 , the depth of the potential at @xmath4 . the form of this interaction in the more realistic version of our model is given in section [ sec - realistic ] . similarly if the mt is pulled outward away from the coupler by a constant external force @xmath8 ( also referred to as a load tension ) , then this force can be derived from the corresponding potential , @xmath9 @xmath10 and @xmath11 are plotted in fig . ( [ potential ] ) . net potential felt by the kinetochore is @xmath12 and @xmath11 are plotted against the instantaneous length of overlap @xmath2 between the outer surface of the mt and inner surface of the coaxial cylindrical coupler . ] we use the symbols @xmath13 and @xmath14 to denote the addition and removal rates of the tubulin subunit from the mt . when subjected to a load tension , according to our postulate ( b ) , the depolymerization rate @xmath15 of the mt decreases with increasing strength of the tension @xmath8 @xcite . more specifically , we assume an exponential decrease of @xmath15 such that @xmath16 where @xmath17 is the depolymerization rate in the absence of any external force and @xmath18 is the characteristic load force at which the mt depolymerization rate is an exponentially small fraction of ( i.e. , a factor of @xmath19 smaller than ) @xmath20 . our motivation for assuming the specific exponential form ( [ eq - fdepbeta ] ) and the key role of the characteristic force @xmath18 will be discussed later in this paper . stochastic movement of the system can be described by a _ hypothetical diffusing particle in an external potential u(x)_. the drift caused by the addition and removal of the tubulin from mt give rise to an additional term @xmath21 to the drift velocity . note that the position of the hypothetical particle ( i.e. , the overlap @xmath5 ) can change by any of the following processes : + ( i ) even in the absence of any external force , the mt has a natural tendency to approach @xmath4 because the system can lower its energy by increasing the overlap between the outer surface of the mt and the inner surface of the coupler . in other words , the potential @xmath10 gives rise an effective force that spontaneously pulls the mt into the coupler . + ( ii ) the external load tension can pull the mt outward with respect to the coupler thereby decreasing the overlap @xmath5 . alternatively , an external force can push the mt into the coupler as long as @xmath5 remains non - zero . + ( iii ) the overlap @xmath5 can decrease , as long as it is non - zero , because of the depolymerization of the mt . + ( iv ) the overlap @xmath5 can increase because of the polymerization of mt . + ( v ) finally , even in the absence of polymerization or depolymerization and force - induced movements , the position @xmath5 can change because of its one - dimensional diffusion ; @xmath22 being the corresponding diffusion constant . therefore , in the overdamped limit , @xmath23 where @xmath24 is the phenomenological coefficient of the viscous drag and @xmath25 is a gaussian white noise . let @xmath26 denote the probability of finding the overlap @xmath5 at time @xmath0 . the fokker planck equation @xcite , @xmath27 that corresponds to the stochastic differential equation ( [ eq - langevin ] ) , is essentially a diffusion equation with an additional drift term where the time - independent , but tension - dependent , net drift velocity is @xmath28 and d is the diffusion constant . the interplay of the four key components are shown clearly in eqn.([myc ] ) . the external load tension @xmath8 competes against @xmath29 while mt depolymerization competes against polymerization . the load tension @xmath8 has two mutually opposite effects on the kinetics : ( i ) it directly tends to decrease @xmath5 by pulling out the mt from the coupler , and ( ii ) it indirectly assists increase of @xmath5 by suppressing @xmath15 that competes against @xmath13 . the equation ( [ main ] ) can also be written as an equation of continuity @xmath30 for the probability density @xmath26 , with the probability current density @xmath31 \label{eq - probcurr}\ ] ] the expression ( [ eq - probcurr ] ) is useful for calculations reported in the sections below . the specific initial condition and the boundary conditions that we impose are motivated by the physical situation that our models are intended to capture . if initial value of @xmath5 is @xmath32 , the corresponding initial condition for ( [ main ] ) can be expressed as @xmath33 throughout this paper we choose @xmath34 , i.e. , maximum overlap between the outer surface of the mt and innser surface of the coaxial cylindrical sleeve - like coupler . the lifetime of an arbitrary kt - mt attachment is defined here as the time @xmath35 taken by the hypothetical particle , that is initially at @xmath4 , to reach @xmath3 _ for the first time_. thus , @xmath35 is essentially a first passage time @xcite that fluctuates from one attachment to another . the average value @xmath36 is the mean life time of the attachment . although @xmath34 initially , @xmath5 does not necessarily decrease monotonically . at any arbitrary instant of time @xmath5 can also increase , just as it can decrease , following the dynamical equations provided @xmath37 . one advantage of the unique initial condition @xmath34 is that the random variation of the lifetime of the kt - mt attachment from one run to another arises strictly from the intrinsically random kinetics and not from the choice of any random initial condition . the distribution of the lifetimes of kt - mt coupler that we have derived does not require any further averaging . in contrast , if a random starting location were selected , one extra averaging ( over sufficiently large number of random initial locations ) would be required to get a meaningful distribution of the lifetimes . since the mt is not allowed to penetrate the kinetochore plate , and the length of the sleeve - like coupler is @xmath1 , the overlap @xmath5 can not be large than @xmath1 . therefore , at @xmath4 we impose a reflecting boundary condition @xmath38 which implies @xmath39 an absorbing boundary condition @xmath40 is imposed so that spontaneous re - formation of the kt - mt is not possible and the time taken for @xmath5 to attain this boundary , starting from its initial value @xmath41 is the first - passage time that we define as the life time of the attachment . ( a ) , for a few different values of @xmath18 , ( in ( a ) ) , and @xmath29 ( in ( b ) ) . the parameter values used for this plot are @xmath42s@xmath43 , @xmath44s@xmath43 . ] ( b ) , for a few different values of @xmath18 , ( in ( a ) ) , and @xmath29 ( in ( b ) ) . the parameter values used for this plot are @xmath42s@xmath43 , @xmath44s@xmath43 . ] we use the laplace inversion method @xcite for extracting the relevant quantitative information from equation ( [ main ] ) . defining the laplace transform of probability density by , @xmath45 equation ( [ main ] ) can be re - expressed in terms of @xmath46 ; solving that equation under the given boundary conditions ( see appendix a for the details ) we get the exact analytical expression for @xmath46 . let @xmath47 denote the probability density , in the laplace space @xmath48 , for the first passage times to reach the position @xmath5 , given that the initial position was @xmath32 . using the relation @xcite @xmath49 between @xmath46 and @xmath47 , and the initial condition @xmath34 , we get @xmath50 } \label{main2}\ ] ] where @xmath51 includes contributions from both the drift velocity @xmath52 and the velocity - like quantity @xmath53 that arises from diffusion . for the calculation of the mean we do not need the distribution in @xmath0-space . instead these can be derived from the relation @xmath54 using the expression ( [ main2 ] ) for @xmath55 , we get @xmath56 an interesting feature of eqn.([eq - meant ] ) is that @xmath57 is like an effective energy barrier ; kt - mt detachment is achieved by crossing this barrier . since in addition to the energetic contributions from @xmath29 and @xmath8 , the barrier @xmath57 gets contribution also from @xmath13 and @xmath14 , this effective barrier is , at least partly , of kinetic origin . external tension @xmath8 influences the kt - mt attachment time @xmath58 in two possible ways . + ( 1 ) @xmath8 decreases the depth of the linear potential well which effectively reduces the mean attachment time . ( 2 ) the same @xmath8 also reduces the depolymerization rate @xmath15 thereby increasing the mean attachment time . + these two effects of the same load tension @xmath8 act against each other . the second effect can dominate over the first at small @xmath8 , thereby giving rise to nonmonotonic variation of @xmath58 with @xmath8 , provided @xmath15 falls sharply with increasing @xmath8 ; such a possibility is ensured if @xmath18 is sufficiently small . therefore , as shown graphically in fig.[mean](a ) , @xmath58 varies nonmonotonically with @xmath8 for both @xmath59pn and @xmath60pn . but , for @xmath61pn , the fall of @xmath15 is not sharp enough to compensate the reduction of the mean attachment time caused directly by @xmath8 . thus , the physical origin of the nonmonotonic variation of @xmath58 with @xmath8 is explained by fig.[mean](a ) in a transparent manner . the nonmonotonic variation of @xmath58 with @xmath8 is displayed also for all the three different values of the parameter @xmath29 in fig.[mean](b ) ; for any given @xmath8 , the larger is the @xmath29 ( i.e. , the deeper is the potential well ) , the longer is the mean life time of the kt - mt attachment . using the symbol @xmath62 to denote the tension that corresponds to the maximum of the average lifetime , from ( [ eq - meant ] ) we get @xmath63 the dependence of @xmath62 on @xmath18 , obtained from ( [ eq - fmax ] ) , is shown in fig.[fig - fmax ] . interestingly , the variation of @xmath62 with @xmath18 is non - monotonic . in the limit @xmath64 , @xmath65 for all @xmath66 . on the other hand , as @xmath67 , @xmath68 for all finite @xmath8 . therefore , in both these limits , @xmath8-dependence of @xmath14 drops out and kt - mt attachment behaves like a slip - bond . the monotonic decrease of @xmath62 with @xmath18 in fig.[mean ] arises from the fact that we have not plotted @xmath58 for even smaller values of @xmath18 where @xmath18 increases with @xmath18 because the corresponding values of @xmath58 turned out to be unphysically long . , the tension that corresponds to the maximum if @xmath58 , is plotted against @xmath18 . ] at this stage we can scrutinize the necessity of the exponential form ( [ eq - fdepbeta ] ) for the tension - dependence of the mt depolymerization rate @xmath14 . as shown in fig.[mean](a ) , a catch - bond - like behavior follows if @xmath15 decreases sufficiently sharply with increasing @xmath8 . on the other hand , if @xmath15 falls very slowly with increasing @xmath8 the system exhibits a slip - bond - like behavior . the exponential decrease assumed in equation ( [ eq - fdepbeta ] ) is an ideal choice for the @xmath8-dependence of @xmath15 because the crossover from catch - bond to slip - bond can be displayed by simply varying @xmath18 which determines the sharpness of the decrease @xmath15 . the series representation of equation ( [ eq - meant ] ) @xmath69 \label{eq - approxmeant}\ ] ] is exact . in the special case @xmath70 eqn.([eq - approxmeant ] ) reduces to @xmath71 which , as expected , arises from kt - mt detachment caused by pure diffusion . the corrections to the diffusive result in the limit @xmath72 , can be estimated from eqn.([eq - approxmeant ] ) . as we clearly stated in the introduction , the aim of the minimal model ( and , to some extent , even the more detailed model ) is not quantitative comparison with the experimental data reported by akiyoshi et al . instead , our aim is to present analytical calculations for a minimal model that provides physical insight into the interplay of opposing forces and competing kinetic processes that , together , determine the stability of the kt - mt coupler . nevertheless , for graphical plots , which often provide a more intuitive understanding , one needs numerical values of the parameters . the largest strength of the load tension applied by akiyoshi et al . @xcite was about @xmath73 pn while the lifetime peaked closed to @xmath74pn . the longest life time that they measured was about @xmath75 minutes . although the mean lifetimes plotted in fig.[mean](a ) are comparable to the measured values for at least one set of parameter values ( corresponding to @xmath76pn ) , the numerical values of the corresponding load tension @xmath8 are much smaller than those applied in the experiment of akiyoshi et al . however , this discrepancy does not invalidate our theory . the magnitudes of the forces in our model depend on the numerical values of the parameters like , for example , @xmath29 , @xmath77 , etc . on the other hand , the corresponding effective values in the experiments of akiyoshi et al.@xcite are not known . therefore , in our plots we have used parameter values that yield lifetimes comparable to those reported in experiments . most probably , our minimal model does not capture all the details of the in - vitro study @xcite . even so , we note that the inclusion of our two postulates is sufficient to generate non - monotonic variation of the lifetimes with load tension . we begin by pointing out that in the continuum formulation treated above , the expression ( [ eq - probcurr ] ) for the probability current density @xmath78 can be recast as @xmath79 with @xmath80 where the effective potential @xmath81 is given by @xmath82 x.\ ] ] note that @xmath83 accounts for the drift caused by the force @xmath84 as well as that arising from the polymerization - depolymerization kinetics of the mt . in this section , we utilize the effective potential @xmath83 to simulate the stochastic movements of kinetochore by using a method of discretization popularized by wang , peskin and elston ( from now onwards referred to as wpe method ) @xcite . following wpe prescription , we discretized the space into @xmath85 cells , each of length @xmath86 ; however , @xmath87 need not be identical to @xmath88 , the separation between the two consecutive binding site of a mt ( see fig.[fig - discretize ] ) . accordingly , the continuous effective potential @xmath83 is replaced by its discrete counterpart @xmath89 x_{j } \label{eq - discreteu}\ ] ] where @xmath90 denotes the position of the center of the @xmath91-th cell . is discretized into @xmath85 cells each of length @xmath92 following the wpe prescription ( see text for details ) . ] next we approximate the continuous movement described by the fokker - planck equation ( [ main ] ) by a master equation in terms of discrete jumps from the center of one cell to that of an adjacent cell either in the forward or backward direction . in this case the expressions for the forward and backward transition rates @xmath93 and @xmath94 ( see fig.[fig - discretize ] ) are given by @xcite @xmath95 @xmath96 where @xmath97 for the simple potential ( [ eq - discreteu ] ) , @xmath93 and @xmath94 are independent of site index @xmath91 . + in computer simulations , the mt is fully inserted into the sleeve in the initial state . the simulation produce the numerical data for the first passage time of the mt tip to exit the coupler . in fig . [ hist ] , histograms of these numerical data are plotted for a few values of force @xmath8 . laplace transform of these distributions are compared in the insets with the corresponding prediction of the exact analytical formula ( [ main2 ] ) . excellent agreement between the theoretical prediction and the simulation data ( albeit in the laplace space ) demonstrates how the theory may be useful in analyzing also the distribution of experimentally measured lifetimes of kt - mt attachments _ in - vitro_. in this section we begin with the assumption that the inner surface of the cylindrically shaped coupler consists of only passive binders . we assume that each binder head engages with the mt by obeying a unit energy function @xmath98 ( see fig.[fig2a ] ) , which has two key parameters : @xmath99 measures free energy drop due to binder affinity for the mt lattice , and @xmath100 describes the activation barrier for transitions between specific mt lattice binding sites @xcite . in other words , @xmath99 is a measure of the strength of the mt - binder affinity while @xmath100 is a measure of the `` roughness '' or friction of the mt - coupler interface . the total potential energy function is given by ( see fig.[fig2a ] ) @xmath101 where @xmath48 is the spacing between consecutive coupler binders ( see fig [ fig2a ] ) . binder spacing is an arbitrary parameter . here we set @xmath102 , where @xmath88 is the distance between consecutive mt binding sites . and @xmath103 ) . the load tension @xmath8 tilts the potential energy landscape @xmath103 . the tension - dependent depolymerization rate @xmath15 increases the effective barrier along the unbinding pathway ( orange line , inset ) by slowing down depolymerization . ] t , b = 0.001a . intermediate regime ( circle ) a = 0.5k@xmath104 t , b = 0.04a , and intermediate regime ( square ) a = 0.6k@xmath104 t , b = 0.2a . high - friction regime ( dashed red line ) is obtained from eq . ( 32 ) with @xmath105 , @xmath106 , @xmath107 pn . ] for @xmath103 in our numerical calculations , we have used the simple expression @xmath108 where @xmath109 , @xmath110 and @xmath111 . the linear and scalar coefficients in eq . ( [ eq5 ] ) arise because we use a fourier series to approximate the well function expression . thus , the mt - coupler interaction is represented more realistically by @xmath103 than by @xmath10 of the minimal model . the friction arising from the potential landscape @xmath103 gives rise to a jump - drift - diffusion process . as we ll point out below , this general version of our model reduces to the minimal version discussed above , and hence effectively to a drift - diffusion process , in a special limiting case . in infinitesimal time interval @xmath112 , the change @xmath113 of the overlap is given by @xmath114dt+ \ell dn_{r}(t)+\sigma dw(t ) \label{eq1}\ ] ] where , as before , the constant @xmath8 is the external load tension and @xmath24 is the effective drag coefficient . @xmath115 accounts for the thermal diffusion of the coupler on the lattice with @xmath116 , and @xmath117 is a poisson counting process describing the mt tip dynamics with rate @xmath118 that is decided by the difference of the mt polymerization rate @xmath13 and mt depolymerization rate @xmath14 . the stochastic differential equation ( [ eq1 ] ) is essentially equivalent to the langevin equation ( [ eq - langevin ] ) ; the form ( [ eq1 ] ) is used in our numerical computations for the convenience of implementation . @xmath119 is characteristic of the structure of the coupler ( coupler length ) whereas its energetics depend on @xmath120 ( i.e. , on the parameters @xmath99 , @xmath100 ) and @xmath8 ; the stochastic kinetics are influenced by the interplay of forces arising from the potential landscape , random brownian forces , and by @xmath13 , @xmath14 . by a combination of standard analytical and numerical methods , we study the trends of variation of @xmath121 with ( a ) @xmath8 , as well as , the ( b ) size , ( c ) energetics , and ( d ) kinetics of the coupler . it is difficult to derive an exact analytical expression for the mean first passage time corresponding to eq . ( [ eq1 ] ) that would be valid in all parameter regimes . therefore , we explore two limiting cases for which explicit approximate solutions can be obtained : ( a ) _ low - friction regime _ ( @xmath122 ) , and ( b ) _ high - friction regime _ ( @xmath123 ) . in the _ low - friction regime _ the coupler can easily rearrange its position relative to the mt ; in this regime @xmath121 approximated well by the expression in equation ( [ eq - meant ] ) in the _ high - friction regime _ large @xmath100 leads to strong local pinning that practically stalls it . the only mode available for detachment of the coupler is via depolymerization of the mt . consequently , in this regime @xmath124 the exponential increase of @xmath121 with @xmath8 in the high - friction regime can be regarded as an artifact of the unphysical ( or , unrealizable ) nature of this regime . unless explicitly stated otherwise , the parameter values that we have used for our numerical calculations in this section are given in the table [ default ] . ccc parameter description & symbol & values tested + mt binding site spacing & @xmath88 & 8/13 nm @xcitemaximal coupler length & @xmath1 & 50 nm @xcitemaximal number of coupler binders & @xmath119 & 15 - 65 @xcitepolymerization rate & @xmath13 & @xmath125 @xmath126@xcitemaximal depolymerization rate & @xmath20 & @xmath127 @xmath126 @xcitecritical depolymerization force & @xmath18 & @xmath128 pn ( estimated here)external load force & @xmath8 & @xmath129 pn force of attractive kt - mt interaction & @xmath29 & @xmath130 pn mt lattice / binder binding energy & @xmath99 & @xmath131 @xcitebinder activation barrier & @xmath100 & @xmath132 ( estimated here)diffusion constant & @xmath22 & @xmath133 nm@xmath134s@xmath43 @xciteeffective drag coefficient & @xmath24 & @xmath135 pns@xmath136m@xmath43@xcite [ default ] . ( a)-(b ) @xmath121 for two sets of maximal rates of mt depolymerization with the ratio @xmath137 fixed . in ( a ) the data are obtained for an _ intermediate _ regime coupler , with @xmath138 k@xmath104 t , @xmath139 , and in ( b ) for a shorter coupler in the _ low - friction _ regime with @xmath140 k@xmath104 t , @xmath141 . ( c)-(d ) @xmath121 for _ intermediate _ regime couplers . in ( c ) are shown data with two different strengths of friction caused by two values of @xmath100 , and in ( d ) for two different binding energies given by @xmath99 . the mt rates are @xmath142 s@xmath43 , @xmath143 s@xmath43 for ( c)-(d ) . ( e ) @xmath121 plotted against @xmath8 for three different lengths of the _ low - friction _ coupler from ( a ) with @xmath140 k@xmath104 t , @xmath144 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . error bars mark standard deviation.,title="fig : " ] + . ( a)-(b ) @xmath121 for two sets of maximal rates of mt depolymerization with the ratio @xmath137 fixed . in ( a ) the data are obtained for an _ intermediate _ regime coupler , with @xmath138 k@xmath104 t , @xmath139 , and in ( b ) for a shorter coupler in the _ low - friction _ regime with @xmath140 k@xmath104 t , @xmath141 . ( c)-(d ) @xmath121 for _ intermediate _ regime couplers . in ( c ) are shown data with two different strengths of friction caused by two values of @xmath100 , and in ( d ) for two different binding energies given by @xmath99 . the mt rates are @xmath142 s@xmath43 , @xmath143 s@xmath43 for ( c)-(d ) . ( e ) @xmath121 plotted against @xmath8 for three different lengths of the _ low - friction _ coupler from ( a ) with @xmath140 k@xmath104 t , @xmath144 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . error bars mark standard deviation.,title="fig : " ] + . ( a)-(b ) @xmath121 for two sets of maximal rates of mt depolymerization with the ratio @xmath137 fixed . in ( a ) the data are obtained for an _ intermediate _ regime coupler , with @xmath138 k@xmath104 t , @xmath139 , and in ( b ) for a shorter coupler in the _ low - friction _ regime with @xmath140 k@xmath104 t , @xmath141 . ( c)-(d ) @xmath121 for _ intermediate _ regime couplers . in ( c ) are shown data with two different strengths of friction caused by two values of @xmath100 , and in ( d ) for two different binding energies given by @xmath99 . the mt rates are @xmath142 s@xmath43 , @xmath143 s@xmath43 for ( c)-(d ) . ( e ) @xmath121 plotted against @xmath8 for three different lengths of the _ low - friction _ coupler from ( a ) with @xmath140 k@xmath104 t , @xmath144 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . error bars mark standard deviation.,title="fig : " ] + , is plotted in the low - friction regime and the corresponding data for the intermediate regime . the parameters are @xmath119 = 65 , a = 0.45 k@xmath104 t , @xmath145 s@xmath43 , @xmath146 s@xmath43 . error bars mark standard deviation.,width=350,height=254 ] for numerical simulations of this version of our model in the intermediate regime we used the parameter values which are within the range reported in the literature@xcite . consequently , computed @xmath36 turn out to be comparable to those observed in the experiments . some of the typical curves are plotted in figs.[fig3 ] . @xmath147 _ effects of variation of rate of kinetics _ the effects of varying the maximal rate of depolymerization @xmath20 , keeping the ratio @xmath148 fixed , is demonstrated in figs.[fig3](a ) and ( b ) . clearly , in such situations , the kt - mt attachment can exhibit catch - bond - like or slip - bond - like behavior depending on the value of @xmath20 . @xmath147 _ effects of variation of parameters that characterize energetics _ the effects of the energetics , namely the influences of the parameters @xmath99 and @xmath100 that determine the potential energy landscape , are displayed in figs.[fig3](c ) and ( d ) . the effects of varying the number of binders @xmath119 is shown in fig.[fig3](e ) . in all these cases the nonmonotonic variation of @xmath121 with @xmath8 seems to be a generic feature , except that @xmath121 may become too large to be observable for some values of the parameters ( see , for example , fig.[fig3](e ) . @xmath147 _ tension - induced switching from depolymerization to polymerization _ fig.[fig3d ] shows the key role of the external tension @xmath8 in deciding the direction of the velocity . suppose the parameter values are chosen such that @xmath20 dominates over @xmath13 in the absence of external tension thereby ensuring depolymerization of the mt . when the external tension is applied , the depolymerization is suppressed . increasing @xmath8 can eventually cause such strong suppression of the depolymerization that , beyond a certain strength of @xmath8 , @xmath13 dominates over @xmath15 . in such a scenario , the mt switches from the depolymerization mode to polymerization mode at a threshold strength of @xmath8 . in this paper , we have presented theoretical models of kt - mt attachment at three levels of detail starting with a minimal version that captures only the key components and essential processes . these models mimic kt - mt attachments reconstituted _ in - vitro _ from purified components extracted from budding yeast @xcite . exploiting the simplicity of the minimal version of the model , we have calculated the distribution of the lifetimes of the kt - mt attachments analytically and compared the results with the corresponding simulation data . equation ( [ main2 ] ) encapsulates the main result of this part of the manuscript . the strength of the minimal model is that the results can be derived analytically . in spite of its simplicity , this minimal model provides deep insight into the roles of opposing forces and competing kinetics in the tension - dependence of the life time of the kt - mt attachments . the second version of the model is an extension of the minimal model ; it incorporates some more realistic features of the interaction between the mt and the coupler . for the second model we have succeeded in calculating analytically only the mean lifetime of the kt - mt attachments under reasonable schemes of approximation . the analytical expression ( [ eq - meant ] ) for the mean life time of the kt - mt attachment in the minimal model has been derived from the corresponding full distribution ( [ main2 ] ) of the life times , the nonmonotonic variation of the mean lifetime of the kt - mt attachments with external tension exhibited over wide range of values of the model parameters in both the models is consistent with the similar trend observed in the _ in - vitro _ experiments @xcite . the realistic potential landscape in the second model does not alter this qualitative trend . we have also demonstrated that the nonmonotonicity of the variation of @xmath58 with @xmath8 depends on the rapidity of decrease of @xmath15 with increasing @xmath8 . by varying @xmath18 , which determines the sharpness of the decrease of @xmath15 , we have shown how the catch - bond - like behavior crosses over to a slip - bond - like behavior . moreover , our analytical calculations also show that , unlike other conventional catch - bonds , the catch - bond - like behavior of the kt - mt attachment arises from the interplay of not only forces derivable from the potential landscape , but also by a competition of the kinetics of polymerization and depolymerization of the mt . in fact , the barrier against the breakdown of the attachment gets contributions from both opposing forces and competing kinetic processes . the slip - bond - like monotonic decrease of @xmath121 with @xmath8 observed in some of the parameter regimes ( whose physical origin has been explained in terms of the nature of @xmath8-dependence of the depolymerization rate @xmath15 ) , might be detectable in experiments under conditions different from those used by akiyoshi et al.@xcite . as expected , we have found that the molecular motors enhance stability or tend to destabilize the kt - mt attachment depending on whether the force it generates opposes or assists the external tension . the latter awaits experimental confirmation . moreover , our analytical predictions on the distribution of the lifetimes will be useful in the future in analyzing the statistical properties of the lifetimes provided the _ in - vitro _ experiment on the reconstituted kt - mt attachment is repeated sufficiently large number of times . in principle , the models presented here can be further extended by including active force generators like molecular motors @xcite . however , so far such active force generators have not been detected at the kt - mt interface in budding yeast although such force - generating motors have been found in other cells including , for example , mammalian cells . but , in contrast to budding yeast , where each kinetochore can attach to only one mt , multiple mts can attach with each kinetochore in a mammalian cell . since modeling of kt - mt attachment in this paper is focussed almost exclusively on budding yeast , we have not discussed the effects of molecular motors in the main text . thus , the extended model in appendix b is not a complete model of kt - mt attachment in mammalian cells ; it is intended to indicate how the effects of molecular motors can be incorporated within the broad theoretical framework developed in this paper . in near future we hope to extend our model to kt - mt attachments in mammalian cells . however , at present , the main hurdle in this modeling is the lack of our current understanding as to how multiple mts attached to the same kt coordinate their kinetics @xcite . the `` sleeve model '' developed originally by hill @xcite was adapted by joglekar and hunt @xcite , to account for the phenomenon of `` directional instability '' @xcite ( also referred to as chromosome oscillation ) observed during mitosis in vertebrates . in those cells each kt can attach with several mt ( typically , up to 35 ) and this feature was also incorporated in the model studied by joglekar and hunt by carrying out computer simulations . our work differs significantly from that of joglekar and hunt @xcite . first , our model is intended to account for the qualititative features of the results obtained from a much simpler _ in - vitro _ system where only one mt is attached to only a single kt that is subjected to an externally applied load force that , in our model , is assumed to alter the rate of mt depolymerization itself . moreover , because of their motivation in studying directional instability , joglekar and hunt @xcite put major emphasis on monitoring the _ spatial _ displacements that characterize this instability . in contrast , the quantity of our main interest is the life _ time _ of the kt - mt attachments . the spatial excursions of the kinetochores was also the main quantity of interest of gardner et al.@xcite who studied the effects of tension - mediated regulation of the kinetics of kinetochore mts by computer simulation of a related model . the scope of the more detailed model studied by scholey et al.@xcite was even broader . our force - balance equations are similar , at least in principle , to the force - balance equations formulated by scholey et al.@xcite . however , instead of the spatial displacements of the kinetochores , the lifetimes of the kt - mt attachments are the main focus of our investigation . most of the earlier models were based on the assumption that the coupler is a stable sleeve or ring . in recent years , a new class of models have been developed on the basis of an altogether different assumption regarding the nature of the coupler . in this scenario @xcite microtubule associated proteins ( maps ) are assumed to make transient attachments with the mt ( i.e. , each map attaches to the mt for a brief duration before getting detached from it ) . the entropic contribution to the force is significant and slight variation in the mean lifetime of the individual transient attachments of the maps can have large effect on the lifetime of the kt - mt attachment . in order to account for their experimental observations , akiyoshi et al . @xcite developed a 2-state kinetic model . this model was motivated by the formal analogy with the models of ` catch - bonds ' in other systems . the growing and shrinking states of the mt were argued to be analogs of the strongly- and weakly - bound states in the models of catch - bonds . in their 4-parameter kinetic model , two parameters were the rates of transitions between the states of growth and shrinkage of the mt . the remaining two parameters were the rates of detachment of the kinetochore from the mt while the latter is in the growing and shrinking phases , respectively . postulating exponential dependence of the type @xmath149 for each of the four rate constants ( n=1,2,3,4 ) , on the external tension @xmath8 , akiyoshi et al . extracted the numerical values of the 8 parameters @xmath150 ( n=1,2,3,4 ) corresponding to the best fit to their experimental data @xcite . in their model , akiyoshi et al.@xcite did not explicitly treat the kinetics of the polymerization and depolymerization , by attachment and detachment of the successive subunits ( more precisely , the @xmath151 tubulin dimers ) , during the growing and shrinking phases of the mt , respectively . therefore , the force - dependence of the four rate constants that provide the best fit to their experimental data do not indicate the corresponding force - dependences of the rates of polymerization and depolymerization of the mt . moreover , the potential energy of interaction between the mt and the coupler as well as the effect of the external tension on this potential energy landscape are not incorporated explicitly in their model . thus , the model of akiyoshi et al.@xcite does not directly demonstrate the interplay of the opposing forces and competing kinetics . since neither any structural features of the coupler , nor the nature of its interaction with the mt enters explicitly into akiyoshi et al.s @xcite model , their model can not be used to study the effects of the size and composition of the coupler or that of the nature of the mt - coupler interaction . neither can it be used to account for the special features of hybrid couplers . in contrast , in this paper we have used our models to study the effects of varying ( i ) the size of the coupler , ( ii ) relative population of active force generators ( molecular motors ) and passive binders , as well as ( iii ) the depth and roughness of the potential of interaction between the coupler and the mt . catch bond formed by the fimbriae of e - coli bacteria with target cell surface and that formed by eukaryotic p - selectin and integrin receptors with their ligands have been studied extensively in recent years @xcite to understand their mechanism . in this paper we have focussed exclusively on the kt - mt attachments to understand the mechanism of its catch - bond - like behavior @xcite using simple theoretical models . one distinct feature of the mt is that , unlike other ligands , these stiff filaments are capable of generating push and pull by their polymerization and depolymerization , respectively . in particular , the pulling force generated by a mt is powered by the hydrolysis of gtp bound to the tubulin subunits . as we have shown in this paper , the competing polymerization and depolymerization of mt gives rise to a unique feature of the catch - bond - like behavior of kt - mt attachments where the stable state can be a non - equilibrium ( local ) steady state rather than a local minimum of the free energy of the ligand - receptor system . we hope the modeling methodology developed here can be usefully adapted to understand the catch - bond - like behavior of depolymerizing actin under load tension observed recently by carrying out _ in - vitro _ experiments with atomic force microscope ( afm ) @xcite . * acknowledgements : * dc thanks frank jlicher for valuable comments . this work has been supported in part by csir ( india ) through a senior research fellowship ( aks ) , by nsf through the grant dms- 1358932 ( bs ) , by mbi , the ohio state university , through the nsf grant dms 0931642 ( bs and dc ) , by iit kanpur through dr . jag mohan garg chair professorship ( dc ) , by serb ( india ) through j.c . bose national fellowship ( dc ) , by dbt ( india ) through a research grant ( dc ) , and by the visitors program of the max - planck institute for the physics of complex systems , dresden ( dc ) . from equation ( [ main ] ) we get @xmath152 for @xmath153 , equation ( [ main1 ] ) has following solution . @xmath154 similarly , for @xmath155 , solution of equation ( [ main1 ] ) is , @xmath156 where @xmath157 and @xmath158 are given by @xmath159 with @xmath160 . imposing the absorbing boundary condition at @xmath3 on @xmath46 we find @xmath161 now @xmath46 and it s first derivative must satisfy the following matching conditions at @xmath162 : + @xmath163 @xmath164 reflecting boundary condition at @xmath4 gives , @xmath165 the four unknowns @xmath166 , @xmath167 , @xmath168 and @xmath169 can be obtained by solving the three equations ( [ b1])-([eq - alg3 ] ) . as we ll see later , we need @xmath46 only for @xmath170 . therefore , we give the expression for @xmath166 only in the region @xmath153 : @xmath171 with @xmath172 . and @xmath167 can be find by using equation [ b1 ] . thus , for @xmath153 @xmath173\ ] ] where the @xmath48-dependence enters through the @xmath48-dependence of @xmath174 ( see eqn.([eq - defnu ] ) that enters into the expression ( [ eq - defnv ] ) for @xmath175 . so far no motor protein has been detected at the kinetochore in budding yeast , which is the object of our modeling in this paper , but , in most of the eukaryotes ( e.g. , mammalian cells ) such motors have been found in the kinetochores each of which , however , attaches to more than one mt . on the other hand , not more than a single mt can attach to each kt in budding yeat . nevertheless , we extend our model further by including motor proteins @xcite , which are active force generators , in addition to the passive binders in the model coupler . we hope to extend this hybrid coupler , in near future , by integrating more than one mt to model kt - mt attachments of mammalian cells . as indicated in recent experiments @xcite , the outermost layer of the _ hybrid _ coupler in mammalian cells consists of passive binders while the innermost layer , that is adjacent to the kt , is composed of active force generators . in the model of hybrid coupler we represent the active force generators by molecular motors ( see fig [ fig4]a ) . as we show in this section , the nature of the response of the kt - mt attachment to the external tension now depends on whether the motors are plus - end directed ( e.g. , kinesin ) or minus - end directed ( e.g. , dynein ) . we use the subscripts or superscripts @xmath176 and @xmath177 to refer to the plus - end and minus - end directed motors , respectively . following the general trend in the literature on molecular motors @xcite . we also postulate a linear force - velocity relation for the individual motors : @xmath178 where @xmath179 and @xmath180 are the stall force and maximal velocity for the plus - end directed and minus - end directed motors , respectively , whereas @xmath181 are the corresponding instantaneous velocities . for an _ intermediate _ regime hybrid coupler with ( @xmath182 ) and without ( @xmath183 ) minus - end motors with @xmath184 k@xmath104 t , @xmath185 , @xmath142 s@xmath43 , @xmath143 s@xmath43 . ( c ) breaking load , f@xmath186 for a low - friction hybrid coupler versus motor densities ; minus - end motors ( squares ) and plus - end motors ( circles ) . average speed vs force for varying densities of minus - end motors . stall forces for each motor density fraction . parameters for ( c)-(d ) are @xmath187 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . for all panels @xmath188.,title="fig:",width=331,height=230 ] for an _ intermediate _ regime hybrid coupler with ( @xmath182 ) and without ( @xmath183 ) minus - end motors with @xmath184 k@xmath104 t , @xmath185 , @xmath142 s@xmath43 , @xmath143 s@xmath43 . ( c ) breaking load , f@xmath186 for a low - friction hybrid coupler versus motor densities ; minus - end motors ( squares ) and plus - end motors ( circles ) . average speed vs force for varying densities of minus - end motors . stall forces for each motor density fraction . parameters for ( c)-(d ) are @xmath187 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . for all panels @xmath188.,title="fig : " ] for an _ intermediate _ regime hybrid coupler with ( @xmath182 ) and without ( @xmath183 ) minus - end motors with @xmath184 k@xmath104 t , @xmath185 , @xmath142 s@xmath43 , @xmath143 s@xmath43 . ( c ) breaking load , f@xmath186 for a low - friction hybrid coupler versus motor densities ; minus - end motors ( squares ) and plus - end motors ( circles ) . average speed vs force for varying densities of minus - end motors . stall forces for each motor density fraction . parameters for ( c)-(d ) are @xmath187 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . for all panels @xmath188.,title="fig : " ] for an _ intermediate _ regime hybrid coupler with ( @xmath182 ) and without ( @xmath183 ) minus - end motors with @xmath184 k@xmath104 t , @xmath185 , @xmath142 s@xmath43 , @xmath143 s@xmath43 . ( c ) breaking load , f@xmath186 for a low - friction hybrid coupler versus motor densities ; minus - end motors ( squares ) and plus - end motors ( circles ) . average speed vs force for varying densities of minus - end motors . stall forces for each motor density fraction . parameters for ( c)-(d ) are @xmath187 , @xmath145 s@xmath43 , @xmath146 s@xmath43 . for all panels @xmath188.,title="fig : " ] we start with the force balance equation which does not include random fluctuations for the overlap velocity @xmath189 where @xmath24 is the effective coupler drag coefficient and @xmath190 is the velocity of the mt tip with respect to a space - fixed frame of reference . the active force term @xmath191 with the motor density function @xmath192 where @xmath193 is the standard heaviside step function and @xmath194 nm , corresponds to the total horizontal length of the coupler that can be populated by active components ( in three - dimensions this corresponds to one layer of motors working around a mt with 13 protofilament tracks , with one motor per track ) . note that in the symbol @xmath195 , the subscript minus ( - ) / plus ( + ) denote the force generated by the minus - end / plus - end directed motors that tend to increase / decrease the overlap @xmath5 . we have expression ( [ eq - fplusminus ] ) for @xmath195 where @xmath179 and @xmath180 are the stall force and maximal velocity for the plus - end directed and minus - end directed motors , respectively , whereas @xmath181 are the corresponding instantaneous velocities . next we express @xmath181 in terms of @xmath196 . using @xmath197 , in case of minus - end directed motors @xmath198 similarly , if only plus - end - directed motors are present , @xmath199 the coupler movement described by eq . ( [ deteqn ] ) is fully deterministic . next , we write down a stochastic differential equation ( sde ) that would , upon averaging , correspond to the deterministic equations written above . suppose over a small time interval @xmath204 the number of subunits ( an @xmath151 tubulin dimer ) added and removed from the tip of the mt by polymerization and depolymerization are @xmath205 , an independent homogenous poisson process . we capture the effects of random brownian forces through the noise @xmath206 which is assumed to be a gaussian stochastic process , which also includes the effects of fluctuations both in the chemical reactions and mechanical stepping involved in each cycle of the individual motors . thus , finally , the equation for the hybrid coupler overlap reads @xmath207dt\nonumber\\ & + \ell dn_{r}(t)+\sigma dw(t).\end{aligned}\ ] ] in fig . [ fig4 ] we plot results for a hybrid coupler with varying density fractions of motors at the active interface , measured by @xmath208 . the non - monotonic nature of @xmath121 is preserved when the active components are added ( see fig . [ if the active force generators are minus - end directed ( for example , dyneins ) , the force exerted by them oppose load force . consequently , the mean lifetime @xmath121 os the kt - mt attachment increases as motor fractions , @xmath209 become larger ( see figs . [ fig4]c ) . for a given @xmath209 , the extent of such motor - induced extra stabilization of hybrid couplers can be sensitive to the binding energy of the passive components measured by @xmath99 ( see fig [ fig4]c ) . despite enhancing the stability , higher numbers of load opposing motors can also lower the ability of the coupler to efficiently track rescued mt tips , as noted by the slower coupler velocities in fig . [ fig4]d . plus - end directed motors ( for example , cenp - e kinesin ) , which oppose overlap between the mt and the coupler , generate force that assist the external load . if the active force generators in the hybrid coupler are such plus - end directed motors , neither the coupler velocities and nor the breaking force @xmath210 show significant variation as @xmath209 is varied under a given tension , provided that there are sufficient numbers of passive binder components supporting attachment . 00 h. lodish , a. berk , s l. zipursky , p. matsudaira , d. baltimore , and j. darnell , _ molecular cell biology , 4th edition _ , ( w. h. freeman , 2000 ) . i. m. cheeseman and a. desai , nature rev . * 9 * , 33 ( 2008 ) . e. karsenti and i. vernos , science * 294 * , 543 ( 2001 ) . mcintosh , m.i . molodtsov and f.i . ataullakhanov , d. chowdhury , phys . rep . * 529 * , 1 ( 2013 ) . d. chowdhury , biophys . j. * 104 * , 2331 ( 2013 ) . g. civelekoglu - scholey , d.j . sharp , a. mogilner and j.m . scholey , biophys . j. * 90 * , 3966 ( 2006 ) . joglekar and a.j . hunt , biophys . j. * 83 * , 42 ( 2002 ) . gardner , c.g . pearson , b.l . sprague , t.r . zarzar , k. bloom , e.d . salmon and d.j . odde , mol . cell * 16 * , 3764 ( 2005 ) . quart . * 45 * , 147 - 207 ( 2012 ) . c.s . peskin and g.f . oster , biophys . j. * 69 * , 2268 ( 1995 ) . b. akiyoshi , k. k. sarangapani , a. f. powers , c. r. nelson , s. l. reichow , h. arellano - santoyo , t. gonen , j. a. ranish , c. l. asbury and s. biggins , nature * 468 * , 576 ( 2010 ) . , ann rev biophys , * 37 * , 399 - 416 ( 2008 ) . h. wang , c. peskin and t. elston , j. theo . biol . * 221 * , 491 ( 2003 ) . h. wang , t. c. elston , j. stat . phys . * 128 * , 35 ( 2007 ) . t. hill , proc . * 82 * , 4404 ( 1985 ) . s. westermann , d.g . drubin and g. barnes , annu . rev . biochem . * 76 * , 563 ( 2007 ) . asbury , j.f . tien and t.n . davis , trends in cell biol . * 21 * , 38 ( 2011 ) . buttrick and j.b.a . millar , chromosome res . * 19 * , 393 ( 2011 ) . e.a . foley and t.m . kapoor , nat . . cell biol . * 14 * , 25 ( 2013 ) . a. d. franck , a. f. powers , d. r. gestaut , t. gonen , t. n. davis , c. l. asbury , nat cell biol * 9 * , 832 ( 2007 ) . s. redner , _ a guide to first - passage processes _ , ( cambridge university press , 2001 ) . v. balakrishnan , _ elements of nonequilibrium statistical mechanics _ , ( ane books pvt . ltd . , 2008 ) b. shtylla and j. p. keener , siam j. appl . math * 71 * 1821 ( 2011 ) . a. p. joglekar , d. c. bouck , j. n. molk , k. s. bloom and e. d. salmon , nat . cell biol . * 8 * , 581 ( 2006 ) . a. p. joglekar , k. bloom and e. d. salmon , curr . biol . * 19 * , 694 ( 2009 ) . x. wan , r. p. oquinn , h. l. pierce , a. p. joglekar , w. e. gall , j. g. deluca , c. w. carroll , s.t . liu , t. j. yen , b. f. mcewen , p. t. stukenberg , a. desai and e.d . salmon , cell * 137 * , 672 ( 2009 ) . j. lawrimore , k. s. bloom , e. d. salmon , , j cell biol . * 195 * , 573 ( 2011 ) . sharp , g.c . rogers and j.m . scholey , nature * 407 * , 41 - 47 ( 2000 ) . j.m . scholey and a. mogilner , _ mitotic spindle motors _ , in : _ molecular motors _ , ed . m. schliwa ( wiley - vch , 2003 ) . h. maiato , j. deluca , e. d. salmon , and w. c. earnshaw , j. cell science * 117 * , 5461 ( 2004 ) . s. dumont , e. d. salmon , t. j. mitchison , science * 337 * , 355 ( 2012 ) . s. gonen , b. akiyoshi , m. g. iadanza , d. shi , n. duggan , s. biggins , t. gonen , nat struct mol biol * 2012 * , 925 ( 2012 ) . w. thomas , annu . * 10 * , 39 ( 2008 ) . thomas , curr . * 19 * , 50 ( 2009 ) . sokurenko , v. vogel , w.e . thomas , cell host & microbe * 4 * , 314 ( 2008 ) . o.v . prezhdo and y.v . pereverzev , acc . 42 * , 693 ( 2009 ) . evans and d.a . calderwood , science * 316 * , 1148 ( 2007 ) . c. zhu , j. lou and r.p . mcever , biorheology * 42 * , 443 ( 2005 ) . c. zhu , annals biomed 36 * , 604 ( 2008 ) . mcever and c. zhu , annu . * 26 * , 363 ( 2010 ) . p. sundd , m.k . pospieszalska , l.s.l . cheung , k. konstantopoulos and k. ley , biorheology * 48 * , 1 ( 2011 ) . p. sundd , m.k . pospieszalska and k. ley , mol . immunol . * 55 * , 59 ( 2013 ) .

we present submillimeter array 880 m dust polarization observations of six massive dense cores in the dr21 filament . the dust polarization shows complex magnetic field structures in the massive dense cores with sizes of 0.1 pc , in contrast to the ordered magnetic fields of the parsec - scale filament . the major axes of the massive dense cores appear to be aligned either parallel or perpendicular to the magnetic fields of the filament , indicating that the parsec - scale magnetic fields play an important role in the formation of the massive dense cores . however , the correlation between the major axes of the cores and the magnetic fields of the cores is less significant , suggesting that during the core formation , the magnetic fields below 0.1 pc scales become less important than the magnetic fields above 0.1 pc scales in supporting a core against gravity . our analysis of the angular dispersion functions of the observed polarization segments yields the plane - of - sky magnetic field strengths of 0.41.7 mg of the massive dense cores . we estimate the kinematic , magnetic , and gravitational virial parameters of the filament and the cores . the virial parameters show that in the filament , the gravitational energy is dominant over magnetic and kinematic energies , while in the cores , the kinematic energy is dominant . our work suggests that although magnetic fields may play an important role in a collapsing filament , the kinematics arising from gravitational collapse must become more important than magnetic fields during the evolution from filaments to massive dense cores .

1703.02566

resent observations of molecular clouds have revealed a two - step scenario of star formation that supersonic flows first compress clouds into parsec - long filaments , and then gravitational bound filaments collapse and fragment into dense cores @xcite . in this scenario , massive dense cores ( @xmath1 0.1 pc , @xmath2 20 m@xmath3 ) formed at intersections of filaments are the sites of clustered and massive star formation @xcite . massive dense cores are magnetized with magnetic field strengths of 0.11 mg @xcite . since the star - formation rate in dense gas is only few percents of the rate of a free - fall collapse , magnetic fields are suggested to be an important ingredient supporting massive dense cores against self - gravity @xcite . the orientation between cloud morphology and magnetic fields reveal information about the role of magnetic fields in star formation . in a strongly magnetized cloud , ions are easier to move along the field lines than perpendicular to the field lines . simulations of magnetized turbulent gas show that if the turbulent energy is stronger than the gravitational energy , turbulence extends gas along the field lines and forms filaments parallel to magnetic fields ( see figure 2 of * ? ? ? * ) . on the other hand , if the gravitational energy is stronger than the turbulent energy , the mass contraction along the field lines results in filaments perpendicular to magnetic fields @xcite . that is , strong magnetic field models predict either - parallel - or - perpendicular alignment between filaments and magnetic fields , depending on the competition between turbulent pressure and gravitational pressure . in contrast , the filaments formed with weak magnetic fields tends to be randomly aligned with respect to magnetic fields @xcite . such alignment of strong magnetic field models and misalignment of weak magnetic field models are also presented in the simulations not including turbulence @xcite . in observations , the correlation between the orientations of clouds and magnetic fields is found to be either - parallel - or - perpendicular in filamentary clouds from few to tens pc scales @xcite and in massive cores at 0.10.01 pc scales @xcite , suggesting a dynamically important role of magnetic fields in various physical scales ( see @xcite @xcite for a review ) . observations of the polarized dust emission from magnetically aligned dust grains have been proven to be the most efficient way to reveal magnetic field structures in molecular clouds @xcite . for massive dense cores with typical sizes of 0.1 pc and distances of few kiloparsecs , several interferometric observations have been performed to spatially resolve the magnetic field structures ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? ? * ; * ? ? ? * also see zhang et al . 2014 for a review of sma polarization legacy project ) . the observations reveal diverse magnetic field structures from ordered hour - glass shapes ( e.g. g31.41 , @xcite @xcite ) to chaotic distributions ( e.g. g5.89 , @xcite @xcite ) . the magnetic energy in a massive core is typically equal or less than the energies of rotation , outflows , and radiation @xcite . in this paper , we present a submillimeter array ( sma ) study of six massive dense cores in the dr21 filament . dr21 filament is the densest and most massive region in the cygnus x complex @xcite , hosting several well - studied high - mass star - forming regions like dr21 and dr21(oh ) @xcite at a distance of 1.4 kpc @xcite . _ herschel _ observations showed that the ridge of dr21 filament has a length of 4 pc and a mass of 15000 m@xmath3 , connected by several sub - filaments with masses between 130 to 1400 m@xmath3 @xcite . molecular line studies suggested continuous mass flows from the sub - filaments onto the dr 21 filament , driving the global gravitational collapse of the filament @xcite . dust continuum studies found 12 massive dense cores in the dr21 filament @xcite . the massive dense cores drive active outflows @xcite and masers @xcite , indicating recent high - mass star formation . thirty - two fragments with sizes of a few thousands au were detected in a sample of seven cores , whereas some of the fragments were proposed to be precursors of ob stars @xcite . interferometric molecular line observations reveal small - scale converging flows within the massive dense cores , which may initiate the dense structures of the cores @xcite . magnetic fields of the dr21 filament at parsec scale have been mapped through single - dish polarimetric observations @xcite , revealing an uniform structure of magnetic fields aligned perpendicular to the filament . zeeman observations of the cn lines found a 0.40.7 mg magnetic field strength of the line - of - sight component in dr21(oh ) @xcite . in contrast to the uniform magnetic fields at pc scales , interferometric dust polarization observations revealed complex magnetic fields in the dr21(oh ) @xcite . here , we present sma dust polarization observations of massive dense cores cyg - n38 ( alias dr21(oh)-w ) , cyg - n43 ( w75s - fir1 ) , cyg - n44 ( dr21(oh ) ) , cyg - n48 ( dr21(oh)-s ) , cyg - n51 ( w75s - fir2 ) , and cyg - n53 . the molecular line data of our observations will be represented in our upcoming paper ( paper ii , ching et al . in prep . ) section 2 describes the observations and data reduction . section 3 presents the dust polarization maps . analysis and discussion of the core property and magnetic fields are presented in section 4 . in section 5 , we draw the main conclusions . the sma polarization @xcite observations were carried out toward six massive dense cores in the dr21 filament from 2011 to 2015 in the 345 ghz band . the observations of the cyg - n44 were previously published by @xcite and @xcite . table [ obs_table ] lists the dates , array configurations , number of antennas , polarimeter mode , and calibrators of the observations of the other five sources . the pointing centers of the sources are listed in table [ map_table ] . the sources were part of the sample of the sma polarization legacy project and were not reported in @xcite except cyg - n44 . the observations before 2013 were carried out in the single receiver mode , and the observations after 2013 were carried out with dual receivers . the dual - receiver mode offers a @xmath4 improvement in the signal - to - noise ratio ( s / n ) of continuum polarization measurement than the single - receiver mode . the bandpass calibrators were also served as the polarization calibrator in each observation . in some observations , when the data of bandpass calibrators were not good enough for polarization calibration , we adopted the leakage table from adjacent days to calibrate polarization . the absolute flux was determined from observations of planets or planetary moons , and the typical flux uncertainty in sma observations was estimated to be @xmath1 20@xmath5 . the data were calibrated in the idl mir package for flux , bandpass and time - dependent gain , and were exported to miriad for further processing . the intrinsic instrumental polarizations ( i.e. , leakage ) of the lower sideband and the upper sideband were calibrated independently with the miriad task gpcal to an accuracy of 0.1@xmath5 @xcite and were removed from the data . the continuum visibilities were averaged from channels absent of line emission to produce the stokes @xmath6 , @xmath7 , and @xmath8 maps . self - calibration for sources with strong stokes @xmath6 continuum emission ( s / n @xmath2 60 ) was performed to refine the gain solutions . the stokes @xmath6 , @xmath7 , and @xmath8 maps were independently deconvolved using the clean algorithm . finally , the polarized intensity , position angle , and polarization percentage along with their uncertainties were derived from the stokes @xmath6 , @xmath7 , and @xmath8 maps using the miriad task impol . map_table ] lists the synthesized beams and rms noises of the maps presented in this paper . the maps were made with the natural weighting to maximize the sensitivity . for the cyg - n43 and cyg - n51 data , we applied a gaussian taper of 2@xmath9 fwhm to enhance the polarization detection . the baselines spanned from 9 k@xmath10 to 85 k@xmath10 , providing a resolution of 3@xmath94@xmath9 and a largest scale of 10@xmath9 with 50% flux recovery @xcite . the on - source time is at least 3 hour per target , resulting in a rms noise level less then 2 mjy beam@xmath11 in the stokes @xmath7 and @xmath8 maps . due to the limited dynamic range of the sma , the rms noise of the continuum stokes @xmath6 emission was significantly higher than that of the stokes @xmath7 and @xmath8 emission . in this paper , we set a polarization cutoff level at s / n equal to 3 . the majority of polarization detections have s / n better than 5 , thus the typical uncertainty in the polarization angle is @xmath1 6@xmath12 @xcite . to illustrate the magnetic fields of the dr21 filament , we also present the polarization observations of james clerk maxwell telescope ( jcmt ) . we obtained the data from the scuba polarimeter legacy catalog compiled by @xcite , which are sampled on a 10@xmath9 grid with a resolution of 20@xmath9 . the noise level of the jcmt stokes @xmath7 and @xmath8 data is @xmath1 10 mjy beam@xmath11 . figure [ fig_dustpol ] presents the sma 880 @xmath0 m and jcmt 850 @xmath0 m polarization maps of the dr21 filament and six massive dense cores . the jcmt map shows that the dr21 filament is elongated in the north south direction . the northern part of the filament is tiled in the northeast direction , and the most southern tip of the filament is tiled in the southeast direction . the magnetic fields of the dr21 filament are mainly in the east west direction , and thus almost perpendicular to the filament . the variation in the direction of magnetic fields is relatively small , except that in the most southern tip of the filament , the magnetic fields are significantly changed to the southeast northwest direction and become parallel to the filament . more detailed studies of the jcmt polarization map of dr21 filament are represented in @xcite and @xcite . the sma maps at 3@xmath94@xmath9 ( @xmath1 0.03 pc or 5000 au ) resolution reveal compact structures of the massive dense cores . owing to the filtering effect of interferometers , the sma is more sensitive to the small - scale magnetic fields in the cores , whereas the jcmt is more sensitive to the large - scale magnetic field of the filament . in contrast to the relatively uniform magnetic fields of the filament , the sma maps display complex structures of the magnetic fields of the cores . in addition , the sma maps show significant depolarization effect toward the center of core , which could be resulted from the distortion of magnetic fields at high column densities @xcite . the dust emission and magnetic fields of cyg - n44 have been studied in several works @xcite . here we focus on the sma maps of the other five massive dense cores . @xmath13 cyg - n38 : the dust emission of cyg - n38 shows a north south elongated peak , consistent with the n@xmath14h@xmath15 10 emission of the core @xcite . in contrast to the east - west orientated jcmt polarization segment , the sma polarization segments of cyg - n38 show northwest - southeast oriented magnetic fields in the northern and southern parts of the core and north - south oriented magnetic fields in the center . the polarized emission is weak at the gaps between the northwest - southeast oriented and north - south oriented magnetic fields . the depolarization gaps might be caused by a perturbation in dust alignment or by an overlap of two components of magnetic fields with different orientations . since our molecular line data show that the line - of - sight velocity in the northern and southern parts of the core is different from that in the center of the core ( paper ii ) , the northwest - southeast oriented and the north - south oriented segments may belong to two independent components of magnetic fields . @xmath13 cyg - n43 : the dust emission of cyg - n43 shows an arc - like structure with the peak emission concentrated at the western end of the arc . similar to cyg - n38 , the polarization segments of cyg - n43 show one group of north - south oriented magnetic fields in the center of the core , and one group of northwest - southeast oriented magnetic fields in the northern and southern parts of the core . depolarization gaps exist at the places between the two groups of polarization segments , again suggesting that the north - south magnetic fields and the northwest - southeast magnetic fields may not be directly connected to each other . @xmath13 cyg - n48 : cyg - n48 has the most extended dust emission in the dr21 filament . the sma map shows that cyg - n48 is composed of several sources , similar to the `` s '' shaped morphology of the 3 mm continuum map @xcite . cyg - n48 also has the most extended polarized emission . the magnetic fields of cyg - n48 rotate smoothly from the northwest - southeast direction in the center of the core to the east - west direction in the east of the core . the change of the magnetic fields from the northwest - southeast direction to the east - west direction appears to be consistent with the curvature of the `` s '' shaped morphology of the core . @xmath13 cyg - n51 : cyg - n51 is significantly elongated along the north south direction , following the large scale filament . the direction of magnetic fields in cyg - n51 is similar to the east west direction of magnetic fields in the filament , revealing that the orientation between magnetic fields and dust emission is consistently perpendicular from filament to core . above the fifth contour in the continuum emission , very few polarization segments are detected , suggesting that the magnetic fields in the center of cyg - n51 could be different from the east west fields revealed by the sma and jcmt maps . @xmath13 cyg - n53 : cyg - n53 is only partially resolved and hence has a round shape with a centrally concentrated peak . although cyg - n53 appears to be a single object in our sma map , 7 fragments in cyg - n53 with sizes of few thousands au have been found in the 1 mm continuum map with higher resolution @xcite . compared to the other cores , cyg - n53 has the weakest polarized emission . the weak polarized emission in the sma map is consistent with the weak polarized emission in the jcmt map . figures [ fig_jcmtsma_south ] and [ fig_jcmtsma_north ] present the jcmt maps , we used the scuba2 commissioning data ( proposal i d m11bec30 ) from the jcmt science archive , which are sampled on a 4@xmath9 grid with a resolution of 15@xmath9 , better than the scuba pol data . ] and sma combined continuum maps at 880 @xmath0 m of the cyg - n38 , cyg - n44 , and cyg - n48 in the south of the filament and the cyg - n43 , cyg - n51 , and cyg - n53 in the north of the filament , respectively . to produce the combined maps , we first scaled the jcmt map at 850 @xmath0 m to the wavelength of 880 @xmath0 m assuming a dust emissivity index of 2.0 , and then combined the scaled jcmt map and sma maps using the feather task in casa and the linmos task in miriad . the resulted resolution and noise level of the combined cyg - n38 , cyg - n44 , and cyg - n48 map are 3.9@xmath9 and 40 mjy beam@xmath11 . the resulted resolution and noise level of the combined cyg - n43 , cyg - n51 , and cyg - n53 map are 4.6@xmath9 and 30 mjy beam@xmath11 . the noise levels of the combined maps are mainly contributed from the scuba2 data . figure [ fig_jcmtsma_south ] shows that the northwestern edge of cyg - n48 is connected to the southeastern edge of cyg - n38 , and the northeastern edge of cyg - n38 is connected to the southwestern edge of cyg - n44 . since the emission between cyg - n44 and cyg - n48 is fainter than the emission between cyg - n38 and cyg - n44 and the emission between cyg - n38 and cyg - n48 , the three cores seem to be aligned in a bow - like structure . this bow - like structure connecting the three cores are also presented in the h@xmath16co@xmath17 and n@xmath18h@xmath17 maps of the dr21 filament @xcite . figure [ fig_jcmtsma_north ] shows that cyg - n43 , cyg - n51 , and cyg - n53 are aligned on the northern ridge of the dr21 filament . the eastern edge of cyg - n43 is connected to the southern edge of cyg - n51 . the emission between cyg - n51 and cyg - n53 is weak in the jcmt map , suggesting that cyg - n53 is more isolated than the other cores . we performed dendrograms @xcite using the _ astrodendro _ software on the jcmt and sma combined maps to characterize the structures of the cores . since all the cores are located on the ridge of the filament , we chose a lower limit of 7@xmath19 for the dendrogram to detect a structure . the required difference to separate structures of the dendrogram was set to be 1@xmath19 , and the minimum area of a structure was one synthesis beam . figures [ fig_jcmtsma_south]b and [ fig_jcmtsma_north]b show the structures identified by the dendrogram . the fwhm sizes along the major axes , the fwhm sizes along the minor axes , the mean sizes ( @xmath20 ) , and the position angles ( @xmath21 ) of the structures computed by the dendrogram are listed in table [ phy_table ] , along with the integrated flux of the structures after a subtraction of the background emission below the cutoffs of the structures . to estimate the errors in the fwhm sizes and in @xmath21 , we performed bootstrap method @xcite by adding additional observational noises to the jcmt and sma combined maps and running dendrograms on those simulated maps . the procedure was repeated for 100 times , and the errors were estimated from the sample of 100 fwhm sizes and 100 @xmath21 identified by dendrograms . the size and flux of cyg - n44 selected by the dendrogram are 11.1@xmath9 and 21.4 jy , compatible with the results of 10.4@xmath9 and 18.8 jy derived from the sma map @xcite . the bright sources mm1 and mm2 in cyg - n44 @xcite can also be separated by the dendrogram . although cyg - n43 , cyg - n48 , and cyg - n51 show clear multiple intensity peaks in the sma maps , the dendrogram failed to identify fragments of the sources owing to the resolution of the combined maps . under the assumptions of optical thin emission and isothermal core , the source mass @xmath22 is related to the integrated flux @xmath23 as @xmath24 where @xmath25 is the gas - to - dust ratio , @xmath26 is the distance of the source , @xmath27 is the dust opacity , and @xmath28 is the planck function at a dust temperature @xmath29 . we adopt a gas - to - dust ratio of 100 , a distance of 1.4 kpc , and a dust opacity of 1.5 @xmath30 g@xmath11 of the cool ( 1030 k ) and dense ( @xmath1 10@xmath31 @xmath32 ) dust mantles @xcite . we adopt a temperature of 20 k for the cores since herschel maps show a @xmath1 1622 k dust temperature of the cores in the dr21 filament @xcite , except a higher temperature of 30 k for the hot core cyg - n44 @xcite . assuming a radius equal to the @xmath20 of the source @xcite , the averaged column and volume density of the source are estimated as follows : @xmath33 @xmath34 the derived masses , column densities , and volume densities are listed in table [ phy_table ] , along with the errors derived from the propagation of the errors in @xmath23 and @xmath20 . although the derived errors are typically smaller than 10% , the uncertainties in the flux calibration , gas - to - dust ratio , distance , dust emissivity and more critically , the temperature may lead to an overall 30%50% uncertainty . the masses estimated from our jcmt and sma combined maps are about a factor of two to four lower than those estimated from single - dish observations @xcite , mainly because we performed dendrograms above a cutoff of 7@xmath19 to separating cores and therefore the core sizes estimated in this work are about half of the sizes in single - dish maps . since the sma is more sensitive to small - scale magnetic fields , we computed the uncertainty - weighted averages of sma polarization angles ( @xmath35 ) to evaluate the averaged orientations of the small - scale magnetic fields of the massive dense cores . figure [ fig_pa_hist ] shows the distributions of the sma polarization angles of the cores with the derived @xmath35 listed in table [ phy_table ] . the distributions of the sma polarization angles have standard deviations of @xmath1 3040@xmath36 , indicating complex magnetic field structures in the cores . in the meantime , the jcmt polarization segments at the peaks of the sources ( @xmath37 , table [ phy_table ] ) were derived to represent the large - scale magnetic fields . to show the difference between the large- and small - scale magnetic fields of the cores , the position angles of @xmath35 and @xmath37 are plotted in figures [ fig_jcmtsma_south]b and [ fig_jcmtsma_north]b . in cyg - n53 , the angle between @xmath35 and @xmath37 is 83@xmath36 , suggesting that the small - scale magnetic fields are almost perpendicular to the large - scale magnetic fields . in the other five cores , @xmath35 are aligned within 45@xmath36 to @xmath37 . in particular , cyg - n48 and cyg - n51 have almost parallel @xmath35 and @xmath37 . in the statistical study of high - mass star - forming regions , @xcite found that 60% of the massive dense cores have small - scale magnetic fields aligned within 40@xmath36 to large - scale magnetic fields , and 40% of the cores have small - scale magnetic fields perfectly perpendicular ( 80@xmath3690@xmath36 ) to large - scale magnetic fields . interestingly , despite that the sample of our work is small , we obtain similar conclusion as @xcite : most massive cores have small- and large - scale magnetic fields aligned within 45@xmath36 and the rest of cores have perpendicular small- and large - scale magnetic fields . figure [ fig_pa_core_b ] shows the correlations between @xmath21 , @xmath37 , and @xmath35 of the six massive dense cores in the dr21 filament . @xmath21 represents the direction of the major axis of cores derived from dendrogram , @xmath37 represents the orientation of magnetic fields at a 0.1 pc scale , and @xmath35 represents the averaged orientation of magnetic fields at a resolution of 0.03 pc . in the panel of @xmath21 versus @xmath37 , the six cores are separated into two groups : one group of parallel alignment and one group of perpendicular alignment , indicating a strong either - parallel - or - perpendicular correlation between @xmath21 and @xmath37 . the correlation between @xmath37 and @xmath35 is weaker : while four cores have parallel @xmath37 and @xmath35 and one core has perpendicular @xmath37 and @xmath35 , cyg - n38 seems to have @xmath37 neither parallel nor perpendicular to @xmath35 . among the three panels , the correlation between @xmath35 and @xmath21 is much less significant that two of the six cores show no preference in the alignment between @xmath35 and @xmath21 . @xcite found an either - parallel - or - perpendicular alignment between the magnetic fields of cores and the elongation of few - pc - long clumps . the strong correlation between @xmath37 and @xmath21 of our data suggests an important role of magnetic fields during the formation of massive dense cores , in agreement with the results of @xcite . however , the weak correlation between @xmath35 and @xmath21 seems to suggest that the small - scale magnetic fields are not as important as the large - scale magnetic field in supporting a core against gravity . while the either - parallel - or - perpendicular alignment appears to be a universal property from clouds to cores , why do the alignment break below the scales of cores ? to explain the break of the either - parallel - or - perpendicular alignment , here we propose a picture that at scales above 0.1 pc , the magnetic fields are strong enough to guide the gas contraction forming structures either aligned or perpendicular to the magnetic fields . when the densities of core scales become large enough to overcome the magnetic fields , gravitational collapse distorts the alignment between the cores and fields , leading to the misalignment below the scales of cores . in this picture , the misalignment between magnetic fields and cores below 0.1 pc scales could also be the origin of the observed misalignment between outflows and magnetic fields @xcite . to evaluate the magnetic field strength in the plane of the sky from dust polarization observations , the chandrasekhar fermi ( hereafter cf ; @xcite @xcite ) equation is the most widely used method . the cf equation suggests that the ratio of turbulence to magnetic field strength would lead to a similar level of variation in the magnetic fields as well as in the velocities , @xmath38 where @xmath39 is the strength of the magnetic field , @xmath40 is the variation about @xmath39 , @xmath41 is the velocity dispersion along the line of sight , and @xmath42 = @xmath39/@xmath43 is the alfv@xmath44n speed at density @xmath45 . recently , a statistical modification of the cf equation has been proposed , which avoids inaccurate estimates of field strengths due to a large - scale field model in the cf method @xcite and includes the signal integration across telescope beam and through line - of - sight depth of the source @xcite . assuming that the plane - of - sky component of magnetic fields is composed of a large - scale ordered component @xmath46 and a small - scale turbulent component @xmath47 , @xcite suggests that if the turbulence correlation length @xmath48 is much smaller than the thickness of the cloud @xmath49 , the ratio of @xmath47 to @xmath46 can be evaluated from the dispersion function of the observed polarization angles ( i.e. , angular dispersion function ) , @xmath50\rangle \simeq \sum_{j=1}^\infty a_{2j}^{\prime}l^{2j } + \left [ \frac{n}{1+n\langle b_0 ^ 2\rangle/\langle b_t^2\rangle}\right ] \\ \times\left\{\frac{1}{n_1}\left[1-e^{-l^2/2(\delta^2 + 2w_1 ^ 2)}\right ] + \frac{1}{n_2}\left[1-e^{-l^2/2(\delta^2 + 2w_2 ^ 2)}\right]\right . - \frac{2}{n_{12}}\left[1-e^{-l^2/2(\delta^2+w_1 ^ 2+w_2 ^ 2)}\right ] \right\ } , \label{eq_adf}\end{gathered}\ ] ] where @xmath51 is the difference between polarization angles measured for all pairs of polarization segments separated by a distance @xmath52 , the summation is a taylor expansion representing the structure in the @xmath46 that does not involve turbulence , @xmath53 and @xmath54 are the widths ( i.e. , the fwhm beam divided by @xmath55 ) of the synthesis beam and the low - frequency filtering effect , and @xmath56 is the number of turbulent cells along the line of sight obtained by @xmath57 @xmath58 @xmath59 @xmath60 the corresponding normalized signal - integrated turbulence autocorrelation function is @xmath61 \left [ \frac{1}{n_1 } e^{-l^2/2(\delta^2 + 2w_1 ^ 2 ) } \right . + \frac{1}{n_2 } e^{-l^2/2(\delta^2 + 2w_2 ^ 2 ) } - \frac{2}{n_{12 } } e^{-l^2/2(\delta^2+w_1 ^ 2+w_2 ^ 2)}\right].\end{gathered}\ ] ] since @xmath47 is the source of perturbation in @xmath46 , @xmath47 is identical to the @xmath40 in the cf equation ( equation 4 ) . therefore , one can apply the @xmath62 derived from equation 5 to evaluate the magnetic field strength on the plane of sky based on the cf equation @xmath63^{-1/2}. \label{eq_cf}\ ] ] we adopt the statistical analysis of angular dispersion function @xcite to estimate the magnetic field strength in the massive dense cores of the dr21 filaments . the polarization segments of cyg - n53 are too few to perform the analysis . figure [ fig_adf ] shows the angular dispersion functions of the other five cores in the dr21 filament using the sma polarization segments . owing to the limited angular resolution , the angular dispersion function is close to zero when the length scale @xmath52 is smaller than the beam . the angular dispersion functions of the cores other than cyg - n44 smoothly increase with the length scale and reach a maximum at 5@xmath910@xmath9 . since the angular dispersion function at small distance is dominated by the turbulent magnetic fields and at large distance dominated by the ordered magnetic fields , the maximum at 5@xmath910@xmath9 indicates a characteristic scale of @xmath1 0.05 pc ( 10@xmath64 au ) that the magnetic fields change from being turbulent into ordered structures . at a large distance , the angular dispersion functions of the four cores can be divided into two categories . in cyg - n44 , cyg - n48 , and cyg - n51 , the angular dispersion functions at scales @xmath2 8@xmath9 are close to random fields ( @xmath65 ; @xcite @xcite ) , likely representing a slow transition of magnetic fields from random structures in the cores to the ordered structure of the filament . in cyg - n38 and cyg - n43 , the maximum of angular dispersion function is higher than the random field value , and the decrease of angular dispersion function is steep with a decline of @xmath1 0.4 within a distance of 4@xmath9 . why the the angular dispersion functions of cyg - n38 and cyg - n43 are different from those of the other three cores ? cyg - n38 and cyg - n43 both may have two groups of polarization segments in different orientations ( see section 3.1 ) . the polarization segments within each group contribute low angular dispersions at small and large distances , whereas the angular dispersion measured from one group to the other group contributes a high value at an intermediate distance , resulting in the low - high - low profiles of the angular dispersion functions in cyg - n38 and cyg - n43 . in addition to the low - high - low profiles , the distributions of the data also suggest that the angular dispersion functions of cyg - n38 and cyg - n43 are resulted from the two groups of magnetic fields rather than random fields . the gray boxes in figure [ fig_adf ] represent the interquartile distributions ( 25%75% ) of the angular dispersion functions . the angular dispersion functions measured within each group should give a narrower interquartile distribution than those measured from one group to anther . therefore , the narrower interquartile distributions at small and large distances and the wider interquartile distributions at intermediate distances of cyg - n38 and cyg - n43 data are in agreement with the angular dispersion functions from two groups of magnetic fields . we use the nonlinear least - squares marquardt levenberg algorithm to evaluate the parameters of @xmath48 , @xmath66 , and @xmath67 in equation [ eq_adf ] , and then use the cf equation ( equation [ eq_cf ] ) to derive the magnetic field strength in the plane of the sky . for our observations with beams of 3@xmath94@xmath9 and a low - frequency filter of 10@xmath9 , we use @xmath53 of 1@xmath685 and @xmath54 of 4@xmath682 in the fitting . we use the mean fhmw derived from dendrograms as the effective thickness @xmath69 of our sources . we estimate the @xmath70 from the velocity dispersion of the h@xmath16cn 43 emission in our data ( paper ii ) , since the emission of h@xmath16cn is well correlated with the dust emission in the massive dense cores of the dr21 filament @xcite . the fitting range is from the beam to 15@xmath9 , and the parameters @xmath67 are reduced to its first order @xmath71 because the fitting range is small . due to the steep drop in the angular dispersion function of cyg - n43 , the fitting of cyg - n43 failed to converge unless the maximum fitting distance narrowed down to 10@xmath9 . although the fitting of cyg - n43 can not reproduce the angular dispersion function at large distance , the derived magnetic field strength of cyg - n43 should still be reliable , since the fitting of @xmath66 is mainly regulated by the angular dispersion at small distance and @xmath66 is the only output of the fitting required in the cf equation . the best fits of angular dispersion functions are shown in figure [ fig_adf ] , and the fitted parameters are listed in table [ phy_table2 ] . the correlation lengths @xmath48 of the cores are from 2@xmath682 to 3@xmath688 ( 15 to 30 mpc ) , well resolved by the @xmath53 of 1@xmath685 . the number of turbulent cells @xmath56 along the line of sight of the sources is about 57 , and the @xmath66 of the sources is about 35 , both suggesting that the local turbulent magnetic fields are stronger than the ordered large - scale magnetic fields . the @xmath66 of the filament is 0.16 , significantly smaller than those of the cores . considering that the @xmath46 in the calculations of filament and cores are the same component of the ordered large - scale magnetic field , the increase of @xmath66 indicates an enhancement of the turbulent magnetic fields from the filament to the cores . the derived magnetic field strengths on the plane of sky are @xmath1 0.56 mg in cyg - n38 , @xmath1 0.42 mg in cyg - n43 , @xmath1 1.71 mg in cyg - n44 , @xmath1 0.48 mg in cyg - n48 , and @xmath1 0.46 mg in cyg - n51 . using the same sma data , the 1.71 mg of cyg - n44 in this work derived using the dispersion function for interferometry data @xcite is more accurate than the 2.1 mg in @xcite derived using the dispersion function for single - dish data @xcite . in addition , 1.71 mg is closer than 2.1 mg to the values of 0.91.3 mg derived from the bima polarization data @xcite , 1.7 mg derived from the velocity dispersion study of the jcmt data @xcite , and 1.2 mg derived from the carma polarization data @xcite . for the other four cores , the strengths of 0.40.6 mg are close to the strength of the dr21 filament . to estimate the total magnetic field strength @xmath72 , we adopt the zeeman observations at a resolution of 23@xmath9 of the cn lines toward cyg - n44 which probe two line - of - sight magnetic field strengths ( @xmath73 ) in two velocity components : @xmath73 = -0.36 mg at @xmath74 = -4.7 km s@xmath11 and @xmath73 = -0.71 mg at @xmath74 = -0.9 km s@xmath11 @xcite . since the -4.7 km s@xmath11 component is found to be associated with the cyg - n44 core and the -0.9 km s@xmath11 is the systematic velocity of the filament , the line - of - sight field strengths of cyg - n44 and the dr21 filament are suggested to be -0.36 mg and -0.71 mg , respectively @xcite . for the other four cores , we assume an energetic equal partition of magnetic fields ( @xmath75 ) . the derived total magnetic field strengths are listed in table [ phy_table2 ] . considering the assumption of the energetic equal partition and the dependance of the cf equation ( equation [ eq_cf ] ) to the density which has a 30%-50% uncertainty due to the temperature ( section 3.3 ) , we estimate that the uncertainty in the total magnetic field strength is about 50% of the value . figure [ fig_b_den ] shows the @xmath73 versus the volume density of the massive dense cores and the dr21 filament , assuming the most probable value of @xmath76 . the magnetic field strengths of the cores are systematically smaller than the strength of the filament , indicating that the true @xmath73 of the cores may be stronger than what we assumed . using the data points of the five massive dense cores , a single power - law fit gives a slope of 0.54 @xmath77 0.30 . the slope of 0.54 is mainly accounted for the factor of @xmath78 in the cf equation ( equation [ eq_cf ] ) , and the large uncertainty in the fit is resulted from the @xmath41 and @xmath62 which are not scaled with the volume density . the slope of how magnetic fields scale with gas density offers an important diagnosis : a slope of 2/3 is expected for weak magnetic field model with flux conservation @xcite and 0.440.5 is predicted in the ambipolar diffusion model at high density regimes ( n@xmath79 @xmath80 @xmath32 , @xcite @xcite ; @xcite @xcite ) . the recent comprehensive study of zeeman observations gives a value of 0.65 @xmath77 0.05 @xcite , which is within 1@xmath19 to our value of 0.54 @xmath77 0.30 . owing to the @xmath78 factor in equation [ eq_cf ] , the magnetic field strengths derived from the cf equation are not sensitive to study the scaling between magnetic field strength and gas density . we compute the parameters of the sonic mach number @xmath81 , the alfv@xmath44n mach number @xmath82 , and the ratio of thermal to magnetic pressures @xmath83 as the following : @xmath85 @xmath86 @xmath87 where @xmath88 is the sound speed at temperature @xmath29 using adiabatic index @xmath89 and mean molecular weight @xmath90 , @xmath91 is the one - dimensional velocity dispersion . the derived parameters are listed in table [ phy_table3 ] . the dr21 filament has supersonic and sub - alfv@xmath44nic mach numbers , revealing that magnetic fields are more important than turbulence in the filament . the @xmath83 of the filament is less than one , indicating that the magnetic pressure is stronger than the thermal pressure . since the linewidth in massive cores includes components of infall , rotation , and turbulence ( see 4.3.2 ) , here we adopt the @xmath92 of cores as the broadening of non - thermal motions , instead of turbulence . the supersonic and trans- to super - alfv@xmath44nic mach numbers of massive cores indicate strong non - thermal motions . the cores typically have a @xmath83 value less than one , suggesting that the magnetic pressure , though weaker than the non - thermal pressure , is stronger than the thermal pressure . the relation of @xmath93 is the same as the results based on the zeeman observations toward cyg - n44 @xcite . to estimate the relevance of the magnetic fields with respect to the gravitaty , we derive the mass - to - flux ratio @xmath94\left[b_{total}/\mu g\right]^{-1}$ ] in units of the critical value @xmath95 @xcite . the derived mass - to - flux ratios of the massive cores and the dr21 filament ( see table [ phy_table3 ] ) are about equal or larger than one , suggesting that both in the cores and in the filament , the magnetic fields can not provide enough support against gravitational collapse . for cyg - n44 , the mass - to - flux ratio of 3.4 is consistent with the values of 2.810 estimated from independent methods @xcite . the virial theorem is a useful tool for analyzing the effects of kinematics , magnetic fields , and self - gravity in molecular clouds @xcite , which can be written as @xmath96 here @xmath6 is a quantity proportional to the inertia of the cloud , and the sign of @xmath97 determines whether the cloud is expanding ( @xmath98 ) or contracting ( @xmath99 ) . for a cloud with a fixed mass @xmath22 , the term @xmath100 is the total kinetic energy in the cloud including both thermal and non - thermal ( turbulence , rotation , infall , and etc . ) the term @xmath101 is the surface kinetic term . the term @xmath102 is the magnetic energy for a cloud with no force from ambient magnetic field . the gravitational term @xmath103 of a spherical cloud with uniform density and radius @xmath104 is @xmath105 the kinetic and gravitational terms of a filamentary cloud are different from those of a spherical cloud @xcite in that @xmath106 @xmath107 where @xmath108 is the length of the filamentary cloud . the derived ratios of @xmath109 and @xmath110 are listed in table [ phy_table3 ] where we adopt a mass of 15210 @xmath111 and a length of 4.1 pc for the dr21 filament @xcite . for the massive dense cores , the typical values of @xmath109 @xmath2 1 @xmath2 @xmath110 reveal a relation of @xmath112 , suggesting that kinematic energy is more important than magnetic energy during the formation of massive dense cores . although the uncertainty in @xmath72 could be as large as 50% of the value , considering that the @xmath109 is at least five times larger than the @xmath110 , the relation of @xmath113 would not change regardless of the uncertainty in @xmath72 . in contrary to the relation of @xmath113 of the massive dense cores , the filament has the @xmath110 significantly larger than the @xmath109 and therefore @xmath114 . the transition from the @xmath114 in filament to the @xmath113 in cores reveals a decrease of magnetic energy and/or an increase of kinematic energy from the parent filament to the daughter cores . the variation of the magnetic and kinematic energies from filament to cores is also revealed in the transition from @xmath115 of the filament to the @xmath116 of the cores obtained from the analysis of the angular dispersion functions . in star formation models , the diffusion of magnetic flux out of cores is expected through the process of ambipolar diffusion arising from the decoupling of neutral particles and magnetic fields in weakly ionized gas @xcite or reconnection diffusion arising from the magnetic reconnection in turbulent gas @xcite . in observations , the presence of turbulent ambipolar diffusion in the massive cores of the dr21 filament is suggested from the different linewidths between ionic and neutral molecular line profiles @xcite , and the the infall motions , rotation motions , and outflows of the cores @xcite might explain the increase of kinematic energy . the @xmath117 listed in table [ phy_table3 ] represents whether the cloud is expanding ( @xmath118 ) or contracting ( @xmath119 ) without considering the surface term @xmath101 . the @xmath117 of the dr21 filament is less than one , in agreement with the global collapse inferred from molecular line observations @xcite . however , the @xmath120 of the massive cores seems to be inconsistent with the signatures of star - forming activities supported by the high level of fragments @xcite , infall @xcite , and outflows @xcite of the cores . how to explain the the discrepancy between the derived virial parameters and the fact that there is star formation in the cores ? the width of the dr21 filament is 0.34 pc @xcite , and the mean radius of our six cores is 0.068 pc . given @xmath121 , the @xmath122 changes by a factor of 5 when gas collapses from filament to core . the collapse can transfer the @xmath122 in the filament to the @xmath123 and @xmath124 of cores , resulting in an increase of @xmath117 from filament to core . in addition , a large portion in the @xmath123 of core could be in the form of kinematic energy , instead of turbulent energy . the velocity dispersions of 1.22 km s@xmath11 of the cygnus x massive cores is significantly larger than the virial velocity dispersions @xmath125 0.4 km s@xmath11 of a 20 m@xmath3 and 0.1 pc core with the geometric factor @xmath126 equal to unity @xcite . cyg - n48 and cyg - n53 exhibit infall signatures in the h@xmath16co lines @xcite , whereas cyg - n43 , cyg - n44 , and cyg - n48 show rotation - like structures in our molecular line data ( @xcite @xcite , paper ii ) . the infall and rotation motions could be the origin of the large linewidths in the cygnus x massive cores . to estimate the contribution of rotational energy to @xmath123 , we assume solid - body rotation of core with rotational velocity @xmath127 and rotational energy @xmath128 . adopting the velocity gradient of @xmath1 30 km s@xmath11 pc@xmath11 from our data as @xmath127 , the ratio of @xmath129 to @xmath123 is in the order of unity . since the @xmath123 of the core is mostly contributed by kinematic energy , the turbulent energy should be smaller than @xmath122 . the large reduction of @xmath122 from filament to core may lead to the misalignment between @xmath21 and @xmath35 through infall and rotation motions of the cores . the correlation between the magnetic field orientation inferred from dust polarization maps and the intensity gradient of stokes @xmath6 dust continuum emission contours has been proposed to evaluate how gravity regulates magnetic fields @xcite . based on the assumption that a change in intensity gradient is a measurement for the resulting direction of motion in the mhd force equation , the magnetic field orientation should be aligned to the intensity gradient when gravity overwhelms magnetic fields , and misalignment between the magnetic field orientation and the intensity gradient is suggested when gravity is relatively less important to magnetic fields . figure [ fig_den_pol ] displays magnetic field segments and intensity gradient segments of the six massive cores . figure [ fig_den_pol_hist ] shows stacked histograms of position angles between magnetic field segments and intensity gradient segments . the rotation - like cores cyg - n43 , cyg - n44 , cyg - n48 show misalignment between the magnetic field orientation and the intensity gradient , whereas the magnetic fields in the non - rotation cores cyg - n38 , cyg - n51 , cyg - n53 tend to be aligned with the intensity gradient , suggesting that gravity is important in regulating the orientation of magnetic fields in non - rotation cores but less important in rotation - like cores . since the rotational energy of core could be as large as than the gravitation energy , the rotation motion can distort magnetic fields at the core scale , resulting in the misalignment between magnetic fields and the intensity gradient of rotation - like cores . interestingly , the rotation - like cores have @xmath35 parallel to the position angles of velocity gradient ( @xmath130 , table 3 ) inferred from our molecular line data . figure [ fig_core_sma_vg ] shows correlations between @xmath35 , @xmath21 , and @xmath130 of the sources in the sma polarization legacy project with clear velocity gradients of rotation - like structures at @xmath1 0.1 pc scale ( g5.89 @xcite @xcite ; w51e8 @xcite @xcite , @xcite @xcite ; w51n @xcite @xcite , @xcite @xcite ; g192.12 @xcite @xcite ; g35.2 @xcite @xcite ; g240.31 @xcite @xcite ) . the rotation - like cores show a random alignment between @xmath35 and @xmath21 , but strong either - parallel - or - perpendicular correlation between @xmath35 and @xmath130 , suggesting that rotation plays a more important role than gravity in determining the orientation of magnetic fields . overall , the comparisons between magnetic fields and the intensity gradient as well as the correlation between @xmath21 , @xmath35 , and @xmath130 reveal a picture that magnetic fields of non - rotating core are regulated by gravity , and magnetic fields of rotating - like core are regulated by rotation motion , consistent with the virial paramenter of @xmath112 of the cores . we present submillimeter array 880 @xmath0 m polarization observations of six massive dense cores in the dr21 filament . with the complementary archival polarimetric observations from scupol of the jcmt @xcite , we are able to characterize the magnetic field properties from the filament to the cores . our main results are the following : 1 . the sma dust polarization shows complex magnetic field structures in the massive dense cores , in contrast to the ordered parsec - scale magnetic fields of the filament . significant depolarization effect is found toward the center of core , and the variation in the sma polarization angles is @xmath1 3040@xmath36 in the cores , again both indicating complex magnetic field structures in the cores . the major axis of core appears to be either parallel or perpendicular to the large - scale magnetic fields inferred from the jcmt polarization map . however , the major axis of the cores shows no preference of alignment to the small - scale magnetic fields inferred from the sma polarization map . the transition from the either - parallel - or - perpendicular alignment to the random alignment suggests that in the formation of massive dense cores , the magnetic fields below 0.1 pc scales become less important as compared to gravity than the magnetic fields above 0.1 pc scales . 3 . our analysis of angular dispersion functions of sma polarization segments yields plane - of - sky magnetic field strengths of 1.7 mg in cyg - n44 and about 0.5 mg in other four cores . most of the cores have the angular dispersion functions similar to the dispersion of random magnetic field . the derived number of turbulent cells and the derived ratio of turbulent magnetic fields to ordered magnetic fields are both much greater than unity , indicating that the magnetic fields in the cores are dominated by turbulent components . the virial parameters show that in the filament , the gravitational energy @xmath122 is dominant over magnetic @xmath124 and kinematic @xmath123 energies , and the gravitational collapse can transfer the @xmath122 of the filament to the @xmath123 and @xmath124 of the cores , resulting in the relations of @xmath112 and @xmath120 in the cores . the virial parameters reveal an overall picture of weakly magnetized cores in a strongly magnetized filament , suggesting that the kinematics arising from gravitational collapse must become more important than magnetic fields during the evolution from filament to massive dense cores . part of the data were obtained in the sma legacy project : filaments , magnetic fields , and massive star formation ( pi : qizhou zhang ) . t. c. c. acknowledges the support of the smithsonian predoctoral fellowship and the university consortium of alma - taiwan ( ucat ) graduate fellowship . t. c. c. and s. p. l. are thankful for the support of the ministry of science and technology ( most ) of taiwan through grants 102 - 2119-m-007 - 004-my3 , 104 - 2119-m-007 - 021 , 105 - 2119-m-007 -022 -my3 , and 105 - 2119-m-007 -024 . q. z. acknowledges the support of the si scholarly studies awards `` are magnetic fields dynamically important in massive star formation ? '' j. m. g. acknowledges the support from the micinn ( spain ) aya2014 - 57369- c3 grant and the mecd ( spain ) prx15/00435 travel grant . k. q. acknowledges the support from national natural science foundation of china ( nsfc ) through grants nsfc 11473011 and nsfc 11590781 . ccccccccc 2011 jul 09 & subcompact & 7 & single & mwc349a & uranus & 3c279 & 3c279 & cyg - n38 + 2011 jul 12 & subcompact & 6 & single & mwc349a & uranus & 3c279 & 3c279 on 2011 jul 13 & cyg - n48 , cyg - n53 + 2011 jul 13 & subcompact & 7 & single & mwc349a & mwc349a & 3c279 & 3c279 & cyg - n38 , cyg - n48 , cyg - n53 + 2012 jul 03 & compact & 7 & single & mwc349a & titan & 3c279 , 3c84 & 3c279 on 2012 jul 04 & cyg - n38 , cyg - n48 + 2012 jul 04 & compact & 7 & single & mwc349a & titan & 3c279 & 3c279 & cyg - n38 , cyg - n48 + 2012 jul 05 & compact & 7 & single & mwc349a & titan & 3c279 , 3c84 & 3c84 & cyg - n38 , cyg - n48 + 2012 aug 07 & subcompact & 6 & single & mwc349a & uranus & 3c279 & 3c84 on 2012 aug 08 & cyg - n38 , cyg - n48 + 2012 aug 08 & subcompact & 6 & single & mwc349a & uranus & 3c84 & 3c84 & cyg - n38 , cyg - n48 + 2012 aug 09 & subcompact & 6 & single & mwc349a & uranus & 3c84 & 3c84 on 2012 aug 08 & cyg - n38 , cyg - n48 + 2014 jun 19 & compact & 7 & dual & mwc349a & neptune & 3c279 , 3c454.3 & 3c454.3 & cyg - n43 , cyg - n51 + 2014 jun 21 & compact & 7 & dual & mwc349a & neptune & 3c454.3 & 3c454.3 on 2014 jun 19 & cyg - n43 , cyg - n51 + 2015 jun 08 & subcompact & 7 & dual & mwc349a & titan & 3c279 & 3c279 & cyg - n53 + 2015 jun 12 & subcompact & 7 & dual & mwc349a & neptune & 3c279 , 3c454.3 & 3c279 & cyg - n43 + 2015 jun 19 & subcompact & 7 & dual & mwc349a & titan & 3c279 & 3c279 & cyg - n51 + lcccccccccc cyg - n38 & 20:38:59.11 & 42:22:25.96 & & 9.4/0 & & 3.6 @xmath131 2.9 & 65 & & 15/1.6 & dr21(oh)-w + cyg - n43 & 20:39:00.60 & 42:24:34.99 & & 0/3.1 & & 4.6 @xmath131 3.5 & 55 & & 10/1.6 & w75s - fir1 + cyg - n44@xmath132 & 20:39:01.20 & 42:22:48.50 & & 4.7/0 & & 3.9 @xmath131 3.4 & 56 & & 20/3.1 & dr21(oh ) + cyg - n48 & 20:39:01.34 & 42:22:04.89 & & 7.5/0 & & 3.3 @xmath131 3.0 & 63 & & 13/1.7 & dr21(oh)-s + cyg - n51 & 20:39:02.40 & 42:24:59.00 & & 0/3.5 & & 4.4 @xmath131 3.4 & 54 & & 8.9/1.6 & w75s - fir2 + cyg - n53 & 20:39:02.96 & 42:25:50.99 & & 1.5/2.1 & & 4.4 @xmath131 3.0 & 60 & & 14/1.9 & + ccccccccccccc cyg - n38 & 4.3 @xmath77 0.3 & 11.5 @xmath77 0.4 & 7.8 @xmath77 0.6 & 64 @xmath77 3 & & -27.0 @xmath77 3.7 & 85.0 @xmath77 5.1 & -50.0 @xmath77 36.2 & & 59 @xmath77 4 & 2.2 @xmath77 0.2 & 1.1 @xmath77 0.1 + cyg - n43 & 1.2 @xmath77 0.1 & 14.3 @xmath77 0.3 & 5.7 @xmath77 0.4 & 61 @xmath77 2 & & -74.4 @xmath77 1.7 & -44.5 @xmath77 24.4 & -24.0 @xmath77 38.1 & @xmath1 -31 & 16 @xmath77 1 & 0.7 @xmath77 0.1 & 0.4 @xmath77 0.03 + cyg - n44 & 21.4 @xmath77 0.3 & 11.1 @xmath77 0.5 & 9.4 @xmath77 0.1 & 69 @xmath77 2 & & 13.3 @xmath77 3.7 & 78.2 @xmath77 5.9 & 47.6 @xmath77 30.1 & @xmath1 63 & 167 @xmath77 2 & 5.3 @xmath77 0.2 & 2.5 @xmath77 0.1 + cyg - n44 mm1 & 2.0 @xmath77 0.1 & @xmath133 4.3 @xmath77 0.1 & @xmath133 3.2 @xmath77 0.1 & @xmath133 25 @xmath77 1 & & 52.3 @xmath77 1.3 & & & & 16 @xmath77 1 & @xmath134 3.8 @xmath77 0.2 & @xmath134 4.8 @xmath77 0.4 + cyg - n44 mm2 & 0.2 @xmath77 0.1 & @xmath133 2.5 @xmath77 0.1 & @xmath133 2.1 @xmath77 0.1 & @xmath133 16 @xmath77 1 & & 16.0 @xmath77 9.6 & & & & 2 @xmath77 1 & @xmath134 1.0 @xmath77 0.5 & @xmath134 2.1 @xmath77 0.1 + cyg - n48 & 7.5 @xmath77 0.3 & 14.3 @xmath77 0.2 & 7.2 @xmath77 0.4 & 69 @xmath77 2 & & -71.5 @xmath77 1.0 & -59.2 @xmath77 9.3 & -60.8 @xmath77 44.0 & @xmath1 -72 & 102 @xmath77 4 & 3.3 @xmath77 0.2 & 1.5 @xmath77 0.1 + cyg - n51 & 3.3 @xmath77 0.2 & 16.7 @xmath77 0.3 & 7.4 @xmath77 0.3 & 75 @xmath77 2 & & 7.5 @xmath77 0.7 & -84.9 @xmath77 8.9 & 89.3 @xmath77 35.3 & & 45 @xmath77 3 & 1.2 @xmath77 0.1 & 0.5 @xmath77 0.03 + cyg - n53 & 1.9 @xmath77 0.1 & 4.0 @xmath77 0.3 & 2.6 @xmath77 0.1 & 22 @xmath77 1 & & -14.3 @xmath77 3.4 & 73.7 @xmath77 22.2 & -23.5 @xmath77 18.3 & & 26 @xmath77 1 & 8.3 @xmath77 0.2 & 12.2 @xmath77 0.2 + [ phy_table ] cccccccc cyg - n38 & 26 @xmath77 12 & 3.8 @xmath77 2.4 & -1.9 @xmath77 1.6 & 4.7 @xmath77 0.1 & 1.5 @xmath77 0.2 & 0.56 @xmath77 0.19 & @xmath1 0.69 + cyg - n43 & 26 @xmath77 31 & 2.8 @xmath77 5.3 & 2.5 @xmath77 6.6 & 4.5 @xmath77 0.3 & 1.6 @xmath77 0.4 & 0.42 @xmath77 0.41 & @xmath1 0.51 + cyg - n44 & 22 @xmath77 5 & 2.0 @xmath77 0.2 & 0.8 @xmath77 0.4 & 5.2 @xmath77 0.5 & 2.2 @xmath77 0.7 & 1.71 @xmath77 0.55 & @xmath1 1.75 + cyg - n48 & 15 @xmath77 3 & 4.5 @xmath77 1.2 & 0.1 @xmath77 0.4 & 7.4 @xmath77 2.2 & 1.2 @xmath77 0.2 & 0.48 @xmath77 0.10 & @xmath1 0.59 + cyg - n51 & 22 @xmath77 6 & 2.9 @xmath77 0.6 & 0.1 @xmath77 0.7 & 5.7 @xmath77 0.6 & 1.6 @xmath77 0.2 & 0.46 @xmath77 0.08 & @xmath1 0.56 + dr21 filament@xmath132 & 151 @xmath77 21 & @xmath1 0.16 & & @xmath1 1 & @xmath1 0.8 & @xmath1 0.62 & @xmath1 0.94 + [ phy_table2 ] ccccccccccc cyg - n38 & 0.34 & 1.5 & 0.9 & 7.5 & 2.7 & 0.27 & 2.4 & 1.4 & 0.19 & 3.0 + cyg - n43 & 0.34 & 1.6 & 1.2 & 8.0 & 2.4 & 0.18 & 1.0 & 5.7 & 0.99 & 12.4 + cyg - n44 & 0.42 & 2.2 & 1.6 & 9.0 & 2.4 & 0.14 & 2.3 & 1.2 & 0.20 & 2.6 + cyg - n48 & 0.34 & 1.2 & 0.7 & 6.0 & 3.0 & 0.50 & 4.3 & 0.6 & 0.06 & 1.3 + cyg - n51 & 0.34 & 1.6 & 1.1 & 8.0 & 2.4 & 0.18 & 1.6 & 2.5 & 0.42 & 5.4 + dr21 filament & 0.30@xmath132 & 0.8 & 3.0@xmath135 & 4.6 & 0.5 & 0.02 & 3.4@xmath136 & 0.04 & 0.28 & 0.36 + [ phy_table3 ] m toward the dr21 filament @xcite . the contours show the total dust continuum emission with levels of 5% , 10% , 15% , 20% , 30% , ... , 90% of the peak . the red and orange segments represent the polarization detections at s / n above 3 and between 23 , respectively . the segments show the magnetic field direction with its length proportional to the polarization percentage . the triangles mark the positions of the millimeter sources in @xcite . left and right panels : the sma dust polarization maps at 880 @xmath0 m toward six massive cores in the dr21 filament . in the maps of cyg - n38 , cyg - n43 , cyg - n48 , cyg - n51 , and cyg - n53 , the contour levels are -5 , -3 , 3 , 5 , 10 , 20 , ... , 70 times the rms noises of the maps . the cyg - n44 map is the sma subcompact , compact and extended map in @xcite , which shows contours at 4% , 8% , 18% , 28% , ... , 98% of the peak . the polarized intensity is shown in gray image with the scale bar at top - left corner of the figure in units of jy beam@xmath11 . the blue segments show the sma polarizations of s / n @xmath137 3 with the directions of magnetic fields and the lengths proportional to the polarization percentage . in each panel , the synthesized beam is plotted as a filled grey ellipse . the size of the sma panels is 32@xmath9 , equal to the diameter of sma primary beam at 880 @xmath0 m . ] m continuum maps of the cyg - n38 , cyg - n44 , and cyg - n48 . ( a ) the gray scale in units of jy beam@xmath11 and the contours show the combined dust stokes @xmath6 emission . the contours are plotted at 5% , 10% , 15% , 25% , ... , 95% of the peak . the dotted circles represent the sma primary beams of the sources . the synthesized beam of the combined map is plotted as a filled grey circle . the red , orange , and blue segments are the segments in figure 1 with equal length ( please note that we do not combine the jcmt and sma stokes @xmath7 and @xmath8 maps ) . ( b ) the structures selected by dendrograms overlaid on the jcmt and sma combined map . the contours show the boundries of the structures , and the ellipses are plotted using the major fwhm , minor fwhm , and @xmath21 computed by dendrograms . for each source , the red ( or orange if s / n between 23 ) and blue segments show the orientations of @xmath37 and @xmath35 , respectively . ] , @xmath37 , and @xmath35 . the widths of the shadow zones are the means of the error bars in @xmath37 and @xmath35 . in order to present the correlation , some of the position angles are shifted by 180@xmath12 . ] \rangle$ ] , binned in 1@xmath685 for a nyquist sampling of the 3@xmath94@xmath9 beam . the dots represent the mean values of the data , and the error bars show the standard deviations of the mean values . the gray boxes display the interquartile ranges of the data . the blue line shows the best fit to the data ( equation 5 ) , and the red line shows the ordered component @xmath138 of the best fit . the dotted vertical and horizontal lines note the beam size and the expected value for random magnetic fields , respectively . bottom panels : the dots represent the correlated component of the best fit to the data . the red line shows the correlation due to the beam ( equation 18 in @xcite @xcite ) , and the blue line shows the turbulent component @xmath139 ( equation 10 ) of the best fit . the dashed line marks the zero value . ] ine - of - sight magnetic field strength against volume density of molecular cloud . the red circles and red square denote the massive cores and the dr21 filament in this work . the triangles denote the zeeman observations in @xcite . note that despite the zeeman observations are probing similar density scales , the spatial scales of the zeeman observations are about 10 times larger than our observations . the dotted line represents the model of maximum magnetic field strength scaling with volume density as @xmath140 proposed in @xcite along with its uncertainty shown as the shadow zone . the solid line represents the single power - law fit of the massive cores in the dr21 filament . [ fig_b_den ]

we performed and band surface photometry for e / s0 galaxies in a nearby rich cluster abell 2199 to investigate their color gradients ( ) . our aims are to study statistical properties of the color gradients and , by comparing them with those in less dense environments , to examine environmental dependence of color gradients in elliptical galaxies . we studied the distribution of the color gradients in the cluster ellipticals and found that the mean value of the color gradients is mag / dex , which can be converted to a metallicity gradient ( ) of . the gradient seems to be comparable to that expected by a recent monolithic collapse model . we further studied the relations between the color gradients and global properties of the galaxies . our data suggest that for the galaxies brighter than an band magnitude of mag , which is roughly equivalent to at the distance of the cluster , brighter galaxies tend to have steeper color gradients . also , for the galaxies with effective radii larger than , which nearly corresponds to considering the correlation between galaxy luminosity and effective radius for elliptical galaxies , the galaxies with larger effective radii seem to have steeper color gradients . these trends could appear if elliptical galaxies formed through the monolithic collapse scenario . on the contrary , it is found based on the published data that such trends are not clearly seen for elliptical galaxies in less dense environments , suggesting that elliptical galaxies in sparse environments formed by galaxy mergers , though the distribution of the color gradients is quite similar to that in the rich cluster . in other words , our data and those in the literature suggest that there is an environmental dependence in the relationship between color gradient and global properties of elliptical galaxies , while the distribution of the values of color gradients is nearly independent of galaxy environment . these results do not fully support the view that ellipticals in rich clusters formed through the monolithic collapse while those in sparse environments formed through galaxy mergers , because the latter ellipticals are expected to have color gradients shallower on average than the former . this apparent conflict may be reconciled by taking into account star formation and subsequent chemical enrichment induced by galaxy merger , which may allow a merger remnant to acquire a metallicity gradient and thus compensate the dilution of the existing metallicity gradients in the progenitors by mixture of stars during the galaxy merger .

astro-ph0304404

elliptical galaxies which reside in a cluster show a tight color - magnitude ( cm ) relation ; brighter ellipticals are systematically redder than fainter ones ( e.g. , bower , lucey , & ellis 1992 ) . recently , it has been found that the relation still holds even in distant rich clusters up to @xmath10 ( e.g. , stanford , eisenhardt , & dickinson 1998 ) . based on these observational results , galaxy evolution models taking into account chemical evolution claim that ellipticals in rich clusters formed at high redshift ( e.g. , @xmath11 ) by a monolithic collapse of a gas cloud ( e.g. , kodama & arimoto 1997 ; kodama et al . however there is another formation scenario for elliptical galaxies , i.e. , the hierarchical merging scenario , which may also explain these observational results ( e.g. , kauffmann & charlot 1998 ; shioya & bekki 1998 ) . which of the scenarios dominates the formation of elliptical galaxies has been one of the controversial issues . in order to address this problem , observational properties representing histories of dynamical evolution such as merging histories as well as histories of chemical evolution of galaxies should be focused on . one of the promising candidates is a radial variation of stellar metallicity , i.e. , a metallicity gradient , in an elliptical galaxy . metallicity gradients in elliptical galaxies are expected to represent how the galaxies have assembled . _ in the monolithic collapse scenario _ , a metallicity gradient in an elliptical galaxy can be built up by a dissipative collapse in its formation phase associated with a burst of star formation and a subsequent galactic wind , because of a more extended period of active star formation and more metal enrichment in the inner region ( larson 1974a , b ; carlberg 1984 ; martinelli et al . 1998 ; tantalo et al . 1998 ; kawata 1999 ) . in addition , more massive ellipticals are predicted to have steeper gradients ( larson 1974b ; carlberg 1984 ) . _ in the hierarchical merging scenario _ , stellar populations in the progenitor galaxies are expected to be well mixed in major mergers and the existing metallicity gradients are likely to be diluted ( white 1980 ) . although metallicity gradients may be newly produced due to secondary star formations induced by mergers ( e.g. , mihos & hernquist 1994 ) , these gradients are suggested to be appreciably shallower than those expected by the monolithic collapse scenario ( bekki & shioya 1999 ) . it should be mentioned that in this scenario , the metallicity gradients would be different from galaxy to galaxy due to their different histories of mass assembly and star formation . therefore , no strong correlation between metallicity gradient and galaxy mass would be expected . although more massive ellipticals might have shallower gradients because they are expected to experience more galaxy mergers in the hierarchical merging scheme and their metallicity gradients may be smeared out more seriously , it is opposite to the correlation expected by the monolithic collapse scenario . evolutionary histories of elliptical galaxies can be dependent on galaxy environment . the presence of tight cm relations in rich clusters at high redshift seems to support the monolithic formation at , e.g. , @xmath12 for ellipticals in such high density environments ( e.g. , kodama et al . 1998 ) . even in the cold dark matter ( cdm ) universe , high peaks of density fluctuations at such high redshifts could form giant galaxies and be related to clusters of galaxies along the biased galaxy formation scenario ( peacock 1999 and references therein ) . since it is expected that the mean density of matter is high and objects are gas - rich at those epochs , formations of galaxies are predicted to be accelerated . this implies that elliptical galaxies in dense environments formed through bursts of star formation at high redshift , which mimic the initial starbursts expected in the monolithic collapse scenario , and tend to reside in rich clusters . on the other hand , the field corresponds to the environment without such high peaks of density fluctuations . the density fluctuations are expected to evolve more slowly and collapse at lower redshifts . galaxies would evolve more quiescently through bottom - up processes of smaller sub - galaxies and could settle down to disk galaxies preferentially . in this case , elliptical galaxies are more likely to form through major mergers of disk galaxies at low redshifts ( e.g. , @xmath13 ) . indeed , some recent observational or theoretical studies suggest that elliptical galaxies which formed through major mergers tend to inhabit in field environments and they have experienced mergers inducing star formation until recent epochs such as @xmath14 ( e.g. , baugh et al . 1996 ; governato et al . 1999 ; kauffmann 1996 ; menanteau et al . 1999 ; menanteau , abraham , & ellis 2001 ) . such late star forming activities could make their luminosity - weighted ages at @xmath15 systematically younger ( terlevich & forbes 2002 ) . in summary , if we consider that elliptical galaxies in cluster environments are suggested to have formed through a process like the monolithic collapse while those in sparse environments are through major mergers , the formers are predicted to have steeper metallicity gradients on average than the latter . also , a clear correlation between metallicity gradients and global properties of galaxies ( mass , size , luminosity , etc ) is not expected for ellipticals in sparse environments but expected for cluster ellipticals . to see whether observational data show these kinds of environmental dependence or not will provide strong constraints on the formation processes of galaxies and the roles played by galaxy environments in evolutionary histories of galaxies . although the above issues motivate us to examine metallicity gradients in elliptical galaxies , the problem is how the gradients are to be examined observationally . it is known that nearby elliptical galaxies have color gradients ; colors in an elliptical galaxy gradually become bluer with increasing radius ( e.g. , vader et al . 1988 ; franx , illingworth , & heckman 1989 ; peletier et al . 1990a ; peletier , valentijn , & jameson 1990b ; goudfrooij et al . 1994 ; michard 1999 ; idiart , michard , & de freitas pacheco 2002 ) . although the age metallicity degeneracy prevents us from interpreting the origin of the color gradients to be the metallicity gradients , an evolution of a color gradient with redshift has been studied to disentangle the degeneracy and it is concluded that the primary origin of color gradients in elliptical are not the age gradients but the metallicity gradients both for the elliptical galaxies in field environments and for those in cluster environments ( tamura et al . 2000 ; tamura & ohta 2000 ; saglia et al . 2000 ; see also hinkley & i m 2001 ) . it should be mentioned here that dust extinction may have some effects on a color gradient in an elliptical galaxy ( goudfrooij & de jong 1995 ) . as a matter of fact , if an elliptical galaxy consists of a mixture of stars without any population gradients and diffusely distributed dust , a calculation of the radiative transfer within the galaxy suggests that the color gradients could be reproduced only by the dust effects ( witt , thronson , & capuano 1992 ; wise & silva 1996 ) . however , many elliptical galaxies show radial gradients of metal absorption line index such as mg@xmath16 , fe@xmath17(5270 ) and fe@xmath16(5335 ) as well as the color gradients ( e.g. , carollo , danziger , & buson 1993 ; davies , sadler , & peletier 1993 ; gonzlez 1993 ; kobayashi & arimoto 1999 ; mehlert et al . 2000 ) , and the radial gradients of the metal absorption line indices are unlikely to be created by dust extinction . it has also been suggested that _ on average _ , metallicity gradients in elliptical galaxies as estimated by a population synthesis model from color gradients are consistent with those estimated from absorption line index gradients ( e.g. , peletier et al . 1990a ; davies et al . 1993 ) , although there are exceptions . we thus assume that metallicity effects seem to be the prime cause of color gradients in elliptical galaxies . although both color gradients and metal absorption line strength gradients can be used to trace the metallicity gradients , we study color gradients in this paper because broad band imaging observations are much easier to obtain high @xmath18 data to study the gradients in elliptical galaxies and to enlarge samples including further extension towards high @xmath19 in future . the previous studies suggest that the color gradients in nearby elliptical galaxies do not correlate with any global properties ( mass , luminosity , etc ) of elliptical galaxies ( e.g. , peletier et al . kobayashi & arimoto ( 1999 ) presented the same results by compiling the data of the mg@xmath16 gradients in the literature and constructing a large sample of nearby elliptical galaxies . these studies may support the merging scenario . however , it should be stressed that most of the ellipticals studied reside in low density environments such as field or group . although some elliptical galaxies in virgo cluster have been studied ( peletier et al . 1990a ; goudfrooij et al . 1994 ; michard 1999 ; idiart et al . 2002 ) , virgo cluster is not a rich cluster but a poor cluster with a rather sparse spatial distribution of galaxies . therefore , it is unlikely to be sufficient to know whether there is a dependence of color gradient on galaxy environment or not , and one of the straightforward extensions to address this issue is to study elliptical galaxies in rich clusters . in this study , we examine color gradients in elliptical galaxies in a nearby rich cluster , abell 2199 . it is known that this cluster is significantly richer than virgo cluster and is a spiral poor and centrally concentrated cd cluster ( e.g. , butcher & oemler 1985 ) . it is also considered to be a relaxed cluster based on its regular and symmetrical appearance in x - ray ( e.g. , buote & canizares 1996 ) . although coma cluster is a good candidate as a nearby typical rich cluster and global properties of the member galaxies have been well studied , it has been pointed out to be a merging cluster with two x - ray peaks ( e.g. , caldwell et al . 1993 ; burns et al . 1994 ) and thus the galaxies in coma cluster may have been also disturbed by recent galaxy mergers . we consider that elliptical galaxies in a more regular and relaxed cluster should be studied at first . the layout of this paper is as follows . observation and data reduction are described in the next section . in 3 , sample definition is described and radial profiles of surface brightness and color of the sample galaxies are presented . also , color gradients in the sample galaxies are derived and their statistical properties are investigated . we discuss implications of the results in 4 , and summarize this paper in 5 . we assume @xmath20 km s@xmath21 mpc@xmath21 throughout this paper . imaging observations in @xmath0 and @xmath1 bands were made on june 20 and 21 , 2001 , with tektronix 2k@xmath222k ccd on the university of hawaii 2.2 m telescope . one pixel of the ccd chip corresponds to @xmath23 and thus each frame covers an area of about @xmath24 . observation log is presented in table [ list1 ] . although early - type ( e , e / s0 , and s0 ) galaxies in abell 2199 were primary targets in this observing run , the central region of abell 2634 was also observed as a pilot study to extend our sample to other clusters . ( abell 2634 is also a centrally concentrated cluster but slightly less rich than abell 2199 ( e.g. , pinkney et al . in addition , 3 elliptical galaxies ( ngc 5638 , ngc 5831 , and ngc 7619 ) , of which color gradients were examined in the previous studies , were observed for comparison . a typical seeing size during the observing run was @xmath25 in fwhm . total exposure time at each field is typically 1,800 sec for @xmath0 band and 750 sec for @xmath1 band , each of which was divided into three or four exposures with dithering . all of the imaging data were reduced with iraf in the standard procedure . after bias subtraction , flat - fielding by dome - flat frames , and sky subtraction were carried out in each frame , both of the @xmath0 and @xmath1 band images at each field were registered by sub - pixel shifts . next , slight differences in flux among the images in each band were corrected by multiplying an image by a constant , and the fwhms of the point spread functions ( psfs ) were matched by convolutions of images with gaussian kernels . these images were combined with the 3 @xmath26 clipping algorithm . after a stacked image was made in each band , a psf size in one band was matched to that in the other band by smoothing with gaussian kernels . a resulting psf size is typically @xmath27 in fwhm . the photometric calibration was done by using the standard stars in landolt ( 1992 ) . aperture photometry was carried out after a subtraction of a local residual sky background around each galaxy , which was determined by `` mode '' in an annulus with an inner radius of @xmath28 and a width of @xmath29 ( 50 pix ) . in order to check the accuracy of the photometry and sky subtraction , growth curves of the objects were compared with those taken from hypercat : several galaxies in the clusters as well as ngc 5638 , ngc 5831 , and ngc 7619 have the data of aperture photometry . it is found that there seem to be zero - point offsets between our growth curves and those taken from hypercat , though the amounts of the offsets are @xmath30 mag in each band . this may imply that the observational condition was slightly non - photometric . however , if we correct for the offsets in each field , a tight cm relation consistent with that observed in coma cluster is obtained both in abell 2199 and in abell 2634 . we calculated an average amount of the zero - point shift on each night and it is corrected for by a constant offset in the following analyses . since we focus on a color gradient in a galaxy in this study , such zero - point offset does not cause any serious effects . the target fields were chosen based on the catalogs of early - type galaxies in the clusters by lucey et al . 40 early - type ( e , e / s0 , and s0 ) galaxies in abell 2199 and 11 early - type galaxies in abell 2634 are included in the fields ; most of the galaxy morphology is originally from butcher & oemler ( 1985 ) and rood & sastry ( 1972 ) , while for several galaxies which had not been classified in the literature , we provided morphological types to the galaxies based on eye inspection . although we may be able to study color gradients of these morphologically selected galaxies , it is worth noting that there is a variety in spheroidal - to - total luminosity ratio among galaxies in each morphological type ( simien & de vaucouleurs 1986 ) . in addition , since the clusters studied here are relatively distant , visual classifications of morphology of fainter galaxies may be less reliable . in order to reduce contamination of disk dominated galaxies in a sample , we made azimuthally averaged radial surface brightness profiles in @xmath0 and @xmath1 bands of each galaxy , and performed decompositions of the profiles into bulge and disk components to estimate a bulge - to - total luminosity ratio ( @xmath31 ) . the radial surface brightness profiles are obtained along ellipses fitted to the isophotes with the ellipse task in the stsdas package with a radial sampling of 0@xmath3222 ( 1 pix ) along major axis . in fitting an ellipse to an isophote , ellipticity is set to be a free parameter . galaxy center is fixed to a centroid of a galaxy in the @xmath1 band image and position angle is also fixed to that of an ellipse within which half of the galaxy light is included . even though position angle is allowed to be a free parameter , no significant differences in the surface brightness profiles appear . in the following analyses , all of the radial profiles and the parameters related to radius are expressed in terms of the equivalent radius of an ellipse : @xmath33 , where @xmath34 and @xmath35 are the semi - major and semi - minor axes of the ellipse . the ellipse task employs an iterative algorithm of rejecting pixels significantly deviant from an average on each ellipse . this enables us to eliminate effects of contamination of background / foreground objects at least to some extent . however , since bright objects in the neighbor of a target galaxy may have serious effects on the fitted ellipses , the neighbors are masked out beforehand . it is found that effective radii and color gradients are nearly invariant to whether these masking processes are applied or not and the differences are comparable to or smaller than errors in the individual measurements of the parameters . in practice , some galaxies are `` embedded '' in an envelope of a large galaxy in the neighbor . in order to dig a galaxy out of the envelope , we model the neighboring galaxy using the bmodel task in the stsdas package and subtract it . in most cases , an estimated effective radius and color gradient in such an `` embedded '' galaxy only slightly depend on whether the subtraction process is applied or not , though there are a few galaxies in which the discrepancies can amount to @xmath36% in effective radius and @xmath37 mag / dex in color gradient . although we include galaxies in the following analyses and adopt the parameters estimated after the subtraction processes , they are flagged in the results to show that the subtraction processes are employed and the measurements of effective radii and color gradients may be more uncertain . the decomposition was performed by fitting a surface brightness profile with the sum of an @xmath38 law bulge and an exponential disk profiles so as to minimize @xmath39 . we considered seeing effects on the surface brightness profiles using the formulation by pritchet & kline ( 1981 ) . we chose to sample early - type galaxies of which @xmath31s in @xmath0 band are estimated to be larger than 0.6 . this criterion isolates galaxies earlier than e / s0 or @xmath40 according to simien & de vaucouleurs ( 1986 ) , where @xmath31s of nearby galaxies are also derived in @xmath0 band . in this selection procedure , 31 galaxies are picked out in abell 2199 , while @xmath41% ( 3/25 ) of the ellipticals ( e and e / s0 ) and @xmath42% ( 6/15 ) of the s0 galaxies among the 40 ellipticals and s0s were excluded . in abell 2634 , 7 galaxies are picked out and two of the elliptical galaxies and one of the s0 galaxies were excluded . these fractions are consistent with those expected from the dispersion of the relation between morphological type and @xmath31 suggested by simien & de vaucouleurs ( 1986 ) . it is suggested according to the decomposition that in almost all the sample galaxies , spheroidal components dominate in surface brightness at radii where radial profiles of color are obtained reliably and color gradients are calculated . however , for two of the galaxies in abell 2199 , their surface brightness profiles suggest that disk light starts dominating at a radius smaller than the outer cut - off radius in deriving a color gradient ( see [ clg ] for detail of outer cut - off radius ) . therefore , we excluded these two galaxies . hence , the sample in abell 2199 consists of 29 early - type galaxies . the r band images of the sample galaxies are shown in the top panels of figure 1 , and their observational properties are listed in table [ list2 ] . the total magnitudes were obtained by using sextractor ( bertin & arnouts 1996 ) and the `` best '' magnitudes were adopted . a @xmath2 color was calculated within an aperture whose radius is set to an effective radius of the galaxy in question . a central velocity dispersion ( @xmath43 ) and an mg@xmath44 index are taken from lucey et al . ( 1997 ) . the surface brightness profiles of these sample galaxies are shown in the middle panels of figure 1 . an error bar attached to each of the data points includes an photometric error , a local sky subtraction error , and a dispersion of surface brightness along each fitted ellipse . a local background around each object and its subtraction error were estimated by calculating a standard deviation around the mode in an annulus with an inner radius of @xmath45 and a width of @xmath29 ( 50 pix ) . in figure 1(a ) and 1(b ) , the sample galaxies in abell 2199 and those in abell 2634 , respectively , are presented in order of @xmath1 band total magnitude . the three galaxies ( ngc 5638 , ngc 5831 , and ngc 7619 ) which were taken for calibration purposes are shown in figure 1(c ) . effective radii ( @xmath46 ) of the sample galaxies were derived by regression lines best fitted to the @xmath1 band surface brightness profiles . in fitting a regression line , the data points in the outermost regions of which @xmath18s are low were removed beforehand . in addition , the innermost regions probably flattened by the seeing effects were not used . the resulting fitting regions are shown by the two vertical solid lines in the middle panels of figure 1 . the effective radii obtained are presented in table [ list2 ] . it is mentioned that if the innermost regions are removed properly , they are almost insensitive to the inner and outer cutoff radii . the effective radii correlate with the @xmath1 band total magnitudes : more luminous galaxies have larger effective radii , which is related to the fundamental plane . ( this correlation is referred to `` scaling relation '' hereafter . ) most of the sample galaxies have @xmath1 band total magnitudes between 14 mag and 16 mag , and in this magnitude range an effective radius typically varies from 6@xmath47 to @xmath48 . before obtaining radial color profiles in the sample galaxies , it is worth addressing any trends seen in their two dimensional color distributions . we made the @xmath2 color maps of the galaxies after the seeing sizes in both of the images were matched . in figure 2(a ) and 2(b ) , the color maps of the sample galaxies in abell 2199 and abell 2634 , respectively , are presented in order of @xmath1 band total magnitude of a galaxy from left to right and from top to bottom . the three calibration galaxies ( ngc 5638 , ngc 5831 , and ngc 7619 ) are presented in figure 2(c ) . a cross mark in each panel shows the centroid of a galaxy in @xmath1 band . the regions with @xmath0 band surface brightness brighter than @xmath49 23 mag / arcsec@xmath50 in the galaxies are indicated in these color maps and gaussian smoothing with a kernel of @xmath51 was performed for the clearer presentation . these are the _ pseudo _ color maps ; a range of color adopted in each map is shown by the wedge at the right side of the panel and is different from galaxy to galaxy . we note that the features like vertical lines seen in bo 24 and ngc 7720 are caused by bad columns on the ccd surviving due to a small dithering scale in the horizontal direction at the fields . it is found that almost all the galaxies show axisymmetric color distributions independent of morphology and luminosity . this demonstrates that their radial color profiles and the color gradients represent the color distributions well . the radial profiles of @xmath2 color of the sample galaxies are shown in the bottom panels in figure 1 . each color profile was constructed by subtracting the surface brightness profile in an @xmath1 band image from that in a @xmath0 band image . both of these surface brightness profiles were made along the same ellipses fitted to isophotes in the @xmath1 band image . an error bar attached to each data point in the color profile includes a photometric error , a local sky subtraction error , and a dispersion of colors along each elliptical isophote . most of these color profiles between the inner and outer cut - off radii ( see next subsection for detail ) seem to be linear , while some of them appear to bend in the outermost regions . deeper data are essential to confirm whether the bends are real or not and to define a color profile of a galaxy out to a larger radius . a slope of a radial color profile , that is , a color gradient in each galaxy is derived by the least square fitting method using a portion of a color profile between an inner and outer cut - off radii , which are defined as follows . the outer cut - off radius is defined to be a radius at which @xmath1 band surface brightness of a galaxy is 22.5 mag / arcsec@xmath50 . this typically corresponds to a radius at which an error of @xmath2 color amounts to @xmath49 0.1 mag . an alternative definition of the outer cut - off radius could be effective radius . however , since a less luminous early - type galaxy tends to have a smaller effective radius , the number of data points available to fit a regression line to derive a color gradient is significantly reduced and thus the number of galaxies in which color gradients are derived reliably is also reduced . if a threshold in surface brightness is adopted to define the cut - off radius , the effects of the reductions are mitigated it should be stressed here that the color profiles in most of the sample galaxies are well approximated by lines even if the regions beyond their effective radii are included . for these reasons , we have chosen to use not an effective radius but a radius defined by the surface brightness for the outer cut - off radius . the inner cut - off radius is necessary to avoid seeing effects on color profiles . colors in the inner regions of a galaxy are affected more seriously by seeing than those in the outer regions , which can cause some amount of change in color gradient . in order to understand seeing effects on color gradients in our data and to define the inner cut - off radius optimal for our study , galaxies simulated with the mkobject task in the artdata package are put onto blank fields in the stacked images . the model galaxies are assumed to have de vaucouleurs law profiles , each of which is convolved with a gaussian kernel with the same fwhm as that in the image . an integrated @xmath2 color in a model galaxy is assumed to be 1.5 mag . an `` intrinsic '' color gradient is given to a model galaxy by adopting a smaller effective radius in @xmath1 band than that in @xmath0 band , and three cases of color gradient ( @xmath52 , and @xmath530.3 mag@xmath54dex ) are simulated . since seeing effects on color gradients may depend on angular size and ellipticity of galaxy ( e.g. , peletier et al . 1990a ) , the parameters of the model galaxies are sampled from the ranges covered by the sample galaxies : apparent magnitude in @xmath1 band ( 14 , 15 , and 16 mag ) , effective radius ( 1@xmath47 , 3@xmath47 , and 6@xmath47 ) and ellipticity ( 0.1 , 0.4 , and 0.7 ) . the magnitude and effective radius were assigned to a model galaxy taking into account the scaling relation at the distance of the cluster and the typical dispersion around the relation . it should be also kept in mind that psfs are not necessarily circular in real images . since such non - circular psfs may have effects on color gradients , they are also to be considered in the simulations . in order to see whether psfs are circular in our data or not , we picked up bright stellar objects using sextractor based on the `` class star '' parameter and examined their shapes . it is found that psfs tend to be slightly deviant from circles and the ellipticities are between 0 and 0.1 , and a typical value is @xmath55 . for position angle , it is suggested that the seeing effects on a color profile are more serious when the position angle of a non - circular psf is perpendicular to that of a galaxy ( e.g. , michard 1999 ) . accordingly , we assume a psf with an ellipticity of 0.1 and a position angle perpendicular to that of a model galaxy to simulate the worse case . the model galaxies are analyzed in the same way as the real galaxies including corrections for differences in seeing size between the @xmath0 and @xmath1 images , and their color profiles and color gradients are obtained . the simulations are executed in all the stacked images . in the following , several characteristics of the seeing effects on color profiles and color gradients are mentioned based on the results of the simulations . \(1 ) first , basic trends of the seeing effects on radial color profiles are described in a qualitative way . if a seeing size in a @xmath0 band image is larger than that in an @xmath1 band image , a color in the innermost portion after correction for the difference in seeing size between the images , which is referred to `` convolved color '' hereafter , becomes redder than the intrinsic color . the convolved color becomes bluer monotonically with increasing radius , and in the outer region , it becomes bluer than the intrinsic color . ( the turnover occurs around an effective radius of a galaxy . ) in other words , a color gradient steeper than the intrinsic value would be obtained due to the seeing effects . if a seeing size in an @xmath1 band image is larger , the convolved color becomes slightly bluer than the intrinsic one near the galaxy center and becomes redder with radius . after it overtakes the intrinsic color , the color profile has a peak and the convolved color becomes bluer with radius . eventually , it becomes bluer than the intrinsic color in the outer region . even in this case , a color gradient is estimated to be steeper than the intrinsic one , unless all the data points including the innermost one are used to derive a color gradient from the color profile . \(2 ) in the following , we mention the seeing effects more quantitatively together with dependencies of the seeing effects on parameters of the model galaxies . the deviation of the color gradient affected by seeing from the intrinsic value is larger for a galaxy with a smaller effective radius , i.e. , with a more compact light distribution . a typical amount of the deviation is estimated to be @xmath56 mag / dex , @xmath57 mag / dex and @xmath58 mag / dex for an effective radius of @xmath59 , @xmath60 , and @xmath48 , respectively . \(3 ) the deviation is larger for a galaxy with a steeper intrinsic color gradient . along with the variation of the intrinsic color gradient from @xmath61 to @xmath62 mag / dex , a typical amount of the deviation increases from @xmath56 mag / dex to @xmath58 mag / dex . \(4 ) if the intrinsic color gradient and effective radius are fixed to some values , there is no clear dependency of the seeing effects on ellipticity of the model galaxy . basically , a larger inner cut - off radius tends to make the seeing effects on a color gradient less serious . however , the number of the data points with high @xmath18s available to fit the regression line to the color profile would be also reduced and thus the error in the measurement of color gradient would be larger . this is likely to be more serious for the less luminous and thus smaller galaxy . considering these effects , we decided to compromise and adopt @xmath63 as the inner cut - off radius . it has to be mentioned that some residual seeing effects could still exist and the color gradient might be estimated to be steeper than the intrinsic one . the typical amount of the deviation is predicted to be @xmath64 mag@xmath54dex , @xmath56 mag@xmath54dex , and @xmath65 mag@xmath54dex for a galaxy with an effective radius of @xmath59 , @xmath60 , and @xmath48 , respectively . it is worth mentioning that each of these deviations is comparable to or smaller than the typical error in the individual measurement of color gradient in the model galaxy with each of the effective radii . finally , some comments are mentioned on the three galaxies of which color gradients were studied by peletier et al . ( 1990a ) ( ngc 5638 and ngc 5831 ) and by franx et al . ( 1989 ) ( ngc 7619 ) . in figure 3 , the radial profiles of @xmath2 color obtained in this study and those in the previous studies are compared in the upper panels , and the differences are presented in the lower panels . the uncertainty of the @xmath66 at each radius is comparable to or larger than the error bar in the color profile obtained in this study . color gradients of these galaxies are re - estimated by using the data for the color profiles in the literature and the outer cut - off radii defined in this study , while we adopted the inner cut - off radii defined in the literature . the resultant color gradients are compared in table [ comptab ] . for ngc 5638 and ngc 5831 , the values agree reasonably with each other . for ngc 7619 , although the difference in color gradient seems to be large , it is unlikely to be significant if the large error in the color profile presented by franx et al . ( 1989 ) is considered . in figure 4 , histograms of @xmath2 color gradients of the sample galaxies in abell 2199 and abell 2634 are shown . a width of each bin in the histograms is taken to be 0.05 mag / dex of @xmath3 , which is the typical uncertainty of the derived individual color gradient . a red solid line and a red dashed line depict the histograms for the galaxies in abell 2199 and in abell 2634 , respectively . we calculated a mean value of color gradient and a standard deviation of each distribution , and summarized the results in table [ values ] . the mean value of the whole sample of abell 2199 and abell 2634 is @xmath67 mag / dex with the standard deviation of @xmath68 . a median value is also close to this . although less luminous ( @xmath69 mag ) and smaller ( @xmath70 ) galaxies show steeper color gradients on average , this is unlikely to be significant in the histogram or the table , if the characteristics of the seeing effects found in the simulations as well as the larger uncertainty in measurement of color gradient are taken into account . these results are independent of whether the `` embedded '' galaxies are excluded from the samples or not . we next examine the relations between the color gradients and global properties of the cluster galaxies . in figure 5 , the @xmath2 color gradients in the sample galaxies are plotted against @xmath1 band total magnitudes , effective radii , central velocity dispersions , and mg@xmath44 indices . the @xmath1 band total magnitude and the effective radius are based on our data . for the @xmath1 band total magnitudes , the error includes the calibration error . for the effective radii , the error is based on 1 @xmath26 fitting error in slope of the regression line to the surface brightness profile . the central velocity dispersion , mg@xmath44 index , and their errors are taken from lucey et al . open circles denote the galaxies embedded in the envelope of a neighboring galaxy of which color gradients were obtained after the neighboring galaxy was modeled and subtracted . red symbols indicate the galaxies with both the effective radius larger than @xmath9 and the @xmath1 band total magnitude brighter than 15 mag . as seen in figure 5 , there are no clear correlations when we consider the whole sample . however , in the relation between the color gradients and @xmath1 band total magnitudes of the galaxies in abell 2199 , it is suggested that there is a discontinuity in the relation around @xmath71 mag , that is , the relation in the magnitude range brighter than @xmath71 mag may be different from that in the fainter magnitude range . for the galaxies brighter than @xmath71 mag , a weak correlation may exist : brighter galaxies have steeper color gradients . ( the correlation probability is estimated to be larger than 99% based on the spearman s rank correlation coefficient , while the linear correlation coefficient is estimated to be 0.6 . ) in the lower luminosity range , on the other hand , it is hard to see whether the correlation exists or not due to the larger uncertainties in the color gradients . the @xmath1 band apparent magnitude of 15 mag corresponds to an absolute magnitude of @xmath72 mag and thus @xmath73 ( e.g. , lin et al . 1996 ) . similar features appear to exist in the relation between the color gradients and effective radii : there seems to be a break around an effective radius of @xmath60 , which roughly corresponds to @xmath74 mag considering the scaling relation , and for the galaxies with effective radii larger than @xmath9 , the galaxies with larger effective radii tend to have steeper color gradients . these trends are preserved if the `` embedded '' galaxies ( open circles in figure 5 ) are excluded from the diagrams . although it is unclear for the galaxies in abell 2634 whether such trends exist or not due to the small number of the galaxies , the data points appear to be consistent with the trends seen for the galaxies in abell 2199 . it should be stressed here that these trends are not expected by the seeing effects , because the seeing effects could make color gradients steeper for less luminous and thus smaller galaxies . almost the same trends are seen when the inner cut - off radius is adopted to be @xmath75 . on the other hand , they disappear when the cut - off radius is defined to be @xmath76 , probably because of the seeing effects . although similar trends are expected to be seen in the relations with the central velocity dispersions and mg@xmath44 indices , no clear trends are seen . we suggest that this is at least partly due to the relatively large errors of central velocity dispersion and mg@xmath44 index . in fact , these quantities from lucey et al . ( 1997 ) only weakly correlate with the @xmath1 band total magnitudes and effective radii obtained from our data . if evolutionary histories of elliptical galaxies depend on galaxy environment , their metallicity gradients and thus color gradients could also have some environmental dependences . as described in 1 , recent studies either observationally or theoretically suggest that ellipticals which reside in rich clusters tend to have formed through the formation processes quite similar to the monolithic collapse , in which more massive ellipticals have been suggested to have steeper metallicity gradients ( larson 1974b ; carlberg 1984 ) . on the other hand , elliptical galaxies in less dense environments are likely to have formed through major mergers , which are expected to dilute the existing metallicity gradients and any correlations between metallicity gradients and global properties of the galaxies . consequently , a significant difference in statistical property of color gradient between elliptical galaxies in rich clusters and those in the sparse environments is expected . the motivation of this study is to examine these issues observationally by adding the data of color gradients in elliptical galaxies in rich clusters , which have been rarely studied . the mean value of the color gradient obtained in the ellipticals in abell 2199 and abell 2634 is @xmath4 mag / dex . using a simple stellar population model by vazdekis et al . ( 1996 ) , this color gradient is converted to a metallicity gradient ( @xmath5 ) of @xmath77 by assuming an age of 12 gyr . on the other hand , numerical simulations based on the monolithic collapse scenario by carlberg ( 1984 ) and by kobayashi ( 2002 ) predicted a metallicity gradient of @xmath78 and @xmath79 , respectively , which are significantly steeper than our estimated metallicity gradient . however , it has been recently suggested that if a proto - galactic gas cloud has small scale density fluctuations expected in the cdm universe rather than a uniform density , an elliptical galaxy could form by the initial starburst following the successive assembly of low mass and gas - rich sub - clumps , and the metallicity gradient and @xmath2 color gradient in the galaxy could be consistent with our results ( kawata 2001 ) . in his simulation , a giant elliptical with a @xmath0 band absolute magnitude of @xmath80 mag at @xmath15 was studied , and its metallicity gradient and @xmath2 color gradient were estimated to be @xmath81 and @xmath82 mag / dex , respectively . these values are consistent with the gradients in the brightest galaxies obtained in our study . if that process is more realistic in the monolithic collapse , our results may imply that elliptical galaxies in rich clusters tend to have formed through a process like the monolithic collapse . our data suggest that for the galaxies in the clusters brighter than @xmath74 mag and those with effective radii larger than @xmath83 , more luminous and larger galaxies tend to have steeper color gradients . this tendency can be seen in the results by numerical simulation based on the monolithic collapse scenario ( larson 1974b ; carlberg 1984 ; see also kawata & gibson 2002 ) . thus it also seems to support the view that the cluster ellipticals formed by the monolithic collapse scenario . since the trend found is rather weak and may not be very convincing , it is important to study ellipticals in other rich clusters . it is interesting that the correlation seen for the galaxies brighter than @xmath71 mag does not seem to continue to the lower luminosity range . it has been suggested that ellipticals less luminous than @xmath8 have observational properties ( isophotal shape , kinematics , etc ) different from more luminous ones ( e.g. , bender , burstein & faber 1992 ; faber et al . 1997 ) and thus have different formation mechanisms . the discontinuity found here may be another piece of the evidence , even though the color gradients in the less luminous galaxies obtained in this study are still uncertain . deeper observations in sub - arcsecond seeing conditions are necessary for more detailed study . we are now in a position to compare the cluster ellipticals with those in less dense environments . in figure 4 , the histograms of the color gradients of the galaxies in abell 2199 and abell 2634 are compared with those of 42 nearby elliptical galaxies studied by michard ( 1999 ) , 39 elliptical galaxies by peletier et al . ( 1990a ) , and 17 elliptical galaxies by franx et al . some of the ellipticals studied in the literature are located in nearby clusters . most of them are virgo ellipticals , and a handful of ellipticals in the other poor clusters ( fornax , awm 4 , and awm 5 ) or rich clusters ( coma cluster , pegasus cluster , abell 2029 , and abell 2162 ) are included . we also calculated the mean value of the color gradient and the standard deviation in a sub - sample of the cluster ellipticals , and the results are presented in table [ values ] . the distributions of the color gradients for abell 2199 and abell 2634 seem to be quite similar to those by peletier et al . ( 1990a ) and by franx et al . ( 1989 ) , though it is slightly shifted to a steeper side than the distribution by michard ( 1999 ) . color gradients do not well correlate with each other ( michard 1999 ) . ] the similarity in the distribution suggests that distribution of color gradient is almost independent of galaxy environment . next , using the published data , color gradients in nearby elliptical galaxies are plotted against the @xmath0 band absolute magnitudes ( figure 6(a ) ) and effective radii ( figure 6(b ) ) . the @xmath0 band absolute magnitudes and effective radii in the diagram by idiart et al . ( 2002 ) were taken from michard ( 1999 ) and michard & marchal ( 1994 ) . in the other diagrams , the data presented in each literature were used . since effective radii measured along major axis are tabulated in peletier et al . ( 1990a ) and goudfrooij et al . ( 1994 ) , we converted them to the equivalent radii using the ellipticities in the literature . solid circles indicate ellipticals in sparse environments such as field or group . open circles and open squares denote ellipticals in virgo cluster and those in the other poor clusters , respectively . open triangles indicate those in rich clusters . ( most of them are the brightest cluster galaxies . ) it should be emphasized that virgo cluster is not a rich cluster but a poor cluster in which a spatial distribution of galaxies is known to be rather sparse , and color gradients in only a few ellipticals in a cluster as rich as abell 2199 have been studied so far . we note that a @xmath0 band absolute magnitude of @xmath84 mag and an effective radii of 2.5 kpc roughly correspond to an @xmath1 band magnitude of 15 mag considering a typical color of an elliptical galaxy and an angular size of 3@xmath47 at the distance of the clusters . figure 6 suggests that , contrary to the galaxies in abell 2199 and abell 2634 , there are no clear trends between the color gradients and galaxy luminosities or effective radii for ellipticals in the less dense environments . this is consistent with the conclusion by kobayashi & arimoto ( 1999 ) based on the studies of the radial gradients of metal absorption line index . although there seems to be a marginal trend in the study by peletier et al . ( 1990a ) that for the cluster ellipticals ( open symbols ) , brighter and larger ellipticals have steeper color gradients , it is still worth mentioning that the color gradients in the non - cluster ellipticals ( solid symbols ) do not correlate with their global photometric properties . in summary , while it is suggested that distribution of color gradient in dense environment such as in abell 2199 is not significantly different from that in sparse environment , there seems to be an environmental dependence in relation between color gradients and global properties of ellipticals : more luminous and larger ellipticals have steeper color gradients in dense environment , but any such trend is clearly weaker in less dense environments . these results do not fully support the view that ellipticals in rich clusters formed through the monolithic collapse while those in sparse environments formed through galaxy mergers , because the latter ellipticals are expected to have color gradients shallower on average than the formers . this apparent conflict may be reconciled if star formation and subsequent chemical enrichment induced by the mergers are taken into account . it is suggested by numerical simulations that if a galaxy merger occurs , a significant fraction of gas in the progenitor galaxies falls into the central regions of the merger remnant and thus star formation activities are also expected to be more active in the inner regions ( e.g. , barnes & hernquist 1991 ) . consequently , stellar populations in the inner regions would be more chemically enriched and thus the merger remnant could acquire a metallicity gradient ( mihos & hernquist 1994 ) . if this effect compensates the dilution of the metallicity gradient existing in the progenitors by mixing stars during the merger , a resultant metallicity gradient in the merger remnant would not become significantly shallower . nevertheless , it is still difficult to address which is more important , the acquisition or the dilution , to determine a metallicity gradient in a merger remnant , because it is likely to depend on initial condition of galaxy merger . it is also possible that a galaxy has experienced several mergers before the present time and thus the metallicity gradient is defined by its whole merging history . it is interesting to point out here that a similar trend to that found in this work has been suggested for a sample of bulges of spiral galaxies . in balcells & peletier ( 1994 ) , @xmath85 and @xmath86 color gradients in 18 nearby spiral bulges were studied . their sample consists of bulges of spiral galaxies earlier than sc , and barred spirals and dusty spirals are excluded . they found that for the bulges more luminous than an @xmath1 band absolute magnitude of @xmath87 mag , the bulge luminosities correlate with the color gradients in the sense that more luminous bulges have steeper color gradients . no clear correlations were found between the bulge color gradients and the total luminosities or any of the parameters related to disk light . since the @xmath86 color gradients in the bulges vary from @xmath88 mag / dex to @xmath89 mag / dex for an increase of the bulge luminosities by 2 magnitudes , the correlation seems to be quite similar to the trend found in this study both qualitatively and quantitatively , though the cluster ellipticals studied here are @xmath90 magnitude more luminous on average than the bulges . it is also suggested by balcells & peletier ( 1994 ) that there is a correlation between the bulge color gradients and bulge luminosities for the less luminous bulges ( @xmath91 ) , but it is opposite to that for the more luminous ones : less luminous bulges have steeper color gradients . although it is unclear in our study whether there is a correlation for the cluster ellipticals in the low luminosity range , our data do not exclude the possibility that there is a discontinuity in the relation around @xmath71 mag . these common features may suggest that luminous bulges of spiral galaxies have formation histories more similar to ellipticals in rich clusters . however , once the environmental dependence found in this work for elliptical galaxies is taken into account , it does not seem to be simply explained because spiral galaxies tend to be in sparse environments , while no clear trends between color gradients and galaxy luminosities have been found for ellipticals in sparse environments . more observational and theoretical works would be necessary to understand what this similarity implies . we performed @xmath0 and @xmath1 band surface photometry for e / s0 galaxies in abell 2199 and abell 2634 , and studied their @xmath2 color gradients . in order to sample spheroid dominated galaxies , the decomposition of the surface brightness profiles into @xmath38 bulge and exponential disk profiles was performed and the sample selection was based on the @xmath0 band bulge - to - total luminosity ratios . resultantly , 29 galaxies in abell 2199 and 7 galaxies in abell 2634 were sampled . we evaluated the seeing effects on the color gradient by simulating with artificial galaxies having various parameters and defined the inner cut - off radius in deriving color gradients to mitigate the effects . the mean value of the color gradient is estimated to be @xmath4 mag / dex , which can be converted to a metallicity gradient of @xmath92 . this gradient seems to be consistent with that expected by a recent monolithic collapse model . we further studied the relations between the @xmath2 color gradients and global properties of the sample galaxies . it is found that for the galaxies in abell 2199 brighter than @xmath71 mag , which roughly corresponds to @xmath8 at the distance of the cluster , there may be a correlation in the sense that more luminous galaxies have steeper color gradients . also , for the galaxies with effective radii larger than @xmath9 , which nearly corresponds to @xmath71 mag considering the scaling relation , the galaxies with larger effective radii seem to have steeper color gradients . the ellipticals in abell 2634 appear to follow the same trends , though the sample is small . these trends also suggest that elliptical galaxies in rich clusters formed through the monolithic collapse scenario . since this is the first systematic study of color gradients in ellipticals in a rich cluster , it is important to study ellipticals in other rich clusters and to see whether the characteristics of the color gradients found in this study are generally seen or not . we compared the results with the color gradients in elliptical galaxies in less dense environments using data from the literature . it is found that no clear trends between color gradient and galaxy luminosity or effective radius are seen , suggesting that elliptical galaxies in sparse environments formed through galaxy mergers , while the distributions of the color gradients seem to be quite similar to those obtained in abell 2199 and abell 2634 . these results suggest that while distribution of color gradient is almost independent of galaxy environment , there may be an environmental dependence in relation between color gradients and global properties of ellipticals : more luminous and larger ellipticals have steeper color gradients in dense environment , but any such trend is clearly weaker in less dense environments . this does not fully support the view that ellipticals in rich clusters formed through the monolithic collapse while those in sparse environments formed through galaxy mergers , because the latter ellipticals are expected to have color gradients shallower on average than the formers . this apparent conflict may be reconciled if we take into account star formation and subsequent chemical enrichment induced by the mergers . since the chemical enrichment could proceed more in the inner regions of the merger remnant , a metallicity gradient could be newly formed and it may compensate the dilution of the metallicity gradient . the correlation between the color gradient and the galaxy luminosity ( or the effective radius ) found for the brighter cluster ellipticals can not be seen for the fainter ellipticals ; this may imply the difference of formation process between them , though it should be confirmed by data for the fainter ellipticals with a higher spatial resolution . on the other hand , the relation seems to be similar , both qualitatively and quantitatively , to that previously found for luminous bulges of spiral galaxies . this may indicate that cluster ellipticals and bulges of spiral galaxies have similar formation mechanisms . we appreciate the support from members of the university of hawaii observatory during the observations . we are grateful to the anonymous referee for useful comments and suggestions to improve our paper . we would like to thank c. kobayashi and n. arimoto for fruitful discussions . this research made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under a contract with the national aeronautics and space administration . figure 1 a top panel shows @xmath1 band image of a sample galaxy . in figure 1(a ) and 1(b ) , the sample galaxies in abell 2199 and those in abell 2634 are presented in order of apparent magnitude in @xmath1 band , respectively . the three calibration galaxies ( ngc 5638 , ngc 5831 , and ngc 7619 ) are presented in figure 1(c ) . value of @xmath31 and morphology are indicated in parentheses . in a middle panel , azimuthally averaged surface brightness profiles in @xmath0 band ( open circles ) and in @xmath1 band ( filled circles ) are shown . the data points in the @xmath1 band profile between the two vertical lines are used to fit a regression line to derive an effective radius of a galaxy . the best - fit line and the lines with the slopes @xmath93 deviant from the best - fit are shown . dashed line indicates the outer cutoff radius in deriving a color gradient . in a bottom panel , a radial profile of @xmath2 color abscissa refers to a logarithmic radius along major axis . the two vertical lines indicate the inner and outer cutoff radii . the regression line best fit to the color profile and the lines with the slopes @xmath94 deviant from the best - fit are shown . figure 2 color maps of the sample galaxies . in figure 2(a ) , the sample galaxies in abell 2199 are presented , and in figure 2(b ) , those in abell 2634 are shown . the three galaxies ( ngc 5638 , ngc 5831 , and ngc 7619 ) are presented in figure 2(c ) . these maps are put in order of @xmath1 band magnitude of a galaxy from left to right and from top to bottom . a cross mark in each panel shows the centroid of the @xmath1 band image . these are the _ pseudo _ color maps , and a range of color adopted in each map , which is shown by the wedge on the right side of the panel , is different from galaxy to galaxy . we note that vertical lines seen in bo 24 and ngc 7720 are artificial and are caused by bad columns on the ccd . figure 3 comparison of radial color profile for ngc 5638 , ngc 5831 , and ngc 7619 . two panels are shown for each galaxy . in the upper panels , the @xmath2 color profiles obtained in this study are shown by filled circles with error bars , and those taken from the previous studies ( peletier et al . ( 1990a ) for ngc 5638 and ngc 5831 and franx et al . ( 1989 ) for ngc 7619 ) are shown by open circles . zero points of the color profiles in the previous studies are arbitrarily shifted for clearer presentation . in the lower panel , a color difference between the two data at each radius is shown . the uncertainty of the @xmath66 can be comparable to or larger than the error bars attached to the color profile obtained in this study . the two vertical lines indicate the inner and outer cutoff radii used to derive a color gradient . one of them at a smaller radius shows the inner cut - off radius defined in the literature . another one indicates the outer cutoff radius defined in this paper . note that a mean value of the difference , which is presumably caused by the zero - point error of our photometry , was removed because it has no effect on color gradient . figure 4 histograms of color gradients . red solid line and red dashed line indicate histograms for the sample galaxies in abell 2199 and in abell 2634 , respectively . blue line , green line , and cyan line show histograms for the galaxies studied by michard ( 1999 ) , by peletier et al . ( 1990a ) and by franx et al . ( 1989 ) , respectively . a width of each bin in the histograms is taken to be 0.05 mag / dex of @xmath95 , which is comparable to an error typically seen in an individual measurement of @xmath2 color gradient in our data . figure 5 relations between color gradients and global properties of the sample galaxies . left panels are the diagrams for the galaxies in abell 2199 , and right panels are for the galaxies in abell 2634 . the total magnitudes and effective radii are based on our data . the central velocity dispersions and mg@xmath44 indices are taken from lucey et al . open stars denote the brightest cluster galaxies ( ngc 6166 in abell 2199 and ngc 7720 in abell 2634 ) . open circles indicate the galaxies embedded in the envelope of a neighboring galaxy and their color gradients were obtained after the neighbor was modeled and subtracted . red symbols show the galaxies with both an effective radius larger than @xmath9 and an @xmath1 band total magnitude brighter than 15 mag . solid lines in the diagrams of the color gradients and total magnitudes and effective radii for the galaxies in abell 2199 are shown as references of the trends . in the relation between color gradient and effective radius , the brightest cluster galaxy is not taken into account in the reference line because it seems to be deviant from the trend . the same lines are also shown in the right panels for comparison . figure 6 relation between color gradients and @xmath0 band absolute magnitudes ( figure 6(a ) ) and that between color gradients and effective radii ( figure 6(b ) ) for nearby elliptical galaxies based on the published data . the data are basically taken from each literature , but the @xmath0 band absolute magnitudes and effective radii in the diagram by idiart et al . ( 2002 ) are taken from michard ( 1999 ) and michard & marchal ( 1994 ) . solid circles indicate ellipticals in sparse environments such as field or group . open circles indicate ellipticals in virgo cluster and open squares are ellipticals in the other poor clusters ( fornax , awm 4 , and awm 5 ) . open triangles indicate those in rich clusters ( coma cluster , pegasus cluster , abell 2029 , and abell 2162 ) . it is noted that one virgo elliptical and one fornax elliptical in goudfrooij et al . ( 1994 ) are not included here because their color gradients are studied in @xmath96 or @xmath97 color , not in @xmath98 color as the other ellipticals are studied . + 16 28 36.7 & 39 33 58 & 1260 & 750 & 1.1 & 0.9 & ngc 6166 ( 9100 ) , bo 15 ( 8782 ) , + & & & & & & bo 24 ( 10182 ) , bo 41 ( 9323 ) , + & & & & & & bo 66 ( 9209 ) , bo 73 ( 7993 ) , + & & & & & & bo 74 ( 8548 ) , bo 95 ( 8259 ) + 16 27 46.7 & 39 34 21 & 1860 & 780 & 1.1 & 0.9 & bo 25 ( 9372 ) , bo 46 ( 9149 ) + 16 29 33.9 & 39 48 52 & 1860 & 780 & 1.0 & 1.2 & bo 19 ( 7833 ) , bo 26 ( 9176 ) + 16 28 53.4 & 39 28 01 & 1920 & 750 & 0.9 & 0.9 & bo 34 ( 8679 ) , bo 51 ( 10164 ) + 16 28 59.6 & 39 41 26 & 1800 & 720 & 1.2 & 1.1 & bo 38 ( 8397 ) , bo 45 ( 9400 ) , + & & & & & & bo 107 ( 9163 ) , bo 113 ( 8477 ) + 16 27 40.0 & 39 23 00 & 1260 & 600 & 1.1 & 1.0 & ngc 6158 ( 8980 ) , bo 70 ( 8777 ) , + & & & & & & bo 96 ( 9253 ) + 16 27 52.9 & 39 15 33 & 1800 & 600 & 1.0 & 1.0 & bo 5 ( 8744 ) , bo 20 ( 9638 ) , + & & & & & & bo 69 ( 8956 ) + 16 27 03.7 & 39 31 57 & 1800 & 960 & 1.1 & 0.8 & bo 8 ( 10159 ) + 16 25 40.0 & 39 36 20 & 660 & 300 & 1.1 & 1.2 & rs 8 ( 8952 ) + 16 28 18.3 & 38 48 43 & 1860 & 600 & 1.1 & 1.1 & rs 72 ( 9288 ) + 16 31 18.5 & 39 08 53 & 1500 & 600 & 0.9 & 1.0 & rs 162 ( 8856 ) , rs 163 ( 8854 ) + & & & & & & + + 23 38 28.8 & 27 01 53 & 540 & 180 & 0.9 & 0.9 & ngc 7720 ( 9060 ) , ic 5342 ( 9274 ) , + & & & & & & bo 11 ( 9502 ) , bo 27 ( 8378 ) , + & & & & & & bo 82 ( 9552 ) , d 75 ( 9828 ) , + & & & & & & lgc 125 ( 9983 ) + & & & & & & + + 14 29 40.4 & 03 14 00 & 360 & 150 & 0.8 & 0.9 & ngc 5638 ( 1676 ) + 15 04 07.0 & 01 13 11 & 420 & 120 & 0.9 & 0.8 & ngc 5831 ( 1656 ) + 23 20 14.7 & 08 12 22 & 540 & 150 & 0.9 & 0.8 & ngc 7619 ( 3762 ) + lccccccccc ngc 6166 & 1.0 & cd & 14.9 & 13.2 & 1.59 & 1.66 & 0.319 & 2.475 & @xmath99 + bo 15 & 0.9 & e & 15.9 & 14.3 & 1.59 & 0.88 & 0.287 & 2.240 & @xmath100 + bo 24 & 0.7 & s0 & 16.1 & 14.3 & 1.69 & 0.55 & 0.325 & 2.459 & @xmath101 + bo 41 & 0.8 & e / s0 & 16.7 & 15.0 & 1.67 & 0.14 & @xmath102 & @xmath102 & @xmath103 + bo 66 & 0.6 & s0 & 16.8 & 15.2 & 1.68 & 0.05 & 0.300 & 2.282 & @xmath104 + bo 73 & 0.8 & e & 17.1 & 15.5 & 1.61 & 0.18 & 0.269 & 2.220 & @xmath105 + bo 74 & 0.8 & e & 16.4 & 14.9 & 1.63 & 0.14 & 0.267 & 2.391 & @xmath106 + bo 95 & 0.8 & e & 17.0 & 15.4 & 1.61 & 0.32 & 0.275 & 2.195 & @xmath107 + bo 25 & 0.9 & e & 16.2 & 14.6 & 1.55 & 0.73 & @xmath102 & @xmath102 & @xmath108 + bo 46 & 0.6 & e & 16.6 & 14.9 & 1.70 & 0.26 & @xmath102 & @xmath102 & @xmath109 + bo 19 & 0.8 & s0 & 16.1 & 14.3 & 1.77 & 0.40 & 0.266 & 2.380 & @xmath110 + bo 26 & 0.9 & e / s0 & 16.2 & 14.6 & 1.71 & 0.48 & 0.275 & 2.310 & @xmath111 + bo 34 & 1.0 & e / s0 & 16.4 & 15.0 & 1.53 & 0.80 & 0.222 & 1.995 & @xmath112 + bo 51 & 0.8 & s0 & 17.1 & 15.6 & 1.54 & 0.39 & @xmath102 & @xmath102 & @xmath113 + bo 38 & 0.8 & e & 17.0 & 14.9 & 2.15 & 0.49 & 0.300 & 2.243 & @xmath114 + bo 45 & 0.7 & e & 17.0 & 14.9 & 2.19 & 0.49 & 0.300 & 2.252 & @xmath115 + bo 107 & 0.6 & s0@xmath116 & 18.0 & 15.9 & 2.19 & @xmath530.08 & @xmath102 & @xmath102 & @xmath117 + bo 113 & 0.8 & e@xmath116 & 18.2 & 16.2 & 2.10 & 0.53 & @xmath102 & @xmath102 & @xmath118 + ngc 6158 & 1.0 & e & 15.3 & 13.6 & 1.70 & 1.02 & 0.277 & 2.280 & @xmath99 + bo 70 & 0.8 & s0@xmath116 & 17.2 & 15.6 & 1.58 & 0.30 & @xmath102 & @xmath102 & @xmath119 + bo 96 & 0.7 & e@xmath116 & 17.5 & 15.9 & 1.70 & 0.14 & @xmath102 & @xmath102 & @xmath120 + bo 5 & 0.9 & e & 15.6 & 13.9 & 1.62 & 0.96 & 0.284 & 2.312 & @xmath110 + bo 20 & 0.9 & e & 16.1 & 14.4 & 1.67 & 0.59 & 0.304 & 2.274 & @xmath101 + bo 69 & 0.7 & s0 & 16.3 & 14.6 & 1.70 & 0.39 & @xmath102 & @xmath102 & @xmath121 + bo 8 & 0.9 & e & 15.5 & 14.0 & 1.55 & 0.99 & 0.269 & 2.146 & @xmath122 + rs 8 & 0.9 & e & 15.4 & 13.8 & 1.56 & 0.76 & 0.297 & 2.395 & @xmath123 + rs 72 & 0.9 & e & 15.6 & 13.9 & 1.72 & 0.74 & 0.304 & 2.285 & @xmath124 + rs 162 & 1.0 & e & 15.8 & 14.1 & 1.66 & 0.77 & 0.288 & 2.378 & @xmath125 + rs 163 & 0.6 & s0 & 16.1 & 14.4 & 1.71 & 0.33 & @xmath102 & @xmath102 & @xmath126 + ngc 7720 & 0.9 & e & 15.0 & 13.3 & 1.76 & 1.11 & 0.332 & 2.520 & @xmath127 + ic 5342 & 0.9 & e & 15.9 & 14.2 & 1.77 & 0.81 & 0.301 & 2.345 & @xmath106 + bo 11 & 0.9 & e & 16.0 & 14.2 & 1.79 & 0.95 & 0.324 & 2.335 & @xmath128 + bo 27 & 0.9 & s0 & 16.3 & 14.5 & 1.81 & 0.72 & 0.302 & 2.316 & @xmath109 + bo 82 & 0.9 & e & 17.1 & 15.4 & 1.80 & 0.34 & 0.284 & 2.263 & @xmath129 + d 75 & 0.7 & e & 17.1 & 15.3 & 1.79 & 0.39 & 0.297 & 2.273 & @xmath130 + lgc 125 & 1.0 & e@xmath116 & 17.9 & 16.2 & 1.67 & 1.08 & @xmath102 & @xmath102 & @xmath131 + ngc 5638 & 0.9 & e & 13.5 & 11.9 & 1.51 & 1.48 & 0.317 & 2.201 & @xmath100 + ngc 5831 & 0.8 & e & 13.6 & 12.0 & 1.62 & 1.44 & 0.289 & 2.220 & @xmath99 + ngc 7619 & 1.0 & e & 13.5 & 11.8 & 1.67 & 1.30 & 0.336 & 2.528 & @xmath122 + llccc abell 2199 & ( all ) & 29 & @xmath4 & 0.03 + & ( @xmath134 mag ) & 20 & @xmath135 & 0.02 + & ( @xmath69 mag ) & 9 & @xmath136 & 0.06 + & ( @xmath137 ) & 16 & @xmath135 & 0.03 + & ( @xmath70 ) & 13 & @xmath138 & 0.05 + abell 2634 & ( all ) & 7 & @xmath128 & 0.07 + michard et al . ( 1999 ) & ( all ) & 42 & @xmath139 & @xmath53 + & ( virgo ) & 15 & @xmath140 & @xmath53 + peletier et al . ( 1990 ) & ( all ) & 39 & @xmath141 & 0.03 + & ( virgo ) & 10 & @xmath142 & 0.02 + & ( rc ) & 5 & @xmath143 & 0.05 + franx et al . ( 1989 ) & ( all ) & 17 & @xmath139 & 0.05 + & ( pc ) & 5 & @xmath144 & 0.06 +

in this article we describe a canonical way to expand a certain kind of-colored regular graphs into closed-manifolds by adding cells determined by the edge - colorings inductively . we show that every closed combinatorial-manifold can be obtained in this way . when , we give simple equivalent conditions for a colored graph to admit an expansion . in addition , we show that if a-colored regular graph admits an-skeletal expansion , then it is realizable as the moment graph of an-dimensional closed-manifold .

math0609557

in 1998 , goresky , kottwitz and macpherson @xcite established the gkm theory , which gives a direct link between equivariant topology and combinatorics ( or more precisely , a link between gkm manifolds and gkm graphs ) . the most essential properties of a gkm manifold @xmath5 , such as the equivariant cohomology and betti numbers of @xmath5 , can thus be explicitly expressed in terms of its corresponding gkm graph . a lot of work on this subject has been carried out since then ( @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite ) . for example , a series of papers by guillemin and zara showed that much geometrical and topological information of a gkm manifold can be read out from the corresponding gkm graph . in the gkm theory , the action group is a torus . alternately , when the action group is chosen to be a mod 2-torus , there is an analogous theory , namely the mod 2 gkm theory ( see @xcite or @xcite ) , which has successfully been applied to find the lower bound of the number of fixed points and to study the equivariant cobordism classification and the smith problem in @xcite . .2 cm the work of this article is mainly motivated by the gkm theory . however , we shall carry out our work in the mod 2 category . thus , let us first recall some basic facts about the mod 2 gkm theory , which inspired us to study colored regular graphs . suppose that @xmath6 ( a mod 2-torus of rank @xmath7 ) , which is also an @xmath3-dimensional linear space over @xmath8 . let @xmath9 be the universal principal @xmath10-bundle , where @xmath11 is the classifying space of @xmath10 . it is well - known that @xmath12 and @xmath13\ ] ] is a @xmath8-polynomial ring over @xmath14 , where the @xmath15 s are one - dimensional generators . .2 cm now suppose that @xmath5 is a closed manifold with an effective @xmath10-action such that the fixed point set @xmath16 is finite , which implies @xmath17 ( see @xcite ) . then the _ borel construction _ @xmath18 is defined as the orbit space of the induced diagonal action @xmath19 on @xmath20 , i.e. , @xmath21 so that the equivariant cohomology of @xmath5 is @xmath22 as shown in @xcite , to the @xmath10-manifold @xmath5 , one can always assign a regular graph @xmath23 with a label @xmath24 @xmath25 for each edge @xmath26 in @xmath23 . the gkm theory points out that when the @xmath10-manifold @xmath5 satisfies certain conditions , its equivariant cohomology is a free module over @xmath27 and can be explicitly read out from @xmath28 : @xmath29~|~f(v)\equiv f(w)\mod\alpha(e)~\text { for } e\in e_v\cap e_w\}\ ] ] where @xmath30 denotes the vertex set of @xmath23 and @xmath31 denotes the set of all edges in @xmath23 that are adjacent to @xmath32 . .2 cm in the extreme case where @xmath33 , the associated graph @xmath23 with @xmath16 as its vertex set can be derived as follows . .2 cm every real irreducible representation of @xmath10 is one - dimensional , so it must be a homomorphism @xmath34 . moreover , for each @xmath35 , @xmath36 since @xmath37 is an involution . therefore , by identifying the multiplicative group @xmath38 with @xmath8 , one can identify the set of all real irreducible representations of @xmath10 with @xmath25 . notice that @xmath39 is also an @xmath3-dimensional linear space over @xmath8 . .2 cm now choose a point @xmath40 , and choose a linear identification @xmath41 . then the induced tangent representation at @xmath32 can be expressed as an @xmath3-dimensional real representation @xmath42 , which can be further split into @xmath7 real irreducible representations @xmath43 clearly the set @xmath44 is independent of the choice of @xmath45 . since @xmath10 acts effectively on @xmath5 , @xmath46 are linearly independent in @xmath25 , so they form a basis of @xmath25 . .2 cm for each @xmath47 , the subgroup @xmath48 is of codimension one in @xmath10 , and the @xmath49-dimensional component of the fixed set of @xmath50 acting on @xmath5 is a circle @xmath51 ( c.f . @xcite or @xcite ) . since @xmath52 , the @xmath10-action on @xmath5 induces an involution on @xmath51 , so @xmath51 contains only two fixed points @xmath32 and @xmath53 in @xmath16 . let us label @xmath51 by @xmath47 and choose an orientation on @xmath51 . then one gets two _ oriented _ edges with endpoints @xmath54 and with the same `` label '' @xmath47 . since the labeled oriented edges always appear in pairs with reversing orientations , one can merge each pair into an _ unoriented _ edge , so that each @xmath51 determines one edge with two @xmath10-fixed points as its endpoints ( i.e. , a @xmath49-valent graph ) . then , our desired graph @xmath23 is the union of all these @xmath49-valent graphs such that the vertex set of @xmath23 consists of all fixed points of @xmath16 . @xmath23 is called the _ moment graph _ and the `` labelling map '' @xmath55 is called the _ axial function _ , where @xmath56 denotes the set of all unoriented edges . .2 cm it is obvious that in this simple case , the moment graph @xmath23 is a regular @xmath3-valent graph , and the axial function @xmath57 satisfies the following two conditions : 1 . for each vertex @xmath32 in @xmath23 , if @xmath58 are all edges adjacent to @xmath32 , then @xmath59 linearly span @xmath25 , so they are all nontrivial and distinct ; 2 . for each edge @xmath26 in @xmath23 , let @xmath60 and @xmath61 denote the two endpoints of @xmath26 . if @xmath62 and @xmath63 are the sets of edges adjacent to @xmath32 and @xmath53 respectively , then @xmath64 ( because they form irreducible decompositions of two equivalent tangent representations of @xmath65 at @xmath32 and @xmath53 respectively ) . it is not difficult to verify that this is equivalent to @xmath66 .2 cm conversely , given a finite connected regular graph @xmath23 together with some axial function @xmath67 satisfying the previous two conditions , a natural question is : _ what information on the topology of manifolds can we extract from the pair @xmath68 _ ? this seems to be an open - ended question . recently , some interesting and beautiful work on this subject has been done by some mathematicians . for example , in @xcite guillemin and zara obtained some purely combinatorial analogues of the main results in the gkm theory . in addition , they also obtained a realization theorem for abstract gkm - graphs : for an abstract gkm graph @xmath68 , there exists a complex manifold @xmath5 and a gkm action of the torus @xmath69 on @xmath5 such that @xmath68 is its gkm graph . note that the constructed complex manifold @xmath5 is not compact , not equivariantly formal , and admits no canonical compactification . .2 cm this article will deal with the problem from a different viewpoint , in the mod @xmath70 category . we shall introduce a `` skeletal expansion technique '' for any given abstract @xmath10-colored finite connected regular @xmath3-valent graphs @xmath71 , where @xmath67 satisfies the above two conditions . specifically , we shall give a canonical way to inductively attach cells on @xmath23 to form a cell complex @xmath72 such that the @xmath49-skeleton of @xmath72 is just @xmath23 . we will call this technique the _ skeletal expansion _ , and we shall use it to carry out our work as follows . .2 cm first , we determine under what conditions this procedure of cell - gluing can be performed to the end until one obtains a closed manifold ( see theorem [ th : manifold ] ) . we further show that every closed combinatorial @xmath1-manifold can be obtained in this way ( see theorem [ th : colorable ] ) . .2 cm next , we consider the realization problem : can we always construct a @xmath10-action on an @xmath3-dimensional closed manifold such that its moment graph is exactly the given @xmath10-colored regular @xmath3-valent graph @xmath71 ? to do this , our strategy is first to try to construct the orbit space of an action from @xmath71 . then by the work of davis and januszkiewicz @xcite on the reconstruction of small covers ( which is stated briefly below ) , we shall use the obtained orbit space to construct the desired @xmath10-action . we shall show that under certain conditions , any abstract @xmath10-colored finite connected regular @xmath3-valent graph @xmath71 is realizable as the moment graph of an @xmath3-dimensional closed @xmath10-manifold ( see theorem [ realization ] ) . .2 cm an _ @xmath3-dimensional convex polytope _ @xmath73 is just an @xmath3-dimensional compact manifold which is the intersection of a finite set of half - spaces in @xmath74 . it naturally becomes a cell decomposition of @xmath75 and has a minimum set of defining half - spaces , each of which corresponds to an @xmath1-dimensional face of @xmath76 . denote the set of all these codimension one faces by @xmath77 . if every vertex is surrounded by exactly @xmath7 faces in @xmath77 , then @xmath76 is called a _ simple _ convex polytope . clearly , the @xmath49-skeleton of each @xmath3-dimensional simple convex polytope is a regular @xmath3-valent graph @xmath78 . .2 cm if @xmath76 is an @xmath3-dimensional simple convex polytope , a _ characteristic function _ is a map @xmath79 such that the @xmath7 faces in @xmath77 adjacent to each vertex are sent to @xmath7 linearly independent vectors in @xmath10 . each such @xmath80 is dual to an axial function @xmath81 , such that for each edge @xmath82 and those faces @xmath83 containing @xmath26 , @xmath84 . it is easy to see that both @xmath85 and @xmath86 are determined by each other . .2 cm an @xmath3-dimensional closed manifold @xmath5 is called a _ small cover _ over @xmath76 if it admits an effective and locally standard @xmath10-action such that the orbit space @xmath87 . in @xcite , davis and januszkiewicz observed that for each simple convex polytope @xmath76 and each characteristic function @xmath80 , there is a canonical way to construct a small cover @xmath88 over @xmath76 , and every small cover over @xmath76 can be reconstructed in this way . moreover , @xmath78 is exactly the moment graph for @xmath88 and @xmath86 is the corresponding axial function . .2 cm in the case of small covers , reconstructing the orbit spaces and @xmath10-manifolds from moment graphs is simple . in fact , the orbit space @xmath89 of a small cover @xmath5 is bounded by @xmath90 with a very nice regular cell decomposition corresponding to @xmath91 , and this cell complex is topologically dual to a triangulation of @xmath90 . the skeletal expansion of the moment graph @xmath68 of @xmath5 will exactly reproduce the pre - defined cell decomposition on @xmath92 , so that this can lead us to recover the orbit space @xmath76 and its characteristic function @xmath85 , and thus @xmath5 can be reconstructed . more generally , if @xmath5 is not a small cover , the problem of reconstructing the @xmath10-manifold @xmath5 and its orbit space compatible with the moment graph becomes much more complicated . we remark that if the orbit space @xmath89 is a compact @xmath3-manifold with boundary such that the pre - image of each component of the boundary of @xmath89 contains a fixed point , then the skeletal expansion of the moment graph @xmath68 will produce a cell decomposition of the boundary of @xmath89 . .2 cm * acknowledgements . * the authors would like to thank m. masuda , f. luo and x. r. zhang for very helpful discussions leading to the results in dimensions two and three . the authors also would like to thank the referee , who gave many valuable suggestions and comments . in this section , we will first define some standard notation and state our main results . suppose @xmath6 and @xmath23 is a finite connected regular @xmath3-valent graph . @xmath10-coloring _ is a map @xmath93 such that at each vertex , the colors of the @xmath7 edges adjacent to it are linearly independent vectors ( here _ the color of an edge @xmath26 _ is simply the vector @xmath94 ) . notice that for some technical reason , the edge colors are in @xmath95 , but _ not _ in @xmath10 . if in addition the total image of @xmath67 contains only @xmath7 vectors @xmath96 , then we call @xmath67 a _ `` pure '' @xmath10-coloring _ , which is just the _ ordinary edge coloring _ ( see fig . [ fg : coloring ] ) . -colored graphs ] for each subgraph @xmath97 , let @xmath98 denote the linear space spanned by all colors of edges in @xmath99 , and call it _ the color of @xmath99_. let @xmath100 if @xmath72 is a triangulation of a manifold @xmath101 and the graph @xmath23 is its @xmath49-skeleton , then each @xmath102-dimensional simplex @xmath103 can be characterized by the @xmath49-skeleton of @xmath104 , which is a subgraph of @xmath23 . motivated by this idea , we introduce the following concept of @xmath102-nest for colored graph . we say that a proper subgraph @xmath105 is a _ @xmath102-dimensional colored nest _ ( or just a _ @xmath102-nest _ ) associated with @xmath68 if 1 . @xmath99 is connected ; 2 . @xmath106 ; 3 . for every connected subgraph @xmath107 , @xmath108 . if @xmath109 , and they are both colored nests , then we say that @xmath99 is a _ face _ of @xmath110 . let @xmath111 denote the set of all colored nests of @xmath68 . we call @xmath112 the _ @xmath102-dimensional colored nest - skeleton _ ( or just the _ @xmath102-skeleton _ ) of @xmath111 . suppose @xmath25 is spanned by @xmath7 vectors @xmath96 . let @xmath113 denote the @xmath114-exterior algebra generated by these vectors , which consists of all square - free ( i.e. , we require @xmath115 in @xmath116 ) @xmath8-polynomials of @xmath96 , and let @xmath117 denote its linear subspace spanned by all degree @xmath102 monomials , namely @xmath118 then @xmath119 , so the color of each edge can also be considered as an element in @xmath113 . furthermore , for any @xmath120-dimensional linear subspace @xmath121 , choose a basis @xmath122 for @xmath123 , one can easily verify that @xmath124 is independent of the choice of @xmath122 . therefore , for each connected subgraph @xmath97 , @xmath125 determines a unique homogeneous @xmath8-polynomial @xmath126 . in particular , one can extend the coloring @xmath67 to a `` coloring '' @xmath127 clearly a @xmath128-nest @xmath32 is just a vertex of @xmath23 . a @xmath49-nest @xmath26 is an edge in @xmath56 together with its two ends . a @xmath70-nest @xmath129 is a maximal arc or maximal circle such that @xmath130 . if for each edge @xmath131 with endpoints @xmath61 and each edge @xmath132 adjacent to @xmath32 , there exists one and only one edge @xmath133 adjacent to @xmath53 such that @xmath134 then we call @xmath67 a _ `` good '' coloring_. clearly `` pure '' colorings are always `` good '' colorings . notice that the @xmath135 s can be considered as elements in @xmath136 . assume that the colors of edges adjacent to each vertex are linearly independent , then the above equation is equivalent to @xmath137 , and is also equivalent to @xmath138 where @xmath139 denotes the set of all edges in @xmath23 adjacent to a vertex @xmath140 . this also implies that there must be a unique bijection @xmath141 such that for any @xmath142 , @xmath143 the collection @xmath144 is called a _ connection _ of @xmath71 . notice also that if @xmath5 is an @xmath3-dimensional closed manifold with an effective and locally standard @xmath10-action such that the fixed point set @xmath16 is finite , then the orbit space @xmath89 is a nice manifold with corners ( see , @xcite and @xcite ) and each axial function on its moment graph is in fact a `` good '' coloring . [ lm : regular ] suppose @xmath145 and @xmath68 is a finite connected @xmath10-colored regular graph . then @xmath67 is `` good '' if and only if each @xmath102-nest of @xmath111 is a connected regular @xmath102-valent subgraph . suppose @xmath67 is `` good '' . fix a @xmath102-nest @xmath99 . for any vertex @xmath140 , let @xmath146 denote the set of all edges _ inside _ @xmath99 adjacent to @xmath140 , and let @xmath147 denote the linear space spanned by @xmath148 . now let us show that for any two vertices @xmath149 , @xmath32 connected by an edge @xmath150 , @xmath151 . if this is not true , suppose without loss of generality that there is an edge @xmath152 such that @xmath153 . then the connection of @xmath67 points out an edge @xmath154 adjacent to @xmath32 such that @xmath155 since @xmath156 and @xmath99 is maximal , @xmath133 must be an edge inside @xmath99 , so @xmath157 and @xmath158 , which implies @xmath159 a contradiction . since @xmath99 is connected , this implies that @xmath147 is the same for all vertices @xmath140 in @xmath99 . therefore for any vertex @xmath140 in @xmath99 , @xmath160 . however , this is a @xmath102-dimensional linear space , thus the valence of @xmath140 in @xmath99 must be always @xmath102 . notice that for @xmath161 , the colors on @xmath162 edges emerging from the same vertex are always linearly independent so they span a @xmath162-dimensional linear space . on the other hand , suppose every @xmath70-nest is a regular @xmath70-valent graph . for each edge @xmath131 with endpoints @xmath61 and each edge @xmath132 adjacent to @xmath32 , let @xmath163 be the graph formed by edges with colors inside @xmath164 together with their vertices , and let @xmath129 be its connected component containing @xmath132 and @xmath131 . then @xmath129 is a @xmath70-nest , so the valence of @xmath129 at @xmath53 equals two . this implies that there is exactly one edge @xmath165 adjacent to @xmath53 , such that @xmath166 . hence @xmath67 is `` good '' . the proof of lemma [ lm : regular ] actually shows that if @xmath67 is `` good '' , then for any linear subspace @xmath123 , the edges in @xmath167 together with their vertices also forms a disjoint union of some regular graphs . recall that a _ regular cell decomposition _ of a topological space @xmath101 is a cell decomposition @xmath72 such that 1 . for each cell @xmath103 , let @xmath168 be the unit _ open _ ball with dimension equal to @xmath169 , and let @xmath170 be its boundary sphere , then the characteristic map @xmath171 extends to a homeomorphism @xmath172 where @xmath173 ; 2 . for each cell @xmath103 , its boundary sphere @xmath104 is the union of some cells in @xmath72 . suppose @xmath174 and @xmath101 is a topological space which has a @xmath102-dimensional finite regular cell decomposition @xmath72 . if there is a one - to - one correspondence @xmath175 preserving dimensions and face relations , then @xmath176 is called a _ @xmath102-skeletal expansion of @xmath68 _ , and @xmath177 is called a _ @xmath10-colored cell decomposition of @xmath101_. if @xmath178 and @xmath101 is a closed manifold , then @xmath101 is called a _ @xmath0-colorable manifold_. if @xmath23 is the @xmath49-skeleton of an @xmath3-dimensional simple convex polytope @xmath76 , and @xmath179 is dual to a characteristic function of @xmath76 ( c.f . section 1 ) , then @xmath91 with the induced cell decomposition is an @xmath1-skeletal expansion of @xmath68 . obviously , @xmath49-skeletal expansions always exist . one can simply choose the vertices of @xmath23 as the @xmath128-cells and choose the edges in @xmath56 as the @xmath49-cells . [ lm : skeletal ] suppose @xmath145 and the coloring @xmath67 is `` good '' . then @xmath68 always has a @xmath70-skeletal expansion . by lemma [ lm : regular ] , each @xmath70-nest is a connected regular @xmath70-valent graph . therefore all 2-nests must be embedded circles . furthermore , one can glue discs to each of these circles in the graph to get a @xmath70-skeletal expansion . if @xmath145 and @xmath180 is `` pure '' , then @xmath68 always has a @xmath70-skeletal expansion . notice that a @xmath70-nest in this case must be a simple closed curve in @xmath23 consisting of an even number of edges with alternating edge - coloring . we call them _ bi - colored circles_. higher dimensional skeletal expansions can be built up inductively using the following lemma , which can be easily proved similarly . [ lm : induction ] suppose that @xmath67 is `` good '' , @xmath68 has a @xmath102-skeletal expansion @xmath176 , and @xmath175 is a one - to - one correspondence . for each @xmath181 edges @xmath182 sharing one common vertex @xmath32 , let @xmath99 be the unique @xmath183-nest containing @xmath184 , and let @xmath185 if each such @xmath186 is a @xmath102-sphere , then one can glue a @xmath183-cell to each @xmath102-sphere and obtain a @xmath183-skeletal expansion of @xmath68 . in general , @xmath101 is only a topological space but not a manifold , so it would be quite natural to ask the following three questions : 1 . when will a finite connected regular @xmath3-valent graph @xmath23 with a `` good '' @xmath0-coloring admit an @xmath1-skeletal expansion ? 2 . if @xmath68 admits an @xmath1-skeletal expansion @xmath176 , when will @xmath101 be a closed manifold ? 3 . what kind of @xmath1-dimensional manifold can be @xmath187-colorable ? in the next two sections , we will give a complete answer to these questions in dimensions @xmath70 and @xmath188 . for dimensions greater than @xmath188 , question ( q1 ) is still open , due to the difficulties in recognizing regular cell decompositions of @xmath189 . however , questions ( q2 ) and ( q3 ) are solved completely by the following two theorems . [ th : manifold ] suppose @xmath6 , @xmath68 is a @xmath10-colored finite connected regular @xmath3-valent graph in which @xmath67 is a `` good '' coloring , and @xmath176 is an @xmath1-skeletal expansion of @xmath68 . then @xmath101 is an @xmath1-dimensional closed manifold . an @xmath1-dimensional _ closed combinatorial manifold _ is a connected topological space with a finite @xmath1-dimensional regular cell decomposition , such that the link complex of each @xmath102-cell is an embedded @xmath190-sphere . notice that every closed differentiable manifold is a closed combinatorial manifold ( see e.g. @xcite ) , and every closed combinatorial manifold is a closed ( topological ) manifold . [ th : colorable ] every @xmath1-dimensional closed combinatorial manifold is @xmath4-colorable . moreover , one can choose `` pure '' colorings for doing this . it is not difficult to see that this is also a necessary condition for being colorable , namely an @xmath1-dimensional closed manifold is @xmath4-colorable _ if and only if _ it is a closed combinatorial manifold . although a manifold constructed as above does not naturally carry any group actions , in some sense , it is associated with the boundary of the orbit space of a manifold with group actions . thus we can pose the following realization problem : 1 . let @xmath68 be a @xmath0-colored finite connected regular @xmath3-valent graph such that @xmath67 is a `` good '' coloring . under what conditions will @xmath68 be realizable as the moment graph of a @xmath0-action on an @xmath3-dimensional closed manifold ? we know by theorem [ th : manifold ] that if @xmath68 admits an @xmath1-skeletal expansion @xmath176 , then @xmath101 is an @xmath1-dimensional closed manifold . by a simple trick , we can modify @xmath101 into the boundary of another manifold @xmath191 and turn @xmath191 into the orbit space of a @xmath10-manifold . this leads us to the following result . [ realization ] suppose @xmath68 is a @xmath0-colored finite connected regular @xmath3-valent graph such that @xmath67 is a `` good '' coloring . if @xmath68 admits an @xmath1-skeletal expansion @xmath176 , then it is realizable as the moment graph of some @xmath0-action on an @xmath3-dimensional closed manifold . although we have not achieved a complete answer for the realization problem , it is very tempting to make the following conjecture : for each @xmath0-colored finite connected regular @xmath3-valent graph @xmath68 with @xmath67 being a `` good '' coloring , @xmath68 can be realized as the moment graph of some @xmath0-action on an @xmath3-dimensional closed manifold . let us first examine the case @xmath193 , which is surprisingly simple . [ th : surface ] suppose @xmath194 , @xmath23 is a finite connected regular @xmath188-valent graph , and @xmath180 is a `` good '' coloring . then @xmath68 has a @xmath70-skeletal expansion @xmath176 in which @xmath101 is a closed surface . by lemma [ lm : skeletal ] , @xmath68 must have a @xmath70-skeletal expansion @xmath176 . moreover , since each edge is used exactly twice by @xmath70-cells , this gives a pairing of arcs on the boundaries of those @xmath70-cells . therefore , gluing these cells together along their boundaries will give a closed surface . as an example of application , the two graphs in fig . [ fg : coloring ] both satisfy the above conditions . it is not difficult to see that the surface corresponding to fig . [ fg : coloring ] ( a ) is a 2-sphere , while the surface for fig . [ fg : coloring ] ( b ) is a real projective plane @xmath195 . every closed surface @xmath101 is @xmath192-colorable . moreover , one can choose planar graphs and `` pure '' colorings for doing this . the first sentence is in fact a corollary of theorem [ th : colorable ] . notice that the graph used in fig . [ fg : surface ] for a real projective plane @xmath195 is exactly the same as fig . [ fg : coloring ] ( b ) . the graphs for building up closed surfaces are shown in fig . [ fg : surface ] , where @xmath196 is a chosen basis for @xmath192 . by theorem [ th : surface ] , these graphs all have @xmath70-skeletal expansions , and the corresponding topological space will automatically become closed surfaces . now let us verify the topological type for these surfaces . the graph for @xmath197 is just a simple exercise . let us examine the graph for @xmath198 ( the connected sum of @xmath37 copies of @xmath199 ) . the graph has @xmath200 vertices , @xmath201 edges and @xmath202 bi - colored circles , which are listed below : @xmath203 [ cols="^,^,<",options="header " , ] @xmath203 suppose @xmath176 is its @xmath70-skeletal expansion . the @xmath70-cells bounded by @xmath204 and @xmath205 can not have opposite orientations on all their common boundary arcs , so @xmath101 must be a non - orientable surface , while its euler characteristic is equal to @xmath206 . thus @xmath101 is indeed a non - orientable surface with genus @xmath102 . the graphs for @xmath198 and @xmath207 are actually constructed from the graphs for @xmath208 ( torus ) and @xmath209 ( real projective plane ) by doing _ connected sums_. suppose @xmath210 is a regular @xmath188-valent graph with `` pure '' coloring @xmath211 and @xmath212 is its @xmath70-skeletal expansion ( @xmath213 or @xmath70 ) . if @xmath214 are two edges in @xmath215 respectively with the same color , then one can cut them open , reconnect the four ends in the other way and form a new regular @xmath188-valent graph @xmath23 with `` pure '' coloring @xmath216 . suppose @xmath176 is its @xmath70-skeletal expansion , then it is not difficult to see that @xmath101 is exactly the connected sum of @xmath217 and @xmath218 . now let us consider the case @xmath220 . [ th:3-manifold ] suppose @xmath221 , @xmath23 is a connected regular @xmath222-valent graph and @xmath180 is a `` good '' @xmath10-coloring . then @xmath68 has a @xmath188-skeletal expansion @xmath176 ( by theorem [ th : manifold ] @xmath101 must be a closed @xmath188-manifold ) if and only if : @xmath223 first let us use @xmath224 to denote ( only in this proof ) the number of @xmath102-nests in @xmath23 . now suppose @xmath68 has a @xmath188-skeletal expansion @xmath176 . then by theorem [ th : manifold ] , @xmath101 is a closed @xmath188-manifold . so it is not difficult to see that @xmath225 is equal to the euler characteristic of @xmath101 , which must be @xmath128 . since @xmath23 is a regular @xmath222-valent graph , @xmath226 so we get @xmath227 . on the other hand , suppose that @xmath227 . first choose a @xmath70-skeletal expansion @xmath228 for @xmath68 according to lemma [ lm : induction ] . for each three edges @xmath229 sharing a common vertex , suppose @xmath99 is the unique @xmath188-nest containing them and suppose @xmath230 . let @xmath231 denote the number of @xmath102-nests in @xmath99 . then the euler characteristic of @xmath232 is equal to @xmath233 , since @xmath232 is a @xmath70-skeletal expansion of @xmath99 . however , @xmath232 is in fact a closed surface , thus @xmath234 . because @xmath99 is a regular @xmath188-valent graph without loop edge , @xmath235 . thus , @xmath236 because any @xmath102 edges ( @xmath237 , or @xmath188 ) sharing a common vertex determine a unique @xmath102-nest , we know that every vertex is the face of exactly four @xmath188-nests , and every @xmath70-nest is the face of exactly two @xmath188-nests . therefore summing up the above inequality over all @xmath188-nests , one obtains @xmath238 our assumption implies that the inequalities here must all be equalities . consequently those @xmath239 must all be spheres . hence by lemma [ lm : induction ] , one can fill these spheres with some @xmath188-cells and get a @xmath188-skeletal expansion @xmath176 . by theorem [ th : manifold ] , @xmath101 must be a closed @xmath188-manifold . to understand intuitively why @xmath101 must be a closed @xmath188-manifold ( without invoking theorem [ th : manifold ] ) , we can consider the local picture near a vertex @xmath32 ( see fig . [ fg : local ] ) . choose some local coordinates such that @xmath32 becomes @xmath240 and the adjacent edges become ( locally ) @xmath241 , @xmath242 , @xmath243 , @xmath244 respectively . suppose @xmath245 . for each @xmath246 , there is exactly one @xmath70-cell @xmath247 adjacent to both @xmath248 and @xmath249 , let us glue it to @xmath248 and @xmath249 ( locally ) as follows : @xmath250 for each @xmath251 , there is exactly one @xmath188-cell @xmath252 adjacent to three edges @xmath248 , @xmath249 and @xmath253 , and we can glue it to @xmath248 , @xmath249 and @xmath253 ( locally ) as follows : @xmath254 the possibility of this arrangement implies that every vertex @xmath32 has a neighborhood in @xmath101 homeomorphic to @xmath255 . thus the combinatorial nature of @xmath176 implies that @xmath101 is a closed @xmath188-manifold . -skeletal expansion ] on the other hand , one can also use this canonical method to verify the @xmath10-colorability of a general @xmath188-manifold , by finding cell decomposition that locally looks like the above canonical picture near each vertex . every closed @xmath188-manifold is @xmath219-colorable . moreover , one can use only four @xmath188-cells to do this . notice that the first half of the statement is also a corollary of theorem [ th : colorable ] , since every closed @xmath188-manifold has a ( in fact unique up to diffeomorphism ) differential structure ( see e.g. @xcite ) , which implies that every closed @xmath188-manifold is a closed combinatorial manifold . notice also that for @xmath188-manifolds the graph may _ not _ be planar anymore , and may contain double - edges . recall the idea of heegaard splitting ( see e.g. @xcite ) : suppose @xmath256 are two genus-@xmath37 handle - bodies ( here we require @xmath257 ) , @xmath258 are two genus-@xmath37 closed orientable surfaces , and @xmath259 is a homeomorphism , then one can glue @xmath260 to @xmath261 along their boundaries using @xmath262 and get a closed @xmath188-manifold @xmath263 . every closed @xmath188-manifold @xmath101 can be obtained in this way and the corresponding @xmath264 is called a _ heegaard splitting _ for @xmath101 . ] each genus-@xmath37 handle - body @xmath265 can be further decomposed into two balls @xmath266 in a canonical way along @xmath267 disjoint non - separating discs @xmath268 . let us decompose @xmath261 and @xmath260 in this way . for notational convenience , let @xmath269 @xmath270 and let @xmath271 ( see fig . [ fg : heegaard ] ) . notice that the topological type of @xmath263 remains unchanged when changing @xmath262 isotopically , while these @xmath272 will be moved around on @xmath273 when doing so . clearly one can isotopically adjust @xmath262 a little bit , such that each @xmath272 intersects these @xmath274 transversely in a finite number of points . moreover , one can isotopically adjust @xmath262 a little bit further , adding some redundant turns to those @xmath272 , such that for any @xmath275 , @xmath276 and all these @xmath274 and @xmath272 together cut @xmath273 into small _ disc _ regions . now the points in @xmath277 cut these @xmath274 and @xmath272 into edges , and these edges together form a finite connected regular @xmath222-valent graph @xmath278 . suppose @xmath279 is a basis for @xmath10 , and @xmath280 is the dual basis in @xmath25 , namely @xmath281 . define a characteristic function @xmath282 and then define the edge coloring @xmath180 as follows : @xmath283 if and only if @xmath284 . in fact , each @xmath285 lies within the boundary of exactly three of @xmath286 . if the fourth ball is @xmath287 , then @xmath288 . it is easy to see that the set of colors of the four edges around each vertex of @xmath23 is exactly @xmath280 , so @xmath67 is a `` pure '' coloring . let us calculate the @xmath188-skeletal expansion of @xmath68 . since @xmath67 is a `` pure '' coloring , the set of all 2-nests of @xmath68 consists of exactly all the bi - colored circles of @xmath23 . by the construction of @xmath23 as described above , it is easy to see that each bi - colored circle of @xmath23 is either the boundary of a disc region of @xmath289 or one of those @xmath274 ( bounds @xmath290 in @xmath261 ) or @xmath272 ( bounds @xmath291 in @xmath260 ) . let those @xmath292 and disc regions be the @xmath70-cells . then one may construct a @xmath70-skeletal expansion @xmath228 of @xmath71 , where @xmath218 is obtained by gluing each @xmath293 to @xmath294 along the boundary , such that @xmath295 is identified with @xmath296 . since an edge @xmath297 if and only if @xmath298 , all edges in each @xmath299 form a connected regular @xmath188-valent subgraph @xmath300 with @xmath301 , and @xmath302 are all the @xmath188-nests . hence by adding @xmath303 as the @xmath188-cells , one obtains a @xmath188-skeletal expansion @xmath304 . clearly @xmath305 is just the original closed @xmath188-manifold @xmath101 . as an example , fig . [ fg : graph ] ( a ) shows the colored graph corresponding to the heegaard splitting of @xmath189 in fig . [ fg : heegaard ] . for a more fancy picture of this graph [ fg : graph ] ( b ) . ] another example is @xmath306 . [ fg : example ] shows a nice heegaard splitting , such that every small region on @xmath273 is a disc ( the picture for @xmath260 is omitted ) , and fig . [ fg : examplegraph ] is the corresponding colored graph . ] ] similarly to the case of simple convex polytopes , the colorings here are also induced by characteristic functions on the @xmath1-dimensional faces . however , the cell decompositions here are not dual to any triangulation , so this is different from the case of simple convex polytopes or the graphs constructed in the proof of theorem [ th : colorable ] . in fact a triangulation for a closed @xmath188-manifold has at least two @xmath188-simplices , so it has at least five vertices , which implies that its dual regular cell decomposition has at least five @xmath188-cells . because of the combinatorial nature of @xmath176 , for any cell @xmath307 with dimension @xmath308 and any two points @xmath309 inside @xmath310 ( away from @xmath311 ) , @xmath149 has a neighborhood homeomorphic to @xmath312 if and only if @xmath32 does so . therefore we only need to verify that every point in @xmath101 sufficiently close to a vertex has a neighborhood homeomorphic to @xmath312 . now choose a vertex @xmath32 , we will show that @xmath32 has a neighborhood homeomorphic to the space @xmath313 which is homeomorphic to @xmath312 . the argument is quite similar to the case @xmath220 in section 3 . choose a local coordinate such that @xmath32 becomes @xmath314 and the adjacent edges become ( locally ) @xmath315 respectively . suppose @xmath245 . for each @xmath316 , there is a unique @xmath102-nest @xmath317 containing @xmath318 , and by lemma [ lm : regular ] these @xmath317 enumerate all @xmath102-nests containing @xmath32 . suppose @xmath319 is the @xmath102-cell corresponding to @xmath317 . by an induction on @xmath102 one can adjust the local coordinates such that each of these @xmath319 becomes ( locally ) @xmath320 the possibility of this arrangement implies that every vertex @xmath32 has a neighborhood in @xmath101 homeomorphic to the cone @xmath321 , which is homeomorphic to @xmath312 . thus the combinatorial nature of @xmath176 implies that every point in @xmath101 has a neighborhood homeomorphic to @xmath312 . @xmath101 is connected since @xmath23 is connected , and @xmath101 is compact because it has only finitely many cells . therefore @xmath101 is a closed @xmath1-manifold . before proving theorem [ th : colorable ] , let us recall some notation . if a cell @xmath310 is a face of cell @xmath322 and @xmath323 , we write @xmath324 . now suppose that @xmath72 is a finite regular cell decomposition of @xmath101 , then the _ barycentric subdivision _ @xmath325 is a simplicial complex , such that every @xmath102-simplex in @xmath325 corresponds to a sequence of cells @xmath326 , we denote such a simplex by @xmath327 $ ] , and call it a _ flag_. notice that such a simplicial complex always exists and is a triangulation of @xmath101 . when @xmath72 itself is a triangulation of @xmath101 , @xmath327 $ ] is just the convex hull spanned by @xmath328 , where @xmath329 denotes the barycenter of @xmath310 . for each cell @xmath307 , the _ link complex _ is @xmath330\in\mathrm{sd}~k~|~ \sigma\prec\sigma_0\prec\cdots\prec\sigma_k\}.\ ] ] if @xmath101 is connected and has a finite regular cell decomposition @xmath72 , and @xmath331 ( @xmath332 is the union of all cells in @xmath333 ) , then @xmath101 is called an _ @xmath1-dimensional closed combinatorial manifold_. notice that if @xmath72 satisfies this condition , then @xmath325 also satisfies this condition . finally for each cell @xmath307 , the _ dual block _ is @xmath334\in\mathrm{sd}~k~|~ \sigma=\sigma_0\prec\cdots\prec\sigma_k\}.\ ] ] if @xmath72 satisfies the previous condition , then these dual blocks are all cells ( in fact @xmath335 is the topological cone on @xmath333 ) , and they also form a finite regular cell decomposition of @xmath101 , we denote it by @xmath336 . in particular , the @xmath49-skeleton of this dual cell decomposition is a regular graph . it is clear that @xmath337 is a one - to - one correspondence _ reversing face relations _ , and @xmath338 . suppose that @xmath339 is a finite regular cell decomposition of @xmath101 satisfying the condition @xmath340 let @xmath341 be its barycentric subdivision . then @xmath342 also satisfies the above condition , thus one can define a dual regular cell decomposition @xmath343 . now let us define a coloring @xmath67 for the @xmath49-skeleton of @xmath72 ( denoted by @xmath23 ) as follows . let @xmath96 be a linear basis for @xmath25 . for each @xmath49-cell @xmath344\in\gamma$ ] ( here @xmath345 ) , suppose that @xmath346 then let @xmath347 . because @xmath348 reverses face relations , for each vertex @xmath349 , say @xmath350 $ ] ( here @xmath351 ) , there are exactly @xmath3 edges in @xmath72 adjacent to @xmath32 , namely @xmath352,~k=0,\ldots , n.\ ] ] since @xmath353 , this implies that @xmath68 is a finite regular @xmath3-valent graph with `` pure '' @xmath10-coloring ( see fig [ fg : dual ] ) . now for any @xmath120-cell @xmath354 $ ] ( here @xmath355 and @xmath356 ) , let @xmath357 be the @xmath49-skeleton of @xmath104 . then a vertex @xmath358 $ ] ( here @xmath359 ) is a vertex of @xmath357 if and only if @xmath360 and an edge @xmath361 $ ] adjacent to @xmath362 is an edge of @xmath357 if and only if @xmath363 , namely @xmath364 @xmath365 this implies that @xmath357 is a connected regular @xmath120-valent subgraph of @xmath23 with edge - colors in @xmath366 , hence @xmath357 is an @xmath120-nest in @xmath111 . therefore the correspondence @xmath367 , sending each cell in @xmath72 to the @xmath49-skeleton of its boundary complex , is a well defined map which preserves dimensions and face relations . @xmath368 is also a one - to - one correspondence . in fact , for any @xmath120-nest @xmath99 , suppose @xmath350 $ ] is a vertex in @xmath99 and @xmath369 are the @xmath120 edges in @xmath99 adjacent to @xmath32 , with each @xmath370 $ ] . then any inverse image of @xmath99 under @xmath368 must contain @xmath32 and these edges , and there does exists a unique @xmath120-cell containing @xmath369 , namely @xmath371 $ ] ( here @xmath372 . clearly @xmath373 $ ] is the one and only cell that is sent to @xmath99 by @xmath368 . thus @xmath368 must be a one - to - one correspondence . as an example of calculation , let us reproduce the `` cubic '' graph for @xmath90 as a @xmath374 . take a regular cell decomposition @xmath339 for @xmath90 with two @xmath375-cells at each dimension @xmath376 : @xmath377 notice that @xmath378 if and only if @xmath379 . therefore @xmath380 contains @xmath381 vertices , each of which has the form @xmath382,~ \epsilon_0,\ldots,\epsilon_n\in\{1,2\}.\ ] ] to give the graph a more fancy appearance , identify each such vertex with @xmath383 two vertices @xmath384 and @xmath385 are connected by an edge in @xmath72 of the form @xmath386\ ] ] ( colored by @xmath387 ) if and only if @xmath388 while for all other @xmath389 , @xmath390 . this implies that we can also identify each such edge with the line segment @xmath391 hence one obtains the @xmath49-skeleton of the @xmath3-dimensional cube @xmath392^{n+1}\subset \mathbb r^{n+1},\ ] ] and an edge of this graph is colored by @xmath387 if and only if it is parallel to the @xmath15-axis . when @xmath193 , this equals to the graph for @xmath197 in fig . [ fg : surface ] . there is in fact a more general way of doing skeletal expansions , which we call the `` generalized skeletal expansion '' . for example , take the colored graph in fig . [ fg : coloring ] ( a ) , one can also glue mbius bands instead of discs to each of the bi - colored circles , and then the resulting manifold will be a closed non - orientable surface with genus @xmath222 . to describe this expansion more accurately , we will use an induction on the dimension . the generalized @xmath49-skeletal expansion of a regular graph @xmath23 with `` good '' coloring @xmath67 is just @xmath23 itself . for @xmath393 , a generalized @xmath102-skeletal expansion of @xmath68 is a space @xmath394 together with a defining sequence @xmath395 of gluing operations such that @xmath23 can be extended to @xmath394 by @xmath395 and each @xmath396-nest with @xmath397 can be extended to a closed @xmath398-dimensional manifold . if @xmath68 admits a generalized @xmath102-skeletal expansion @xmath399 , then @xmath395 also induces a generalized @xmath102-skeletal expansion for each @xmath183-nest of @xmath68 . by @xmath400 we denote the disjoint union of the generalized @xmath102-skeletal expansions of all @xmath183-nests in @xmath23 . suppose that @xmath68 admits a generalized @xmath102-skeletal expansion @xmath399 such that @xmath401 is the boundary of a @xmath183-dimensional compact manifold @xmath402 ( may not be connected ) . then one can glue @xmath403 onto @xmath394 to form a new space @xmath404 . add this operation to @xmath395 to form a new sequence of operations @xmath405 , and we call @xmath406 a _ generalized @xmath183-skeletal expansion _ of @xmath68 . clearly , in each dimension , ordinary skeletal expansions ( if they exist ) are uniquely determined by the colored graphs up to cell isomorphisms , while the generalized skeletal expansions are not so . as an example , let us study the colored regular @xmath222-valent graph @xmath23 shown in fig . [ fg : general ] . it admits a @xmath70-skeletal expansion but admits no @xmath188-skeletal expansion , since by theorem [ th:3-manifold ] @xmath407 in fact @xmath23 has five @xmath188-nests . when doing ( ordinary ) @xmath70-skeletal expansion on their disjoint union , two of them ( both colored by @xmath408 ) yield @xmath195 , while the other three ( colored by @xmath409 , @xmath410 and @xmath411 respectively ) yield @xmath197 . however , one can still glue three @xmath188-cells to those @xmath197 , glue a @xmath412 to the two @xmath195 , and finally glue the boundary surfaces of all these blocks together according to the chosen ( ordinary ) @xmath70-skeletal expansion . the resulting space of this generalized @xmath188-skeletal expansion is homeomorphic to @xmath413 . this graph @xmath23 even has a generalized @xmath222-skeletal expansion , since @xmath413 is the boundary of the compact @xmath222-manifold @xmath414 . -skeletal expansion ] for the generalized skeletal expansion , we would like to propose the following conjecture : suppose that @xmath23 is a regular @xmath3-valent graph with a `` good '' @xmath0-coloring @xmath67 . then @xmath68 always admits a generalized @xmath3-skeletal expansion @xmath415 such that @xmath101 is a compact @xmath3-manifold with boundary . finally , let us consider the realization problem of reconstructing @xmath3-dimensional @xmath10-manifolds from @xmath3-valent @xmath10-colored graphs , where @xmath6 . in the previous section , we showed that under certain conditions , a @xmath10-colored graph @xmath68 admits a skeletal expansion into a closed manifold @xmath101 . if @xmath101 can be used as the boundary of a simple convex polytope @xmath76 , then @xmath67 determines a characteristic function @xmath85 on @xmath76 . furthermore , by the reconstruction technique of small covers in @xcite , one can use @xmath85 and the product bundle @xmath416 to construct a small cover @xmath5 such that the moment graph of @xmath5 is exactly the given @xmath10-colored graph @xmath68 . this observation provided much insight into the study of the realization problem . let @xmath23 be a finite connected regular @xmath3-valent graph with a `` good '' @xmath10-coloring @xmath67 . suppose that @xmath71 admits an @xmath1-skeletal expansion @xmath176 . then by theorem [ th : manifold ] , @xmath101 is an @xmath1-dimensional closed manifold . first , let us prove the special case in which @xmath101 is the boundary of a compact @xmath3-dimensional manifold @xmath417 . suppose that @xmath418 is a @xmath102-nest with @xmath174 . we recall that the color of an edge @xmath26 is a nontrivial element @xmath94 in @xmath25 . let @xmath419 denote the set of all edges in @xmath99 . then @xmath420 is a co - rank @xmath102 subgroup of @xmath10 . notice that @xmath421 if @xmath99 is a 0-nest ( i.e. , a single vertex ) . on the other hand , the @xmath422-skeletal expansion of @xmath99 bounds a ball , and it corresponds to a @xmath102-dimensional open cell @xmath423 in @xmath101 . moreover , for each point @xmath424 , there is a unique nest @xmath99 such that @xmath425 . now suppose that @xmath417 admits a trivial action of @xmath10 . consider the trivial principal @xmath10-bundle @xmath426 over @xmath417 , where the bundle projection is denoted by @xmath427 , and the action of @xmath10 on @xmath205 is defined by @xmath428 as follows : @xmath429 . then we define an equivalence relation @xmath430 on @xmath205 as follows : @xmath431 if and only if clearly @xmath5 is obtained by gluing @xmath381 copies of @xmath417 together along their boundaries . to prove that @xmath101 is indeed an @xmath3-dimensional manifold , it suffices to show that each point @xmath436 $ ] with @xmath439 has a neighborhood homeomorphic to @xmath74 . let @xmath436\in x$ ] with @xmath439 . then there is a @xmath102-nest @xmath99 such that @xmath440 . since all points inside a component of @xmath441 can have homeomorphic neighborhoods , we need only to prove that @xmath436 $ ] has a neighborhood homeomorphic to @xmath442 when @xmath443 is very close to a vertex . without loss of generality , one assumes that @xmath444 is a vertex of @xmath23 . then there are @xmath7 @xmath1-nests containing @xmath32 , denoted by @xmath445 respectively . choose a local coordinate chart @xmath446 near @xmath32 in @xmath417 such that @xmath32 is mapped into the origin of @xmath442 , each @xmath447 is mapped into an @xmath1-dimensional open cone @xmath448 and @xmath449 is mapped into an @xmath3-dimensional cone @xmath450 each @xmath451 is a subgroup of rank 1 , so we can write @xmath452 . obviously , @xmath453 forms a basis of @xmath10 . then one can define a linear action @xmath454 by @xmath455 using this action , one can extend @xmath456 into a continuous surjection @xmath457 defined by @xmath458 it is not difficult to see that @xmath459 if and only if @xmath460 . therefore , @xmath461 induces a homeomorphism from a neighborhood of @xmath436 $ ] to @xmath442 . the action @xmath462 of @xmath10 on @xmath205 also induces a natural action @xmath463 on @xmath5 . if @xmath99 is a @xmath102-nest , let @xmath464 , then @xmath10 acts on @xmath465 with kernel @xmath466 . in fact , @xmath465 contains @xmath467 copies of @xmath423 and the action of @xmath10 on @xmath465 permutes these copies . also , the fixed point set of @xmath468 acting on @xmath5 contains @xmath469 as a component . in particular , for an edge @xmath26 with two ends @xmath470 , the fixed point set of @xmath471 acting on @xmath5 contains a circle which is the union of @xmath472 ( two arcs ) and @xmath473 ( two fixed points ) . this also implies that the moment graph of the @xmath10-manifold @xmath5 is just @xmath71 . now let us consider the general case in which @xmath101 is not the boundary of a compact @xmath3-dimensional manifold . choose an @xmath1-nest @xmath99 , by the construction of @xmath101 , then @xmath99 corresponds to an open @xmath1-cell @xmath423 of @xmath474 and the @xmath1-skeletal expansion of @xmath99 is exactly the closure @xmath475 of @xmath423 , which is an @xmath1-disc . now let us remove a very small open @xmath1-ball @xmath168 in the interior of @xmath475 . next , taking another copy @xmath476 of @xmath101 and forgetting those combinatorial structures @xmath71 and @xmath72 on it , one randomly removes a small open @xmath1-ball @xmath477 in @xmath476 . then one glues @xmath478 and @xmath479 together along their boundaries . this is actually a connected sum between two copies of @xmath101 , so the resulting manifold @xmath191 becomes the boundary of an @xmath3-dimensional compact manifold @xmath417 . in this operation , we see that the @xmath1-skeletal expansion @xmath475 of @xmath99 is exactly replaced by the connected sum between @xmath475 and @xmath476 . thus , @xmath191 becomes actually a generalized @xmath1-skeletal expansion of @xmath71 such that @xmath99 corresponds to the interior of @xmath479 rather than @xmath423 . it is not difficult to see that the above argument still works for this @xmath417 . this completes the proof . it should be pointed out that if we start from _ any _ principal @xmath10-bundle over @xmath417 ( not just the trivial one ) , then this argument still works , i.e. , we can always obtain a @xmath10-manifold ( c.f . @xcite , @xcite , @xcite , and @xcite ) . in the case where @xmath193 , theorem [ realization ] gives a complete answer to question ( q4 ) . this is because @xmath71 in this case always admits a 2-skeletal expansion , and the resulting surface bounds a 3-manifold if and only if its euler characteristic is even . for every finite connected regular 3-valent graph @xmath23 with a `` good '' @xmath10-coloring @xmath67 where @xmath194 , there is a 3-dimensional @xmath10-manifold @xmath5 such that @xmath68 corresponds to its moment graph . for every finite connected regular 4-valent graph @xmath23 with a `` good '' @xmath10-coloring @xmath67 where @xmath221 , if @xmath480 , then there is a 4-dimensional @xmath10-manifold @xmath5 such that @xmath68 corresponds to its moment graph .

in an earlier paper we developed a parton bubble model ( pbm ) for rhic , high - energy heavy - ion collisions . pbm was based on a substructure of a ring of localized bubbles ( gluonic hot spots ) which initially contain 3 - 4 partons composed of almost entirely gluons . the bubble ring was perpendicular to the collider beam direction , centered on the beam , at midrapidity , and located on the expanding fireball surface of au au central collisions ( 0 - 10% ) at=200 gev . the bubbles emitted correlated particles at kinetic freezeout , leading to a lumpy fireball surface . for a selection of charged particles ( 0.8 gev / c 4.0 gev / c ) , the pbm reasonably quantitatively ( within a few percent ) explained high precision rhic experimental correlation analyses in a manner which was consistent with the small observed hbt source size in this transverse momentum range . we demonstrated that surface emission from a distributed set of surface sources ( as in the pbm ) was necessary to obtain this consistency . in this paper we give a review of the above comparison to central au au collisions . the bubble formation can be associated with gluonic objects predicted by a glasma flux tube model ( gftm ) that formed longitudinal flux tubes in the transverse plane of two colliding sheets of color glass condensate ( cgc ) , which pass through one another . these sheets create boost invariant flux tubes of longitudinal color electric and magnetic fields . a blast wave gives the tubes near the surface transverse flow in the same way it gave transverse flow to the bubbles in the pbm . in this paper we also consider the equivalent characteristics of the pbm and gftm and connect the two models . in the gftm the longitudinal color electric and magnetic fields have a non - zero topological charge density . these fields cause a local strong cp violation which effects charged particle production coming from quarks and anti - quarks created in the tube or bubble . # 1#2#3#4#1 * # 2 * , # 3 ( # 4 )

1001.1132

our interest in and the eventual development of the pbm goes back many years to the early nineteen eighties . van hove s work@xcite in the early eighties considered the bubbles as small droplets of quark - gluon plasma ( qgp ) assumed to be produced in ultra - relativistic collisions of hadrons and/or nuclei . his work recognized that experimental detectors only see what is emitted in the final state at kinetic freezeout . therefore final state predictions of his string model work on plasma bubbles or any other theory required specific convincing predictions that are observable in the final state which can be measured by experiment . his work predicted that the final state rapidity distribution dn / dy of hadrons would exhibit isolated maxima of width @xmath4 ( single bubble ) or rapidity bumps of a few units ( due to those cases producing a few bubbles ) . these rapidity regions would also contain traditional signals of qgp formation . even this early work by van hove needed his bubbles near the surface of the final state fireball in order to experimentally measure the particles emitted in the final state at kinetic freezeout . we and other searched for these van hove bubbles , but no one ever found any significant evidence for them . our next paper@xcite was our final attempt to make a theoretical treatment of the single bubble case ( similar to the van hove case ) which could be experimentally verified . it is possible that with enough statistics one in principle could find single to a few bubble events . we developed a number of event generators which could possibly provide evidence for striking signals resulting from these bubbles . one should note that in that paper we had included in sec . 11 mathematical expressions describing how charged pions are effected by strong color electric and color magnetic fields which are present and parallel in a cp - odd bubble of metastable vacuum . this section was motivated by the work of kharzeev and pisarski@xcite . for example eq . 3 - 4 of the section show the boosts of the @xmath5 and the @xmath6 we estimated . thus in 2000 we were already interested in and were investigating and publishing predictions for this phenomenon . when we considered the rhic quantum interference data in 2000@xcite it became clear that the fireball surface was rapidly moving outward . this implied a very large phase space region covered by the fireball , and also implied that it would be very unlikely for there to be only one isolated bubble ( small source size ) with a large amount of energy sitting on the surface . it seemed more likely that there would be many bubbles or gluonic hot spots around the expanding surface . this led to a paper@xcite which was our earlier version of ref.@xcite . in that paper we concluded that the behavior of the hanbury - brown and twiss ( hbt ) measurements@xcite should be interpreted as evidence of a substructure of bubbles located on the surface of the final state fireball in the central rapidity region at kinetic freezeout . the hbt radii were decreasing almost linearly with transverse momentum and implying to us a source size of @xmath72 fm radii bubbles were on the surface and could be selected for if we considered transverse momenta above 0.8 gev / c . these momenta would allow sufficient resolution to resolve individual bubbles of @xmath72 fm radii . we further concluded that these hbt quantum interference observations were likely due to phase space focusing of the bubbles pushed by the expanding fireball . thus the hbt measurements of source size extrapolated to transverse momenta above 0.8 gev / c were images of the bubbles . the hbt correlation has the property of focusing these images on top of each other for a ring of bubbles transverse to and centered on the beam forming an average hbt radius . thus our model was a ring of bubbles sitting on the freezeout surface with average size @xmath72 fm radius perpendicular to the beam at mid - rapidity . the phase space focusing would also lead to angular correlations between particles emitted by the bubbles and be observable . 1 shows the geometry of the bubbles . for the background particles that account for particles in addition to the bubble particles we used unquenched hijing@xcite with its jets removed . we assumed that most of the jets were eliminated from the central events because of the strong jet quenching observed at rhic@xcite . we investigated the feasibility of using charged - particle - pair correlations as a function of angles which should be observable due to the phase space focusing of the particles coming from individual bubbles . at that point we fully expected that these correlations would be observable in the star detector@xcite at rhic if one analyzed central au au collisions . correlation analyses are powerful tools in detecting substructures . historically substructures have played an important role in advancing our understanding of strong ( non perturbative ) interactions . our earlier paper explained the general characteristics of the angular correlation data , and was also consistent with hbt measurements in a qualitative manner . this motivated us to develop a reasonably quantitative model , the parton bubble model@xcite which is discussed in the following . in this publication@xcite we developed a qcd inspired parton bubble model ( pbm ) for central ( impact parameter near zero ) high energy heavy ion collisions at rhic . the pbm is based on a substructure consisting of a single ring of a dozen adjoining 2-fm - radius bubbles ( gluonic hot spots ) transverse to the collider beam direction , centered on the beam , and located at or near mid - rapidity . the at or near mid - rapidity refers to the boost invariant region that exist in the rhic heavy ion collisions which spans 2 - 4 units of rapidity . because of this spread in @xmath8 , we will use the relative spread of particles in @xmath8 ( @xmath9 ) in order to capture the angular correlations along the beam axis . the ring resides on the fireball blast wave surface ( see fig . 1 ) . we assumed these bubbles are likely the final state result of quark - gluon - plasma ( qgp ) formation since the energy densities produced experimentally are greater than those estimated as necessary for formation of a quark - gluon - plasma . thus this is the geometry for the final state kinetic freezeout of the qgp bubbles on the surface of the expanding fireball treated in a blast wave model . the average behavior of emitted final state particles coming from the surface bubbles at kinetic freezeout is given by the ring of twelve bubbles formed by energy density fluctuations near the surface of the expanding fireball of the blast wave . one should note that the blast wave surface is moving at its maximum velocity at freezeout ( 3c/4 ) . for central events each of the twelve bubbles have 3 - 4 partons per bubble each at a fixed @xmath10 for a given bubble . this number of partons was determined from correlation data analyzed by the star experiment@xcite . the transverse momentum ( @xmath2 ) distribution of the charged particles is similar to pqcd but has a suppression at high @xmath2 like the data . the bubble ring radius of our model was estimated by blast wave , hbt and other general considerations to be approximately 8 fm . the bubbles emit correlated charged particles at final - state kinetic freezeout where we select a @xmath2 range ( 0.8 gev / c @xmath1 @xmath2 @xmath1 4.0 gev / c ) in order to increase signal to background . the 0.8 gev / c @xmath2 cut increases the resolution to allow resolving individual bubbles which have a radius of @xmath72 fm . this space momentum correlation of the blast wave provides us with a strong angular correlation signal . pythia fragmentation functions@xcite were used for the bubbles fragmentation that generate the final state particles emitted from the bubbles . a single parton using pythia forms a jet with the parton having a fixed @xmath8 and @xmath10 ( see fig . the 3 - 4 partons in the bubble which shower using pythia all have a different @xmath8 value but all have the same @xmath10 ( see fig . the pbm explained the high precision au au central ( 0 - 10% ) collisions at @xmath11 200 gev@xcite ( the highest rhic energy ) . we utilize a two particle correlation function in the two dimensional ( 2-d ) space of @xmath12 vs @xmath9 . the azimuthal angle @xmath10 of a particle is defined by the angle of the particle with respect to the vertical axis which is perpendicular to the beam axis and is measured in a clock - wise direction about the beam . @xmath12 is the difference , @xmath13 - @xmath14 , of the @xmath10 angle of a pair of particles ( 1 and 2 ) . the pseudo - rapidity @xmath8 of a particle is measured along one of the beam directions . @xmath9 is the difference , @xmath15 - @xmath16 , of the @xmath8 values of a pair of particles ( 1 and 2 ) . the two dimensional ( 2-d ) total correlation function is defined as : @xmath17 where s(@xmath18 ) is the number of pairs at the corresponding values of @xmath18 coming from the same event , after we have summed over all the events . m(@xmath18 ) is the number of pairs at the corresponding values of @xmath18 coming from the mixed events , after we have summed over all our created mixed events . a mixed event pair has each of the two particles chosen from a different event . we make on the order of ten times the number of mixed events as real events . we rescale the number of pairs in the mixed events to be equal to the number of pairs in the real events . this procedure implies a binning in order to deal with finite statistics . to enhance comparison we use the same binning in our simulations as is used by the star high precision experimental analyses@xcite . the division by m(@xmath18 ) for experimental data essentially removes or drastically reduces acceptance and instrumental effects . if the mixed pair distribution was the same as the real pair distribution c(@xmath18 ) would be one for all values of the binned @xmath18 . in the correlations used in this paper we select particles independent of its charge . the correlation of this type is called a charge independent ( ci ) correlation . this difference correlation function has the defined property that it only depends on the differences of the azimuthal angle ( @xmath12 ) and the beam angle ( @xmath9 ) for the two particle pair . thus the two dimensional difference correlation distribution for each bubble which is part of c(@xmath18 ) is similar for each of our 12 bubbles and will image on top of each other . the pbm fit to the angular correlation data was reasonably quantitative to within a few percent ( see fig . 4 and fig . 5 ) ) . the correlation functions we employed ( like the hbt correlation functions ) have the property that for the difference in angles ( difference of momentum for hbt ) these correlations will image all 12 bubbles on top of each other . this leads to average observed angles of approximately @xmath19 in @xmath12 , @xmath20 in @xmath9 , originating from a source size of about @xmath72 fm radius which is consistent with the hbt correlation . thus the pbm generates @xmath12 vs @xmath9 charged - particle - pair correlations for charged particles with @xmath2 in the range 0.8 gev / c to 4.0 gev / c as displayed in fig . 4 and fig . we show the ci correlation in fig . 5 a reasonably quantitative successful comparison with data . it should be noted that the spread in @xmath9 of the bubble is determined by the longitudinal momentum distribution of the 3 - 4 partons that make up the bubble . this distribution which is quassian in nature was adjusted to the observed @xmath9 width of correlation data analyzed by the star experiment@xcite . furthermore the model results were consistent with the hanbury - brown and twiss ( hbt ) observations@xcite that the observed source radii determined by quantum statistical interference were reducing by a considerable factor with increasing transverse momentum ( @xmath2 ) . the hbt radii are interpreted to reduce from @xmath76 fm at @xmath2 @xmath70.2 gev / c to @xmath72 fm at @xmath2 @xmath71 gev / c for our @xmath2 range . the generally accepted explanation for this behavior is that as @xmath2 increases radial flow increasingly focused the viewed region of the final state into smaller volumes . if just one small hbt size bubble were emitting all the correlated particles , this phenomena would lead to large spikes of particles emitted at one limited @xmath10 angular region in individual events . this is definitely not observed in the au au or other collision data at rhic . therefore , a distributed ring of small sources around the beam as assumed in the pbm is necessary to explain both the hbt results@xcite and the correlation data@xcite . the particles emitted from the same bubble are virtually uncorrelated to particles emitted from any other bubble except for momentum conservation requirements . an away side peak in the total correlation is built up from momentum conservation between the bubbles . the pbm was also recently extended to pbme@xcite which is identical to the pbm for central collisions ( 0 - 5% ) . for centralities running from 30 - 80% jet quenching is not strong enough to make jets negligible therefore , a jet component was added which was based on hijing calculations . this jet component accounts for more of the correlation as one moves toward peripheral bins , and explains all of it for the most peripheral collisions . the pbme explained in a reasonably quantitative manner ( within a few percent ) the behavior of the recent quantitative experimental analysis of charge pair correlations as a function of centrality@xcite . this further strengthened the substantial evidence for bubble substructure . the agreement of the pbm and the pbme surface emission models with experimental analyses strongly implied that at kinetic freezeout the fireball was dense and opaque in the central region and most centralities ( except the peripheral region ) in the intermediate transverse momentum region ( 0.8 @xmath1 @xmath2 @xmath1 4.0 gev / c ) . thus we conclude that the observed correlated particles are both formed and emitted from or near the surface of the fireball . in the peripheral region the path to the surface is always small . the ci correlation used in fig . 4 and the pbm@xcite is also defined as the sum of two charged - particle - pair correlations . they are the unlike - sign charge pairs ( us ) and the like - sign charge pairs ( ls ) . this charge independent correlation ( ci ) is defined as the correlation made up of charged - particle - pairs independent of what the sign is , which is the average of the us plus the ls correlations . thus ci = ( us + ls)/2 . therefore the ci is the total charge pair correlation observed in the experimental detection system within its acceptance . the ci was generated for @xmath0 = 200 gev central ( 0 - 5% ) au au collisions in the @xmath2 range 0.8 gev / c to 4.0 gev / c at rhic . when the ci was compared to the corresponding experimental analyses@xcite a reasonably quantitative agreement within a few percent of the observed ci was attained . see ref.@xcite sec . 4.2 , ref.@xcite sec . v , and ref.@xcite sec . vi b. this analysis and our previous work@xcite was done without using a jet trigger . in fact jet production in the central region was negligible due to strong jet quenching@xcite . as one can see in fig . 4 the central ci correlation of @xmath12 on the near side ( @xmath12 @xmath1 @xmath21 ) , is sharp and approximately jetlike at all @xmath9 . in contrast the @xmath9 dependence is relatively flat . in the central production the average observed angles are approximately @xmath19 in @xmath12 and @xmath20 in @xmath9 . it is of interest to note that the sharp jetlike collimation in @xmath12 persists at all centralities ( 0 - 80% ) . the elongation in @xmath9 persists in the 0 - 30% centrality range and then decreases with decreasing centrality becoming jetlike in the peripheral bins . these characteristics are in reasonable quantitative agreement with the fits of the pbm and the pbme to experimental data analyses@xcite . the difference of the us and ls correlations is defined as the charge dependent ( cd ) correlation ( cd = us - ls ) . the 2-d experimental cd correlation for centralities ( 0 - 80% ) has a jetlike shape which is consistent with pythia jets ( vacuum fragmentation ) for all centralities . this clearly implies that both particle hadronization and emission occur from the fireball surface region as explained in ref.@xcite sec . a glasma flux tube model ( gftm)@xcite that had been developed considers that the wavefunctions of the incoming projectiles , form sheets of color glass condensates ( cgc)@xcite that at high energies collide , interact , and evolve into high intensity color electric and magnetic fields . this collection of primordial fields is the glasma@xcite , and initially it is composed of only rapidity independent longitudinal color electric and magnetic fields . an essential feature of the glasma is that the fields are localized in the transverse space of the collision zone with a size of 1/@xmath22 . @xmath22 is the saturation momentum of partons in the nuclear wavefunction . these longitudinal color electric and magnetic fields generate topological chern - simons charge@xcite which becomes a source for particle production . the transverse space is filled with flux tubes of large longitudinal extent but small transverse size @xmath7@xmath23 . particle production from a flux tube is a poisson process , since the flux tube is a coherent state . as the partons emitted from these flux tubes locally equilibrate , transverse flow builds due to the radial flow of the blast wave@xcite . the flux tubes that are near the surface of the fireball get the largest radial flow and are emitted from the surface . as in the parton bubble model these partons shower and the higher @xmath2 particles escape the surface and do not interact . these flux tubes are in the present paper considered strongly connected to the bubbles of the pbm . the method used to connect the pbm and the gftm is described and discussed in the next section v. @xmath22 is around 1 gev / c thus the transverse size of the flux tube is about 1/4 fm . the flux tubes near the surface are initially at a radius @xmath75 fm . the @xmath10 angle wedge of the flux tube is @xmath71/20 radians or @xmath7@xmath24 . thus the flux tube initially has a narrow range in @xmath10 . the large width in the @xmath9 correlation which in the pbm depended on the large spread in @xmath9 of the bubble partons results from the independent longitudinal color electric and magnetic fields that created the glasma flux tubes . how much of these longitudinal color electric and magnetic fields are still present in the surface flux tubes when they have been pushed by the blast wave will be a speculation of this paper for measuring strong cp violation ? it has been noted that significant features of the pbm that generated final state correlations which fit the experimentally observed correlation data@xcite are similar to those predicted by the gftm@xcite . therefore in the immediately following section we assume a direct connection of the pbm and the gtm , give reasons to justify it , and then discuss its consequences , predictions and successes . the successes of the pbm have strongly implied that the final state surface region bubbles of the pbm represent a significant substructure . in this subsection we show that the characteristics of our pbm originally developed to fit the precision star au au correlation data in a manner consistent with the hbt data ; implies that the bubble substructure we originally used to fit these previous data is closely related to the gftm . a natural way to connect the pbm bubbles to the gftm flux tubes is to assume that the final state at kinetic freezeout of a flux tube is a pbm final state bubble . thus the initial transverse size of a flux tube @xmath71/4 fm has expanded to the size of @xmath72 fm at kinetic freezeout . with this assumption we find consistency with the theoretical expectations of the gftm . we can generate and explain the triggered ridge phenomenon and data ( see sec . vi ) , thus implying the ridge is connected to the bubble substructure . we can predict and obtain very strong evidence for the color electric field of the glasma from comparing our multi - particle charged particle correlation predictions , and existing experimental correlation publications ( see sec . vii ) . we have also predicted correlations which can be used to search for evidence for the glasma color magnetic field ( sec . of course an obvious question that arises is that since a flux tube is an isolated system does a pbm bubble , that we assume is the final state of a flux tube at kinetic freezeout , also act as an isolated system when emitting the final state particles ? the near side correlations signals since the original pbm@xcite have always come virtually entirely from particles emitted from the same bubble . thus each bubble has always acted as an isolated system similar to the behavior of a flux tube . in heavy ion collisions at rhic there has been observed a phenomenon called the ridge which has many different explanations@xcite . the ridge is a long range charged particle correlation in @xmath9 ( very flat ) , while the @xmath12 correlation is approximately jet - like ( a narrow gaussian ) . there also appears with the ridge a jet - like charged - particle - pair correlation which is symmetric in @xmath9 and @xmath12 such that the peak on the jet - like correlation is at @xmath9 = 0 and @xmath12 = 0 . the @xmath12 correlation of the jet and the ridge are approximately the same and smoothly blend into each other . the ridge correlation is generated when one triggers on an intermediate @xmath2 range charged particle and then forms pairs between that trigger particle and each of all other intermediate charged particles with a smaller @xmath2 down to some lower limit . the first case we will study in this paper is a trigger charged particle between 3.0 to 4.0 gev / c correlated with all other charged particles which have a @xmath2 between 1.1 gev / c to 3.0 gev / c . in this paper we will investigate whether the pbm can account for the ridge once we add a jet trigger to our pbm generator@xcite . however this trigger will also select jets which previously could be neglected because there was such strong quenching@xcite of jets in central collisions . a jet trigger had not been used in the pbm comparison to all previous data . we use hijing@xcite merely to determine the expected number of jets to add for our added jet trigger . these jet particles were added to our pbm generator . thus our pbm generator now had hijing generated background particles , bubbles of the pbm and added jet particles from hijing . we have already shown that our final state particles come from hadrons at or near the fireball surface . we reduce the number of jets by 80% which corresponds to the estimate that only the parton interactions on or near the surface are not quenched away , and thus at kinetic freezeout form and emit hadrons which enter the detector . this 80% reduction is consistent with single @xmath25 suppression observed in ref.@xcite . we find for the reduced hijing jets that 4% of the au au central events ( 0 - 5% ) centrality at @xmath11 200 have a charged particle with a @xmath2 between 3.0 and 4.0 gev / c with at least one other charged particle with its @xmath2 greater than 1.1 gev / c coming from the same jet . the addition of the jets to the pbm generator provides the appropriate particles which are picked up by the trigger in order to form a narrow @xmath9 correlation signal at 0 which is also a narrow signal in @xmath12 at 0 ( fig . this narrow jet signal is present in the data and is what remains of jets after 80% are quenched away . we then form two - charged - particle correlations between one - charged - particle with a @xmath2 between 3.0 to 4.0 gev / c and another charged particle whose @xmath2 is greater than 1.1 gev / c . the results of these correlations are shown in fig . 6 is the ci correlation for the 0 - 5% centrality bin with the just above stated @xmath2 selections on charge pairs . since we know in our monte carlo which particles are emitted from bubbles and thus form the ridge , we can predict the shape of the ridge for the above @xmath2 cut by plotting only the correlation formed from pairs of particles that are emitted by the same bubble ( see fig . the charge pair correlations are virtually all emitted from the same bubble . those charge pair correlations formed from particles originating from different bubbles are small contributors which mainly have some effect on the away side correlation . 7 is the ridge signal which is the piece of the ci correlation for the 0 - 5% centrality of fig . 6 , after removing all other pairs except the pairs emitted from the same bubble . a very important aspect of the gftm is the boost that the flux tubes get from the radial flow of the blast wave . this boost is the same all along the flux tube and only depends on how far away from the center axis of the blast wave the flux tube is . in the pbm the partons in the bubble received a boost in @xmath2 from the radial flow field in the blast wave ; that after final state fragmentation of the bubbles gave the generated particles a consistent @xmath2 spectrum with the data . since the boost from the blast wave depended on the position of the bubble in the radial flow field , there should be a correlation between partons @xmath2 within a bubble . if one redistributed the partons with their @xmath2 boosts uniformly among the bubbles , the over all results for the generated correlations without a trigger would be unchanged . one should note none of our prior work ( i.e.pbm@xcite or pbme@xcite ) contained a trigger . we have only added a trigger here to treat the `` triggered ridge '' which obviously requires it . in our treatment of the gftm ( sec . vii ) we have removed the trigger . once we require a trigger demanding higher @xmath2 particles , we start picking out bubbles which have more radial flow ( harder particles ) . thus correlated particles which pass our @xmath2 selection and trigger come almost entirely from the harder bubbles with more radial flow , while the softer bubbles which were subject to less radial flow mainly generate low enough @xmath2 particles that become background particles or are lost to the correlation analysis due to @xmath2 selection . figure 6 shows the result of our trigger and @xmath2 selection . if one redistributed the partons with their @xmath2 radial flow boost among all the bubbles , then all bubbles become equal and there is no longer soft and hard bubbles . with this change we can generate bubble events . we find that both the non - triggered correlation and the @xmath12 correlation are virtually unchanged in the region less than about @xmath26 ( see fig . it seems that we move from the situation where we have a few bubbles with a lot of correlated particles above our @xmath2 cut to a lot of bubbles that have few correlated particles above our cuts . however there is a difference in the triggered correlation s away side or @xmath12 near the @xmath27 peak . in ref.@xcite we discuss this away side effect and attributed this peak to momentum conservation . this is however not the whole story . the away side peak is generated by the geometry of the bubble ring which requires that for every bubble there is another bubble nearly @xmath27 away that is emitting particles . these symmetry requirements on the geometry contribute more to the away side peak especially for the case without a trigger or redistributed random partons . this geometry effect is what causes the away side correlation of elliptic flow in a peripheral heavy ion collision . in such a collision there is a higher energy density aligned with the reaction plane . on one side of the reaction plane there are more particles produced because of increased energy density , and because of the geometry of the situation . on the other side of the reaction plane there are also more particles produced for the same reasons . thus there is a correlation between particles that are near each other on one side of the reaction plane and a correlation between particles that are on both sides of the reaction plane . momentum conservation requires particle communication by bouncing into each other , while geometry has no such communication . the right hand does not need to know what the left hand is doing . however , if physics symmetry makes them do the same thing then they can show a correlation . the away side peak depends on the fragmentation of the away side bubble . for the triggered case where there is a hard bubble with a strong correlation of parton @xmath2 inside the bubble ( gftm like ) , we will have more correlated particles adding to the signal on the near side . on the other hand the away side bubble will be on average a softer bubble which will have a lot less particles passing the @xmath2 cuts and thus have a smaller signal . if we consider the case of a random parton @xmath2 bubble the particles passing the @xmath2 cuts should be very similar on the near and the away side . 8 we compare the two correlations generated by the correlated and the random cases . we compare the @xmath12 correlation for the two cases ( solid is correlated and open is random)for each of the five @xmath9 bins which cover the entire @xmath9 range ( 0.0 to 1.5 ) . the vertical correlation scale is not offset and is correct for the largest @xmath9 bin , 1.2 @xmath28 1.5 , which is the lowest bin on the figure . as one proceeds upwards to the next @xmath9 bin the correlation is offset by + 0.05 . this is added to the correlation of each subsequent @xmath9 bin . the smallest @xmath9 bin on top of fig . 8 has a + 0.2 offset . a solid straight horizontal line shows the offset for each @xmath9 bin . each solid straight horizontal line is at 1.0 in correlation strength . we see that the away side correlation in the @xmath12 region greater than about @xmath26 is larger for the random case . particles which come from bubbles that satisfy the trigger ( 3.0 @xmath1 @xmath2 @xmath1 4.0 gev / c ) are harder and have a different @xmath2 distribution compared to the case without trigger requirements . in a given triggered event if one could select particles that come from the bubble or ridge , one would find that the @xmath2 spectrum is harder than the average @xmath2 spectrum . in fig . 9 we show how one can by trigger choose particles that are rich with ridge particles . we consider a typical trigger particle and the associated regions in the available experimental @xmath10 and @xmath8 ranges . the ridge particles we consider are separated from the trigger particle by 0.7 in @xmath8 . this is done to eliminate particles coming from jet production that could also be associated with the trigger particle . the bulk of the ridge particles lie within @xmath29 of the @xmath10 of the trigger particle . the width of the @xmath10 spread is @xmath12 = @xmath30 . therefore on a trigger by trigger basis we accumulate particles from the ridge region ( see fig . 9 ) and form a @xmath2 spectrum of charged particles . we then form a ratio of the @xmath2 spectrum for the ridge region particles divided by the @xmath2 spectrum for all particles in central au au events . this ratio is highly dependent on our ridge cut area . we can virtually remove this dependency by normalizing the two @xmath2 spectra to have the same counts for the 2.0 gev / c bin . figure 10 is this ratio which can be considered a correlation of particles as a function of @xmath2 in the ridge region compared to particles from the event in general . in fig . 10 we form this ratio or correlation for the bubbles which have a correlated @xmath2 among the partons ( gftm like ) . we also show the correlation for bubbles which have a random @xmath2 distribution among partons in the bubble . for the random case the @xmath2 spectrum does not differ much from the average @xmath2 spectrum while there is a big difference for the flux tube like case . this correlation could be easily measured in the rhic data . triggered angular correlation data showing the ridge was presented at quark matter 2006@xcite . figure 11 shows the experimental @xmath12 vs. @xmath9 ci correlation for the 0 - 10% central au au collisions at @xmath11 200 gev ; requiring one trigger particle @xmath2 between 3 and 4 gev / c and an associated particle @xmath2 above 2.0 gev / c . the yield is corrected for the finite @xmath9 pair acceptance . for the pbm generator , we then form a two - charged - particle correlation between one charged particle with a @xmath2 between 3.0 to 4.0 gev / c and another charged particle whose @xmath2 is greater than 2.0 gev / c . the results of this correlation is shown in fig . figure 11 shows the corrected pair yield determined in the central data whereas fig . 12 shows the correlation function generated by the pbm which does not depend on the number of events analyzed . we can compare the two figures , if we realize that the away side ridge has around 420,000 pairs in fig . 11 while in fig . 12 the away side ridge has a correlation of around 0.995 . if we multiply the correlation scale of fig . 12 by 422,111 in order to achieve the number of pairs seen in fig . 11 , the away side ridge would be at 420,000 and the peak would be at 465,000 . this would make a good agreement between the two figures . the correlation formed by the ridge particles is generated almost entirely by particles emitted by the same bubble . we have shown in all our publications that the same side correlation signals are almost entirely formed by particles coming from the same bubble . thus we can predict the shape and the yield of the ridge for the above @xmath2 trigger selection and lower cut , by plotting only the correlation coming from pairs of particles that are emitted by the same bubble ( see fig . 13 ) . in ref.@xcite it was assumed that the ridge yield was flat across the acceptance while in fig . 13 we see that this is not the case . therefore our ridge yield is approximately 35% larger than estimated in ref.@xcite . finally we can plot the jet yield that we had put into our monte carlo . we used hijing@xcite to determine the number of expected jets , and then reduced the number of jets by 80% . this assumes that only the parton interactions on or near the surface that form hadrons at kinetic freezeout are not quenched away and thus enter the detector . this 80% reduction is consistent with single @xmath25 suppression observed in ref.@xcite . this jet yield is plotted in fig . 14 where we subtracted contributions from the bubbles and the background particles from fig . 11 . the strong cp problem remains one of the most outstanding puzzles of the standard model . even though several possible solutions have been put forward it is not clear why cp invariance is respected by the strong interaction . it was shown however through a theorem by vafa - witten@xcite that the true ground state of qcd can not break cp . the part of the qcd lagrangian that breaks cp is related to the gluon - gluon interaction term @xmath3 . the part of @xmath3 related to cp violations can be separated into a separate term which then can be varied by multiplying this term by a parameter called @xmath31 . for the true qcd ground state @xmath31 = 0 ( vafa - witten theorem ) . in the vicinity of the deconfined qcd vacuum metastable domains@xcite @xmath31 non - zero could exist and not contradict the vafa - witten theorem since these are not the qcd ground state . the metastable domains cp phenomenon would manifest itself in specific correlations of pion momenta@xcite . the glasma flux tube model ( gftm)@xcite considers the wavefunctions of the incoming projectiles , form sheets of cgc@xcite at high energies that collide , interact , and evolve into high intensity color electric and magnetic fields . this collection of primordial fields is the glasma@xcite . initially the glasma is composed of only rapidity independent longitudinal ( along the beam axis ) color electric and magnetic fields . these longitudinal color electric and magnetic fields generate topological chern - simons charge@xcite through the @xmath3 term and becomes a source of cp violation . how much of these longitudinal color electric and magnetic fields are still present in the surface flux tubes when they have been pushed by the blast wave is a speculation of this paper for measuring strong cp violation ? the color electric field which points along the flux tube axis causes an up quark to be accelerated in one direction along the beam axis , while the anti - up quark is accelerated in the other direction . so when a pair of quarks and anti - quarks are formed they separate along the beam axis leading to a separated @xmath5 @xmath6 pair along this axis . the color magnetic field which also points along the flux tube axis ( which is parallel to the beam axis ) causes an up quark to rotate around the flux tube axis in one direction , while the anti - up quark is rotated in the other direction . so when a pair of quarks and anti - quarks are formed they will pickup or lose transverse momentum after the bubble is boosted radial outwards due to radial flow of the blast wave . these changes in @xmath2 will be transmitted to the @xmath5 @xmath6 pairs . _ it is important to note that the cp violating asymmetries in @xmath5 and @xmath6 momenta arise through the witten - wess - zumino@xcite _ term . the quarks and anti quarks which , later form the @xmath5 and @xmath6 , directly respond to the color electric and color magnetic fields and receive their boosts at the quark and anti quark level before they are pions . these boosts are transmitted to the pions at hadronization . thus no external magnetic field is required in the methodology followed in this paper . the following references demonstrate this:@xcite . _ _ to represent the color electric field effect we assume as the first step for the results being shown in this paper ; that we generate bubbles which have an added boost of 100 mev / c to the quarks in the longitudinal momentum which represents the color electric effect . the @xmath5 and the @xmath6 which form a pair are boosted in opposite directions along the beam axis . in a given bubble all the boosts are the same but vary in direction from bubble to bubble . for the color magnetic field effect , we give 100 mev / c boosts to the transverse momentum in an opposite way to the @xmath5 and the @xmath6 which form a pair . for each pair on one side of a bubble one of the charged pions boost is increased and the other is decreased . while on the other side of the bubble the pion for which the boost was increased is now decreased and the pion which was decreased is now increased by the 100 mev / c . all pairs for a given bubble are treated in the same way however each bubble is random on the sign of the pion which is chosen to be boosted on a given side . this addition to our model is used in the simulations of the following subsections . above we saw that pairs of positive and negative pions should show a charge separation along the beam axis due to a boost in longitudinal momentum caused by the color electric field . a measure of this separation should be a difference in the pseudorapidity ( @xmath9 ) of the opposite sign pairs . this @xmath9 measure has a well defined sign since we defined this difference measuring from the @xmath6 to the @xmath5 . in order to form a correlation we must pick two pairs for comparison . the pairs have to come from the same bubble ( a final state of an expanded flux tube ) since we have shown by investigating the events generated by the model that pairs originating from different bubbles will not show this correlation . therefore we require the @xmath5 and the @xmath6 differ by @xmath29 or less in @xmath10 . let us call the first pair @xmath32 . the next pair ( @xmath33 having the same @xmath10 requirement ) has to also lie inside the same bubble to show this correlation . thus we require that there is only @xmath34 between the average @xmath10 of the pairs . this implies that at most in @xmath10 no two pions from a given bubble can differ by more than @xmath19 . in fig . 15 we show two pairs which would fall into the above cuts . @xmath32 and @xmath33 are positive in fig . however if we would interchange the @xmath5 and @xmath6 on either pair the value of its @xmath9 would change sign . finally the mean value shown on fig . 15 is the mid - point between the @xmath5 and the @xmath6 where one really uses the vector sum of the @xmath5 and the @xmath6 which moves this point toward the harder pion . considering the above cuts we defined a correlation function where we combine pairs each having a @xmath9 . our variable is related to the sum of the absolute values of the individual @xmath9 s ( @xmath35 ) . we assign a sign to this sum such that if the sign of the individual @xmath9 s are the same it is a plus sign , while if they are different it is a minus sign . for the flux tube the color electric field extends over a large pseudorapidity range therefore let us consider the separation of pairs @xmath36 greater than 0.9 . for the numerator of the correlation function we consider all combinations of unique pairs ( sign ( @xmath35 ) ) from a given central au au event divided by a mixed event denominator created from pairs in different events . we determine the rescale of the mixed event denominator by considering the number of pairs of pairs for the case @xmath37 lying between @xmath38 and @xmath39 for events and mixed events so that the overall ratio of this sample numerator to denominator is 1 . by picking @xmath38 @xmath1 @xmath37 @xmath1 @xmath39 we make sure we are not choosing pairs from the same bubble . for a simpler notation let ( sign ( @xmath35 ) ) = @xmath40 which varies from -4 to + 4 since we have an over all @xmath8 acceptance -1 to + 1 ( for the star tpc detector for which we calculated ) . the value being near @xmath41 4 can happen when one has a hard pion with @xmath2 of 4 gev / c ( upper cut ) at @xmath8 = 1 with a soft pion @xmath2 of 0.8 gev / c ( lower cut ) at @xmath8 = -1 combined with another pair ; a hard pion with @xmath2 of 4 gev / c at @xmath8 = -1 with a soft pion with @xmath2 of 0.8 gev / c at @xmath8 = 1 . in fig . we show the correlation function of opposite sign charge - particle - pairs paired and binned by the variable @xmath40 with a cut @xmath36 greater than 0.9 between the vector sums of the two pairs . the events are generated by the pbm@xcite and are charged particles of 0.8 @xmath1 @xmath2 @xmath1 4.0 gev / c , and @xmath42 @xmath1 1 , from au au collisions at @xmath11 200 gev . since we select pairs of pairs which are near each other in @xmath10 they all together pick up the bubble correlation and thus these pairs of pairs over all show about a 4% correlation . in the 1.0 @xmath1 @xmath40 @xmath1 2.0 region the correlation is 0.5% larger than the -2.0 @xmath1 @xmath40 @xmath1 -1.0 region . this means there are more pairs of pairs aligned in the same direction compared to pairs of pairs not aligned . this alignment is what is predicted by the color electric field effect presented above . in fact if one has plus minus pairs all aligned in the same direction and spread across a pseudorapidity range , locally at any place in the pseudorapidity range one would observe an increase of unlike sign charge pairs compared to like sign charge pairs . at small @xmath9 unlike sign charge pairs are much larger than like sign charge pairs in both the pbm and the data which agree . 10 , 11 , and 14 of ref.@xcite . figure 11 of ref.@xcite compares the total correlation for unlike - sign charge pairs and like - sign charge pairs in the precision star central production experiment for au au central collisions ( 0 - 10% centrality ) at @xmath0=200 gev , in the transverse momentum range 0.8 @xmath1 @xmath2 @xmath1 2.0 gev / c@xcite . the unlike - sign charge pairs are clearly larger in the region near @xmath12 = @xmath9 = 0.0 . the increased correlation of the unlike - sign pairs is 0.8% @xmath41 0.002% . figure 10 of ref.@xcite shows that the pbm fit to these data gives the same results . figure 14 of ref.@xcite shows that the cd = unlike - sign charge pairs minus like - sign charge pairs is positive for the experimental analysis . therefore the unlike - sign charge pairs are considerably larger than the like - sign charge pairs . in fact this effect is so large and the alignment is so great that when one adds the unlike and like sign charge pairs correlations together there is still a dip at small @xmath9 and @xmath12 see fig . 4 and fig . 7 of the present paper . the statistical significance of this dip in the two high precision experiments done independently from different data sets gathered 2 years apart@xcite is huge . it would require a fluctuation of @xmath43 to remove the dip . this dip is also not due to any systematic error since both of the just cited precision experiments carefully investigated that possibility ; and found no evidence to challenge the reality of this dip . this highly significant dip ( @xmath7@xmath43 ) means that like - sign pairs are removed as one approaches the region @xmath9 = @xmath12 = 0.0 at a rate much larger than regular jet fragmentation . jet fragmentation does not have a dip in its ci correlation thus this is very strong evidence for the predicted effect of the color electric field . after considering the color electric effect we turn to the color magnetic effect which causes up quarks to rotate around the flux tube axis in one direction , while the anti - up quarks rotate in the other direction . so when a pair of quarks and anti - quarks are formed they will pickup or lose transverse momentum . these changes in @xmath2 will be transmitted to the @xmath5 and @xmath6 pairs . _ it was previously shown in sec . vii a that the cp violating asymmetries in the @xmath5 and the @xmath6 momenta arise through the witten - wess - zumino term . the quarks and anti quarks which , later form the @xmath5 and the @xmath6 , directly respond to the color electric and color magnetic fields and receive their boosts at the quark and anti quark level before they hadronize into pions . these boosts are transmitted to the @xmath5 and the @xmath6 . _ in order to observe these differential @xmath2 changes one must select pairs on one side of the bubble in @xmath10 and compare to other pairs on the other side of the same bubble which would lie around @xmath30 to @xmath44 away in @xmath10 . we defined a pair as a plus particle and minus particle with an opening angle ( @xmath31 ) of @xmath45 or less . we are also interested in pairs that are directly on the other side of the bubble . we require they are near in pseudorapidity ( @xmath9 @xmath1 0.2 ) . the above requirements constrain the four charged particles comprising both pairs to be contained in the same bubble and be close to being directly on opposite sides of the bubble ( a final state expanded flux tube ) . the difference in @xmath2 changes due to and predicted by the color magnetic effect should give the @xmath5 on one side of the bubble an increased @xmath2 and a decreased @xmath2 on the other side , while for the @xmath6 it will be the other way around . this will lead to an anti - alignment between pairs . in fig . 17 we show two pairs which would fall into the above cuts . both pairs are at the limit of the opening angle cut @xmath46 and @xmath47 equal @xmath45 . the @xmath2 of the plus particle for pair number 1 is 1.14 gev / c , while the minus particle is 1.39 gev / c . thus @xmath48 is equal to -0.25 gev / c . the @xmath2 of the plus particle for pair number 2 is 1.31 gev / c , the minus particle is 0.91 gev / c and @xmath49 is equal to 0.40 gev / c . finally the mean value shown on fig . 17 is the mid - point between the @xmath5 and the @xmath6 where one really uses the vector sum of the @xmath5 and the @xmath6 which moves this point toward the harder pion . considering the above cuts we defined a correlation function where we combine pairs each having a @xmath50 . our variable is related to the sum of the absolute values of the individual @xmath50 s ( @xmath51 ) . we assign a sign to this sum such that if the sign of the individual @xmath50 s are the same it is a plus sign , while if they are different it is a minus sign . for the flux tube the color magnetic field extends over a large pseudorapidity range where quarks and anti - quarks rotate around the flux tube axis , therefore we want to sample pairs at the different sides of the tube making a separation in @xmath10 ( @xmath12 ) between @xmath30 to @xmath44 . we are interested in sampling the pairs on the other side so we require the separation in @xmath8 ( @xmath9 ) be 0.2 or less . for the numerator of the correlation function we consider all combinations of unique pairs ( sign ( @xmath51 ) ) from a given central au au event divided by a mixed event denominator created from pairs in different events . we determine the rescale of the mixed event denominator by considering the number of pairs of pairs for the case @xmath36 lying between 1.2 and 1.5 plus any value of @xmath37 for events and mixed events so that the overall ratio of this sample numerator to denominator is 1 . by picking this @xmath9 bin for all @xmath37 we have around the same pair count as the signal cut with the @xmath12 correlation of the bubbles being washed out . for a simpler notation let ( sign ( @xmath51 ) ) = @xmath52 which we plot in the range from -4 to + 4 since we have an over all @xmath2 range 0.8 to 4.0 gev / c . thus the maximum magnitude of @xmath50 s is 3.2 gev / c which makes @xmath52 have a range of @xmath416.4 . however the larger values near these range limits occur very rarely . in fig . 18 we show the correlation function of opposite sign charged - particle - pairs paired and binned by the variable @xmath52 with a cut @xmath36 less than 0.2 between the vector sums of the two pairs , and with @xmath30 @xmath1 @xmath37 @xmath1 @xmath44 . the events are generated by the pbm@xcite and are charged particles of 0.8 @xmath1 @xmath2 @xmath1 4.0 gev / c , with @xmath42 @xmath53 1 , from au au collisions at @xmath11 200 gev . since we select pairs of pairs which are near each other in @xmath10 ( @xmath30 to @xmath44 ) they all together pick up the bubble correlation and thus these pairs of pairs over all show about a 0.4% correlation . in the -4.0 @xmath1 @xmath52 @xmath1 -1.0 region the correlation increases from 0.4% to 1% , while in the 1.0 @xmath1 @xmath52 @xmath1 4.0 region the correlation decreases from 0.4% to -0.2% . this means the pairs of pairs are anti - aligned at a higher rate than aligned . this anti - alignment is what is predicted by the color magnetic field effect presented above . in fact the anti - alignment increases with @xmath2 as the ratio bubble particle to background increases . however these predicted color magnetic effects have not been searched for yet and therefore there is no experimental evidence for them . experimental investigations of star tpc detector charged particle correlations provide a method of obtaining experimental confirmation of the anti - alignment caused by the color magnetic field . in this article we summarize the assumptions made and the reasoning that led to the development and construction of our parton bubble model ( pbm)@xcite , which successfully explained the charge - particle - pair correlations in the central ( 0 - 10% centrality ) @xmath11 200 gev au au data@xcite . the pbm is also consistent with the au au central collision hbt results . this is presented and discussed in sec . 4 of ref.@xcite and sec . i and ii of this paper . in sec . iii we discussed extending our model , which was a central collision model , to be able to treat the geomery of bubble production for 0 - 80% collision centralities ( pbme@xcite ) such as measured and analyzed in rhic data@xcite . the bubbles represent a significant substructure of gluonic hot spots formed on the surface of a dense opaque fireball at kinetic freezeout . the origins of these hot spots come from a direct connection between our parton bubble model ( pbm@xcite ) and the glasma flux tube model ( gftm)@xcite ) . in the gftm a flux tube is formed right after the initial collision of the au au system . this flux tube extends over many units of pseudorapidity ( @xmath9 ) . a blast wave gives the tubes near the surface transverse flow in the same way it gave flow to the bubbles in the pbm . this means the transverse momentum ( @xmath2 ) distribution of the flux tube is directly translated to the @xmath2 spectrum . in the pbm this flux tube is approximated by a sum of partons which are distributed over this same large @xmath9 region . initially the transverse space is filled with flux tubes of large longitudinal extent but small transverse size @xmath7@xmath23 . the flux tubes that are near the surface of the fireball get the largest radial flow and are emitted from the surface . as in the parton bubble model these partons shower and the higher @xmath2 particles escape the surface and do not interact . @xmath22 is around 1 gev / c thus the size of the flux tube is about 1/4 fm initially . the flux tubes near the surface are initially at a radius @xmath75 fm . the @xmath10 angle wedge of the flux tube @xmath7 1/20 radians or @xmath7@xmath24 . in sec . v ( also see sec . iv ) we connect the gftm to the pbm : by assuming the bubbles are the final state of a flux tube at kinetic freezeout , and discussing evidence for this connection which is further developed in sec . vi and sec . vii . with the connection of the pbm to the gftm two new predictions become possible for our pbm . the first is related to the fact that the blast wave radial flow given to the flux tube depends on where the tube is initially in the transverse plane of the colliding au au system . the tube gets the same radial boost all along its longitudinal length . this means that there is correlated @xmath2 among the partons of the bubble . in this paper we consider predictions we can make in regard to interesting topics and comparisons with relevant data which exist by utilizing the parton bubble model ( pbm@xcite ) , and its features related to the glasma flux tube model ( gftm)@xcite ) . topic 1 : the ridge is treated in sec . topic 2 : strong cp violation ( chern - simons topological charge ) is treated in sec . we show in sec . vi that if we trigger on particles with 3 to 4 gev / c and correlate this trigger particle with an other charged particle of greater than 1.1 gev / c , the pbm can produce a phenomenon very similar to the ridge@xcite . ( see figs . we then selected charged particles inside the ridge and predicted the correlation that one should observe when compared to the average charged particles of the central au au collisions at @xmath0 = 200 gev@xcite . in sec . vi d ( comparison to data ) : triggered experimental angular correlations showing the ridge were presented at quark matter 2006@xcite . figure 7 shows the experimental @xmath12 vs. @xmath9 ci correlation for 0 - 10% central au au collisions at @xmath0 = 200 ; requiring one trigger particle @xmath2 between 3 to 4 gev / c and an associated particle @xmath2 above 2.0 gev / c . the yield is corrected for the finite @xmath9 pair acceptance . for the pbm generator , we then form a two charged particle correlation between one charged particle with a @xmath2 between 3.0 to 4.0 gev / c and another charged particle whose @xmath2 is greater than 2.0 gev / c . these are the same trigger conditions as in ref.@xcite which is shown in fig . 11 , that shows the corrected pair yield in the central data . 12 shows the correlation function generated by the pbm which does not depend on the number of events analyzed . the two figures were shown to be in reasonable agreement when compared as explained previously in sec . vi d. in fig . 13 we show the ridge signal predicted by the pbm for very similar data but with 0 - 5% centrality . figure 14 shows the extraction of the jet signal . explanations are given in the text . the second prediction is a development of a predictive pionic measure of the strong cp violation . the gftm flux tubes are made up of longitudinal color electric and magnetic fields which generate topological chern - simons charge@xcite through the @xmath3 term that becomes a source of cp violation . the color electric field which points along the flux tube axis causes an up quark to be accelerated in one direction along the beam axis , while the anti - up quark is accelerated in the other direction . so when a pair of quarks and anti - quarks are formed they separate along the beam axis leading to a separated @xmath5 and @xmath6 pair along this axis . the color magnetic field which also points along the flux tube axis ( which is parallel to the beam axis ) causes an up quark to rotate around the flux tube axis in one direction , while the anti - up quark rotates in the other direction . so when a pair of quarks and anti - quarks are formed they will pickup or lose transverse momentum . these changes in @xmath2 will be transmitted to the @xmath5 and @xmath6 pairs . the above pionic measures of strong cp violation are used to form correlation functions based on four particles composed of two pairs which are opposite sign charge - particle - pairs that are paired and binned . these four particle correlations accumulate from bubble to bubble by particles that are pushed or pulled ( by the color electric field ) and rotated ( by the color magnetic field ) in a right or left handed direction . the longitudinal color electric field predicts aligned pairs in a pseudorapidity or @xmath9 measure . the longitudinal color magnetic field predicts anti - aligned pairs in a transverse momentum or @xmath50 measure . the observations of these correlations would be a strong confirmation of this theory . the much larger unlike - sign pairs than like - sign pairs in the pbm and the data ; and the strong dip of the ci correlation at small @xmath9 ( see fig . 4 and sec . vii c for full details ) shows very strong evidence supporting the color electric alignment prediction in the @xmath0 = 200 gev central au au collision data analyses at rhic@xcite . this highly significant dip ( @xmath7@xmath43 ) means that like - sign pairs are removed as one approaches the region @xmath9 = @xmath12 = 0.0 . thus this is very strong evidence for the predicted effect of the color electric field . the color magnetic anti - aligned pairs in the transverse momentum prediction , treated in sec . vii d as of now has not been observed or looked for . however our predicted specific four charged particle correlations can be used to search for experimental evidence for the color magnetic fields . our success in demonstrating strong experimental evidence for the expected color electric field effects from previously published data suggests that the unique detailed correlations we have presented for searching for evidence for the predicted color magnetic field effects should be urgently investigated . if we are lucky and the predicted color magnetic effects can be confirmed experimentally we would have strong evidence for the following : \1 ) cp is violated in the strong interaction in isolated local space time regions where topological charge@xcite is generated . \2 ) the glasma flux tube model ( gftm ) which was evolved from the color glass condensate ( cgc ) would be found to be consistent with a very significant experimental check . \3 ) the parton bubble model event generator ( pbm ) is clearly closely connected to the gftm . the bubble substructure strongly supported by the pbm is likely due to the final state of the flux tube at kinetic freezeout . this research was supported by the u.s . department of energy under contract no . de - ac02 - 98ch10886 . the author thank william love for valuable discussion and assistance in production of figures . finally this manuscript is in memory of my fellow collaborator prof . sam lindenbaum which help form many of the ideas of bubbles in heavy ion collisions . 99 l. van hove , z. phys . c 27 , 135 ( 1985 ) . lindenbaum , r.s . longacre , j. phys . g 26 , 937 ( 2000 ) . d. kharzeev , r.d . pisarski , phys . rev . d 61 , 111901 ( 2000 ) . c. adler _ et al . 87 , 082301 ( 2001 ) . lindenbaum , r.s . longacre m. kramer , eur . j. c. 30 , 241 ( 2003 ) . lindenbaum , r.s . longacre , eur . j. c. 49 , 767 - 782 ( 2007 ) . j. adams _ et al . _ , phys . rev . c. 71 , 044906 ( 2005 ) , s.s . adler _ et al . _ , phys rev . 93 , 152302 ( 2004 ) . x.n . wang and m. gyulassy , phys . d 44 , 3501 ( 1991 ) . k. adcox _ et al . 88 , 022301 ( 2001 ) . nucl . instrum . meth . a 499,(2003 ) , c. adler et al . , 433 - 436 , k.h . ackermann et al.(star collaboration ) , 624 - 632 , m. anderson et al . , 659 - 678 , f.s . bieser et al . , 766 - 777 . t. sjostrand , m. van zijil , phys . rev . d 36 , 2019 ( 1987 ) . j. adams _ et al . c 75 , 034901 ( 2007 ) . lindenbaum and r.s . longacre , phys . rev . c 78 , 054904 ( 2008 ) . abelev _ et al . _ , arxiv:0806.0513[nucl - ex ] . j. adams _ et al . 91 , 172302 ( 2003 ) . a. adare _ et al . lett . 101 , 232301 ( 2008 ) . a. dumitru , f. gelis , l. mclerran and r. venugopalan , nucl . a 810 , 91 ( 2008 ) . l. mclerran and r. venugopalan , phys . rev . d 49 , 2233 ( 1994 ) ; phys . d 49 , 3352 ( 1994 ) ; phys . d 50 , 2225 ( 1994 ) . t. lappi and l. mclerran , nucl . phys . a 772 , 200 ( 2006 ) . f. gelis and r. venugopalan , acta phys . b 37 , 3253 ( 2006 ) . d. kharzeev , a. krasnitz and r. venugopalan , phys . lett . b 545 , 298 ( 2002 ) . s. gavin , l. mclerran and g. moschelli , eur . j. c. 62 , 277 - 280 ( 2009 ) . n. armesto , c. salgado , u.a . wiedemann , phys rev . 93 , 242301 ( 2004 ) . p. romatschke , phys . c. 75 , 014901 ( 2007 ) . e. shuryak , phys . rev . c. 76 , 047901 a. dumitru , y. nara , b. schenke , m. strickland , phys . c 78 , 024909 ( 2008 ) . pantuev , arxiv:0710.1882[hep - ph ] . r. mizukawa , t. hirano , m. isse , y. nara , a. ohnishi , j. phys . g 35 , 104083 ( 2008 ) . wong , phys . rev . c. 78 , 064905 ( 2008 ) .. r.c . hwa , arxiv:0708.1508[nucl - th ] . j. putschke , j. phys . 34 , s679 ( 2007 ) . c. vafa , e. witten , phys . 53 , 535 ( 1984 ) , nucl . b 234 , 173 ( 1984 ) . d. kharzeev , r.d . pisarski , m.h.g . tytgat , phys . 81 , 512 ( 1998 ) . c. vafa , e. witten , nucl . phys . b 233 , 422 ( 1983 ) . j. wess , and b. zumino , phys . lett . 37b , 95 ( 1971 ) l.e . finch _ et al . c 65 , 014908 ( 2001 ) .

we have studied the effect of impurities on the critical temperature for a model of ybco involving pairing both in the cuo planes and in the cuo chains . in this model pairing in the planes is due to phonons , while coulomb repulsion induces in the chains an order parameter with opposite sign . due to the anticrossing produced by hybridization between planes and chains , the order parameter changes sign on a single sheet of the fermi surface resulting in nodes in the gap . we find that , in our model , the critical temperature is much less sensitive to impurities than in standard d - wave models . one reason is that impurities produce essentially plane - plane and chain - chain scattering , which does not affect the critical temperature . is reduced by the scattering between parts of the fermi surface which have opposite signs for the order parameter , just as in standard d - wave . in our model this is due to plane - chain scattering . we have found that this scattering , whatever its origin , will be smaller by a factor of order t ( that is hybridization coupling over fermi energy ) compared to plane - plane and chain - chain scattering . accordingly the sensitivity of to impurities in our model is reduced by a similar factor , compared to the d - wave situation . in the specific case which we have studied in details and which reduces to the two - band model , we have found a further reduction of the sensitivity of to impurities with a behaviour which can vary continuously from s - wave like to d - wave like depending on the parameters . we expect a similar behaviour and reduction to occur in the general case .

cond-mat9603131

the mechanism responsible for high @xmath1 superconductivity is still highly controversial @xcite . while a large part of the theoretical effort is based on the hypothesis that cooper pairs are formed in high @xmath1 compounds , there is no agreement on the physical origin of the pairing interaction nor on the symmetry of the pair wavefunction . these two questions are actually intimately related . indeed repulsive interactions , such as produced by spin fluctuations @xcite , require an order parameter which changes sign on the fermi surface in order to produce pairs . this leads , in the simplest hypothesis , to pairing with d - wave symmetry if we assume singlet pairing . on the other hand purely attractive interactions lead to pairing with s - wave symmetry since any change of sign is unfavorable in this case . therefore an experimental determination of the order parameter symmetry should greatly help to identify the physical interaction responsible for pair formation : although it would not be enough to provide a unique identification , it would strongly narrow the remaining possibilities . for this reason a large part of the recent experimental work has been aimed toward providing a clear signature for the symmetry of the order parameter . quite surprisingly recent experiments on @xmath3 designed with this purpose of identifying the symmetry have given clear cut , but contradictory answers @xcite . indeed the observation of a sizeable josephson current @xcite in a c - axis tunnelling junction between @xmath3 and pb is quite difficult to reconcile with a pure d - wave symmetry while it is in full agreement with s - wave symmetry . similarly the fact that there is no angular dependence in the critical current of ybco - ybco grain boundary junctions in the a - b plane @xcite goes clearly in the direction of an s - wave interpretation . on the other hand a number of experiments are in favor of a d - wave symmetry . many experiments , including tunnelling , nmr , raman scattering , photoemission , penetration depth @xcite , have shown the existence of low energy excited states . actually these experiments are compatible with a strongly anisotropic s - wave order parameter . or more simply one may look for extrinsic effects and wonder if these states do not arise from surface effects or defects , which would provide quasi - normal regions . however the existence of a linear t dependence of the penetration depth over a large range of temperature , simultaneously in cristals and films , with the same slope @xcite , makes an extrinsic interpretation for all these experiments unlikely ( while this kind of explanation may very well be valid for some of them ) . moreover some experiments specifically designed to check if the order parameter changes sign over the fermi surface have given positive answers . these are the corner squid experiments @xcite which give a clear indication for a change of sign of the order parameter between the a and the b axis , and the observation of a spontaneous magnetization corresponding to a half magnetic flux quantum in 3 grain - boundary josephson junctions @xcite which implies a @xmath4 shift , in clear agreement with d - wave symmetry . on the other hand the d - wave interpretation is not free of problems . for example recent experiments show that some thermodynamical superconducting properties are markedly anisotropic ( the a and b - axis results are different ) . indeed the penetration depth in good ybco cristals display a strong anisotropy of the penetration depth @xcite , the specific heat anomaly at @xmath1 of the parent compound @xmath5 ( with same @xmath1 as ybco ) has a marked anisotropy as a function of the orientation of an applied magnetic field @xcite . it is difficult to ascribe these anisotropies to the weak orthorhombic distorsion , and it is more likely that the cuo chains play a significant role in the superconducting properties . one of the most conspicuous problem of the d - wave interpretation is the weak sensitivity of the critical temperature of @xmath3 to the presence of impurities . indeed any kind of impurities , whether magnetic or not , produces in d - wave superconductors @xcite an effect analogous to pair - breaking by magnetic impurities in standard s - wave superconductors @xcite . in particular the critical temperature @xcite decreases rapidly with increasing impurity concentration following the abrikosov - gorkov law @xcite , and superconductivity disappears at a critical concentration . in contrast all samples of @xmath3 seem to have a @xmath1 around 90 k. it is difficult to believe that all samples ( including the earlier ones ) are clean enough to affect only weakly the critical temperature , whereas all microscopic studies show that there are always more structural defects than what is generally admitted . moreover ion @xcite or electron @xcite irradiation experiments have shown a rather weak sensitivity of the critical temperature of @xmath3 on the inverse lifetime deduced from resistivity measurements @xcite . actually zn impurities are known @xcite to have a depressing effect on @xmath6 but this can be interpreted as a standard pair - breaking effect since the environment of a zn impurity , located in the @xmath7 planes , is known to be magnetic @xcite . in order to solve these contradictions we have proposed recently @xcite for @xmath3 a model which mixes s - wave and d - wave features . in our model , in addition to the @xmath7 planes , the cuo chains play an essential role . the pairing interaction within the planes is attractive ( it can be for example produced by phonons ) . on the other hand the pairing interaction between planes and chains is repulsive ( it can be produced by coulomb interaction ) . in this way the order parameter has opposite signs on the planes and on the chains . moreover we include the hybridization between planes and chains , which corresponds physically to take into account the possibility for an electron to jump from planes to chains or vice - versa . naturally the coupling responsible for this hybridization is fairly small , but it is a well known feature of all band structure calculations @xcite . it is of importance only when the plane and the chain band intersect . in this case it leads to an anticrossing in the dispersion relations , and similarly to an anticrossing of the fermi surfaces , wherever the ( uncoupled ) pieces of the fermi surface related to plane and chain cross . as a result , when we move on a given sheet of the fermi surface , we go from a part which corresponds physically to a plane electron , to a part which corresponds physically to a chain electron . since the order parameter has opposite signs for plane and chain electrons , this implies that the order parameter changes sign on a given sheet of the fermi surface and therefore has nodes on this sheet by continuity . therefore our model provides an order parameter which is quite analogous to a d - wave order parameter and it can in this way explain @xcite all the experiments in favor of d - wave symmetry . on the other hand it does not have a d - wave symmetry since the nodes , which occur at the intersection between plane and chain bands , have no reason to satisfy @xmath8 = @xmath9 , and the average of the order parameter has no symmetry reason to be zero . hence there is for example no problem with the nonzero josephson current in @xmath3 - pb junctions along the c - axis @xcite . moreover since the attractive in - plane interaction and the repulsive interaction between plane and chain help each other , there is no problem to explain the high value of the critical temperature of @xmath3 @xcite . naturally our model is specific of @xmath3 , but we can think that it can be generalized to other compounds where the role of the chains can be played by other parts of the structure , such as the bio planes in bssco . on the other hand there is no possibility of this kind in lsco and accordingly we do not expect experiments to display in this compound the same physical features as in ybco . our model has common features with many other models . as a two band model it is quite similar to the two band models introduced by suhl , matthias and walker and others @xcite to describe superconductivity in transition metals . the possibility of an interband repulsive interaction in two band models was already introduced by kondo @xcite . more recently a two band s - n model has been introduced by abrikosov and klemm @xcite to account for raman scattering data . the idea of an interband repulsion has been put forward by various groups @xcite recently in the context of high @xmath1 superconductivity in order to show that experiments displaying a change of sign of the order parameter did not necessarily imply a spin fluctuation mechanism . our model is also similar to the one proposed by abrikosov @xcite , where there are attractive and repulsive interactions within a single band , leading to a change of sign of the order parameter within this band . finally , as noted above , chain - plane hybridization is a standard feature of band structure calculations @xcite and there have been various suggestions of the importance of the chains in the physics of @xmath10 whether for intrinsic or for extrinsic reasons @xcite , these models allowing for the electron to jump between planes and chains . in this paper we consider the effect of impurities on the critical temperature in our model . we will actually restrict ourselves to non magnetic impurities . indeed magnetic impurities are easily included , but they will naturally lead to pair - breaking and produce a @xmath1 following the ag law as in other models . hence magnetic impurities can not discriminate between various models and we ignore them for simplicity . the conclusion of our study is that , in our model , for generic parameters , the critical temperature is much less sensitive to impurities than in standard d - wave models . therefore we might say that , with respect to the sensitivity of the critical temperature to non magnetic impurities , our model behaves as a weak d - wave model . this happens for two independent reasons . first the reduction of @xmath1 is due to plane - chain scattering , which is weak compared to plane - plane scattering . next the fact that we have a two band model provides further possibilities for a weak impurity sensitivity . in section ii we present our model and we calculate the critical temperature of the clean superconductor as a first step toward the calculation of the impurity dependence , which is dealt with in section iii . finally section iv is devoted to a comparison with experimental results and to our conclusion . let us first define specifically our model . we will describe the motion of an electron in planes and chains by the following hamiltonian : @xmath11 where @xmath12 and @xmath13 are creation operators in the plane and in the chain band respectively . the first term corresponds to an isolated plane with dispersion relation @xmath14 the second one to isolated chains with dispersion relation @xmath15 and the last term describes hopping between planes and chains . actually this independent electron hamiltonian does not correspond precisely to the situation found in @xmath3 . indeed @xmath3 is build by stacking up sets made of two @xmath7 planes and one cuo chains plane . therefore a better description is obtained by the hamiltonian : @xmath16 where all quantities are understood to depend on @xmath8 and @xmath9 and summations run also over @xmath17 the indices 1 and 2 number the @xmath7 planes and the index n numbers the stacks . introducing the even and odd plane band operators @xmath18 by @xmath19 = ( @xmath20 @xmath21 @xmath22 @xmath23 , and taking the fourier transform in the z direction , we obtain : @xmath24 where a dependence and summation on @xmath25 is now understood , and t @xmath26 = t @xmath27 and t @xmath28 = t @xmath29 , with c being the size of the unit cell along the z direction . band structure calculations @xcite give a crossing of the chain band fermi surface with the odd plane band fermi surface , whereas there is no crossing with the even plane band . therefore the even plane band will not play an interesting physical role in our model and we will forget it for simplicity , but it should clearly be retained in a very realistic calculation . this leads us back to the hamiltonian eq.(1 ) where the plane band is actually the odd plane band . the hopping term has a @xmath25 dependence as given by t @xmath28 . however , if the even plane band happens to cross the chain band , we would have no problem in extending our model . this would be particularly easy in the two extreme cases where the even and odd plane bands are nearly degenerate ( t @xmath30 @xmath31 t @xmath32 ) or well separated ( t @xmath30 @xmath33 t @xmath32 ) since we would be led again to the hamiltonian eq.(1 ) . with respect to pairing interactions , as we have already indicated , we take an attractive pairing interaction in the plane which can be for example due to phonon exchange . on the other hand we assume a repulsive pairing interaction between plane and chain . a natural physical origin for this term is the coulomb repulsion , since screening is certainly not very efficient in ybco because it is a weak metal . moreover screening is probably very ineffective along the z direction since hopping in this direction is small . we note however that hopping is physically necessary in order to have a pairing term between planes and chains . finally there is little experimental evidence for an attractive pairing interaction in the chains . however we will assume its existence for the sake of generality . we include also the standard intraband coulomb repulsion . let us first consider the critical temperature of the clean superconductor . in a first step we ignore the hopping term t @xmath34 the eliashberg equations at the critical temperature are : @xmath35 @xmath36 here @xmath37 and @xmath38 are the order parameter and the renormalization function at the matsubara frequency @xmath39 primed quantities refer to the chains while unprimed ones correspond to the planes ; @xmath40 and @xmath41 are the effective frequency dependent interactions due to phonon exchange while @xmath42 @xmath43 and k , k are the intraband and interband coulomb repulsion with cut - off @xmath44 of order of the fermi energy . while the interactions that we consider here have certainly a wavevector dependence in the real superconductor , we have no precise idea of what they are nor do we have any physical reason to believe that this dependence is strong . therefore , for the sake of simplicity and clarity , we have taken isotropic interactions within planes and chains . we can now proceed to the usual reduction of the cut - off from @xmath44 to a frequency @xmath45 a few times a typical phonon frequency . for @xmath46 @xmath47 @xmath48 , the order parameters @xmath37 and @xmath49 become constants @xmath50 and @xmath51 while @xmath38 , @xmath52 @xmath53 1 . these constants are obtained from eq.(4 ) and ( 5 ) as : @xmath54 @xmath55 where r = ln @xmath56 / @xmath57 . when these results are carried into eq.(4 ) and ( 5 ) , one obtains the same equations , except that the cut - off is now @xmath45 and @xmath42 @xmath43 and k , k are replaced by renormalized coulomb interactions @xmath58 @xmath59 and k * , k * given by : @xmath60 and similar expressions for @xmath59 and k * obtained by exchanging primed and unprimed quantities . naturally one recovers the standard coulomb pseudopotential when there is no interband interaction . it seems safe to assume @xmath61 - kk @xmath47 0 , otherwise we would have pairing from s - wave coulomb interaction alone ( when pairing from coulomb interaction is considered , this is usely for higher angular momenta ) . in this case renormalization does not change the sign of the various quantities . we see that , in addition to its direct effect , interband interaction decreases the effective intraband repulsion , leading to an increase of the critical temperature as it should @xcite . in the following we will restrict ourselves to the weak coupling limit where @xmath37 and @xmath49 can be taken as constants and mass renormalization is negligible . indeed , as it will be clear in the following , we do not expect strong coupling effects to modify qualitatively our conclusions . in this case @xmath1 depends only on the combinations @xmath62 @xmath63 @xmath64 - @xmath65 and @xmath66 @xmath63 @xmath67 - @xmath68 in addition to k * and k*. from now on we will omit the star for all the renormalized coupling constants . then , if we let x = ln ( @xmath69 / @xmath1 ) where @xmath69 is a typical phonon frequency ( or rather 1/4 of a typical phonon frequency @xcite ) , @xmath1 is obtained from : @xmath70 which is the standard two band result @xcite . although one expects physically that coulomb interaction reduces the isotope effect , this is not obvious from eq.(8 ) and ( 9 ) . this has been shown to be true by kondo @xcite in some limiting cases . we show in the appendix that this property holds quite generally . as a preliminary to the calculation with impurities , let us finally consider how the above calculation of the critical temperature is modified when we take into account the hybridization t @xmath71 between planes and chain . as indicated above , we take isotropic interactions and order parameters in the planes and in the chains . corresponding to the effective interaction hamiltonian : @xmath72 the order parameter satisfy : @xmath73 @xmath74 however because of the hybridization , the plane and chain operators @xmath75 and @xmath76 do not correspond to the eigenstates of @xmath77 the unitary transformation : @xmath78 diagonalizes the hamiltonian into @xmath79 = @xmath80 @xmath81 @xmath82 @xmath83 where we have set for convenience @xmath84 @xmath63 c and @xmath85 @xmath86 d . the energies @xmath87 and @xmath88 of the hybridized bands are given by 2 @xmath89 = @xmath90 + @xmath91 @xmath21 [ ( @xmath90 - @xmath91 @xmath92 + 4 t @xmath93 @xmath94^{1/2 } $ ] and @xmath95 = @xmath96 = @xmath97 - @xmath98 - @xmath99^{1/2 } $ ] @xmath63 @xmath100 , @xmath101 = - @xmath102 = @xmath97 - @xmath103 - @xmath99^{1/2 } $ ] @xmath63 @xmath104 , with again the k dependence understood everywhere . when we carry out this transformation in the mean field hamiltonian : @xmath105 we will obtain interband pairing terms , such as @xmath106 @xmath107 , coupling the hybridized bands . however we will consider that t @xmath71 is large enough so that these bands are well separated . specifically this means that we assume @xmath45 @xmath31 t @xmath71 ( otherwise we should make a much more careful strong coupling treatment , which would probably not bring anything new qualitatively ) . in this case the above pairing terms will be negligible because we can not have two electrons ( k,1 ) and ( -k,2 ) at the fermi surface with opposite wavevectors , but belonging to different bands ( we will justify more specifically this approximation in the next section ) . therefore the transformation eq.(13 ) gives : @xmath108 with @xmath109 = @xmath110 @xmath111 + @xmath112 @xmath113 and @xmath114 = @xmath110 @xmath113 + @xmath115 @xmath111 . naturally this band diagonal expression for the hamiltonian is obvious once the interband pairing terms are neglected , and we could have written it immediately . our essential point is that , starting from isotropic interactions in planes and chains , we obtain a specific anisotropy for the order parameter @xmath109 and @xmath114 . in particular we obtain nodes at the fermi surface since we have managed to have @xmath110 and @xmath112 with opposite signs , and @xmath111 goes from 0 to 1 when we move at the fermi surface of a given band . therefore we have an order parameter which is d - wave like , although we have assumed an attractive pairing in the planes . we have then as usual @xmath116 @xmath117 @xmath118 @xmath117 @xmath119 @xmath47 = - @xmath120 @xmath121 from eq.(11),(12 ) and ( 13 ) this leads to : @xmath122 @xmath123 since the bands are well separated , the * k * summations will be around the fermi surface of each band with cut - off @xmath45 for @xmath124 . the integration perpendicular to the fermi surface will give the standard factor x = ln ( 1.13 @xmath125 @xmath45 ) and we are left with summations along the fermi surface : @xmath126 @xmath127 where the order parameter @xmath128 should not be confused with the dirac function @xmath129 . for @xmath130 = 0 we have @xmath131 @xmath132 @xmath133 + @xmath134 and @xmath131 @xmath132 @xmath135 + @xmath134 for @xmath136 = 0 . moreover for most of the fermi surface we have @xmath90 @xmath31 @xmath91 or @xmath91 @xmath31 @xmath90 leading to @xmath131 @xmath132 @xmath137 or 1 . this corresponds merely to the result one obtains without hybridization . subtracting from eq.(18 ) and ( 19 ) their counterpart without hybridization , we are left with summations over quantities which are essentially nonzero only in the anticrossing region . in this small region it is convenient to take , at fixed @xmath138 @xmath90 and @xmath91 as new variables instead of @xmath8 and @xmath139 the jacobian of the transformation is j = @xmath140 @xmath141 @xmath142 @xmath143 ) = @xmath144 @xmath145 x @xmath146 @xmath144 where @xmath145 and @xmath146 are the fermi velocities of the plane and chain band at the crossing point . we are then led to evaluate : @xmath147 \nonumber\end{aligned}\ ] ] with @xmath148 = @xmath149 or @xmath150 or @xmath151 this is easily done by taking @xmath152 @xmath71 and @xmath153 @xmath71 as new variables and extending the boundaries of the integrals to infinity . the result is the same for the three integrals , namely - @xmath4 @xmath154 @xmath71 @xmath155 when this result is carried into eq.(18 ) and ( 19 ) we find naturally the first order correction to the coupling constants @xmath156 = @xmath157 g , @xmath158 = @xmath159 g , k = @xmath159 k and k = @xmath157 k , where @xmath157 and @xmath159 are the total density of states of the ( uncoupled ) plane and chain bands . the hybridization changes these coupling constants respectively into @xmath156 - @xmath160 ( g+k ) , @xmath158 - @xmath160 ( g+k ) , k - @xmath160 ( g+k ) and k - @xmath160 ( g+k ) . here @xmath160 is an effective density of states due to the hybridization and given by @xmath160 = @xmath161 @xmath162 where @xmath163 is the average of @xmath154 @xmath71 @xmath144 over @xmath25 ( we have taken into account that there are 4 crossings in the brillouin zone ) . naturally since we have @xmath160 @xmath31 @xmath157 , @xmath159 in our model these changes are small . we consider now the effect on the critical temperature of impurities located in the planes and in the chains . we will assume that the impurity potential scatters electrons within the planes or within the chains , but that it does not scatter them from plane to chain . our physical motivation is naturally that planes and chains are physically well separated , which corresponds to the fact that the hopping term t @xmath71 is small . therefore the potential for scattering from plane to chain is reduced by a factor of order t @xmath71 / @xmath164 compared to the intraplane or intrachain scattering ( where @xmath164 is a typical hopping term within planes or chains ) . this physical point is essential for the validity of our result . including this small plane - chain scattering does not make any problem , but for sake of clarity and simplicity we will not do it explicitly and we will rather come back to this point after we have obtained the effect of intraplane and intrachain scattering . we will also take an isotropic scattering within plane and chain , consistently with what we have done for the interactions . here again we will come back later on to the effect of a possible anisotropy . this leads us to the following impurity potential : @xmath165 we treat this potential within the born approximation since , for the calculation of the critical temperature , making use of the t - matrix approximation is merely equivalent to renormalize the scattering potential . there is however another effect produced by going beyond the born approximation , which will be considered at the end of this section . within the born approximation and after impurity averaging , only contributions to the self - energy corresponding to scattering twice on the same impurity are retained @xcite . since the potential eq.(20 ) does not scatter electrons from plane to chain , the impurities contribution to the self - energy has only components within planes or chains . they are given by @xcite : @xmath166 where @xmath167 and @xmath168 are the number of impurities per unit volume in the planes and in the chains , @xmath169 are the pauli matrices in nambu space and g is the temperature green s function in plane - chain representation . the green s function is related to the self - energy by : @xmath170 where @xmath171 is the green s function in the absence of the impurity self - energy . explicitly we set : @xmath172 @xmath173 where @xmath174 @xmath132 @xmath175 @xmath4 t is the matsubara frequency ( we omit systematically for clarity the index n in the matsubara frequency and all frequency dependent quantities ) , @xmath176 is the pauli matrix in plane - chain space and @xmath177 is the off - diagonal self - energy due to the pairing interaction ( we have set @xmath178 @xmath63 @xmath90 and @xmath179 @xmath63 @xmath91 ) . in order to invert @xmath180 it is convenient to go the hybridized band representation , or more precisely to make the transformation which diagonalizes @xmath180 when the off - diagonal part @xmath181 of the self - energy is zero . for @xmath182 the transformation matrix is just @xmath183 @xmath184 where @xmath183 is the transformation considered in the preceding section . in the present case we have merely to formally generalize it by replacing @xmath185 by @xmath185 - i @xmath186 for the particule - particule part in nambu space and by @xmath185 + i @xmath186 in the hole - hole part . we keep the same notation for this transformation , but the @xmath183 are now complex . the complete transformation matrix is now @xmath187 = ( @xmath183 + @xmath188 ) @xmath184 /2 + ( @xmath183 - @xmath188 ) @xmath189 /2 . actually we have @xmath80 , @xmath110 @xmath31 t since we assume that the hybridized bands are well separated ( @xmath80 or @xmath110 are typically of the same order of magnitude in the important cases as it will be clear below ) . moreover as usual we will be only concerned with matsubara frequencies @xmath174 at most of order @xmath80 and @xmath110 and therefore negligible compared to t . this implies that , when it multiplies a quantity of order @xmath80 or @xmath110 @xmath190 we can take for @xmath183 its zero order value , namely the real value found in the preceding section . this leads to the following expression @xmath191 = @xmath192 @xmath180 a for the full green s function in the hybridized representation ( @xmath192 is the transpose of a ) : @xmath193{\delta } _ { ij}-{a}_{ki}{\tilde{\delta } } _ { k}{a}_{kj}{\tau } _ { 1 } \label{eq25}\end{aligned}\ ] ] where : @xmath194}^{1/2}\approx 2{e}_{1,2}-2i{\tilde{\omega } } _ { 1,2 } \label{eq26}\end{aligned}\ ] ] and we have set : @xmath195}^{1/2 } } \label{eq27}\end{aligned}\ ] ] now the dominant contributions to the summations of g over the wavevector * k * will come from the vicinity of the fermi surface of the hybridized bands . in this case either @xmath196 is large and all the other matrix elements of @xmath191 are small ( of order @xmath110 typically ) , or @xmath197 is large and the other elements are small . it is then easy to see that when @xmath197 is large , we obtain g by simply inverting the ( 1,1 ) block of @xmath198 that is have @xmath199 @xmath53 @xmath200^{-1 } $ ] @xmath201 @xmath202 . the other terms of @xmath203 are smaller by a factor of order @xmath110 over energy separation of the hybridized bands ( for fixed * k * ) , that is a factor of order @xmath204 or less . we have a similar result when @xmath196 is large . taken together this means that we can take @xmath199 as block diagonal , that is @xmath199 = @xmath205 @xmath206 with : @xmath207 where @xmath208 = @xmath209 @xmath210 , that is explicitly @xmath211 = @xmath212 @xmath131 + @xmath213 @xmath214 and @xmath215 = @xmath212 @xmath214 + @xmath213 @xmath216 we can now calculate the impurity self - energy @xmath80 . since @xmath199 is block diagonal we have @xmath217 = @xmath209 @xmath218 . when we carry this result into eq.(21 ) , we can perform the integration perpendicularly to the fermi surface ( that is integrate over @xmath87 or @xmath88 ) since it converges within an energy range of order @xmath219 we are then left with summations along the fermi surface , just as in the calculation of the critical temperature in the preceding section . this leads to : @xmath220 where we have taken into account that @xmath208 @xmath31 @xmath221 since we are only interested in the calculation of the critical temperature . we can then proceed as in the preceding section : for most of the fermi surface we have @xmath178 @xmath31 @xmath179 or @xmath179 @xmath31 @xmath178 leading to @xmath222 @xmath132 @xmath137 or 1 . this corresponds to the result without hybridization which is : @xmath223 where @xmath224 = @xmath4 @xmath225 @xmath226 @xmath227 with @xmath228 @xmath63 @xmath157 and @xmath229 @xmath63 @xmath230 when we calculate @xmath231 - @xmath232 we are left with summations over nonzero quantities only in the anticrossing region . we take then @xmath233 - @xmath234 / t and @xmath233 + @xmath179 ) / 2 t as new variables and extend the boundaries of the integrals to infinity . the resulting expression simplifies under exchanging @xmath178 and @xmath179 . we obtain that the diagonal parts of @xmath231 and @xmath232 are equal which implies @xmath235 = @xmath174 + @xmath224 sign @xmath174 . on the other hand we find from the off - diagonal part : @xmath236 \label{eq31}\end{aligned}\ ] ] where u = @xmath233 - @xmath237 - @xmath238 + 4 @xmath239 @xmath94^{1/2 } $ ] , and a similar result for @xmath213 . this equation displays explicitly an essential point for our result : the correction to the self - energy due to hybridization is small , of order @xmath160 / @xmath240 , compared to the self - energy in the absence of hybridization . this correction corresponds physically to the scattering from plane to chain induced by hybridization . however , without hybridization , we are in the standard s - wave situation and impurities do not change the critical temperature . the only effect on @xmath1 will come from the correction due to hybridization . since this correction is small , we expect the change of @xmath1 to be small with respect to what one would get with a simple d - wave order parameter . although the integration in eq.(31 ) can be performed analytically in the general case , the result is not simple . therefore , in order to obtain reasonably simple calculations , we will continue our quantitative investigation only in the particular case where @xmath241 = @xmath242 @xmath63 @xmath243 which implies @xmath244 = @xmath245 @xmath63 @xmath246 . actually this equality is always valid when the matsubara frequency @xmath174 is large enough compared to @xmath241 and @xmath242 . this is in particular always valid when the impurity concentration is small enough so that @xmath247 @xmath31 @xmath1 . therefore the case we consider is quite reasonable . in this case the equations for @xmath248 and @xmath249 simplify into : @xmath250 @xmath251 we have finally to calculate from eq.(12 ) and ( 13 ) the self - energies @xmath252 and @xmath253 due to the pairing interactions . we have already obtained the off - diagonal part of the green s function , and the calculation is essentially the same as for the off - diagonal part of the impurity self - energy . one finds : @xmath254-{n}_{2}k\pi t\sum\nolimits\limits_{n}^ { } [ { { \tilde{\delta } } _ { 2 } \over { \left|{\tilde{\omega } } \right|}^{}}+{{n}_{t } \over { n}_{2}}{{\tilde{\delta } } _ { 1}-{\tilde{\delta } } _ { 2 } \over { \left|{\tilde{\omega } } \right|}^ { } } ] \label{eq34}\end{aligned}\ ] ] @xmath255-{n}_{1}k\pi t\sum\nolimits\limits_{n}^ { } [ { { \tilde{\delta } } _ { 1 } \over { \left|{\tilde{\omega } } \right|}^{}}+{{n}_{t } \over { n}_{1}}{{\tilde{\delta } } _ { 2}-{\tilde{\delta } } _ { 1 } \over { \left|{\tilde{\omega } } \right|}^ { } } ] \label{eq35}\end{aligned}\ ] ] however these equations merely mean that we have to calculate the critical temperature with the coupling constants modified by the hybridization , as we have found at the end of the preceding section . with these coupling constants , eq.(34 ) and ( 35 ) take the simple form : @xmath256 @xmath257 from eq.(32 ) and ( 33 ) we have : @xmath258 @xmath259 where we have set @xmath260 = @xmath261 @xmath160 / @xmath228 and @xmath262 = @xmath261 @xmath160 / @xmath229 . as expected the inverse lifetime @xmath261 due to scattering within planes and chains has disappeared from the equations , and only are left @xmath260 and @xmath262 which describe physically the scattering from plane to chain due to hybridization . when this is carried into eq.(36 ) and ( 37 ) and the summations on matsubara frequency is carried out , one finds : @xmath263-k[{\delta } _ { 2}x+({\delta } _ { 1}-{\delta } _ { 2}){k ' \over k+k'}k(\rho ) ] \label{eq40}\end{aligned}\ ] ] @xmath264-k'[{\delta } _ { 1}x+({\delta } _ { 2}-{\delta } _ { 1}){k \over k+k'}k(\rho ) ] \label{eq41}\end{aligned}\ ] ] where x = ln ( 1.13 @xmath45 / @xmath1 ) and @xmath265 = @xmath266 + @xmath267 ) - @xmath268 , with @xmath267 = @xmath269 + @xmath270 and we have used the fact that in our case k / k = @xmath260 / @xmath262 . eq.(40 ) and ( 41 ) are essentially what one obtains for the effect of the impurities in a two band model @xcite . this is easy to understand physically since , because of hybridization , the in - plane scattering potential allows effectively an electron to scatter from plane to chain for example . however our essential point is that the corresponding effective inverse scattering time @xmath117 @xmath63 @xmath260 + @xmath262 is strongly reduced , compared to plane - plane or chain - chain scattering , because the hybridization is small . we note also that , in the general case , our equations eq.(31 ) do not reduce to a simple two - band model , although we expect the physics to be similar . let us briefly review the existing literature @xcite for the consequences of eq.(40 ) and ( 41 ) for the critical temperature , and add some results for the case which is of interest for us , namely the general situation with repulsive interactions . from eq.(40 ) and ( 41 ) one gets an equation for x which is more conveniently rewritten as : @xmath271 here @xmath272 and @xmath273 are the solutions of eq.(9 ) . more precisely eq.(9 ) has two physically relevant solutions when @xmath274 @xmath47 kk. the lower one @xmath272 gives the physical critical temperature . the higher one @xmath273 corresponds to an unstable superconducting state where the order parameter has the same sign in the two bands , and therefore the interband coupling decreases the critical temperature with respect to the uncoupled situation instead of increasing it ( in the uncoupled limit @xmath272 = @xmath275 and @xmath273 = @xmath276 ; we assume for example @xmath156 @xmath47 @xmath158 @xcite ) . @xmath116 kk , the interband coupling is too strong for the state with the ill - chosen signs of the order parameter to exist and the unstable solution disappears ( there is still a mathematical solution with @xmath273 @xmath116 0 but it must be rejected physically because it does not satisfy the weak coupling condition x @xmath33 1 anymore ) . finally we have ( @xmath274 - kk ) @xmath277 = ( @xmath156 k + @xmath158 k + 2 kk ) / ( k+k ) in eq.(42 ) . it is easy to see that one has always @xmath272 @xmath116 @xmath277 @xmath116 @xmath273 for @xmath274 @xmath47 kk , while @xmath273 @xmath116 @xmath277 @xmath116 @xmath272 is satisfied for @xmath274 @xmath116 kk . now for fixed @xmath117 , @xmath265 is a positive , increasing function of x + ln @xmath278 , with @xmath265 @xmath279 0 for x @xmath279 - @xmath280 and @xmath265 @xmath53 x + ln @xmath278 for x @xmath279 @xmath280 . then it is easy to see graphically that , starting from @xmath272 for @xmath117 = 0 , x as given by eq.(42 ) is an increasing function of @xmath117 . therefore , in all possible cases , the critical temperature decreases when the impurity concentration increases . for @xmath274 @xmath47 kk the increase of x saturates at @xmath277 leading to a large concentration limit @xmath281 for the critical temperature given , as found by kusakabe @xcite , by : @xmath282 this saturation corresponds physically to the situation where strong impurity scattering has made the order parameter completely isotropic @xmath252 = @xmath283 while ( @xmath252 - @xmath253 ) @xmath265 takes an independent nonzero value . the result eq.(43 ) is then easily rederived directly from eq.(40 ) and ( 41 ) . therefore in this case the superconducting state survives very strong impurity scattering by becoming isotropic . this situation is analogous to the fate of very dirty anisotropic s - wave superconductors @xcite and naturally quite different from the standard behaviour of a d - wave superconductor . we note incidentally that , when one goes from the pure system to the very dirty one , one goes from @xmath252 @xmath253 @xmath116 0 to @xmath252 @xmath253 @xmath47 0 , which implies that , if @xmath156 @xmath284 k @xmath47 @xmath158 + k for example , the chains become gapless @xmath253 = @xmath137 @xmath285 some specific impurity concentration ( given from eq.(40 ) and ( 41 ) by @xmath265 = @xmath286 - kk ) and x = @xmath287 - kk ) ) . however this case @xmath274 @xmath47 kk requires not too small coupling constants @xmath156 and @xmath158 in both bands . in ybco it is not clear at all that the cu0 chains alone have a superconducting tendency . if we have @xmath158 @xmath53 0 , the condition @xmath274 @xmath47 kk can not be satisfied . for @xmath274 @xmath116 kk there is no saturation in the increase of x and the critical temperature goes to zero when the impurity concentration is increased , as found by kusakabe @xcite . however when we substitute in eq.(42 ) the large x behaviour of @xmath288 we obtain : @xmath289 where @xmath290 is the critical temperature without impurities . therefore @xmath1 goes to zero for a critical impurity concentration and we find in this case a behaviour similar to the standard d - wave result . the corresponding critical value @xmath291 for @xmath117 is given by : @xmath292 \label{eq45}\end{aligned}\ ] ] the right hand side of eq.(45 ) is positive and it goes from zero ( for @xmath156 - @xmath158 = k - k , or k = 0 ) to infinity ( for @xmath156 @xmath158 = k k ) . therefore the lowest possible value of @xmath291 is the standard d - wave result @xmath291 = .88 @xmath290 , but it can easily be much higher since it is given by the exponential of the r.h.s . of eq.(45 ) . from this , one would conclude that , for a given @xmath293 our model is much less sensitive to impurities than standard d - waves . however when one looks at the behaviour of @xmath1 as a function of @xmath293 one sees that this large value of @xmath291 is obtained at the end of a long tail where the value of @xmath1 is already quite small , as it can be seen in fig.1 . therefore in order to fully characterize the sensitivity of @xmath1 to impurities , we have also to look at the initial slope found for low impurity concentration . in this regime we have from eq.(42 ) : @xmath294 \label{eq46}\end{aligned}\ ] ] which is the result of moskalenko and palistrant @xcite . as they indicated , ( @xmath272 - @xmath277 ) / ( @xmath272 - @xmath273 ) goes from 0 to 1 , which implies that the decrease of @xmath1 is always less than for the pure d - wave case . the d - wave result is obtained in the limiting case @xmath156 - @xmath158 = k - k or k @xmath279 0 . on the other hand for k @xmath279 0 ( and fixed k ) , the critical temperature does not depend anymore on the impurity concentration ( this result is actually valid whatever the impurity concentration as it can be seen from eq.(42 ) ) . this limiting case is not unphysical since it corresponds to a situation where the density of states @xmath159 in the chains would be very small . we would still have a d - wave type order parameter with change of sign and nodes , but the critical temperature would be completely controlled by the planes and the chains would simply follow . therefore we have the surprising result of a superconducting phase with a d - wave type order parameter which has a critical temperature insensitive to impurities just as an s - wave superconductor . however it is fair to say that we do not expect such a limiting situation to occur in ybco . hence we believe that in our case eq.(46 ) leads to a sensitivity to impurity which is always somewhat reduced with respect to the standard d - wave case , but not by a large factor . an example of the general behaviour of @xmath1 @xmath295 @xmath290 as a function of our scattering rate @xmath117 is given in fig.1 where we have chosen the parameters @xmath156 = 1 , @xmath158 = 0 , k = 0.5 and k takes the values 0.01 , 0.1 , 0.2 , 0.5 , 1.5 and 5 ( k = 1.5 gives exactly an abrikosov gorkov law ) . in order to conclude this section , let us now come back to the various terms we have omitted from the beginning . we did not take into account direct plane - chain scattering by impurities . as we mentionned , we expect physically this scattering to be reduced by a factor @xmath160 / @xmath296 compared to plane - plane or chain - chain scattering which we have treated . including this plane - chain term is just what is done in the standard two - band model when the effect of impurities on @xmath1 is calculated . since we have shown that our calculation lead us to the two - band model , including direct plane - chain scattering would just give us an additional term of the same order as the one we have found ( it would appear as an additional term in @xmath297 ) . therefore our conclusions with respect to the sensitivity of @xmath1 to impurities are unchanged . we have also omitted mixed plane - chain terms @xmath298 for the order parameter . this is justified by our finding that , when the hybridized bands are well separated , the dominant contributions of g come from the block diagonal parts @xmath299 and @xmath300 . hence we have for example @xmath301 = @xmath95 @xmath299 @xmath101 + @xmath102 @xmath300 @xmath96 . however the products @xmath95 @xmath101 and @xmath102 @xmath96 are significantly different from zero only in the anticrossing region . when we calculate the off - diagonal self - energy , we have to sum @xmath301 over * k*. accordingly the result @xmath298 will be smaller by a factor of order @xmath160 / @xmath296 compared to @xmath252 and @xmath302 which justifies our approximation . finally we note that going beyond the born approximation will not only renormalize the impurity cross section . it will also introduce plane - chain terms @xmath303 in the impurity self - energy because , after scattering on a plane impurity , an electron can go to the chains through @xmath301 and scatter on a chain impurity . but we have just seen that , after summation of @xmath301 over * k * , one obtains a result which is smaller by a factor @xmath160 / @xmath296 . this leads to a @xmath303 which is smaller than @xmath304 or @xmath305 by the same factor and hence negligible . the conclusion of our study is that , in our model , the critical temperature is much less sensitive to impurities than it is in standard d - wave models . one basic reason is that we expect physically impurities to produce essentially plane - plane and chain - chain scattering . just as in s - wave superconductors , this scattering does not affect the critical temperature . on the other hand @xmath1 is reduced by scattering between parts of the fermi surface which have opposite signs for the order parameter , just as in standard d - wave . in our model this is due to plane - chain scattering . we have found that this scattering , whatever its origin , will be smaller by a factor of order t / @xmath306 ( that is hybridization coupling over fermi energy ) compared to plane - plane and chain - chain scattering . from the band structure calculations we expect this factor to be typically somewhere between 0.1 and 0.3 . therefore the sensitivity of @xmath1 to impurities in our model is reduced by a similar factor , compared to the d - wave situation . in the specific case which we have studied in details and which reduces to the two - band model , we have found a further reduction of the sensitivity of @xmath1 to impurities with a behaviour which can vary continuously from s - wave like to d - wave like depending on the parameters . from our discussion of its physical origin , we expect a similar behaviour and reduction to occur in the general case . it is unfortunately not possible to make quantitative comparison with experiments . indeed it seems surprisingly quite difficult @xcite to avoid a magnetic character for substitutional impurities in ybco . naturally this occurs when the isolated impurity atom itself has a magnetic moment . but this happens also when a non magnetic impurity acquires a magnetic moment due to its interaction with the environment , as it is the case for zn for example @xcite . naturally an impurity with a magnetic character will produce pair - breaking leading to an abrikosov - gorkov like law in any model , in agreement with what is observed experimentally . hence these kind of experiments can not be used directly to eliminate theoretical models . nevertheless , after these words of caution with respect to a simple - minded interpretation of impurities experimental results , we note that the reduction of @xmath1 down to 13 k for 8% zn is obtained by an increase of the residual resistivity @xcite by a factor of order 10 . the corresponding @xmath307 / @xmath308 should be then of order 20 @xmath1 ( see below ) . this is much more than what is necessary to destroy superconductivity within a d - wave model according to the abrikosov and gorkov law . moreover it is useful to plot the results of ref.17 and ref.18 for the critical temperature as a function of the residual resistivity . this is done on fig.2 . it can be seen that the behaviour of some of our results found in fig.1 is quite similar to the experimental results ( we assume naturally that the residual resistivity is proportional to the scattering rate due to impurities ) . we have naturally enough adjustable parameters to fit them nicely . however such a fit would be rather meaningless because , in addition to the magnetic problem , the scatter in the data is rather important at low impurity content and the interpretation of the data for larger concentration is uncertain ( localisation effects which are not taken into account in our theory might play a significant role ) . if we turn to irradiation experiments , whether by electrons or by light ions , the interpretation is also not an easy one . it is known that high @xmath1 compounds have a critical temperature much more sensitive to irradiation than standard superconducting materials . however , although it is likely that most of the created defects do not have a magnetic character , we can not eliminate , from our knowledge on substitutional impurities , the possibility that some are magnetic . the experimental evidence is controversial in this respect @xcite . then there is some evidence that localisation effects might be important since a metal - insulator transition is observed in ybco under light ion irradiation , and there is no intermediate normal phase between the superconducting phase and the insulating one @xcite . this sensitivity to localisation is easy to understand because of the two - dimensional nature of the @xmath7 planes . localisation effects are not included in our theoretical study . it is also not clear at all that the created disorder can be considered as homogeneous @xcite since the resistivity measurements do not always display a sharp drop at the critical temperature . similarly one might expect that the defects created by irradiation are randomly distributed at the microscopic scale , but it is actually quite likely that the chains are more sensitive to irradiation than the planes . there is finally the obvious problem of having an experimental determination of the quasiparticle lifetime produced by disorder . measuring the increase in resistivity appears the best way to do it @xcite although it is far from perfect since , for example , it measures at best transport lifetime which we expect to be somewhat larger than quasiparticle lifetime . notwithstanding the above problems let us try to interpret the irradiation experiments with our theoretical model , just to see what comes out . from drude s law , with a typical resistivity of 100 @xmath309 cm at @xmath1 and a plasma frequency of 1.1 ev @xcite , we have a typical inverse lifetime @xmath310 / @xmath308 @xmath53 2 @xmath311 since superconductivity disappears in d - wave when the inverse lifetime due to impurities is of order 2 @xmath1 we would expect that an increase of resistivity of 100 @xmath309 cm leads to the suppression of @xmath1 @xcite . this corresponds roughly to a decrease of @xmath1 of 1k per @xmath309 cm ( the ag law is essentially linear ) . the experimental results of ref.15 give a linear decrease of @xmath1 with respect to resistivity with a slope 0.3 k / @xmath309 cm . in ref.14 there is an upward curvature at low resistivity with a maximal initial slope of 0.1 k/ @xmath309 cm . since in our model a reduction by a typical factor 1/10 with respect to the d - wave result corresponds to a typical choice of our parameters , we see that this last experimental result agrees with our expectation . but the result of ref.14 could easily be explained for example by a larger value of the hybridization energy or by a suitable choice of the other parameters of our two - band model . quite generally we have enough parameters to vary in our model so we can get easily agreement with these various experimental results . however it must also be kept in mind that , as discussed above , there are other possible physical processes which we have not taken into account and which will add up to produce a faster decrease of the critical temperature with the resistivity . a clear example of this is found in ref.14 where a more rapid decrease is found for a set of samples and attributed to extrinsic effects , while good samples show an average decrease of 0.03 k/ @xmath309 cm . therefore we can consider an experimental result as an upper bound for our theoretical result , but it may quite well be larger than what we find . clearly it is more difficult to explain in a d - wave model the slow dependence of @xmath1 on resistivity found in ref.14 than it is in our model to explain the somewhat stronger dependence found in ref.15 . in conclusion we have seen that our model is quite coherent with the present experimental evidence . naturally , with respect to this problem of the effect of impurities on the critical temperature , it would be much better to have experiments providing stronger constraints on theoretical models , but this might prove difficult to achieve . let us vary eq.(9 ) with respect to the phonon frequencies @xmath45 and @xmath69 . we assume for example @xmath312 = - @xmath313 / @xmath45 = - @xmath314 @xmath295 @xmath69 @xmath116 @xmath315 the variation of eq.(9 ) gives : where @xmath317 . the isotope effect is reduced if we show that @xmath318 @xmath116 0 . we set @xmath319 = ( @xmath320 ) ( @xmath321 ) - @xmath322 . we have @xmath323 ( @xmath324 ) = @xmath325 @xmath326 / d , and @xmath327 = @xmath65 @xmath326 / d + d ( @xmath61 - kk ) @xmath312 @xmath47 @xmath65 @xmath326 / d ( we assume @xmath61 - kk @xmath47 0 , which implies @xmath328 - @xmath329 ) . since we have y @xmath47 @xmath62 and y @xmath47 @xmath330 and @xmath326 @xmath47 0 it is enough to prove that : @xmath335 = \nonumber \\ - k^*k'^ * [ \lambda ^*-\lambda ^{\prime * } + \mu ^*-\mu ^{\prime * } ] ^{2 } - ( \mu ^*\mu ^{\prime * } - k^*k'^ * ) [ ( \lambda ^*-\lambda ^{\prime * } ) ^{2 } + 4 k^*k'^ * ] < 0 \label{eqa3}\end{aligned}\ ] ] where @xmath336 = 1/ @xmath272 and @xmath333 = 1/ @xmath273 . since we have y @xmath47 @xmath336 @xmath47 y , eq.(48 ) is satisfied and the isotope effect is indeed reduced . there is no reduction only in the limiting case @xmath61 = kk and @xmath337 for a recent review of the controversy about the order parameter symmetry , see r. c. dynes , sol . 92 , 53 ( 1994 ) and j. r. schrieffer , sol . 92 , 129 ( 1994 ) . d. j. scalapino , e. loh and j. e. hirsch , phys . rev . b * 34 * , 8190 ( 1986 ) ; a. kampf and j. r. schrieffer , phys . rev . b*42 * , 7967 ( 1990 ) ; p. monthoux , a. v. balatsky and d. pines , phys.rev.lett.*67 * , 3448 ( 1991 ) . a. g. sun , d. a. gajewski , m. b. maple and r. c. dynes , phys.rev.lett.*72 * , 2267 ( 1994 ) . p. chaudhari and shawn - yu lin , phys.rev.lett.*72 * , 1084 ( 1994 ) . w. n. hardy , d. a. bonn , d. c. morgan , r. liang and k. zhang , phys . rev . lett . * 70 * , 3999 ( 1993 ) ; l. a. de vaulchier , j. p. vieren , y. guldner , n. bontemps , r. combescot , y. lemaitre and j. c. mage , europhys * 33 * , 153 ( 1996 ) . d. a. wollman , d. j. van harlingen , w. c. lee , d. m. ginsberg and a. j. leggett , phys.rev.lett.*71 * , 2134 ( 1993 ) ; d. a. brawner and h. r. ott , phys.rev.b * 50 * , 6530 ( 1994 ) . c. c. tsuei , j. r. kirtley , c. c. chi , l. s. yu - jahnes , a. gupta , t. shaw , j. z. sun and m. b. ketchen , phys.rev.lett . * 73 * , 593 ( 1994 ) d. n. basov _ _ , phys.rev.lett.*74 * , 598 ( 1995 ) . j. buan , b. zhou , c. c. huang , j. z. liu and r. n. shelton , phys.rev . b * 49 * , 12220 ( 1994 ) . p. hirschfeld , p. wlfle and d. einzel , phys.rev.b * 37 * , 83 ( 1988 ) . r. j. radtke , k. levin , h. b. schttler and m. r. norman , phys.rev.b * 48 * , 653 ( 1993 ) . p. muzikar , d. rainer and j. a. sauls , fermi liquid theory of non s - wave superconductivity in lecture notes for the nato advanced study institute : vortices in superfluids to be published by kluwer ( 1994 ) . a. a. abrikosov and l. p. gorkov , zh . . fiz . * 39 * , 1781 ( 1960 ) [ sov . jetp * 12 * , 1243 ( 1961 ) ] . j. lesueur , l. dumoulin , s. quillet and j. radcliffe , j. alloys and compounds * 195 * , 527 ( 1993 ) . j. giapintzakis , d. m. ginsberg , m. a. kirk and s. ockers , phys.rev . b * 50 * , 15967 ( 1994 ) . r. liang , t. nakamura , h. kawaji and m. itoh , physica c * 170 * , 307 ( 1990 ) . t. r. chien , z. z. wang and n. p. ong , phys.rev.lett . * 67 * , 2088 ( 1991 ) . s. zagoulaev , p. monod and j. jegoudez , phys.rev . b * 52 * , 10474 ( 1995 ) . r. combescot and x. leyronas , phys.rev.lett.*75 * , 3732 ( 1995 ) . j. yu , s. massida , a. j. freeman and d. d. koeling , phys . lett . a * 122 * , 203 ( 1987 ) ; w.e . pickett et al.,science * 255 * , 46 ( 1992 ) . b. ashauer , w. lee and j. rammer , z. phys . b * 67 * , 147 ( 1987 ) ; l.n . bulaevskii , o.v . dolgov , m.o . ptitsyn , phys.rev . b * 38 * , 11290 ( 1988 ) . h. suhl , b. t. matthias and l. r. walker , phys.rev.lett . * 3 * , 552 ( 1959 ) . v. a. moskalenko , fiz . * 8 * , 503 ( 1959 ) . j. kondo , prog . * 29 * , 1 ( 1963 ) . t. soda and y. wada , prog . phys.*36 * , 1111 ( 1966 ) . v. a. moskalenko and m. e. palistrant , zh . * 49 * , 770 ( 1965 ) [ sov . jetp * 22 * , 536 ( 1966 ) ] ; c. c. sung and v. k. wong , j. phys . * 28 * , 1933 ( 1967 ) ; w. s. chow , phys.rev . * 172 * , 467 ( 1968 ) ; t. kusakabe , prog . phys.*43 * , 907 ( 1970 ) . a. a. abrikosov , physica c * 182 * , 191 ( 1991 ) ; a. a. abrikosov and r. a. klemm , physica c * 191 * , 224 ( 1992 ) . a. a. golubov , o. v. dolgov , e. g. maksimov , i. i. mazin and s. v. shulga , physica c * 235 - 240 * , 2383 ( 1994 ) . a. a. golubov and i. i. mazin physica c * 243 * , 153 ( 1995 ) . a. i. liechtenstein , i. i. mazin and o. k. andersen , phys . rev . lett.*74 * , 2303(1995 ) . d. z. liu , k. levin and j. maly , phys.rev.b * 51 * , 8680 ( 1995 ) . a. a. abrikosov , physica c * 244 * , 243 ( 1995 ) . z. tesanovic , a. r. bishop and r. l. martin , sol . comm , * 68 * , 337 ( 1988 ) . v. z. kresin and s. a. wolf , phys.rev . b * 46 * , 6458 ( 1992 ) . r. a. klemm and s. h. liu , phys.rev.lett.*74 * , 2343 ( 1995 ) . d. rainer , sol . comm * 6 * , 111 ( 1968 ) . r. combescot , europh.lett.10 , 177 ( 1989 ) ; r.combescot , phys . b * 42 * , 7810 ( 1990 ) . a. a. abrikosov , l. p. gorkov and i. e. dzyaloshinski , methods of quantum field theory in statistical physics ( dover , n.y . , 1975 ) . see for example k. maki in superconductivity , ed.by r. d. parks ( dekker , n. y. 1969 ) ; p.b.allen and b. mitrovic , sol . state phys . 37 , ed.f . seitz , d. turnbull and h. ehrenreich ( academic , n. y. 1982 ) . d. markowitz and l.p . kadanoff , phys.rev . * 131 * , 563 ( 1963 ) . the symmetric case @xmath156 = @xmath158 and k = k has @xmath277 = @xmath273 = @xmath156 - k , and is solved explicitely into two branches @xmath338 + @xmath265 and @xmath339 with the physical solution corresponding to the lower value for x. this degenerate situation is interesting to keep in mind , since the general case can be considered as resulting from the coupling of these two branches when one moves away from this particular case .

we present a detailed examination of the brown dwarf multiples 2mass j08503593 + 1057156 and 2mass j17281150 + 3948593 , both suspected of harboring components that straddle the l dwarf / t dwarf transition . resolved photometry from _ hubble space telescope_/nicmos show opposite trends in the relative colors of the components , with the secondary of 2mass j0850 + 1057 being redder than its primary , while that of 2mass j1728 + 3948 is bluer . we determine near - infrared component types by matching combined - light , near - infrared spectral data to binary templates , with component spectra scaled to resolved nicmos and photometry . combinations of l7 + l6 for 2mass j0850 + 1057 and l5 + l6.5 for 2mass j1728 + 3948 are inferred . remarkably , the primary of 2mass j0850 + 1057 appears to have a later - type classification compared to its secondary , despite being 0.81.2 mag brighter in the near - infrared , while the primary of 2mass j1728 + 3948 is unusually early for its combined - light optical classification . comparison to absolute magnitude / spectral type trends also distinguishes these components , with 2mass j0850 + 1057a being mag brighter and 2mass j1728 + 3948a.5 mag fainter than equivalently - classified field counterparts . we deduce that thick condensate clouds are likely responsible for the unusual properties of 2mass j1728 + 3948a , while 2mass j0850 + 1057a is either an inflated young brown dwarf or a tight unresolved binary , making it potentially part of a wide , low - mass , hierarchical quintuple system .

1011.0798

the l dwarfs and the t dwarfs are the two lowest luminosity spectral classes of very low - mass stars and brown dwarfs known today , encompassing effective temperatures ( t@xmath2 ) from @xmath32200 k down to @xmath3600 k ( @xcite ; see also and references therein ) . spectroscopic studies of these sources reveal atmospheres that are remarkably diverse . l dwarfs having abundant molecular gas species and clouds of condensates in their photospheres , while t dwarfs have relatively cloud - free photospheres and more complex molecular gas species including ch@xmath4 , nh@xmath5 and co@xmath6 @xcite . condensate cloud properties are believed responsible for near - infrared spectral and color variations among equivalently - classified l dwarfs @xcite , temporal variability in late - type dwarfs and dramatic changes in the spectral energy distributions between the l dwarf and t dwarf classes @xcite . non - equilibrium chemistry and atmospheric circulation @xcite coupled with the complex processes of condensate grain formation , growth , circulation and evaporation makes the formation , evolution and influence of condensate clouds in low - temperature atmospheres among the outstanding problems in brown dwarf astrophysics today . the disappearance of photospheric condensates at the transition between the l dwarf and t dwarf classes is one particularly interesting aspect of brown dwarf clouds . this transition occurs over a narrow range of temperatures ( @xmath7t@xmath2 @xmath1 200 k ) and luminosities ( @xmath7@xmath8 @xmath1 0.2 dex ; @xcite ) , and is accompanied by a temporary brightening at 1 @xmath9 ( the `` j - band bump '' ; @xcite ) and enhanced rates of multiplicity @xcite . these effects appear to arise from the rapid depletion of condensate clouds at the l dwarf / t dwarf transition @xcite , although the driving mechanism for that depletion is inadequately explained by current cloud models @xcite . whether arising from the fragmentation of cloud structures @xcite , a sudden increase in sedimentation efficiency @xcite or some other process remains unclear , but cloud depletion at the l dwarf / t dwarf transition has important implications on cloud evolution in other low - temperature atmospheres , such as those of extrasolar planets @xcite . coeval and cospatial multiples are important laboratories for studying this transition , eliminating dependencies on age , composition and distance . so - called `` flux - reversal '' binaries , whose components straddle the l dwarf / t dwarf transition , have verified the 1 @xmath9 brightening as an intrinsic aspect of brown dwarf evolution @xcite . yet resolved spectroscopy of the components of these systems has been difficult to obtain , due largely to the close separations typical of brown dwarf multiples ( @xmath10 @xmath11 20 au ; @xcite ) . in this article , we present a comprehensive analysis of two binaries whose components are suspected to straddle this transition : 2mass j08503593 + 1057156 ( hereafter 2mass j0850 + 1057 ) and 2mass j17281150 + 3948593 ( hereafter 2mass j1728 + 3948 ) . both were resolved with the _ hubble space telescope _ ( hst ) wide field planetary camera 2 ( wfpc2 ; @xcite ) and noted as potential l dwarf / t dwarf transition binaries based on their photometric properties and late ( combined - light ) systemic spectral types of l6 and l7 , respectively . here we combine resolved near - infrared photometry obtained with the hst / near infrared camera and multi - object spectrometer ( nicmos ) and ground - based laser guide star adaptive optics ( lgsao ) imaging with combined - light , near - infrared spectroscopy to infer component spectral types , colors and absolute magnitudes . observations are described in section 2 . in section 3 we describe point spread function ( psf ) fitting of the hst data that yield relative fluxes for each binary in five spectral bands spanning 1.01.8 @xmath9 . in section 4 we describe our spectral fitting procedure and determine component classifications , colors and absolute magnitudes . in section 5 we compare these measures to current absolute magnitude / spectral type and absolute magnitude / color trends . in section 6 we discuss our results in the context of cloud evolution at the l dwarf / t dwarf transition , and explore the possibility of higher - order multiplicity in the case of 2mass j0850 + 1057 . results are summarized in section 7 . the empirical properties of the two binaries examined here are summarized in table [ tab_properties ] . 2mass j0850 + 1057 was initially identified in the two micron all sky survey ( 2mass ; @xcite ) by @xcite , and selected to be the prototype for the l6 spectral subclass . its optical spectrum exhibits 6708 li i absorption , indicating component masses below @xmath30.06 m@xmath12 @xcite . this source was resolved as a 0@xmath1316 binary system by @xcite with hst / wfpc2 , and has been subsequently confirmed as a common proper motion pair from multi - epoch hst and lgsao observations . the large difference in component brightnesses in the wfpc2 f814w band ( @xmath7f814w = 1.47@xmath140.09 ; @xcite ) has suggested a late - type l or early - type t dwarf secondary , although this component has been a persistent outlier in color magnitude diagrams . astrometric parallax measurements by @xcite and @xcite find discrepant distances of 25.6@xmath142.3 pc and 38@xmath146 pc , differing by nearly 2@xmath15 . @xcite identified an unrelated background source that may have skewed the @xcite astrometric measurements , and report a preliminary parallax that is intermediate between these two values . we adopt the @xcite for this study , as was also used by @xcite . @xcite also identified a widely - separated , common proper motion and common distance companion to 2mass j0850 + 1057 , the m5 + m6 binary nltt 20346ab at a projected separation of 4@xmath161 ( 7700 au ) . h@xmath17 and x - ray emission in these m dwarfs , coupled with li i absorption in 2mass j0850 + 1057 , indicate a relatively young age of 0.251.5 gyr for the combined system , and relatively low masses for the 2mass j0850 + 1057 components : 0.04@xmath140.02 m@xmath12 and 0.03@xmath140.01 m@xmath12 based on evolutionary models @xcite . these mass estimates are consistent with weak empirical constraints by @xcite , m@xmath18 = 0.2@xmath140.2 m@xmath12 , as inferred from incomplete astrometric monitoring of its orbit . lccl optical spectral type & l6 & l7 & 1,2 + nir spectral type & l7@xmath142 & l6@xmath141 & 3 + mko @xmath19 & 16.20@xmath140.03 & 15.90@xmath140.08 & 3,4 + mko @xmath20 & 1.85@xmath140.04 & 2.01@xmath140.09 & 3,4 + distance ( pc ) & 38@xmath146 & 24.1@xmath141.9 & 5 + @xmath21 ( km s@xmath22 ) & 26.6@xmath144.5 & 5.1@xmath140.9 & 5 + separation ( mas ) & 157@xmath143 & 131@xmath143 & 6 + ( au ) & 6.0@xmath140.9 & 3.2@xmath140.3 & 5,6 + position angle ( @xmath23 ) & 114.7@xmath140.3 & 27.6@xmath141.2 & 6 + @xmath24 & 1.47@xmath140.09 & 0.37@xmath140.04 & 6,7 + @xmath25 & & -0.25@xmath140.14 & 7 + @xmath26 & 0.78@xmath140.07 & 0.63@xmath140.03 & 8 + li i ? & yes & no & 1,2 + age ( gyr ) & 0.251.5 & @xmath271.5 & 3,9 + combined mass ( m@xmath12 ) & 0.2@xmath140.2 & 0.15@xmath28 & 8 + 2mass j1728 + 3948 was also identified in 2mass by @xcite and classified l7 based on its optical spectrum . this source shows no indication of li i absorption and hence its primary is inferred to have a mass greater than 0.06 m@xmath12 . @xcite and @xcite identified this system as a 0@xmath1313 binary based on hst / wfpc2 observations . the former study found the `` secondary '' to be fainter in the f814w band and brighter in the f1042 m band , the first reported example of a l / t flux reversal binary . multi - epoch hst and lgsao observations reported in and @xcite have verified the common proper motion of the components , with the latter study finding an orbital period of 31@xmath1412 yr , semimajor axis of 5.3@xmath140.8 au ( based on the parallax distance measurement of 24.1@xmath141.9 pc ; @xcite ) and total system mass of 0.15@xmath28 m@xmath12 . however , like 2mass j0850 + 1057 , the orbit of this system has not been fully sampled . assuming a primary temperature of @xmath11450 k based on the combined light optical spectral type @xcite , the absence of li i indicates a system age of at least 1.5 gyr , according to the evolutionary models of @xcite and @xcite . low resolution near - infrared spectral data for 2mass j0850 + 1057 and 2mass j1728 + 3948 were obtained with the 3 m nasa infrared telescope facility ( irtf ) spex spectrograph @xcite on 2008 january 8 ( ut ) and 2006 august 21 ( ut ) , respectively . conditions on both nights were mostly clear , with light cirrus during the 2006 august observations ; seeing was @xmath30@xmath136 at @xmath19-band in both observations . we employed the prism - dispersed mode of spex which provides 0.752.5 @xmath9 continuous spectroscopy at a resolution @xmath29 @xmath30 for the 0@xmath135 slit , and dispersion of 2030 pixel@xmath22 . the slit was aligned to the parallactic angle in all observations . 2mass j0850 + 1057 was observed at an airmass of 1.13 , with four exposures of 150 s each obtained in an abba dither pattern along the slit . 2mass j1728 + 3948 was observed in a similar manner , at an airmass of 1.06 and with four exposures of 180 s each . after each target observation , a0 v stars hd 74721 ( v = 8.71 ) and hd 164899 ( v = 7.91 ) were observed with identical instrument settings and at a similar airmass for telluric absorption and flux calibration . these were followed by internal flat field and ar arc lamps for pixel response and wavelength calibration . data were reduced using the spextool package , version 3.3 @xcite using standard settings ; see @xcite for details . the reduced spectra for both sources are shown in figure [ fig_nirspec ] . the data exhibit classic signatures of l dwarf near - infrared spectra , including a steep 0.81.2 @xmath9 spectral slope ; feh absorption at 1.0 @xmath9 ; unresolved and doublets at 1.13 , 1.17 and 1.22 @xmath9 ; deep h@xmath6o absorption bands centered at 1.4 and 1.9 @xmath9 ; and co absorption at 2.3 @xmath9 @xcite . their overall near - infrared spectral energy distributions are fairly red , consistent with their @xmath31 colors and indicative of cloud opacity capping the 1.25 @xmath9 and 1.65 @xmath9 spectral peaks @xcite . there is no obvious indication of ch@xmath4 absorption at either 1.6 @xmath9 or 2.2 @xmath9 , commonly seen in the combined light spectra of l dwarf / t dwarf pairs ( e.g. , @xcite ) . lcccc -j & 0.673@xmath140.014 & l8.0@xmath140.4 & 0.712@xmath140.009 & l6.9@xmath140.3 + h@xmath6o - h & 0.652@xmath140.010 & l8.2@xmath140.4 & 0.723@xmath140.005 & l5.6@xmath140.2 + ch@xmath4-k & 0.987@xmath140.009 & l4.8@xmath140.3 & 0.955@xmath140.004 & l5.9@xmath140.2 + mean spt & & l7@xmath142 & & l6@xmath141 + we derived near - infrared spectral types for these sources using the h@xmath6o - j , h@xmath6o - h and ch@xmath4-k spectral indices defined in @xcite , and the index - spectral type relations defined in @xcite which have a typical scatter of @xmath31 subtype for l0l8 dwarfs . values are reported in table [ tab_indices ] . the inferred near - infrared types are l7@xmath142 for 2mass j0850 + 1057 and l6@xmath141 for 2mass j1728 + 3948 . these types are later and earlier than the reported optical types for these sources , respectively , albeit consistent within the uncertainties . the larger uncertainty in the near - infrared spectral type of 2mass j0850 + 1057 is driven by a discrepant ch@xmath4-k index value , which indicates an l5 spectral type as compared to l8 from the h@xmath6o indices . this difference may be related to the multiplicity of this system , or its young age and the corresponding low surface gravity of its components . both sources were observed in single orbits with the hst / nicmos nic1 camera as part of program go-9843 ( pi gizis ) . 2mass j0850 + 1057 was observed on 2003 november 9 ( ut ) and 2mass j1728 + 3948 on 2003 september 7 ( ut ) . nic1 is the highest - resolution camera on nicmos with pixel scale 0@xmath13043 and field of view 11@xmath32@xmath3311@xmath32 , providing a well - sampled psf down to the diffraction limit at 1 @xmath9 . both sources were observed through the wide - band filters f110w ( @xmath34 = 1.025 @xmath9 , @xmath35 = 0.6 @xmath9 ) and f160w ( @xmath34 = 1.55 @xmath9 , @xmath35 = 0.4 @xmath9 ) , and the medium - band filters f110 m ( @xmath34 = 1.1 @xmath9 , @xmath35 = 0.2 @xmath9 ) , f145 m ( @xmath34 = 1.45 @xmath9 , @xmath35 = 0.2 @xmath9 ) , and f170 m ( @xmath34 = 1.7 @xmath9 , @xmath35 = 0.2 @xmath9 ) . as shown in figure [ fig_nirspec ] , the f110w and f160w filters sample the prominent @xmath19 and @xmath36 flux peaks of late - type l and t dwarf spectra , while the f110 m , f145 m and f170 m filters sample the 1.1 @xmath9 h@xmath6o , 1.4 @xmath9 h@xmath6o and 1.6 @xmath9 ch@xmath4 absorption bands , respectively . these filter combinations provide discriminating colors for late - type l and t dwarfs ( e.g. , @xcite ; figure [ fig_color ] ) . all data were acquired in multiaccum mode , with two dithered exposures ( 1@xmath135 step ) obtained in each filter . total integration times for each source were 256 s in f110w and f110 m ( 2mass j1728 + 3948 was observed for 288 s in f110 m ) , 320 s in f160w , 704 s in f170 m and 832 s in f145 m . integrations were longest with the f145 m filter because of the deep h@xmath6o absorption present at these wavelengths . raw images were reduced by standard pipeline processing ( calnica , @xcite ) using calibration images ( inflight and model reference files , circa 2002 ) and photometric keywords ( post - cyrogenic ) current as of february 2008 . calnica reduction includes analog - to - digital correction , subtraction of bias and dark current frames , linearity correction , correction for readout artifacts ( the `` bars '' anomaly ) , division by an appropriate flat field image , photometric calibration , cosmic ray identification , and combination of multiaccum frames into a single calibrated image . no correction was made for the temperature dependence of the dark and flat - field reference files . these basic calibrated images ( two images per filter per object ) were used for the analysis . + + + + figure [ fig_images ] displays subsections of the calibrated imaging data for each binary in the five filter bands . two components are resolved in each image , although the increase in psf size in the longer wavelength data results in better psf separation in the f110 m and f110w images as compared to the f170 m images . there are notable differences in the relative component fluxes between these two sources . the eastern component of 2mass j0850 + 1057 is significantly fainter than the western component in all five filter bands , while the two components of 2mass j1728 + 3948 are roughly equal in brightness . we measured aperture photometry for the combined light of each source using the iraf phot routine , employing a 15 pixel radius aperture centered on the brightest component and a 20 - 25 pixel annulus to measure the median background . count rates were converted to vega magnitudes on the cit system using photometric conversion parameters listed in the nicmos data handbook version 8.0 , but without applying aperture corrections . values are listed in table [ tab_nicmos ] , where the uncertainties include shot noise and background uncertainties , as well as a standard 5% calibration uncertainty . relative photometric measurements are discussed below . lcccc|cccc epoch & & + f110w & 17.38@xmath140.07 & 1.15@xmath140.03 & 17.70@xmath140.07 & 18.99@xmath140.08 & 16.76@xmath140.05 & 0.322@xmath140.015 & 17.36@xmath140.05 & 17.68@xmath140.05 + f110 m & 17.49@xmath140.06 & 1.176@xmath140.015 & 17.81@xmath140.06 & 18.99@xmath140.08 & 16.81@xmath140.07 & 0.25@xmath140.03 & 17.44@xmath140.07 & 17.69@xmath140.07 + f145 m & 16.48@xmath140.05 & 1.000@xmath140.015 & 16.84@xmath140.05 & 17.84@xmath140.05 & 15.77@xmath140.05 & 0.517@xmath140.014 & 16.29@xmath140.05 & 16.81@xmath140.05 + f160w & 15.70@xmath140.05 & 0.90@xmath140.05 & 16.09@xmath140.05 & 16.99@xmath140.06 & 15.08@xmath140.05 & 0.461@xmath140.016 & 15.63@xmath140.05 & 16.09@xmath140.05 + f170 m & 15.36@xmath140.05 & 0.89@xmath140.07 & 15.76@xmath140.05 & 16.65@xmath140.07 & 14.77@xmath140.05 & 0.462@xmath140.014 & 15.32@xmath140.05 & 15.78@xmath140.05 + f110w - f160w & 1.68@xmath140.09 & & 1.61@xmath140.09 & 1.86@xmath140.10 & 1.68@xmath140.07 & & 1.74@xmath140.07 & 1.60@xmath140.07 + f110w - f170 m & 2.02@xmath140.09 & & 1.95@xmath140.09 & 2.21@xmath140.10 & 1.99@xmath140.07 & & 2.05@xmath140.07 & 1.91@xmath140.07 + @xmath37 ( mas ) & & + ( au ) & & + pa ( @xmath23 ) & & + relative photometry for the components of each binary were determined using a psf - fitting algorithm similar to that described in @xcite . models for each binary image were generated using psfs calculated with the _ tiny tim _ package , version 6.3 @xcite . psfs were generated for each filter passband ( assuming post - cryocooler aberrations for nic1 ) , and at two pointing positions to account for mirror zonal errors . the spex spectra of each source were used as templates for calculating filter passband effects . individual psf models were generated for a subimage size of 3@xmath32 @xmath33 3@xmath32 , which was resampled at ten times the original resolution to allow for subpixel offsets . a binary psf model for each nicmos image was determined using an iterative image fitting algorithm that finds optimal positions and relative fluxes . initial guesses were found using a simple peak detection algorithm and single - psf subtraction on a 3@xmath32 @xmath33 3@xmath32 subframe of the data image centered on the binary . we then varied the primary position , secondary position , primary flux and secondary flux , in that order , to minimize the statistic @xmath38 where @xmath39 is the data subframe , @xmath40 is the model image , @xmath41 is a masking image used to exclude bad pixels ( @xmath41 = 0 for bad pixels ) , @xmath42 is the mean difference ( to account for residual background flux in the data ) and the sum is performed over all pixels . the iteration step in position was set at 0.1 pixels ( 4.3 mas ) in accord with the subsampling of the psf model ; the iteration in flux values was 1% . iterations were performed in a hierarchical recursive loop , and ceased when the fractional decrease in @xmath43 was less than 10@xmath44 for all parameters . for each source , we adopted as our final relative flux values the mean of the two pointing frames in each filter band , and uncertainties include the difference in individual measurements and a 1% flux sampling uncertainty . final separations and position angles003 . ] were adopted as the mean of all frames , with uncertainties incorporating measurement scatter and a 0.1 pixel sampling uncertainty . the close separation and overlapping psfs of the components of these two systems initially raised concerns that fitting biases for the longer - wavelength images ( e.g. , f170 m ) could lead to skewed component photometry and colors . this was of particular concern for 2mass j0850 + 1057 , whose components have a larger difference in brightness . to test our fitting program , we performed an identical analysis with simulated binary images generated from the model psfs . the simulated images were constructed to have the same component positions and relative magnitudes as determined from the psf fitting ( see below ) as well as relative magnitudes inferred from unconstrained spectral fits ( see section 4.1 ) . in both cases , the fitting algorithm reproduced the input relative magnitudes in all filter bands to within 0.01 mag , even for the f170 m images of 2mass j0850 + 1057 . table [ tab_nicmos ] summarizes the results of the psf fitting analysis and all of the nicmos photometry . we find separations and position angles that are consistent to within 1.5@xmath15 of those measured by using the same nicmos data . the relative magnitudes are all positive , indicating that the primary components are brighter than the secondary components at these wavelengths for both sources . relative component brightnesses show distinct trends . for 2mass j0850 + 1057 , relative magnitudes decrease toward longer wavelengths , from @xmath7f110 m = 1.176@xmath140.015 to @xmath7f170 m = 0.89@xmath140.07 , indicating a secondary that is redder than the primary . this is confirmed by the redder broadband color of the secondary , f110w - f160w = 1.61@xmath140.09 and 1.86@xmath140.10 for 2mass j0850 + 1057a and 2mass j0850 + 1057b , respectively . in contrast , the relative magnitudes of 2mass j1728 + 3948 generally increase toward longer wavelengths ( with the exception of the f145 m band ) indicating a bluer secondary . this trend is also seen in resolved @xmath45 photometry for 2mass j1728 + 3948 by @xcite . figure [ fig_color ] compares the colors of the 2mass j0850 + 1057 and 2mass j1728 + 3948 components to spectrophotometric estimates calculated from l2t6 spex spectral templates ( section 4.1 ) . in general , the components span the full range of colors seen in the l dwarf templates , although 2mass j0850 + 1057b and ( to lesser degree ) 2mass j1728 + 3948a are notably redder in both broadband and narrow - band colors . f110m - f145 m and f110m - f170 m colors , sampling 1.4 @xmath9 h@xmath6o and 1.6 @xmath9 ch@xmath4 features , indicate that none of the components have spectral types t0 or later ; i.e. , they are all l dwarfs . nicmos colors alone are unable to distinguish l subtypes for the individual components . they are , however , cleanly segregated in f110w - f170 m and f110w - f160w colors , which are correlated for the l and t dwarf templates . the binary components follow a sequence of 2mass j1728 + 3948b , 2mass j0850 + 1057a , 2mass j1728 + 3948a and 2mass j0850 + 1057b in this color - color plot , in order of increasing f110w - f170 m and f110w - f160w colors . to infer the component near - infrared spectral types of 2mass j0850 + 1057 and 2mass j1728 + 3948 , we compared their combined - light spex spectra to a suite of empirical binary templates scaled to the observed relative photometry . we followed a procedure similar to that described in @xcite , drawing from a uniform sample of low - resolution , high signal - to - noise spex prism spectra of late - type m , l and t dwarfs . the template sample was restricted to dwarfs later than m7 based on published optical classifications for m and l dwarfs ( tied largely to the schemes of @xcite ) and near - infrared classifications for t dwarfs ( tied largely to the scheme of @xcite ) . about @xmath315% of the m and l dwarfs in our sample have only near - infrared types in the literature , based on various schemes ( e.g. , @xcite ) . we also computed alternate near - infrared spectral classifications directly from the spex data using the @xcite and @xcite index - spectral type relations , following the iterative procedure detailed in @xcite . known ( resolved ) binaries and sources noted to have peculiar spectra or highly uncertain classifications ( @xmath462 subtypes ) were purged from the template sample . the resulting 462 spectra of 438 sources were interpolated onto a common wavelength scale , and synthetic vega magnitudes in the five nicmos filters and mko @xmath47 filters were computed by convolving each spectrum and a kurucz model of vega with the appropriate filter transmission functions ( see @xcite ) . from these individual templates , we produced separate libraries of binary templates for 2mass j0850 + 1057 and 2mass j1728 + 3948 by adding together appropriately scaled pairs . the pairs were initially selected to have secondary component types that were no more than 2 subtypes earlier than the primary type . this produced 137,212 unique combinations . we then computed an uncertainty - weighted mean relative flux scaling between the two components of each system using all five nicmos measurements and @xmath0 photometry- and @xmath36-band photometry for 2mass j1728 + 3948 , @xmath48 = 0.32@xmath140.02 and @xmath49 = 0.45@xmath140.02 . these measurements are redundant and in agreement with broad - band f110w and f160w nicmos photometry , and were therefore not used in the analysis . ] from @xcite ; see table [ tab_properties ] . after scaling the spectra , we required that relative synthetic magnitudes in all six filter bands agree with measured values to within 3@xmath15 . this substantially reduced the number of `` allowable '' templates to 19,042 for 2mass j0850 + 1057 and 1389 for 2mass j1728 + 3948 . comparisons were then made between the source spectral data and allowed binary templates using the weighted chi - squared statistic defined in @xcite , @xmath50\left [ \frac{d[{\lambda}]-{\alpha}t[{\lambda}]}{\sigma_d[{\lambda } ] } \right]^2 \label{equ_chisq}\ ] ] here , @xmath51 $ ] and @xmath52 $ ] are the data and template spectra , respectively ; @xmath53 $ ] is the uncertainty spectrum of the data ; @xmath54 $ ] is a vector of weights satisfying @xmath55 } = 1 $ ] ; @xmath17 is a scaling factor that minimizes @xmath56 ( see equation 2 in @xcite ) ; and the sum is performed over the wavelength ranges @xmath57 = 0.951.35 @xmath9 , 1.451.8 @xmath9 and 2.02.35 @xmath9 to avoid regions of strong telluric absorption . we adopted the same weighting scheme as that used in @xcite and @xcite , with each pixel weighted by its spectral width ( i.e. , @xmath58 ) . templates which provided minimum values of @xmath56 were deemed to be the best fits . however , it was generally the case that other templates gave @xmath56 values only slightly larger than this best - fit template , and are therefore statistically equivalent . we therefore determined average component parameters ( spectral types and relative magnitudes ) and their uncertainties following ( * ? ? ? * equations 46 ) , with an effective degrees of freedom @xmath59 = 253 used for weighting the parameters with the f - test statistic . the resulting best - fitting binary templates are listed in table [ tab_fits_all ] , while figures [ fig_fit0850 ] and [ fig_fit1728 ] display the four best fits to 2mass j0850 + 1057 and 2mass j1728 + 3948 , respectively . these templates reproduce the observed data rather well , including the overall spectral slopes , shapes of the @xmath60 flux peaks , and depths of the h@xmath6o , co and feh absorption bands . by design , each of the spectral combinations listed also reproduce the relative magnitude measurements to within the observational uncertainties . notably , the binary template fits are statistically superior to equivalent single template comparisons . best - fit @xmath56 values for these single templates were 2.20 for 2mass j0850 + 1057 ( comparing to the l5/l6.51 subtype , `` : : '' for @xmath142 subtypes . ] 2massw j1326201 - 272937 ; @xcite ) and 3.90 for 2mass j1728 + 3948 ( comparing to the l5:/l5 2mass j06244595 - 4521548 ; @xcite ) , differing at the @xmath61 99.9% significance level based on the f - test statistic . the significantly improved fits by the binary templates indicate that they are ( as expected ) better representations of the combined light spectra of these sources . llllllccc + sdss j151506.11 + 443648.3 & l7.5@xmath141.5 & l6.9 & 2massw j1553214 + 210907 & l5.5 & l6.1 : & 1.46 & 1.00 + sdss j151506.11 + 443648.3 & l7.5@xmath141.5 & l6.9 & 2massi j0028394 + 150141 & l4.5 & l5.2 : : & 1.57 & 0.28 + sdss j151506.11 + 443648.3 & l7.5@xmath141.5 & l6.9 & sdssp j010752.33 + 004156.1 & l8 & l6.0 & 1.62 & 0.20 + sdss j171714.10 + 652622.2 & l4 & l5.8 : & sdssp j010752.33 + 004156.1 & l8 & l6.0 & 1.85 & 0.03 + sdss j171714.10 + 652622.2 & l4 & l5.8 : & sdss j080959.01 + 443422.2 & l6 & l6.2 : & 1.88 & 0.02 + sdss j171714.10 + 652622.2 & l4 & l5.8 : & sdssp j132629.82 - 003831.5 & l8 : & l6.1 : & 1.94 & 0.01 + 2mass j03101401 - 2756452 & l5 & l6.0 : & sdssp j010752.33 + 004156.1 & l8 & l6.0 & 1.95 & 0.01 + & l7.5@xmath140.0 & l6.9@xmath140.0 & & l5.7@xmath141.2 & l5.8@xmath140.4 & & + + 2mass j01443536 - 0716142 & l5 & l3.9 & sdss j104409.43 + 042937.6 & l7 & l7.1 & 2.41 & 1.00 + 2massw j2224438 - 015852 & l4.5 & l4.0 : & sdss j104409.43 + 042937.6 & l7 & l7.1 & 2.58 & 0.29 + 2mass j03185403 - 3421292 & l7 & l6.5 & 2mass j09054654 + 5623117 & l5 & l5.6 : & 2.64 & 0.23 + 2mass j01443536 - 0716142 & l5 & l3.9 & 2mass j23254530 + 4251488 & l8 & l7.1 & 2.75 & 0.15 + 2mass j03185403 - 3421292 & l7 & l6.5 & 2massi j1305410 + 204639 & l4 : & l6.0 : & 2.92 & 0.06 + 2mass j01443536 - 0716142 & l5 & l3.9 & sdssp j003259.36 + 141036.6 & l8 & l7.9 : & 3.12 & 0.02 + & l5.3@xmath141.0 & l4.6@xmath141.1 & & l6.5@xmath141.1 & l6.8@xmath140.6 & & + the spectra of the best - fit template components also replicate photometric trends from hst and lgsao imaging . for 2mass j0850 + 1057 , each of the templates in figure [ fig_fit0850 ] has a secondary that is distinctly redder than its primary . these secondaries 2massw j1553214 + 210907 ( l5.5/l6.1 : , @xmath31 = 2.03@xmath140.19 ; @xcite ) , 2massi j0028394 + 150141 ( l4.5/l5.2 : : , @xmath31 = 1.95@xmath140.13 ; @xcite ) and sdssp j010752.33 + 004156.1 ( l8/l6.0 , @xmath31 = 2.12@xmath140.07 ; @xcite ) are among the reddest l dwarfs currently known . in contrast , the primaries of these templates have @xmath31 colors that are in line with median values for their spectral types ( e.g. , @xcite ) . what is more surprising , however , is that the best - fit secondaries are generally of _ earlier type _ than the best - fit primaries . this is true for both literature and spex classifications . the mean component classifications reflect this : l7.5 and l5.5@xmath141.2 based on literature classifications and l7 and l6 based on spex classifications ; we adopt the latter combination as the uncertainty - weighted means . this apparent reversal in classifications is particularly remarkable given the 0.81.2 magnitude brightness difference between the primary and secondary . there is no evidence that either component is a t dwarf . for 2mass j1728 + 3948 , the best - fit secondaries are consistently bluer than the primaries , again in line with photometric trends . notably , the components of the binaries shown in figure [ fig_fit1728 ] have comparable fluxes in the 0.91.0 @xmath9 region , supporting the f1042 m flux reversal reported by @xcite , and diverge toward longer wavelengths . for these templates , it is the primary components that are unusually red : 2mass j01443536 - 0716142 ( l5/l4.2 , @xmath31 = 1.92@xmath140.03 ; @xcite ) , 2massw j2224438 - 015852 ( l4.5/l4.5 : , @xmath31 = 2.05@xmath140.04 ; @xcite ) and 2mass j03185403 - 3421292 ( l7/l6.4 , @xmath31 = 2.06@xmath140.07 ; @xcite ) . the mean spectral type of 2mass j1728 + 3948a , l5.5@xmath141.0 from literature classifications and l4.5@xmath141.1 from spex classifications , is also remarkable for being considerably earlier than the combined - light l7 optical classification . it is typically the earlier - type component that dominates optical flux in l and t dwarf binaries . in this case , it appears that the comparable brightnesses of the two components at red optical wavelengths produces a `` blended '' combined - light spectral type . the secondaries of the best - fit templates have normal colors , although both sdss j104409.43 + 042937.6 ( l7/l6.9 ; @xcite ) and 2mass j23254530 + 4251488 ( l8/l7.3 : ; @xcite ) show weak signatures of 1.6 and 2.2 @xmath9 ch@xmath4 absorption in their near - infrared spectra . these features suggest that 2mass j1728 + 3948b is at the threshold of the l dwarf / t dwarf transition , despite its mid - l near - infrared spectral type . we adopt mean component types of l5 and l6.5 for this system . to assess the robustness of our results , we conducted the same template fits for three subsets of relative photometry : the nicmos data alone , and single filter scalings with f110w and @xmath0 photometry . the nicmos fits produce little change in the inferred spectral components , with 2mass j0850 + 1057 still hosting a later - type primary ( l7 + l6 components based on both literature and spex classifications ) and 2mass j1728 + 3948 having a primary that is typed earlier than the combined light spectrum ( l5 + l6.5 for literature classifications , l4 + l7 for spex classifications ) . however , the mean relative @xmath0 magnitudes from these fits 0.950@xmath140.015 for 2mass j0850 + 1057 and 0.54@xmath140.05 for 2mass j1728 + 3948 differ from the measurements of @xcite at the 7@xmath15 and 1.5@xmath15 levels , respectively . inferred brightnesses diverge even more dramatically when a single filter is used to scale the templates . f110w - scaled templates for 2mass j0850 + 1057 indicate component types of l6 + l9 : , with a significantly reduced best - fit @xmath56 value ( 1.12 versus 1.46 , distinct at the 98% confidence level ) . these types are more in line with prior estimates ( e.g. , @xcite ) , but the inferred component magnitudes differ significantly from measured values : @xmath7f160w disagrees at the 8@xmath15 level , while @xmath26 disagrees at the 6.5@xmath15 level . @xmath0-scaled templates for 2mass j0850 + 1057 produce comparably large discrepancies in inferred nicmos photometry . single - filter template fits to 2mass j1728 + 3948 are somewhat more robust , producing similar component types as the baseline template sample ( l5+l7 for f110w fits , l5+l7.5 : for @xmath0 fits ) , although inferred relative magnitudes still deviate by up to 2@xmath15 . we conclude that multi - band photometry is essential to accurately and precisely constrain the component properties of these binary systems , and particular care must be taken when drawing conclusions from single - filter measurements . nevertheless , the general agreement in component types inferred from different multi - band subsets ( e.g. , nicmos+@xmath0 , nicmos only ) indicates that our results are robust . both 2mass j0850 + 1057 and 2mass j1728 + 3948 have parallax distance measurements from @xcite , so it is possible to compare component absolute fluxes to those of comparable field dwarfs . figure [ fig_abs ] displays combined light and component absolute mko @xmath19 and @xmath62 magnitudes versus near - infrared spectral type and @xmath20 color . the mko magnitudes from 2mass j0850 + 1057 and 2mass j1728 + 3948 are based on measurements from @xcite for the former , and synthesized from 2mass photometry and spex spectroscopy for the latter . component @xmath63 fluxes were determined from the template fits , synthesized from the various best - fit template spectra and combined following the same weighting scheme as the average spectral types ( table [ tab_components ] ) . we use the near - infrared spectral types calculated in section 2.2 . the comparison sample was drawn from the compilation of @xcite , where we have either used published near - infrared spectral types listed in that study or types calculated from spex spectra ( where available ) following the same spectral index method described above . lccccc nir spectral type & l7 & l7+l7 & l6 & l5 & l6.5 + @xmath64 & 13.62@xmath140.35 & 14.37@xmath140.35 & 14.75@xmath140.35 & 14.60@xmath140.19 & 14.90@xmath140.19 + @xmath65 & 11.82@xmath140.35 & 12.57@xmath140.35 & 12.77@xmath140.35 & 12.48@xmath140.18 & 13.07@xmath140.18 + @xmath20 & 1.80@xmath140.04 & & 1.98@xmath140.05 & 2.12@xmath140.09 & 1.84@xmath140.09 + @xmath8 & -4.13@xmath140.14 & -4.43@xmath140.14 & -4.52@xmath140.14 & -4.42@xmath140.08 & -4.63@xmath140.08 + t@xmath2 ( k ) & 1770@xmath14200 & 1480@xmath14180 & 1400@xmath14170 & 1570@xmath1480 & 1340@xmath1470 + mass ( m@xmath12 ) & 0.06@xmath140.02 & 0.10@xmath140.03 & 0.05@xmath140.02 & 0.075@xmath140.007 & 0.066@xmath140.008 + focusing first on the absolute magnitude / spectral type comparisons , we find that three of the components2mass j0850 + 1057b , 2mass j1728 + 3948a and 2mass j1728 + 3948b roughly cluster , with absolute magnitudes similar to other l6l8 sources . 2mass j1728 + 3948a , and to a lesser degree 2mass j1728 + 3948b , are somewhat underluminous for their types , particularly at @xmath19-band where they fall @xmath10.5 mag below the absolute magnitude relations of @xcite . in contrast , the primary of 2mass j0850 + 1057 is @xmath11 mag brighter than comparable l6l8 dwarfs , or equivalently has a classification 3 subtypes too late for its measured absolute brightness . this component is even marginally brighter than combined light photometry for 2mass j1728 + 3948 . as such , the discrepancy between brightness and spectral type between the components of the 2mass j0850 + 1057 system appears to be rooted to the unusually late spectral type and/or overluminosity of its primary . in the near - infrared color - magnitude diagrams ( cmds ) , all four components follow the locus of field l dwarfs , tracing the inflection in @xmath20 color and @xmath64 magnitude at the end of the l sequence . in particular , 2mass j0850 + 1057b and 2mass j1728 + 3948b bridge the turnover to bluer near - infrared colors , and the latter appears to be the faintest l dwarf with a parallax measurement currently known . its location on the near - infrared cmds supports our hypothesis that it is on the cusp of becoming a t dwarf ( section 4.2 ) . the components of 2mass j0850 + 1057 are no longer outliers in these plots , in contrast to prior results ( e.g. , @xcite ) . 2mass j0850 + 1057a in particular does not stand out as unusually bright , due largely to the near - vertical locus of l dwarfs in near - infrared cmds . for completeness , we also determined luminosity , t@xmath2 and mass estimates for the components of these binaries using their inferred absolute magnitudes and near - infrared spectral types . bolometric luminosities were calculated using @xmath62-band bolometric corrections ( @xmath66 ) derived from the @xmath66/spectral type relation of @xcite . note that these values may be systematically too high for the unusually red sources 2mass j0850 + 1057b and 2mass j1728 + 3948a . t@xmath2s and masses were estimated using the @xcite cloudy evolutionary models , based on the derived luminosities and age estimates listed in table [ tab_properties ] ; for 2mass j1728 + 3948 , we assumed an age range of 1.57.5 gyr , the upper limit based on the small @xmath21 of this source . the inferred t@xmath2s for 2mass j1728 + 3948a , 2mass j0850 + 1057b and 2mass j1728 + 3948b are consistent with prior estimates for l5l6.5 dwarfs @xcite , while the brightness of 2mass j0850 + 1057a makes it @xmath1300 k warmer than comparable l7 dwarfs . we discuss this component in detail in section 6.2 . our photometric and spectroscopic analyses of the components of 2mass j0850 + 1057 and 2mass j1728 + 3948 have revealed several unusual traits , particularly in their primaries : under- and overluminous fluxes and unusually early and late near - infrared spectral types . these peculiar traits can be related to their unique atmospheric properties . in the case of 2mass j1728 + 3948a , we hypothesize that condensate cloud effects are responsible for shifting this component toward both an earlier near - infrared spectral classification and toward slightly fainter @xmath19-band fluxes . grey extinction from condensate cloud grains in l dwarf photospheres dominate the opacity at the @xmath19- and @xmath36-band flux peaks , as these windows in molecular gas opacity probe deeper into the atmosphere and sample a larger column depth of cloud material @xcite . the @xmath62-band peak , on the other hand , is modulated by both cloud opacity and collision - induced @xmath67 absorption @xcite . as a result , greater condensate opacity tends to produce redder @xmath20 colors and fainter @xmath19-band fluxes @xcite . in addition , contrast in molecular absorption bands is reduced , particularly for the near - infrared h@xmath6o and feh bands to which near - infrared schemes are commonly tied @xcite . veiling of these features can skew near - infrared classifications toward earlier types @xcite . hence , thicker condensate clouds leads to redder near - infrared colors , reduced flux at @xmath19-band , and systematically earlier near - infrared classifications . these conditions accurately reflect the properties of 2mass j1728 + 3948a , and to a lesser degree 2mass j0850 + 1057b . the best - fit primary components for 2mass j1728 + 3948 are consistently red , mid - type l dwarfs whose near - infrared spectral types are consistently earlier than their optical types ( table [ tab_fits_all ] ) . these include the templates 2mass j01443536 - 0716142 ( hereafter 2mass j0144 - 0716 ; l5 optical classification , l4 spex classification ; @xcite ) and 2massw j2224438 - 015852 ( hereafter 2mass j2224 - 0158 ; l4.5 optical classification , l3.5 near - infrared classification ; @xcite ) . both of these sources exhibit indications of thick clouds , based on the detection of linear polarization in the case of 2mass j0144 - 0716 ( 0.58@xmath140.19% at @xmath68-band ; @xcite ) , and spectral model fits and pronounced silicate grain absorption at 911 @xmath9 in the case of 2mass j2224 - 0158 @xcite . unusually red dwarfs such as these have been found to have cooler t@xmath2s for their spectral types @xcite , which can contribute to fainter magnitudes . the older age of the 2mass j1728 + 3948 system , based on the absence of li i absorption , argues that thick condensate clouds , rather than low surface gravity , gives rise to its unusually red color @xcite . thick clouds in 2mass j1728 + 3948a may also explain its comparable brightness at 1 @xmath9 compared to its later - type companion . as discussed above , 2mass j1728 + 3948b appears to be at the threshold of the l dwarf / t dwarf transition , a phase in which condensate clouds are inexplicably dispersed and @xmath19-band fluxes increase . in a sample of unresolved l / t transition binary candidates , @xcite found that sources with comparable component types but redder primaries showed a more pronounced flux reversal than sources with normal or blue primaries . they argued that @xmath19-band fluxes in the primaries of these systems were more suppressed . we may be seeing a similar effect in the 2mass j1728 + 3948ab pair . alternatively , we may be observing the tops of thick clouds that are constrained to the same temperature layer in both sources ( i.e. , t@xmath69 = constant ; @xcite ) . these possibilities should be explored with detailed modeling of resolved component spectra , rather than spectral templates . while thick clouds can explain the unusual faintness of 2mass j1728 + 3948a , thin clouds do not readily explain the unusual brightness of 2mass j0850 + 1057a . thin - cloud l dwarfs , also known as unusually blue l dwarfs ( ubls ) , do exhibit discrepancies between optical and near - infrared spectral types @xcite . there is also evidence that ubls have systematically larger t@xmath2s and/or absolute fluxes for their spectral classifications @xcite . however , 2mass j0850 + 1057a has a normal near - infrared color for its spectral type , and best - fit templates exhibit none of the spectroscopic hallmarks of a ubl ( e.g. , deep h@xmath6o and feh bands , blue near - infrared sed ; @xcite ) . it is therefore unlikely that this source is a ubl with unusually thin clouds . youth may play a role in the unusual brightness of 2mass j0850 + 1057a , arising from the enlarged radius of a still - contracting brown dwarf . the radius of a 1500 k brown dwarf is 25% larger at an age of 250 myr the minimum bound cited by @xcite than at 2 gyr , corresponding to an increase in brightness of roughly 0.5 mag . given the current uncertainties in the distance of this source , such a shift may be sufficient to move 2mass j0850 + 1057a back onto absolute magnitude / spectral type tracks . however , assuming that the two components of this system are coeval , 2mass j0850 + 1057b would also have to be overluminous by roughly the same factor , which does not appear to be the case . moreover , such a correction fails to explain the significant brightness difference between these two comparably - classified sources . we propose an alternative explanation : 2mass j0850 + 1057a is itself an unresolved , near - equal mass binary . such a scenario would explain how this component could appear both brighter and warmer than its equivalently - typed companion , but have an otherwise normal spectral energy distribution . the components of an equal - mass 2mass j0850 + 1057a pair would have absolute magnitudes , luminosities and t@xmath2s fully consistent with empirical trends ( table [ tab_components ] ) . the fact that existing high angular resolution images have not resolved this source requires an angular separation @xmath1150100 mas , or a projected separation @xmath1124 pc at this distance of this system . in fact , long - term dynamic stability requires an even tighter binary . a hierarchical triple is generally stable if the ratio of outer periastron and inner apastron distances : @xmath70 satisfies @xmath71 @xcite . here , @xmath72 @xmath73 1 and @xmath74 are the inner and outer mass ratios . assuming that all three components have nearly equal masses , and that the inner and outer orbits have eccentricities @xmath75 = 0 ( 0.5 ) , equation [ eqn_orbit ] requires a limit on the inner orbit semimajor axis of @xmath76 ( @xmath770.07@xmath78 ) and hence @xmath79 25 mas ( 8 mas ) , or 1 au ( 0.3 au ) , based on the semimajor axis determination of @xcite and the @xcite parallax . such a tight separation is not unusual for brown dwarf multiples ; brown dwarf spectroscopic binaries with comparable separations have already been identified ( e.g. , @xcite ) . moreover , three other tight , hierarchical , brown dwarf triple candidates identified in the literature gliese 569bcd , @xcite , denis j0205 - 1159abc @xcite , and kelu 1abc @xcite exhibit evidence that one component is an unresolved pair , based on radial velocity variations , psf - fitting residuals and spectroscopic features , respectively . this scenario is also consistent with orbital mass constraints from @xcite , as the estimated total mass of a triple 2mass j0850 + 1057 system , 0.15@xmath140.04 m@xmath12 , is closer to the mean ( but weakly constrained ) value of 0.2 m@xmath12 found in that study . indeed , tighter constraints on the total mass of this system from ongoing astrometric monitoring may affirm or refute the presence of a third body . the triple hypothesis can also be tested though radial velocity monitoring ; a @xmath115 km s@xmath22 ( @xmath118 km s@xmath22 ) line shift arising from a pair of 0.05 m@xmath12 brown dwarfs separated by 1.0 au ( 0.3 au ) can be readily detected with current near - infrared instrumentation @xcite . if 2mass j0850 + 1057a is confirmed as a binary , it would complete a remarkable , young , low - mass , hierarchical quintuple system with the double m dwarf nltt 20346ab , encompassing 4 orders of magnitude in separation and composed entirely of objects less massive than 0.15 m@xmath12 @xcite . we have presented photometric and spectroscopic analyses of the late - type l dwarf binaries 2mass j0850 + 1057 and 2mass j1728 + 3948 , aimed at assessing component spectral types , absolute magnitudes and near - infrared colors . multi - band hst / nicmos photometry have revealed distinct trends in the relative colors of these two systems , with 2mass j0850 + 1057b being redder than its primary and 2mass j1728 + 3948b being bluer . neither secondary exhibits narrow - band colors consistent with being a t dwarf . these results are borne out in spectral template fits , using nicmos and @xmath0 resolved photometry , which also determine component near - infrared spectral types of l7 + l6 for 2mass j0850 + 1057 and l5 + l6.5 for 2mass j1728 + 3948 . the early classification of 2mass j1728 + 3948a , its relative faintness at @xmath19 , and its unusually red color can be explained by the presence of thick condensate clouds in its photosphere . the secondary of this system , in contrast , may be losing its photospheric cloud deck as it transitions onto the t dwarf sequence . for 2mass j0850 + 1057 , the surprisingly later spectral type of its bright primary may stem from youth ( inflated radius ) and/or unresolved multiplicity . the latter hypothesis , which would make 2mass j0850 + 1057 part of a low - mass hierarchical quintuple , can be tested through ongoing astrometric monitoring and/or resolved spectroscopic monitoring to search for rv variations . as two resolved ( or partly - resolved ) coeval systems spanning the end of the l dwarf sequence and exhibiting a broad range of cloud properties , 2mass j0850 + 1057 and 2mass j1728 + 3948 remain important laboratories for studying cloud formation and evolution in low - temperature atmospheres . improved parallactic distance measurements including resolution of current distance discrepancies for 2mass j0850 + 1057resolved component spectroscopy , and ongoing photometric and astrometric monitoring will aid in characterizing the clouds , spectral properties and multiplicity of these benchmark brown dwarf systems . the authors would like to thank telescope operators dave griep and bill grolisch and instrument specialist john rayner for their assistance during the irtf observations . we acknowledge helpful comments from trent dupuy and jacqueline faherty on our original manuscript , and thank our referee sandy leggett for her prompt and helpful review . ajb acknowledges support from the chris and warren hellman fellowship ; dcbg acknowledges funding from the john reed fund . this publication makes use of data from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , and funded by the national aeronautics and space administration and the national science foundation . 2mass data were obtained from the nasa / ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . this research has also made use of the simbad database , operated at cds , strasbourg , france ; the m , l , and t dwarf compendium housed at dwarfarchives.org and maintained by chris gelino , davy kirkpatrick , and adam burgasser ; the spex prism spectral libraries , maintained by adam burgasser at http://www.browndwarfs.org/spexprism ; and the vlm binaries archive maintained by nick siegler at http://www.vlmbinaries.org . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . , h. , martn , e. l. , brandner , w. , forveille , t. , delfosse , x. , hulamo , n. , basri , g. , girard , j. , zapatero osorio , m .- r . , stumpf , m. , ghez , a. , valdivielso , l. , marchis , f. , burgasser , a. j. , & cruz , k. 2008 , , 481 , 757 , c. c. , harris , h. c. , vrba , f. j. , guetter , h. h. , canzian , b. , henden , a. a. , levine , s. e. , luginbuhl , c. b. , monet , a. k. b. , monet , d. g. , pier , j. r. , stone , r. c. , walker , r. l. , burgasser , a. j. , gizis , j. e. , kirkpatrick , j. d. , liebert , j. , & reid , i. n. 2002 , , 124 , 1170 , t. r. , knapp , g. r. , leggett , s. k. , fan , x. , golimowski , d. a. , anderson , s. , brinkmann , j. , csabai , i. , gunn , j. e. , hawley , s. l. , hennessy , g. , henry , t. j. , hill , g. j. , hindsley , r. b. , ivezi , . , lupton , r. h. , mcdaniel , a. , munn , j. a. , narayanan , v. k. , peng , e. , pier , j. r. , rockosi , c. m. , schneider , d. p. , smith , j. a. , strauss , m. a. , tsvetanov , z. i. , uomoto , a. , york , d. g. , & zheng , w. 2002 , , 564 , 466 , g. r. , leggett , s. k. , fan , x. , marley , m. s. , geballe , t. r. , golimowski , d. a. , finkbeiner , d. , gunn , j. e. , hennawi , j. , ivezi , z. , lupton , r. h. , schlegel , d. j. , strauss , m. a. , tsvetanov , z. i. , chiu , k. , hoversten , e. a. , glazebrook , k. , zheng , w. , hendrickson , m. , williams , c. c. , uomoto , a. , vrba , f. j. , henden , a. a. , luginbuhl , c. b. , guetter , h. h. , munn , j. a. , canzian , b. , schneider , d. p. , & brinkmann , j. 2004 , , 127 , 3553 , s. k. , geballe , t. r. , fan , x. , schneider , d. p. , gunn , j. e. , lupton , r. h. , knapp , g. r. , strauss , m. a. , mcdaniel , a. , golimowski , d. a. , henry , t. j. , peng , e. , tsvetanov , z. i. , uomoto , a. , zheng , w. , hill , g. j. , ramsey , l. w. , anderson , s. f. , annis , j. a. , bahcall , n. a. , brinkmann , j. , chen , b. , csabai , i. , fukugita , m. , hennessy , g. s. , hindsley , r. b. , ivezi , . , lamb , d. q. , munn , j. a. , pier , j. r. , schlegel , d. j. , smith , j. a. , stoughton , c. , thakar , a. r. , & york , d. g. 2000 , , 536 , l35 , d. 1986 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 627 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . d. l. crawford , 733+ , f. j. , henden , a. a. , luginbuhl , c. b. , guetter , h. h. , munn , j. a. , canzian , b. , burgasser , a. j. , kirkpatrick , j. d. , fan , x. , geballe , t. r. , golimowski , d. a. , knapp , g. r. , leggett , s. k. , schneider , d. p. , & brinkmann , j. 2004 , , 127 , 2948

the wigglez dark energy survey is a large - scale structure survey of intermediate - redshift uv - selected emission - line galaxies scheduled to cover 1000 deg , spanning a broad redshift range . the main scientific goal of the survey is the measurement of baryon acoustic oscillations ( bao ) in the galaxy clustering pattern at a significantly higher redshift than previous studies . the bao may be applied as a standard cosmological ruler to constrain dark energy models . based on the first of the dataset , we present initial results concerning the small - scale clustering of the wigglez targets , together with survey forecasts . the wigglez galaxy population possesses a clustering length mpc , which is significantly larger than uv - selected samples , with a slope . this clustering length is comparable to lyman break galaxies with similar uv luminosities . the clustering strength of the sample increases with optical luminosity , uv luminosity and reddening rest - frame colour . the full survey , scheduled for completion in 2010 , will map an effective volume gpc ( evaluated at a scale mpc ) and will measure the angular - diameter distance and hubble expansion rates in three redshift bins with accuracies . we will determine the value of a constant dark energy equation - of - state parameter , , with a higher precision than existing supernovae observations using an entirely independent technique . the wigglez and supernovae measurements lie in highly complementary directions in the plane of and the matter density . the forecast using the full combination of wigglez , supernovae and cmb datasets is a marginalized error , providing a robust and precise measurement of the properties of dark energy including cross - checking of systematic errors . surveys , cosmology : observations , large - scale structure of universe , galaxies : starburst

0901.2587

the large - scale structure of the universe is one of the pillars of our modern understanding of cosmology , encoding information about the contents and evolution of the universe , the physics of the growth of density fluctuations with time , and the formation and evolution of galaxies within the underlying network of dark matter haloes . in particular , the large - scale clustering pattern of galaxies is sensitive to the properties of the cosmic dark energy component which is currently poorly understood . dark energy influences both the rate of growth of structure and the geometrical distance - redshift relations . one of the cleanest probes of dark energy is to delineate as a function of redshift the apparent tangential and radial size of the baryon acoustic oscillation scale , a known `` standard ruler '' preferred separation imprinted into the galaxy distribution ( cooray et al . 2001 ; eisenstein 2002 ; blake & glazebrook 2003 ; seo & eisenstein 2003 ; hu & haiman 2003 ; linder 2003 ; glazebrook & blake 2005 ) . this cosmological probe is helping to motivate a new generation of massive spectroscopic galaxy surveys . cosmic structure has been mapped out by a succession of galaxy redshift surveys of increasing size and depth . the local universe ( redshifts @xmath16 ) has been studied in exquisite detail by the 2-degree galaxy redshift survey ( 2dfgrs ; colless et al . 2001 ) and the sloan digital sky survey ( sdss ; york et al . the sdss luminous red galaxy component extended this programme to a mean redshift @xmath17 using a specific type of tracer galaxy ( eisenstein et al . indeed , the cosmological conclusions reached should be independent of the galaxy type used , given that the `` bias '' with which galaxies trace the underlying dark matter fluctuations is expected to be a simple linear function on large scales ( coles 1993 ; scherrer & weinberg 1998 ) . in this sense , the choice of the `` tracer population '' of galaxies can be determined by observational considerations such as telescope exposure times , the availability of input imaging data for target selection , and secondary science goals . the wigglez dark energy survey , using the aaomega multi - object spectrograph at the 3.9 m anglo - australian telescope ( aat ) , is designed as the next leap forwards in redshift coverage , targetting the range @xmath18 . the survey is scheduled to cover a sky area of 1000 deg@xmath0 , mapping a cosmic volume @xmath19 gpc@xmath8 sufficient to measure the imprint of baryon oscillations in the clustering pattern at a significantly higher redshift than has been previously achieved by 2dfgrs ( cole et al . 2005 ; percival et al . 2007 ) and sdss ( eisenstein et al , 2005 ; huetsi 2006 ; percival et al . 2007 ; gaztanaga et al the survey redshift range is motivated by the optimal redshift location for testing a cosmological constant model in a spatially - flat universe ( parkinson et al . 2007 ) , which is the sensible initial hypothesis to reject in the dark energy parameter space . the target galaxy population is bright emission - line galaxies selected from uv imaging by the galaxy evolution explorer ( galex ) satellite ( martin et al . 2005 ) . this choice is motivated by the short ( 1-hr ) exposure times required to obtain redshifts at the aat . the survey commenced in august 2006 and is scheduled to finish in july 2010 , using the equivalent of 165 clear nights of telescope time ( 220 awarded nights ) . secondary science goals involve the study of star formation and galaxy evolution as a function of redshift and environment . in this initial study we focus on the small - scale clustering properties of the first @xmath2 of the wigglez sample . the clustering strength is an important parameter in the survey design and cosmological parameter forecasts : the signal - to - noise with which we can recover the galaxy power spectrum depends on the bias of the galaxies with respect to the dark matter fluctuations , which affects the balance between sample variance and shot noise in the power spectrum error budget . these initial clustering measurements allow us to determine the bias parameter and complete the survey forecast . furthermore , the joint uv - optical selection in the redshift interval @xmath20 places the wigglez survey in an interesting location in the parameter space of galaxy evolution . in this context , the clustering strength of a set of galaxies provides a direct indication of the density of the environment or ( equivalently ) the typical mass of the dark matter haloes hosting the galaxies . the clustering strength of uv - selected samples has been studied at low redshift @xmath21 ( milliard et al . 2007 ; heinis et al . 2007 ) and the corresponding rest - frame samples have been selected at much higher redshift @xmath22 through studies of the clustering of lyman break galaxies ( lbgs ; e.g. giavalisco & dickinson 2001 ; ouchi et al . 2001 ; arnouts et al . 2002 ; foucaud et al . 2003 ; adelberger et al . 2005 ; allen et al . 2005 ; ouchi et al . 2005 ; lee et al . 2006 ; yoshida et al . 2008 ) . the wigglez survey samples a redshift range which is intermediate to these previous studies . moreover , the clustering strength of optically - selected star - forming galaxies at high redshift has been studied over small areas by the deep extragalactic evolutionary probe ( deep2 ) project ( coil et al . 2008 ) and the vimos vlt deep survey ( vvds ; meneux et al . 2006 ) . wigglez is mapping an area @xmath23 times larger , and is therefore able to measure accurately the clustering strength of the most luminous star - forming galaxies , for which these smaller surveys are limited by small - number statistics and sample variance . the backdrop to these studies is the recent concept of `` down - sizing '' ( cowie et al . 1996 ; glazebrook et al . 2004 ; van dokkum et al . 2004 ) whereby the stars in more massive galaxies appear to have formed earlier , and the typical mass of the most actively star - forming galaxies is expected to decrease with time . a recent study of lbgs ( yoshida et al . 2008 ) has emphasized the importance of studying the clustering segregration with both uv and optical luminosities , which crudely trace ongoing star formation rate and stellar mass , respectively . wigglez is well - suited for undertaking such studies over the redshift range @xmath18 . the plan of this paper is as follows : in section [ secdata ] we introduce the wigglez survey strategy and target selection and describe the data sample used in this study . in section [ secanalysis ] we describe the methodology used to produce the small - scale clustering measurement including the generation of random ( unclustered ) realizations of the dataset with the correct selection functions and redshift completeness map . we also explain how we derive the statistical error in the clustering measurement . we present the clustering results in section [ secresults ] ( split by redshift , absolute magnitude and rest - frame colour ) together with initial comparisons to other studies . section [ secforecast ] contains the cosmological parameter forecasts for the full wigglez survey , and section [ secconc ] summarizes our conclusions . when converting redshifts to co - moving co - ordinates we assume a spatially - flat universe with cosmological parameters @xmath24 and @xmath25 . the design and implementation of the wigglez dark energy survey will be fully described in a forthcoming `` survey paper '' ( drinkwater et al . 2009 , in preparation ) which will accompany our mid - term data release . we include a brief outline here both for ease of reference and to emphasize the key points relevant to the small - scale clustering analysis . the wigglez survey strategy is to harvest low signal - to - noise spectra of a large number of uv - selected emission - line galaxies in relatively short exposure times ( 1-hr integrations at the aat ) . the survey tolerates a relatively low redshift completeness of @xmath26 but generates a large statistical sample of galaxy redshifts . the survey goal is to cover @xmath27 deg@xmath0 of the equatorial sky , gathering @xmath28 spectra of which @xmath29 are expected to yield successful redshifts . the survey was designed such that the average galaxy number density @xmath30 is related to the amplitude of the galaxy clustering power spectrum @xmath31 on the relevant baryon oscillation scales by @xmath32 , implying that the contributions of sample variance and shot noise to the clustering error are equal . this is the optimal survey strategy for fixed number of galaxies . the wigglez survey area , illustrated in figure [ figsurvey ] , is split into seven equatorial regions to facilitate year - round observing . we require that each region should possess a minimum angular dimension of @xmath33 deg , corresponding to a spatial co - moving scale that exceeds by at least a factor of two the standard ruler preferred scale [ which projects to ( 8.5 , 4.6 , 3.2 , 2.6 ) deg at @xmath34 . the survey coverage within individual regions should also be highly ( @xmath35 ) contiguous , otherwise the significance of the detection of the acoustic features is degraded by convolution with the survey window function . the survey duration is forecast to be @xmath36 clear nights between august 2006 and july 2010 , using the multi - object capability of the 2df positioner system coupled to the aaomega spectrographs ( saunders et al . 2004 ; sharp et al . 2006 ) . galaxy redshifts are obtained from the bright emission lines associated with star - forming galaxies , in particular redshifted [ oii ] 3727 , h@xmath37 4861 and [ oiii ] 4959 , 5007 . low - resolution ( 5 fwhm ) spectra are obtained spanning the ( observed - frame ) wavelength range @xmath38 , hence the majority of successful redshifts in the range @xmath39 are confirmed by multiple emission lines . single - line redshifts are almost invariably [ oii ] , for which we usually resolve the doublet in the range @xmath40 , increasing our confidence in the line identification . redshifts are obtained by visual inspection of each spectrum using the interactive software tool `` runz '' , and are classified by a quality flag @xmath41 , where the range @xmath42 denotes a `` reliable '' redshift ( see colless et al . the fraction of stellar contamination is very small ( @xmath43 ) and we find a similarly low fraction of high - redshift quasar interlopers . the galaxy continuum is typically detected with low signal - to - noise ( an average of @xmath44 per resolution element ) . wigglez targets are chosen by a joint uv - optical selection . the primary selection dataset is the medium imaging survey undertaken by the galaxy evolution explorer ( galex ) uv satellite , which provides typical exposure times of 1500 seconds in two filter bands , @xmath45 ( @xmath46 ) and @xmath47 ( @xmath48 ) . the galex point - spread function is too broad to allow for accurate placement of the spectrograph optical fibres , therefore the uv imaging is cross - matched with optical data . for our ngp regions , the sdss imaging data are used . for our sgp regions , the sdss @xmath49 stripes are too narrow compared to the preferred baryon oscillation scale hence we use imaging data from the second red cluster sequence ( rcs2 ) project instead ( yee et al . 2007 ) . sources are cross - matched between the galex and optical catalogues with a matching tolerance of 2.5 arcsec ( which produces a negligible fraction of incorrect matches ) . in each imaging dataset , the majority of galaxies possess relatively low signal - to - noise ( @xmath50 ) but their detection in both datasets ensures a robust sample . we note that acquisition of our eventual requirement of @xmath51 galex orbits of data is still ongoing . about @xmath26 of this total had been obtained at the end of 2008 . targets are chosen from the uv - optical matched sample using a series of magnitude and colour cuts . these cuts are tuned to optimize the fraction of targets lying at high redshift @xmath52 . firstly the galaxy magnitudes are de - reddened using standard dust corrections based on the local value of @xmath53 measured from the schlegel , finkbeiner & davis ( 1998 ) dust maps . the primary galex selection criterion is a red @xmath54 colour ( @xmath55 or @xmath45 drop - out ) , motivated by the lyman break passing through the @xmath45 filter for @xmath52 , and tuned by looking at galaxy templates . at the depth of the medium imaging survey this colour is noisy , resulting in a significant amount of contamination by low - redshift ( @xmath56 ) galaxies which are partially removed by the additional cuts described below . we also impose a faint uv magnitude limit @xmath57 and an additional signal - to - noise requirement @xmath58 for the detected nuv flux ( which becomes relevant for fields with unusually high dust content or low exposure time ) . the galex field - of - view is circular with radius @xmath59 deg ; we only select sources within the central @xmath60 deg because of concerns over the photometry at the edge of the field . our primary optical selection cuts are derived from sdss @xmath61-band imaging . we require a uv - optical colour in the range @xmath62 based on the expected model tracks of star - forming galaxies . we impose a bright @xmath61-band limit @xmath63 ; the uv - optical colour cut implies that the median optical magnitude of our targets is @xmath64 . finally we increase the high - redshift efficiency by introducing optical colour cuts . different cuts are used for the sdss and rcs2 regions , governed by the available imaging bands and depths . for the sdss regions analyzed in this paper , we apply cuts for those ( brighter ) galaxies with good detections in the sdss @xmath65- and @xmath66-bands . specifically , for targets with @xmath67 and @xmath68 we reject galaxies in the colour space defined by @xmath69 and @xmath70 which is occupied by low - redshift galaxies both theoretically and empirically ( more details will be given in drinkwater et al . 2009 , in preparation ) . the final fraction of @xmath71 galaxies obtained is @xmath72 . the redshift distribution is displayed in figure [ fignz ] . an average of 3 - 4 pointings of the 2df spectrograph per patch of sky is required in order to achieve the required target density of 350 deg@xmath73 . for any observing run the optimal placement of field centres ( based on the current availability of targets ) is achieved using the `` metropolis '' ( simulated annealing ) algorithm ( campbell , saunders & colless 2004 ) . galaxies are prioritized for spectroscopic follow - up on the basis of optical @xmath61-band magnitude , in the sense that fainter targets are observed first . the motivation for this strategy is to combat the potential inefficiency of `` mopping up '' residual galaxies in the final pointing for any patch of sky : the brighter remaining galaxies can be observed in a shorter exposure time by configuring fewer fibres . in this paper we analyze the subset of the wigglez sample assembled from our first observations in august 2006 up until the end of the 08a semester ( july 2008 ) . at this point we had utilized 108 of our allocated nights , of which the equivalent of 70 nights were clear . the available galaxy database included @xmath74 reliable ( @xmath75 ) wigglez unique galaxy redshifts . in this analysis we only use those galaxies lying in the sdss regions of our optical imaging because work is still on - going on the rcs2 portion of the angular selection function . specifically , we include the wigglez 9-hr , 11-hr , 15-hr and 0-hr ( sdss ) regions illustrated in figure [ figsurvey ] . the number of existing aaomega pointings in these regions is ( 42 , 98 , 140 , 48 ) . the numbers of galaxies in each region with reliable redshifts satisfying the final survey selection criteria are ( 5782 , 14873 , 21629 , 4383 ) , constituting a total sample of @xmath76 for this initial analysis ( about @xmath2 of the final sample ) . we quantify the small - scale clustering of the galaxy distribution using a standard set of techniques based on the 2-point correlation function . this statistic compares the number of observed close galaxy pairs to that expected by random chance , as a function of spatial separation . the key requirement is an ensemble of random ( unclustered ) realizations of the survey possessing the same selection function as the observed galaxy distribution . with this in place we convert the data ( d ) and random ( r ) galaxy angle - redshift distributions into a grid of co - moving co - ordinates @xmath77 using an assumed cosmological model ( we use a flat model with @xmath24 ) . we then bin the number of data - data ( dd ) , data - random ( dr ) and random - random ( rr ) pairs in the two - dimensional space of separation perpendicular to the line - of - sight ( denoted by @xmath78 ) and parallel to the line - of - sight ( denoted by @xmath79 ) . this decomposition is motivated by the influence of galaxy peculiar velocities ( redshift - space distortions ) which shift galaxies in @xmath79 , but not in @xmath78 . each of our random realizations contains the same number of targets as the data sample , and is generated by a method described below . the pair counts @xmath80 and @xmath81 are determined by averaging over 10 random realizations . the 2d redshift - space correlation function @xmath82 is derived using the estimator proposed by landy & szalay ( 1993 ) : @xmath83 ( where this last equation assumes an equal number of data and random galaxies ) . we bin galaxy pairs by the absolute value of the line - of - sight separation , i.e. @xmath84 . the `` real - space '' correlation function ( independent of the redshift - space distortion ) can be obtained by summing equation [ eqxiest ] over @xmath79 . we first define the projected correlation function @xmath85 ( davis & peebles 1983 ) : @xmath86 where the factor of 2 extrapolates the result to the full range @xmath87 . if we assume that the real - space correlation function @xmath88 is well - described by a power - law @xmath89 , where @xmath90 is the clustering length , @xmath91 is the slope and @xmath92 , and if we neglect the coherent infall velocities described below , we can then derive @xmath93 where @xmath94 the difficulty with this method is that the measurement of @xmath82 becomes noisy for large @xmath79 and therefore the summation in equation [ eqxiproj ] must be truncated at some @xmath95 , invalidating equation [ eqxireal ] . we therefore adopted the following approach ( similar to the methodology of coil et al . 2008 ) to convert a model real - space correlation function @xmath89 into a projected correlation function which may be compared with the data . in the linear regime , the effect of coherent infall velocities on the correlation function can be modelled by @xmath96 where @xmath97 are the legendre polynomials , @xmath98 and @xmath99 is the angle between @xmath61 and @xmath79 . for a power - law real - space correlation function , @xmath100 \xi_r(r)\end{aligned}\ ] ] where @xmath101 is the redshift - space distortion parameter ( hamilton 1992 ; hawkins et al . 2003 ) and @xmath102 is the linear galaxy bias parameter . we assumed @xmath103 for this model , consistent with our measurements ( see section [ secreddist ] ) , and for each set of trial values @xmath104 we employed the above set of equations to calculate @xmath82 . for each value of @xmath78 we then integrated this function in the @xmath79 direction up to @xmath105 in order to compare with the correlation function measurements . we assumed @xmath106 mpc , and we checked that our results did not depend sensitively on the value of @xmath107 . we treated each of the four survey regions independently , measuring the correlation function and corresponding error . we then constructed the `` combined '' correlation function by averaging the measurements in the four regions with inverse - variance weighting . for convenience , we plot projected correlation functions in this paper as @xmath108 , where @xmath109 we now discuss the generation of the random survey realizations that are required for calculation of the correlation function . this determination of the survey `` selection function '' will be described fully in a forthcoming paper ( blake et al . 2009 , in preparation ) and we give a brief summary here . the calculation begins with the angular selection function of the `` parent '' sample of uv - optical matches . this function is defined firstly by the boundaries of the galex fields and sdss coverage map . secondly , because the uv magnitudes of our sample lie close to the threshold of the galex mis observations , there is a significant incompleteness in the galex imaging that depends on the local dust extinction and galex exposure time . we used the galex number counts as a function of dust and exposure time to calibrate the relation between these quantities and the parent wigglez target density . this angular completeness function is displayed in figure [ figangsel ] for the four survey regions analyzed in this paper . we used this map to produce a series of random realizations of the parent catalogue for each region . the next step is to process these random parent catalogues into random realizations of the redshift catalogue . the spectroscopic follow - up of the parent catalogue comprises a network of overlapping aaomega pointings , with field centres optimized by the simulated annealing algorithm and not lying on a regular grid . the fraction of successful redshifts in each pointing varies considerably depending on weather conditions . furthermore , the redshift completeness within each aaomega field exhibits a significant radial variation due to acquisition errors at the plate edges . in figure [ figredcomp ] we illustrate how the redshift completeness varies across these survey regions by simply taking the ratio of successful redshifts to parent galaxies in each pixel . this is a useful visualization , but in fact the number of unique sectors defined by the overlapping aaomega fields is so large that this determination of the redshift completeness map is very noisy . indeed , some unique sectors contain zero parent galaxies . one possible approach is to smooth this completeness map over larger areas to reduce the poisson noise at the expense of a diminished sensitivity to small - scale completeness variations between aaomega pointings . in this analysis we use an alternative approach , which is to apply the aaomega pointing sequence to each of the random realizations of the parent catalogue , and thereby create an ensemble of random realizations of the redshift catalogue . the aaomega pointing sequence is defined by the right ascension and declination of the field centre together with the number of successful and unsuccessful redshifts obtained for that pointing . within each field centre parent galaxies are chosen randomly to create the synthetic redshift catalogue . it is also necessary to track the sky coverage of the galex data which was contemporaneous with each aaomega pointing . because the acquisition of the galex imaging data is ongoing with the spectroscopic follow - up , the boundaries of the angular mask must be modulated in step with the redshift follow - up . in addition we impose the radial redshift completeness variation across each aaomega field , measured independently for each observing run . the redshift distribution @xmath110 of observed galaxies varies with position in the sky . this is due to the magnitude prioritization described in section [ sectarsel ] . because galaxies with fainter @xmath61-band magnitudes are targeted first , the @xmath110 will be skewed toward higher redshifts for areas of the survey which have been targeted by fewer aaomega observations . this dependence is accounted for in our random catalogues by measuring the magnitude distribution of targeted galaxies as a function of sky position and drawing a random redshift from the correctly weighted @xmath110 . the optical fibres of the 2df spectrograph can not be placed closer together than @xmath111 arcmin , and there is a diminishing probability of observing in a single pointing both members of a close pair of parent galaxies separated by an angular distance of less than 2 arcmin [ a projected spatial distance of @xmath112 mpc at @xmath34 . this restriction will eventually be ameliorated by the requirement of observing each patch of sky with 3 - 4 aaomega pointings to build up the number density of the redshift catalogue . at present , however , there is a deficit of close angular pairs in the redshift catalogue , which artificially suppresses the measured value of the galaxy correlation function on small scales . the close angular pair deficit is illustrated in figure [ figfibcol ] by plotting the ratio @xmath113 as a function of angular separation @xmath99 , where @xmath114 and @xmath115 are the angular correlation functions of the targeted catalogue and the parent catalogue , respectively . this ratio provides the fraction of surviving close pairs . in order to correct the galaxy correlation function for the missing close pairs we increased the contribution of each galaxy pair to the data - data pair count as a function of angular separation by a factor @xmath116 ( the inverse of the quantity plotted in figure [ figfibcol ] ) using a 2-parameter model @xmath117\}/2 $ ] fitted to the data in figure [ figfibcol ] . we note that for a survey with a redshift - dependent galaxy number density @xmath118 , the minimum - variance correlation function measurement for separation @xmath119 is achieved if galaxies are assigned a redshift - dependent weight @xmath120^{-1}$ ] where @xmath121 ( efstathiou 1988 ; loveday et al . 1995 ) . in our case the galaxy number density is sufficiently low that @xmath122 and this weighting makes a negligible difference to the results and we do not use it . the low signal - to - noise spectra obtained by the wigglez survey imply that a small but significant fraction of galaxies are assigned a `` reliable '' ( @xmath42 ) redshift which proves to be incorrect owing to emission - line mis - identification . this is monitored in the survey by allocating a small number of fibres ( typically 3 to 5 out of 400 per pointing ) to re - observe galaxies with existing @xmath42 redshifts . the fraction of repeat observations producing a discrepant redshift may be used to estimate the redshift `` blunder '' rate . there is a significant difference in the reliability of @xmath123 redshifts and @xmath124 redshifts . @xmath123 redshifts ( which represent a fraction @xmath125 of reliable redshifts ) are typically based either on noisy spectra or on a single emission line with no confirming spectral features , whereas @xmath124 redshifts are based on multiple detected emission lines . comparing repeat observations consisting of a @xmath123 redshift and a @xmath124 redshift , assuming that the latter provides the correct redshift identification , we conclude that @xmath126 of @xmath123 redshifts are blunders . this agrees with the internal discrepancy rate amongst repeated pairs of @xmath123 redshifts ( which is @xmath127 , which must be divided by two to obtain the blunder rate per object ) . comparing repeat observations consisting of @xmath124 redshifts we find that only @xmath128 of these redshifts are blunders . the blunder rate for @xmath123 spectra varies significantly with the true galaxy redshift , which determines how many emission lines appear in the observed wavelength range . the dependence is displayed in figure [ figzblund ] based on the comparison of @xmath123 and @xmath129 pairs of repeat observations . the total blunder rate for all reliable ( @xmath42 ) redshifts is below @xmath130 for the range @xmath131 , increasing to @xmath2 by @xmath132 . the redshift blunder rate does not depend on galaxy continuum magnitude . redshift mis - identification reduces the measured value of the galaxy correlation function because a fraction of true close data - data pairs are lost as one or both of the redshifts is randomized . if @xmath133 is the redshift blunder rate , the correction to the correlation function is a constant factor @xmath134 assuming the blunder redshift is randomly distributed . we applied this correction to the measured correlation function to deduce the final value : @xmath135 when measuring the galaxy correlation function for a particular redshift or luminosity range , we re - calculated the redshift blunder rate for the corresponding sample in each region as explained below . we corrected the correlation function for that region using equation [ eqcorrblund ] , before combining together the correlation functions for the different regions . we determined the redshift blunder rate for each region by weighting the blunder probabilities of the @xmath136 individual objects in that region : @xmath137 for objects with @xmath123 we assigned the probability for each object based on its redshift using figure [ figzblund ] . for objects with @xmath124 we assumed a blunder rate of @xmath138 . in order to determine the error in the measured correlation function we must characterize the statistical fluctuations in the data sample . it is well - known that these fluctuations are not well - described by poisson statistics , for which the error in the pair count @xmath139 in a separation bin would be equal to @xmath140 . sample variance , geometrical edge effects and the same galaxy participating in pairs in different separation bins cause the statistical variance of the galaxy pair count to exceed the poisson prediction and induce covariances between the bins . in this analysis we use jack - knife re - sampling to determine the correlation function error . in this technique the dataset is divided into @xmath136 equal - area sub - regions on the sky . the correlation function analysis is repeated @xmath136 times , in each case omitting one of the sub - regions in turn . labelling the different correlation function measurements at separation @xmath119 as @xmath141 from @xmath142 to @xmath143 , the covariance between separation bins @xmath144 and @xmath145 was deduced as : @xmath146 where @xmath147 . the factor @xmath148 in equation [ eqcov ] is required because the jack - knife re - samples are not independent , sharing a high fraction of common sources . we defined the jack - knife samples by splitting each survey region into @xmath149 sub - regions using constant boundaries of right ascension and declination . we tried the alternative technique of using the galex tiles to define the jack - knife regions ; this produced a result that did not differ significantly . future analyses of the wigglez survey clustering will quantify the statistical fluctuations using mock galaxy catalogues constructed from n - body simulations . figure [ figdist2d ] illustrates the dependence of the 2d redshift - space correlation function @xmath82 of equation [ eqxiest ] on the separations @xmath79 and @xmath78 perpendicular and parallel to the line - of - sight for the sample of wigglez galaxies spanning the full redshift range @xmath150 . we measured the correlation function separately for the four independent survey regions and combined the results using inverse - variance weighting . the non - circularity of the contours of constant @xmath151 trace the imprint of galaxy peculiar velocities ; we use linear scales of @xmath78 and @xmath79 in this plot to focus on the large - scale distortions . in particular , for scales @xmath152 mpc the increase in the value of @xmath151 with increasing angle to the line - of - sight @xmath153 for fixed total separation @xmath154 is a signature of coherent galaxy infall and can be quantified to measure the redshift - space distortion parameter @xmath37 ( see section [ secreddist ] ) . we also detect some evidence for `` fingers of god '' , in the form of elongation of the contours of @xmath151 along the @xmath79-axis , due to the virialized motions of galaxies in clusters . there is some similarity here with the results of coil et al . ( 2008 ) figure 7 for luminous blue galaxies , except that our sample size is significantly larger . galaxy peculiar velocities change values of @xmath79 but not @xmath78 . the real - space clustering properties of the galaxies may therefore be deduced by integrating @xmath82 along the @xmath79-axis , as discussed in section [ seccorrest ] . we summed the 2d correlation function for the @xmath150 sample in 5 logarithmic bins of @xmath79 between @xmath155 mpc and @xmath156 mpc . the result is plotted in figure [ figprojall ] for the projected separation range @xmath157 mpc , with errors obtained from the jack - knife re - sampling . the full covariance matrix @xmath158 deduced from the jack - knife re - samples is displayed in figure [ figcov ] by plotting in greyscale the correlation coefficient between two separation bins @xmath66 and @xmath144 : @xmath159 we employed the methodology of section [ seccorrest ] to fit a power - law real - space correlation function @xmath160 to the redshift - space data over the range @xmath161 mpc , by minimizing the @xmath162 statistic using the covariance matrix : @xmath163 where @xmath164 is the vector of offsets between the data and the model , and @xmath165 is the inverse of the covariance matrix . the fitting range was motivated by our wish to estimate the clustering length @xmath90 for which @xmath166 . a power - law provides a good fit to the data with a best - fitting @xmath167 ( for 8 degrees of freedom ) . the marginalized measurements of the power - law parameters are @xmath168 mpc and @xmath5 for the @xmath150 sample . we compare these measurements to previous studies of uv - selected and optically - selected galaxies in section [ secprevious ] . in figure [ figprojreg ] we plot the separate projected correlation function measurements for each of the four survey regions analyzed in this paper . the four regions give consistent results . in order to derive the bias factor of the galaxies with respect to dark matter we generated a model non - linear matter power spectrum at @xmath4 assuming a flat cosmological model with fiducial parameters @xmath169 , @xmath170 , @xmath171 ( where @xmath172 km s@xmath10 mpc@xmath10 ) and @xmath173 , using the `` camb '' software package ( lewis , challinor & lasenby 2000 ) including corrections for non - linear growth of structure using the fitting formula of smith et al . we used this model power spectrum to determine the non - linear matter correlation function @xmath174 at @xmath4 . the resulting correlation function satisfied @xmath175 for @xmath176 mpc , which we assumed as our estimate of @xmath177 , the clustering length of dark matter at @xmath4 . given that the overall amplitude of the power spectrum scales with redshift in the linear regime as @xmath178 , where @xmath179 is the linear growth factor , we can approximate : @xmath180 where @xmath181 . hence the linear bias factor @xmath102 of a population of galaxies with clustering length @xmath90 can be approximated as : @xmath182 our measured clustering length @xmath183 mpc for a sample at median redshift @xmath184 is hence equivalent to a linear bias factor @xmath185 . the peculiar velocities generated by large - scale coherent infall can be parameterized by @xmath101 where @xmath102 is the linear bias parameter ( kaiser 1987 ) . for a flat cosmological constant model with @xmath186 , @xmath187 , and our real - space clustering measurement @xmath188 hence predicts @xmath103 at the median redshift of the sample . the purpose of this section is to demonstrate that our data contains this self - consistent signal of peculiar velocities ( we leave detailed fits for @xmath37 to a further study ) . we may quantify the imprint of peculiar velocities by measuring the quadrupole moment , @xmath189 , of the 2d correlation function ( hamilton 1992 ) . this statistic quantifies the anisotropy evident in figure [ figdist2d ] . if we define the correlation function moment @xmath190 for multipole @xmath191 as : @xmath192 we can then show that @xmath193 - \xi_0(s ) } = \frac { \frac{4}{3 } \beta + \frac{4}{7 } \beta } { 1 + \frac{2}{3 } \beta + \frac{1}{5 } \beta^2 } \label{eqdistq}\ ] ] which is valid for large scales @xmath194 mpc . figure [ figdistq ] plots the measured quantity @xmath189 as a function of separation @xmath119 , together with the prediction of equation [ eqdistq ] for various values of @xmath37 . in order to construct the quantity @xmath189 we measured the 2d redshift - space correlation function in bins of @xmath119 and @xmath195 , and summed over @xmath195 , weighting in accordance with equation [ eqxil ] . the result is consistent with our estimate @xmath196 and constitutes a statistically - significant detection of redshift - space distortions in our sample . our sample of wigglez galaxies is large enough for us to analyze the dependence of the clustering length @xmath90 on redshift , galaxy luminosity and colour . the situation is complicated by our joint uv - optical selection and strong luminosity - redshift correlation , but we can make some comparisons with previous studies . we fix the correlation function slope @xmath197 in this section of the analysis . the variation of the clustering length with redshift is plotted in figure [ figr0vsz ] , dividing all wigglez galaxies in the range @xmath198 into redshift bins of width @xmath199 . the clustering length is roughly constant with redshift for the range @xmath200 , with a trend to a reduced clustering strength at low redshifts . our interpretation of the overall constancy of @xmath201 is that it is a product of two cancelling effects . galaxy luminosity increases with redshift , which would tend to increase clustering length , but at redshifts @xmath52 optically red galaxies , which possess enhanced clustering strengths , are removed from the sample by the optical colour cuts described in section [ sectarsel ] . we also analyzed the clustering in absolute magnitude and rest - frame colour bins . we considered the clustering as a function of rest - frame @xmath45-band and @xmath202-band absolute magnitudes , which are well - matched in wavelength ( for redshift @xmath203 ) to the observed - frame @xmath47-band and @xmath61-band magnitudes which are used to define our target samples . for this initial analysis we assumed redshift - dependent average k - corrections which we applied to all galaxies regardless of colour . these k - corrections were derived using the spectral energy distribution of a lyman break galaxy including an intrinsic dust contribution @xmath204 , which produces a very good match to the redshift - dependence of the average observed @xmath205 colour of the wigglez targets . we note that the @xmath45-band and @xmath202-band absolute magnitudes of our target sample correlate strongly with redshift . this is depicted by figure [ figselbox ] which plots the target selection box in @xmath206 for 4 different redshifts , also indicating the characteristic absolute magnitudes @xmath207 at each redshift obtained from arnouts et al . ( 2005 ) and willmer et al . ( 2006 ) . between @xmath208 and @xmath132 the average value of @xmath209 brightens by 2 magnitudes ( becoming positive at @xmath210 ) and the average value of @xmath211 brightens by 4 magnitudes ( becoming positive at @xmath212 ) . the dependence of the clustering length @xmath90 of the @xmath150 wigglez sample on @xmath213 , @xmath214 and @xmath215 is displayed in the panels of figure [ figr0panel ] . these measurements show that the clustering strength of the sample increases steadily with @xmath202-band absolute magnitude , @xmath45-band absolute magnitude and reddening @xmath215 colour . sub - samples have values of @xmath90 ranging from @xmath216 mpc to @xmath217 mpc . figure [ figr0vsmb ] plots the variation of @xmath90 with @xmath213 for the low - redshift and high - redshift halves of the dataset , divided at @xmath218 . this measurement confirms that at fixed @xmath213 , the clustering length of the sample drops slightly with redshift as the redder galaxies are removed by the colour cuts . coil et al . ( 2008 ) present clustering measurements as a function of galaxy colour and luminosity for the deep2 galaxy redshift survey , which has measured redshifts for @xmath219 galaxies in the range @xmath220 over an area of 3 deg@xmath0 . the deep2 subset of luminous blue galaxies ( coil et al . table 2 , line 5 ) has best - fitting clustering parameters @xmath221 mpc and @xmath222 at @xmath223 ( for a galaxy density @xmath224 mpc@xmath225 and median absolute magnitude @xmath226 assuming @xmath227 ) . these results lie in good agreement with ours . milliard et al . ( 2007 ) and heinis et al . ( 2007 ) present clustering analyses of galex - selected samples . at low redshift ( @xmath228 ) the clustering strength of the uv - selected sample is @xmath229 mpc , corresponding to low - density environments , and shows no dependence on uv luminosity ( indeed , there is tentative evidence for an anti - correlation between @xmath90 and luminosity ) . these results may naturally be compared to clustering measurements of @xmath22 lbgs also selected at rest - frame uv wavelengths ( e.g. giavalisco & dickinson 2001 ; ouchi et al . 2001 ; arnouts et al . 2002 ; foucaud et al . 2003 ; adelberger et al . 2005 ; allen et al . 2005 ; ouchi et al . 2005 ; lee et al . 2006 ; yoshida et al . these results show a qualitatively different conclusion : lbgs are highly clustered and concentrated in overdense regions . furthermore , the clustering strength for galaxies brighter than @xmath230 increases with @xmath45 luminosity , reaching @xmath231 mpc for the most luminous sub - samples . yoshida et al . ( 2008 ) demonstrate that the behaviour of the clustering length @xmath90 depends on a combination of uv and optical luminosities : galaxies bright in optical magnitudes are strongly clustered irrespective of uv magnitude , whereas galaxies faint in optical magnitude have correlation lengths increasing with uv luminosity ( see yoshida et al . fig.15 ) . in figure [ figr0compare ] we overplot the clustering measurements of the @xmath150 wigglez sample as a function of @xmath45 absolute magnitude on the compilation of low - redshift and high - redshift clustering measurements presented by heinis et al . ( 2007 ) . at low @xmath45 absolute magnitudes @xmath232 the clustering strengths of the different uv - selected samples agree well . this absolute magnitude range corresponds to low redshifts @xmath228 in the wigglez sample ( figure [ figselbox ] ) for which we recover a clustering length @xmath233 mpc , similar to samples of low - redshift quiescent star - forming galaxies . at higher @xmath45 luminosities and redshifts , the wigglez clustering strength is more comparable to @xmath6 lbgs rather than @xmath4 uv - selected galaxies . this is expected as the @xmath234 wigglez selection cut becomes effective for @xmath235 and the nature of the resulting wigglez galaxy population changes to merger - induced starbursts . the wigglez sample does not recover the very high values of @xmath90 present in very luminous lbgs at @xmath6 ; we suggest that this may be a result of the wigglez colour cuts selecting against redder galaxies . the clustering amplitude of the wigglez target sample is a required input for forecasting the accuracy with which the full 1000 deg@xmath0 survey will measure the large - scale galaxy power spectrum . the error in the power spectrum measurement @xmath236 at a given redshift @xmath237 and fourier wavenumber @xmath145 is determined by the quantity @xmath238 , where @xmath118 is the galaxy number density and @xmath239 is the galaxy power spectrum amplitude . this quantity fixes the balance between sample variance and shot noise in the measurement error such that @xmath240 where @xmath241 is the total number of independent fourier modes contributing towards the measurement ( e.g. feldman , kaiser & peacock 1994 ; tegmark 1997 ) . the contributions of sample variance and shot noise are equal when @xmath242 . we model the angle - averaged redshift - space linear galaxy power spectrum as a function of @xmath145 and @xmath237 as : @xmath243^\gamma \left ( 1 + \frac{2\beta}{3 } + \frac{\beta^2}{5 } \right ) \label{eqpkmod}\ ] ] where we assume @xmath244 mpc , @xmath245 mpc , @xmath246 and @xmath103 . the second term on the right - hand - side of equation [ eqpkmod ] describes the boost from the galaxy linear bias factor @xmath102 ( equation [ eqbias ] ) using the relation @xmath247 . the third term is the result of redshift - space distortions averaged over angles . we used the cosmological parameters as listed in section [ secr0 ] to produce the @xmath4 dark matter power spectrum : @xmath169 , @xmath170 , @xmath171 and @xmath248 . in order to incorporate the fraction of redshift blunders @xmath249 we reduced the effective value of the power spectrum by a factor @xmath250 [ i.e. increased the value of @xmath251 by a factor @xmath252 . figure [ fignpvsz ] plots the dependence of @xmath238 on redshift for a set of different scales @xmath253 mpc@xmath10 relevant for the detection of baryon acoustic oscillations , assuming a source redshift distribution combining the survey regions plotted in figure [ fignz ] . we further assume a total target density of 350 deg@xmath73 with a @xmath26 redshift completeness . we note that over a significant range of redshifts and scales our large - scale power spectrum measurement will be limited by sample variance rather than shot noise , i.e. @xmath254 . a useful quantity to describe the survey is the scale - dependent `` effective volume '' @xmath255 which is defined by @xmath256 ^ 2 \frac{dv}{dz } \ , dz\ ] ] where @xmath257 is the co - moving volume element . the effective volume represents an optimally - weighted stacking of power spectrum measurements at different redshifts ( tegmark 1997 ) . for scales @xmath258 mpc@xmath10 we find @xmath259 gpc@xmath8 . thus the survey design will achieve the goal of mapping @xmath260 gpc@xmath261 gpc@xmath8 . we can use the effective survey volume to forecast the error in the final survey power spectrum @xmath262 in a fourier bin of width @xmath263 ( tegmark 1997 ) : @xmath264 this prediction is plotted for bins of width @xmath265 mpc@xmath10 in figure [ figpksim ] , in which we divide the power spectrum by the `` no - wiggles '' reference spectrum provided by eisenstein & hu ( 1998 ) in order to delineate clearly the baryon acoustic oscillations . we also generated 100 monte carlo realizations of the final 1000 deg@xmath0 survey using the methods described in blake & glazebrook ( 2003 ) and glazebrook & blake ( 2005 ) . the scatter in the power spectrum measurements across the realizations was very close to that predicted by equation [ eqpkerr ] . we used these monte carlo realizations to assess the accuracy with which the full wigglez survey will measure the tangential and radial standard ruler scale imprinted by the baryon acoustic oscillations via the fitting formula described in blake et al . ( 2006 ) . restricting ourselves to the @xmath266 subset , and first considering an `` angle - averaged '' measured power spectrum @xmath267 , we found that the scatter in the fitted acoustic wavescale was @xmath268 . measuring instead a 2d power spectrum @xmath269 , where @xmath270 and @xmath271 are wavevectors measured perpendicular and parallel to the line - of - sight , the scatters in the tangential and radial fitted wavescales were @xmath272 and @xmath273 , respectively . this latter pair of measurements corresponds to the accuracy of determination of the quantities @xmath274 and @xmath275 at an effective redshift @xmath276 , where @xmath277 is the angular diameter distance , @xmath278 is the high - redshift hubble constant , and @xmath119 is the sound horizon at recombination , i.e. the standard ruler scale . dividing the survey into redshift slices we find that the angle - averaged wavescale may be measured with accuracy @xmath279 in redshift slices @xmath280 . the angle - averaged wavescale measures a quantity proportional to @xmath281 , as discussed by eisenstein et al . ( 2005 ) . these forecasts should be considered a pessimistic lower limit on expected performance for two reasons . firstly we have neglected the cosmological information contained in the overall shape of the galaxy power spectrum , which is divided out in the above analysis to focus on the `` standard ruler '' aspect of the acoustic oscillations . this method produces robustness against systematic errors ( which are expected to affect the shape of the power spectrum but not the oscillatory signature ) . the power spectrum shape carries information about @xmath282 and @xmath283 which further breaks the degeneracy in cosmological distances between these two parameters and the dark energy . secondly we have neglected the improvement offered by `` reconstruction '' of the density field , which sharpens the measurement of the acoustic signature by un - doing ( to first order ) the large - scale coherent galaxy motions which smooth out the acoustic peaks ( eisenstein et al . 2007 ) . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] [ tabmodel ] we investigated improved forecasts using the methodology of seo & eisenstein ( 2007 ) which properly incorporates information from the power spectrum shape , redshift - space distortions and density - field reconstruction . the predicted tangential and radial measurement accuracies for the @xmath266 sample are @xmath284 and @xmath285 , respectively ( and are correlated with a correlation coefficient @xmath286 , further enhancing the power to constrain the cosmological model ) . we assume here that reconstruction can improve the parameters ( @xmath287 , @xmath288 ) defined by seo & eisenstein ( 2007 ) by a factor equal to @xmath289 ( eisenstein , priv . comm . ) . dividing the survey into redshift slices we find that the tangential and radial wavescales may be measured with accuracies @xmath290 for @xmath291 , @xmath292 for @xmath293 and @xmath294 for @xmath295 . this information is collected in table [ tabmodel ] for ease of reference . we used this last set of forecasts with reconstruction in 3 redshift bins to determine the expected accuracy of measurement of a constant equation - of - state @xmath13 of dark energy ( assuming the measurements of @xmath277 and @xmath296 are correlated with coefficient @xmath297 ) . confidence ellipses are displayed in figure [ figomw1 ] in the space of @xmath13 and the matter density @xmath282 for a flat cosmology with fiducial model @xmath298 and @xmath299 . results are shown for each redshift bin separately and for the combination of all 3 bins . in order to generate this figure we have used the 5-year - wmap measurement of the cmb acoustic scale @xmath300 ( komatsu et al . 2009 ) in order to cancel the dependence of the baryon oscillation measurement on the sound horizon at recombination . in figure [ figomw1 ] we have not included any further cmb information or other external datasets . the marginalized errors are @xmath301 and @xmath302 . in figure [ figomw2 ] we add in information from the 5-year - wmap measurement of the cmb shift parameter @xmath303 ( komatsu et al . 2009 ) , including the correlation between @xmath304 and @xmath305 , together with the latest supernovae data from the essence , snls and hst observations ( see wood - vasey et al . 2007 , astier et al . 2006 , riess et al . 2007 , davis et al . 2007 ) . the marginalized errors in the cosmological model from the full combination of datasets are @xmath306 and @xmath307 . the forecast performance of the wigglez survey exceeds that of the current cmb and supernovae data , but the different measurements are also complementary , breaking degeneracies in the @xmath308-plane through independent techniques . disagreement between any pair of datasets would produce the possibility of discovering non - standard physics ( if it exists ) or systematic measurement errors . the final accuracy of @xmath13 constitutes a robust and precise test of the dark energy model . we have measured the small - scale clustering amplitude of high - redshift bright emission - line galaxies using the first @xmath2 of spectra from the aat wigglez dark energy survey ( @xmath309 galaxies in the redshift range @xmath150 ) . we have successfully developed a methodology to generate random realizations of the survey incorporating the currently sparse selection function and redshift incompleteness . we find that : * the wigglez galaxy sample in the redshift range @xmath150 possesses a clustering length @xmath168 mpc and slope @xmath5 . this clustering amplitude significantly exceeds that of uv - selected samples at @xmath310 and agrees well with that of the most luminous blue galaxies observed by the deep2 galaxy redshift survey at @xmath311 . the clustering amplitude of wigglez targets is comparable to that of lyman break galaxies at a similar uv luminosity . * the clustering length of the wigglez targets is approximately constant with redshift for the range @xmath235 . the value of @xmath90 increases with @xmath202-band luminosity , @xmath45-band luminosity , and reddening rest - frame colour . * the redshift - space distortion signature of coherent galaxy motions is detected and its amplitude ( @xmath196 ) is consistent with that predicted from the galaxy bias . we detect some evidence for `` fingers of god '' due to the virialized motions of galaxies in clusters . using these results , we forecast the performance of the full 1000 deg@xmath0 wigglez survey in the measurement of the galaxy power spectrum and cosmological model . we find that : * the survey design is well - tuned to the `` optimal '' mean galaxy number density @xmath312 , where @xmath31 is the amplitude of the galaxy power spectrum on the scales of importance for baryon oscillations . * the survey will delineate the baryon acoustic oscillations in the large - scale clustering pattern in three independent redshift slices , providing measurements of the cosmic distance and expansion rate in each redshift slice with accuracies of @xmath11 . * the resulting measurement of a constant dark energy equation - of - state parameter @xmath13 from the wigglez survey , calibrating the standard ruler using the cmb measurement of the acoustic scale , has a higher precision than provided by current supernovae datasets . these independent dark energy probes lie in a highly complementary direction in the parameter space of @xmath12 and @xmath282 . the full combination of wigglez , supernovae and cmb datasets provides a measurement of the equation of state with accuracy @xmath313 , constituting a robust and precise test of the dark energy model incorporating cross - checking of systematic errors between different probes . the final survey will enable a wide range of scientific investigations into the cosmological model and galaxy evolution . we acknowledge financial support from the australian research council through discovery project grants funding the positions of sb , mp , gp and td . we also thank the university of queensland for supporting the phd scholarship of rj . we acknowledge the efforts of nick jones and david barnes in creating the online wigglez database , and emily wisnioski for incorporating principal component analysis sky subtraction into the data reduction pipeline . we thank karl forster for his assistance in scheduling our galex observations and sebastien heinis , ted wyder and mark seibert for invaluable galex support and discussions . we acknowledge correlation function modelling performed by carlos contreras and ben jelliffe which revealed a mistake in the submitted version of this paper . galex ( the galaxy evolution explorer ) is a nasa small explorer , launched in april 2003 . we gratefully acknowledge nasa s support for construction , operation and science analysis for the galex mission , developed in co - operation with the centre national detudes spatiales of france and the korean ministry of science and technology . finally , the wigglez survey would not be possible without the dedicated work of the staff of the anglo - australian observatory in the development and support of the aaomega spectrograph , and the running of the aat . adelberger k.l . , steidel c.c . , pettini m. , shapley a.e . , reddy n.a . , erb d.k . , 2005 , apj , 619 , 697 allen p.d . , moustakas l.a . , dalton g. , macdonald e. , blake c.a , clewley l. , heymans c. , wegner g. , 2005 , mnras , 360 , 1244 arnouts s. et al . , 2002 , mnras , 329 , 355 arnouts s. et al . , 2005 , apj , 619 , 43 astier p. et al . , 2006 , a&a , 447 , 31 blake c.a , glazebrook k. , 2003 , apj , 594 , 665 blake c.a . , parkinson d. , bassett b. , glazebrook k. , kunz m. , nichol r.c . , 2006 , mnras , 365 , 255 campbell l. , saunders w. , colless m. , 2004 , mnras , 350 , 1467 coil a. et al . , 2008 , apj , 672 , 153 cole s. et al . , 2005 , mnras , 362 , 505 coles p. , 1993 , mnras , 262 , 1065 colless m. et al . , 2001 , mnras , 328 , 1039 cowie l.l . , songaila a. , hu e.m . , cohen j.g . , 1996 , aj , 112 , 839 cooray a. , hu w. , huterer d. , joffre m. , 2001 , apj , 557l , 7 davis m. , peebles p.j.e . , 1983 , apj , 208 , 13 davis t.m . et al . , 2007 , apj , 666 , 716 efstathiou g. , 1988 , in lawrence a. , ed . 3rd iras conf . , comets to cosmology , springer , new york , p.312 eisenstein d.j . , hu w. , 1998 , apj , 518 , 2 eisenstein d.j . et al . , 2001 , aj , 122 , 2267 eisenstein d.j . , 2002 , _ ` large - scale structure and future surveys ' _ in ` next generation wide - field multi - object spectroscopy ' , asp conference series vol . 280 , ed . m.brown & a.dey , p.35 ( astro - ph/0301623 ) eisenstein d.j . et al . , 2005 , apj , 633 , 560 eisenstein d.j . , j . , sirko e. , spergel d.n . , 2007 , apj , 664 , 675 feldman h.a . , kaiser n. , peacock j.a . , 1994 , apj , 426 , 23 foucaud s. , mccracken h.j . , le fevre o. , arnouts s. , brodwin m. , lilly s.j . , crampton d. , mellier y. , 2003 , a&a , 409 , 835 gaztanaga e. , cabre a. , hui l. , 2008 , submitted ( astro - ph/0807.3551 ) giavalisco m. , dickinson m. , 2001 , apj , 550 , 177 glazebrook k. et al . , 2004 , nature , 430 , 181 glazebrook k. , blake c.a . , 2005 , apj , 631 , 1 hamilton a.j.s . , 1992 , apj , 385 , 5 hawkins e. et al . , 2003 , mnras , 346 , 78 heinis s. , treyer m. , arnouts s. , milliard b. , donas j. , gal r. , martin d.c . , viton m. , 2004 , a&a , 424 , l9 heinis s. et al . , 2007 , apjs , 173 , 503 hu w. , haiman z. , 2003 , phys . rev . d , 68 , 063004 huesti g. , 2006 , a&a , 449 , 891 kaiser n. , 1987 , mnras , 227 , 1 komatsu e. et al . , 2009 , apjs , 180 , 330 landy s.d . , szalay a.s . , 1993 , apj , 412 , 64 lee k .- s . , giavalisco m. , gnedin o. , somerville r.s . , ferguson h.c . , dickinson m. , ouchi m. , 2006 , apj , 642 , 63 lewis a. , challinor a. , lasenby a. , 2000 , apj , 538 , 473 linder e.v . , 2003 , d , 68 , 083504 loveday j. , maddox s.j . , efstathiou g. , peterson b.a . , 1995 , apj , 442 , 457 martin d. , et al . , 2005 , apj , 619 , l1 meneux b. et al . , 2006 , a&a , 452 , 387 milliard b. et al . , 2007 , apjs , 173 , 494 ouchi m. et al . , 2001 , apj , 558 , l83 ouchi m. et al . , 2005 , apj , 635 , l117 parkinson d. , blake c.a . , kunz m. , bassett b. , nichol r. , glazebrook k. , 2007 , mnras , 377 , 185 percival w. , cole s. , eisenstein d.j . , nichol r.c . , peacock j.a . , pope a.c . , szalay a.s . , 2007 , mnras , 381 , 1053 riess a. et al . , 2007 , apj , 659 , 98 saunders w. et al . , 2004 , _ ` aaomega : a scientific and optical overview ' _ in ` ground - based instrumentation for astronomy ' , ed . a.moorwood & i.masanori , proceedings of the spie , vol . 5492 , p.389 scherrer r.j . , weinberg d.h . , 1998 , apj , 504 , 607 schlegel d.j . , finkbeiner d.p . , davis m. , 1998 , apj , 500 , 525 sharp r. et al . , 2006 , _ ` performance of aaomega : the aat multi - purpose fiber - fed spectrograph ' _ in ` ground - based and airborne instrumentation for astronomy ' , ed . i.mclean & i.masanori , proceedings of the spie , vol . 6269 , p.14 ( astro - ph/0606137 ) seo h j . , eisenstein d.j . , 2003 , apj , 598 , 720 seo h .- j . , eisenstein d.j . , 2007 , apj , 665 , 14 smith r.e . et al . , 2003 , mnras , 341 , 1311 tegmark m. , 1997 , prl , 79 , 20 van dokkum p. et al . , 2004 , apj , 611 , 703 willmer c. et al . , 2006 , apj , 647 , 853 wood - vasey w. et al . , 2007 , apj , 666 , 694 yee h. , gladders m. , gilbank d. , majumdar s. , hoekstra h. , ellingson e. , 2007 , _ ` the red - sequence cluster surveys ' _ in ` cosmic frontiers asp conference series ' , ed . n.metcalfe & t.shanks , vol . 379 , p.103 ( astro - ph/0701839 ) york d.g . et al . , 2000 , aj , 120 , 1579 yoshida m. , shimasaku k. , ouchi m. , sekiguchi k. , furuswawa h. , okamura s. , 2008 , apj , 679 , 269

the concept of prefactors is considered in order to decrease the complexity of the guruswami sudan interpolation step for generalized reed solomon codes . it is shown that the well - known re - encoding projection due to ktter et al . leads to one type of such prefactors . the new type of sierpinski prefactors is introduced . the latter are based on the fact that many binomial coefficients in the hasse derivative associated with the guruswami sudan interpolation step are zero modulo the base field characteristic . it is shown that both types of prefactors can be combined and how arbitrary prefactors can be used to derive a _ reduced _ guruswami sudan interpolation step . generalized reed solomon codes , guruswami sudan algorithm , list decoding , polynomial interpolation , pascal triangle , sierpinski gasket , binomial coefficients

1309.7901

of generalized reed solomon codes comes in multiple flavors . among the most widely - deployed techniques for decoding up to @xmath0 errors , where @xmath1 is the length and @xmath2 is the dimension of the code , is decoding based on linear shift register synthesis with the berlekamp massey algorithm @xcite . other techniques utilize the extended euclidean algorithm @xcite , newton interpolation @xcite , and bivariate polynomial interpolation @xcite . apart from implementational speedups , e.g. , divide & conquer techniques that allow to implement the berlekamp massey algorithm in @xmath3 $ ] as proposed in @xcite , all these algorithms are in @xmath4 $ ] . decoding beyond @xmath0 is much more complicated and efficient ( polynomial - time ) decoding algorithms for this case have been known only after the invention of the sudan algorithm in 1997 @xcite and the guruswami sudan algorithm in 1999 @xcite . the latter can correct up to @xmath5 errors . as we will see shortly , the guruswami sudan algorithm consists of two steps where one of them an interpolation problem is computationally more involved than the other . extensive efforts to speed up the interpolation problem include exploiting structured matrices @xcite or the underlying algebraic structure @xcite . the technique of re - encoding was introduced in @xcite ; it allows one to make predictions about the structure of the solutions of the interpolation problem and then to exploit this knowledge in order to reduce the problem . our contribution in this paper is to show how the size of the interpolation problem can in many cases be reduced beyond re - encoding , utilizing a simple property of the generalized reed solomon code s base field . in order to provide a framework that unites re - encoding , our proposal , and the combination of both , we introduce the general notion of a - priori known _ prefactors _ and show how such factors can lead to a reduction of the interpolation problem . the rest of this paper is organized as follows . section [ sec : prelim ] covers generalized reed solomon codes as well as the guruswami sudan algorithm with its interpolation and factorization steps . the concept of re - encoding is explained in section [ sec : reenc ] and a theorem delivers the associated _ re - encoding _ prefactors . section [ sec : sierpinski ] represents the main part of the paper . it introduces the new type of _ sierpinski _ prefactors and gives results about their existence and properties . section [ sec : factors ] shows how prefactors of both types can be combined and how they can be used to diminish the size of the interpolation problem . one of our main results is given in this section , i.e. , the reduced interpolation step given by problem [ prob : interpolation_reduced ] . the paper is wrapped up in section [ sec : conclusion ] and two useful algorithms are provided in an appendix . [ def : grs ] for a prime power @xmath6 and @xmath7 with @xmath8 let @xmath9 be an ordered set of distinct elements ( _ code locators _ ) from the finite field @xmath10 and let @xmath11 be an ordered set of nonzero ( not necessarily distinct ) elements ( _ column multipliers _ ) from @xmath10 . then the set of vectors @xmath12,\deg[u(x)]<k\right\}\end{gathered}\ ] ] is a _ generalized reed solomon ( grs ) code _ @xcite . since we frequently assume that _ base field _ @xmath10 , _ code length _ @xmath1 , and _ code dimension _ @xmath2 as well as code locators and column multipliers are fixed , we write @xmath13 for @xmath14 whenever it is convenient . grs codes fulfill the singleton bound with equality , i.e. , their _ minimum distance _ is @xmath15 . @xmath13 is a linear subspace of @xmath16 , and thus it is a linear code . note that _ conventional reed solomon ( rs ) _ codes are special cases of grs codes with @xmath17 and @xmath18 , where @xmath19 has order @xmath1 and @xmath20 @xcite . the state - of - the - art of decoding grs codes is the _ guruswami sudan algorithm ( gsa ) _ it can be divided into two steps : the interpolation step ( problem [ prob : interpolation ] ) and the factorization step ( problem [ prob : factorization ] ) . our focus here is on the interpolation step , which is computationally more involved . let @xmath21 be a codeword , @xmath22 be an error vector of hamming weight @xmath23}}=\varepsilon$ ] , and @xmath24 be the corresponding received vector obtained from the transmission channel . furthermore , let @xmath25 and let @xmath26 be two parameters of the gsa with @xmath27 . with any set @xmath28 we associate the polynomial @xmath29 with @xmath30 if @xmath31 . here , @xmath19 is a primitive element . [ prob : interpolation ] given a received vector @xmath32 and @xmath33 , find a nonzero bivariate polynomial @xmath34 $ ] such that @xmath35\leq r(n-\varepsilon_0)-\nu(k-1)-1{\ensuremath{\triangleq}}{\ensuremath{d_{q_{\nu}}}}\ ] ] for @xmath36 and @xmath37 where @xmath38 . the nested sum in is called the @xmath39th _ mixed partial hasse derivative _ @xcite of @xmath40 . the condition that all @xmath39th hasse derivatives with @xmath41 evaluate to zero for all tuples @xmath42 , @xmath43 , means that these tuples are zeros of multiplicity @xmath44 of @xmath40 . for that reason , we refer to the parameter @xmath44 as the _ multiplicity _ of the gsa . a straightforward analysis shows that the homogeneous linear system of equations associated with problem [ prob : interpolation ] has @xmath45 equations and @xmath46 unknowns . both numbers are exceedingly large even for short grs codes and intermediate parameters @xmath44 and @xmath47 . it can be shown that the linear system has a nonzero solution @xmath40 ( i.e. , it has more equations than unknowns ) as long as @xmath48 naively solving the linear system with gaussian elimination in order to obtain a solution is in @xmath49 $ ] since @xmath50 and @xmath51 . one of the fastest methods to solve the interpolation step is ktter interpolation @xcite , which is in @xmath52 $ ] . a fast algorithm that actually solves the homogeneous linear system was provided in @xcite and is based on results from @xcite . without loss of generality we assume in the following that the columns of the coefficient matrix ( from left to right ) are associated with the unknown coefficients @xmath53 such a coefficient matrix can be set up using algorithm [ alg : gsa_matrix ] in the appendix . the particular ordering of the coefficients will become important in section [ sec : factors ] where we apply our results in order to reduce the coefficient matrix . [ prob : factorization ] given a solution @xmath40 of problem [ prob : interpolation ] , find all factors of the form @xmath54 with @xmath55 $ ] , @xmath56<k$ ] . let us collect all such @xmath57 in a set @xmath58 . from @xmath58 we can calculate the _ result list _ @xmath59 obviously , @xmath60 and @xmath61 . due to the latter fact , we refer to the parameter @xmath47 as the _ list size_. note that all codewords @xmath62 with @xmath63}}>\varepsilon_0 $ ] can be discarded . it is proven in @xcite that @xmath64 if @xmath65}}={\ensuremath{\mathrm{wt_\mathrm{h}}\left[{\ensuremath{\boldsymbol{e}}}\right]}}\leq\varepsilon_0 $ ] . problem [ prob : factorization ] can be solved with time complexity in @xmath66n^2\right]$ ] using a technique from @xcite , but this is not the focus of this paper . it follows from the exposition of the welch berlekamp algorithm in @xcite and the interpretation of justesen and hholdt in ( * ? ? ? * sections 5.2 and 12.2 ) that the gsa simplifies to the sudan algorithm @xcite if we restrict the multiplicity to @xmath67 and that it further simplifies to the welch berlekamp algorithm @xcite if we additionally restrict the list size to @xmath68 . the re - encoding projection in the context of the gsa was first introduced in @xcite and later elaborated for application with interpolation algorithms in @xcite . its key idea is to project the received vector @xmath69 onto a subspace of @xmath16 with zero components at positions indexed by @xmath70 , @xmath71 . not surprisingly , it turns out that this allows to skip @xmath2 of the interpolation points @xmath42 in and thus to reduce the size of the interpolation problem [ prob : interpolation ] from @xmath1 to @xmath72 points . the _ re - encoding projection _ with respect to @xmath70 , @xmath71 , for @xmath14 is defined as the linear map @xmath73 where @xmath74 , such that @xmath75 . the re - encoding projection annihilates the components @xmath76 , @xmath77 , and sets them to zero . it is injective since grs codes fulfill the singleton bound with equality and thus any codeword @xmath78 is uniquely determined by the @xmath2 components @xmath79 , @xmath77 . the remaining @xmath72 components @xmath79 , @xmath80 , can be efficiently found using an erasures - only decoder . @xmath81 is indeed a projection , i.e. , it is idempotent . this can be seen by the fact that all components of @xmath82}}={\ensuremath{\boldsymbol{v}}}+\widetilde{{\ensuremath{\boldsymbol{c}}}}_1 $ ] at positions @xmath77 are zero by definition . but then @xmath83}}\right]}}={\ensuremath{\boldsymbol{v}}}+\widetilde{{\ensuremath{\boldsymbol{c}}}}_1+\widetilde{{\ensuremath{\boldsymbol{c}}}}_2 $ ] , where @xmath84 is zero at the @xmath2 positions in @xmath85 , which is only possible if it is the all - zero codeword . [ ex : reenc1 ] consider the conventional rs code @xmath86 with @xmath87 ( primitive element @xmath88 ) and @xmath89 . for this code , the gsa with multiplicity @xmath90 and list size @xmath91 can correct up to @xmath92 errors . assume the following composition of the received vector : @xmath93 assume further @xmath94 , i.e. , when we apply the re - encoding projection to @xmath32 we obtain @xmath95 and need to correct the @xmath96 using an erasures - only decoder . the ( unique ) result of this procedure is @xmath97 . with this , the re - encoding projection of @xmath32 becomes : @xmath98 \end{array}\ ] ] but then @xmath99={\ensuremath{\boldsymbol{c}}}+\widetilde{{\ensuremath{\boldsymbol{c}}}}+{\ensuremath{\boldsymbol{e}}}$ ] with @xmath100 , hence we have a new received vector with ( as we will see ) favorable properties but still the original error vector : @xmath101 \end{array}\ ] ] the example shows how the re - encoding projection can be used in order to simplify the decoding : first , the received vector @xmath32 is projected , which yields a new received vector @xmath99 $ ] with ( at least ) @xmath2 zero components . then , the gsa is applied to the new received vector in order to obtain the error vector @xmath102 . this can be done more efficiently than decoding of @xmath32 due to the designed properties of @xmath99 $ ] . finally , the error vector together with the original received vector can be used to calculate the transmitted codeword @xmath103 . the following theorem shows that decoding of @xmath99 $ ] with the gsa induces strong structural properties on the bivariate result polynomial @xmath40 of the interpolation step or , more precisely , its constituent univariate polynomials @xmath104 , @xmath36 . it will become clear in section [ sec : factors ] how this can be exploited in order to reduce the size of the interpolation problem [ prob : interpolation ] . [ thm : reenc_factors ] let @xmath14 be a grs code , @xmath70 with @xmath71 , and @xmath105 such that the gsa can correct at most @xmath106 errors . let further @xmath21 , @xmath22 with @xmath23}}\leq\varepsilon_0 $ ] and @xmath24 . when the gsa is applied to @xmath99 $ ] it yields a bivariate result polynomial @xmath34 $ ] whose constituent univariate polynomials @xmath104 , @xmath107 , can be factored as @xmath108 where @xmath109\leq { \ensuremath{d_{q_{\nu}}}}-k(r-\nu){\ensuremath{\triangleq}}{\ensuremath{d_{u_{\nu}}}}$ ] . we emphasize that the _ re - encoding prefactors _ @xmath110 in are fixed a - priori , i.e. , they do not depend on the received vector @xmath32 . this fact will be exploited in section [ sec : factors ] , it basically allows to work with the quotient polynomials @xmath111 instead of the full polynomials @xmath104 . note that we required @xmath51 in section [ sec : prelim ] , hence there generally are polynomials @xmath104 , @xmath112 that can not be factored . assume that @xmath40 is a solution of problem [ prob : interpolation ] for the projected received vector @xmath99 $ ] . that is , the coefficients of @xmath40 fulfill . if we restrict to @xmath113 we obtain @xmath114 since @xmath115 . the factor @xmath116 annihilates all summands of the outer sum except the one for @xmath117 , which yields @xmath118 . but then becomes @xmath119 which means that the @xmath120th to @xmath121th hasse derivatives of the @xmath122 , @xmath123 , evaluate to zero at all @xmath124 , @xmath77 . but then the @xmath124 , @xmath77 , are roots of multiplicity @xmath125 of @xmath122 , which ( after replacing @xmath126 by @xmath127 ) proves our argument . the bound on the degrees of the @xmath111 is easy to see by comparing the degrees of the involved polynomials . [ ex : reenc2 ] for the setting of example [ ex : reenc1 ] , the interpolation step of the gsa could result in the bivariate polynomial @xmath128 with univariate constituent polynomials rcl q_0(x ) & = & 6x^13 + 2x^12 + 6x^11 + x^10 + 2x^9 + 7x^8 + 2x^7 + & & + 4x^6 + 2x^5 + 7x^4 + 3x^3 + 3x^2 + 8x + 5 + q_1(x ) & = & 7x^9 + x^8 + 9x^7 + 5x^4 + 6x^3 + 4x^2 + 1 + q_2(x ) & = & 2x^5 + 10x^4 + z^2 + 2x + 6 + q_3(x ) & = & 1 . from @xmath94 and we obtain @xmath129 and it is easy to verify the factorizations @xmath130 that are given by theorem [ thm : reenc_factors ] . we have seen in this section that the re - encoding projection can be used in order to obtain structured solutions , i.e. , bivariate polynomials @xmath40 , of problem [ prob : interpolation ] . the computational overhead of re - encoding is negligible , as it is confined to a single erasures - only decoding step for a vector with @xmath72 erased symbols . in this section , we introduce a new technique that results in structured solutions of problem [ prob : interpolation ] . in contrast to the re - encoding projection , this approach does _ not _ require any additional computations , it simply exploits basic properties of the grs code s base field @xmath10 . the main idea is to exploit the fact that many of the binomial coefficients in are zero modulo the characteristic of @xmath10 . to see this , let us consider the left - aligned pascal triangles in fig . [ fig : sierpinski ] , where the entries are calculated modulo @xmath131 and modulo @xmath132 , respectively . obviously , the zero entries of the triangles follow regular patterns . note that we use the rather uncommon left - aligned representation of the pascal triangle in order to be able to refer to its columns . the triangle resembles variants of the left - aligned _ sierpinski gasket _ , one of the most basic examples of a self - similar set . the resemblance gets more accurate as more rows of the pascal triangle are considered . in the following , we will refer to a pascal triangle with entries modulo any positive integer @xmath133 as a _ sierpinski triangle _ and we denote it by @xmath134 . now consider the interpolation constraints . the summands of the outer sum are weighted by the binomial coefficients @xmath135 , @xmath136 . these are exactly the binomial coefficients that appear at the first @xmath137 entries in column @xmath126 of a sierpinski triangle . for any @xmath19 , @xmath138 , where @xmath139 $ ] is the characteristic of @xmath10 . thus , summands for which @xmath135 is a multiple of @xmath133 are zero . for given list size @xmath47 and multiplicity @xmath44 , let us assume there is a column @xmath140 , @xmath141 , in @xmath134 such that @xmath142 i.e. , all entries except the first one ( which is @xmath143 ) are zero . we refer to such columns as _ _ zero columns _ _ , hence the degenerate zero column with a single entry @xmath144 can not occur in context of the gsa . however , it will turn out to be useful for the proof of lemma [ lemma : zerocolumnsspoilers ] to include this case in the definition . ] . [ fig : sierpinski ] shows that zero columns actually exist , e.g. , column @xmath145 for @xmath146 in @xmath147 ( fig . [ fig : sierpinski2 ] ) or column @xmath148 for @xmath149 in @xmath150 ( fig . [ fig : sierpinski3 ] ) . assume that @xmath40 is a solution of problem [ prob : interpolation ] for a received vector @xmath32 and consider for zero column @xmath140 . this equation simplifies to @xmath151 since all summands of the outer sum except the first one are annihilated by the zero binomial weights . but then , with the same argumentation as in the proof of theorem [ thm : reenc_factors ] , the @xmath124 , @xmath43 , are roots of multiplicity @xmath152 of @xmath153 and thus @xmath153 can be factored as @xmath154 where @xmath155\leq { \ensuremath{d_{q_{t_0}}}}-n(r - t_0)$ ] . the following lemma specifies the conditions for the existence of a zero column @xmath140 and its location . [ lemma : zerocolumnexists ] let @xmath26 , @xmath51 . if * @xmath156 or * @xmath157 then a zero column does _ not _ exist . otherwise , find the least significant base-@xmath133 digit @xmath158 of @xmath47 such that @xmath159 . then , @xmath160 is a zero column . in particular , @xmath140 is the maximal ( rightmost ) zero column . the proof of lemma [ lemma : zerocolumnexists ] relies on the following well - known theorem about the divisibility of binomial coefficients , which we state in a modified form that is particularly convenient for our purposes . let @xmath161 and @xmath133 prime . then @xmath162 if and only if at least one base-@xmath133 digit of @xmath163 is greater than the corresponding base-@xmath133 digit of @xmath164 . a zero column @xmath140 is defined by property , which can using lucas theorem be equivalently formulated as follows : @xmath140 in base @xmath133 must be such that it has at least one base-@xmath133 digit greater than that of @xmath165 , @xmath166 , at least one base-@xmath133 digit greater than that of @xmath47 . if @xmath156 then @xmath167 are all single - digit numbers and since @xmath168 , @xmath140 can not be greater than @xmath169 , proving the first case . the base-@xmath133 expansion of an integer that fulfills the second case is @xmath170 where @xmath171 . that is , all digits except the leading one assume the maximal possible value @xmath172 , which can not be exceeded by the digits of any integer modulo @xmath133 , particularly not by those of @xmath140 . but the most significant digit of @xmath140 can not be larger than @xmath173 as well since @xmath174 . thus , lucas theorem can not be fulfilled for @xmath175 and @xmath176 , proving the second case . + if the first and second cases do not apply then @xmath47 has at least two digits , at least one of them being smaller than @xmath172 . this guarantees the existence of the least significant such digit , i.e. , @xmath158 . we need to find the greatest @xmath20 , @xmath177 , whose digits @xmath178 , @xmath179 are zero . this is only possible if @xmath180 is a multiple of @xmath181 . set @xmath180 to the greatest such multiple smaller than @xmath44 , that is @xmath182 . let @xmath183 . this subtraction leads to a carry up to digit @xmath184 and thus @xmath185 for all @xmath179 , i.e. , the @xmath186 least significant digits of @xmath187 are maximal . due to the maximality of @xmath180 there can be no @xmath188 for which both @xmath189 for all @xmath179 and @xmath190 hold . by invalidity of the second case there can also be no such @xmath191 with @xmath192 . this shows that lucas theorem is fulfilled for @xmath193 and all @xmath194 , thereby proving the third case . the maximality of @xmath140 follows from the maximality of @xmath180 . [ ex : sier1 ] consider the conventional rs code @xmath195 with @xmath196 and @xmath197 . the characteristic of the code s base field is @xmath198=3 $ ] . for this code , the gsa with multiplicity @xmath199 and list size @xmath200 can correct up to @xmath201 errors . lemma [ lemma : zerocolumnexists ] yields @xmath202 , since the binary expansion of @xmath200 is @xmath203 . we can easily check fig . [ fig : sierpinski3 ] in order to confirm that this is a zero column for @xmath200 . thus , according to , for any result @xmath204 of the interpolation step of the gsa holds the factorization @xmath205 , where @xmath206 ( from ) and @xmath207 ( from @xmath208 ) . it follows directly from lemma [ lemma : zerocolumnexists ] that zero columns can not exist for the special cases @xmath209 ( sudan algorithm ) and @xmath210 ( welch berlekamp algorithm ) of the gsa . the statement of the lemma can be interpreted graphically , see fig . [ fig : zerocolumnexists ] . in the following , we will generalize the concept of zero columns to zero columns with resolvable spoilers . this will reveal additional structure in solutions @xmath40 of problem [ prob : interpolation ] , i.e. , to factorizations of additional univariate polynomials @xmath104 , @xmath211 , besides @xmath153 . assume that for given gsa parameters @xmath212 and base field @xmath10 with @xmath213 $ ] there exists a zero column @xmath140 in @xmath134 . further assume there is a column @xmath214 , @xmath215 , in @xmath134 such that @xmath216 we refer to such a column as _ zero column with spoiler at @xmath217 _ , examples are shown in fig . [ fig : sierpinski2 ] and fig . [ fig : sierpinski3 ] , respectively . if @xmath40 is a solution of problem [ prob : interpolation ] for a received vector @xmath32 then for column @xmath214 , @xmath215 , becomes @xmath218 since all summands except the ones for @xmath219 and @xmath220 are annihilated . but , according to , the second sum evaluates to zero at all @xmath124 , @xmath43 , since @xmath140 is by assumption a zero column . we refer to @xmath217 as a _ resolvable spoiler _ for @xmath214 because the sum associated with @xmath217 vanishes . as a result , we obtain @xmath221 i.e. , the @xmath124 , @xmath43 , are roots of multiplicity @xmath152 of @xmath222 and thus it can be factored as @xmath223 where @xmath224\leq { \ensuremath{d_{q_{t_1}}}}-n(r - t_0)$ ] . it is easy to see that if @xmath225 is a spoiler for @xmath226 , @xmath227 , then it is also resolvable . furthermore , it is easy to see that the concept generalizes to multiple spoilers . this leads to the following recursive definition : consider @xmath134 and @xmath26 . for @xmath228 let @xmath229 and @xmath230 then @xmath127 is a _ zero column with resolvable spoilers _ if and only if @xmath231 . the basic case is @xmath232 if @xmath140 exists . note that zero columns are special cases of zero columns with resolvable spoilers where @xmath233 . the sets of zero columns with resolvable spoilers are non - increasing with @xmath127 , i.e. , @xmath234 . we stress that @xmath235 contains all zero columns with resolvable spoilers in @xmath134 , i.e. , it is the set that we are interested in . [ lemma : zerocolumnsspoilers ] let @xmath213 $ ] and @xmath26 , @xmath51 . if according to lemma [ lemma : zerocolumnexists ] a maximal zero column @xmath140 of @xmath134 exists then the set of zero columns with resolvable spoilers of @xmath134 is @xmath236 otherwise it is @xmath237 . assume that @xmath140 exists . note that it is the maximal zero column not only for @xmath47 but for all @xmath238 since appending an entry to a nonzero column obviously can not make it a zero column . let us set @xmath239 . note that in that case , column @xmath140 is degenerate . we obtain @xmath240 . if we now advance to column @xmath241 it can either have a spoiler at @xmath242 or no spoiler at all , meaning either @xmath243 or @xmath244 . in both cases @xmath245 and we have @xmath246 . in the same manner , column @xmath247 can either have @xmath248 , @xmath249 , @xmath250 , or @xmath251 . in all cases @xmath252 , resulting in @xmath253 . this process continues all the way to column @xmath120 and as a result _ all _ spoilers are resolvable and thus _ all _ columns of @xmath134 are zero columns with resolvable spoilers with respect to @xmath239 , eventually resulting in @xmath254 . now let us gradually increase @xmath255 by one until @xmath256 and recall that @xmath140 stays a maximum zero column . due to the maximality of @xmath140 , any nonzero @xmath257 with @xmath258 is a non - resolvable spoiler and thus such columns @xmath126 must be removed from @xmath259 . but this means that eventually only the columns @xmath126 for which all @xmath260 , @xmath261 , are zero remain in @xmath262 , which ( after substituting @xmath126 by @xmath127 ) proves the statement . if @xmath140 does not exist then there exists no zero column and also no resolvable spoilers , hence @xmath263 . [ ex : sier2 ] lemma [ lemma : zerocolumnexists ] yields maximal zero column @xmath148 for the setting of example [ ex : sier1 ] . fig . [ fig : sierpinski3 ] shows that for @xmath264 we have @xmath265 , i.e. , @xmath264 is a zero column with resolvable spoilers and we can set @xmath266 . for @xmath267 we have @xmath268 and thus it is a zero column with resolvable spoilers as well . this gives @xmath269 . for @xmath270 we have @xmath271 and thus it is a zero column with resolvable spoilers as well . it turns out that for all @xmath272 holds @xmath273 and thus the only zero columns with resolvable spoilers in @xmath150 with respect to @xmath200 are @xmath148 , @xmath264 , @xmath267 , and @xmath270 . it can be readily checked that lemma [ lemma : zerocolumnsspoilers ] confirms this result and delivers @xmath274 . the following map will turn out to be useful in the following , it returns either @xmath127 itself or its greatest spoiler : @xmath275{ll } \max\{\mathcal{s}_\nu^{(\ell)}\ } & \mathcal{s}_\nu^{(\ell)}\neq\emptyset\\ \nu & \mathcal{s}_\nu^{(\ell)}=\emptyset\\ \end{array}\right . \end{array}\right .. \ ] ] [ thm : sier_factors ] let @xmath14 be a grs code and @xmath105 such that the gsa can correct at most @xmath106 errors . let further @xmath21 , @xmath22 with @xmath23}}\leq\varepsilon_0 $ ] and @xmath24 . when the gsa is applied to @xmath32 it yields a bivariate result polynomial @xmath34 $ ] whose constituent univariate polynomials @xmath104 , @xmath276 , can be factored as @xmath277},\ ] ] where @xmath278\leq { \ensuremath{d_{q_{\nu}}}}-n\left(r - g[\nu]\right){\ensuremath{\triangleq}}{\ensuremath{d_{v_{\nu}}}}$ ] . let @xmath276 . since @xmath127 is a zero column with resolvable spoilers , all spoilers @xmath279 with @xmath280 are resolvable , that is , @xmath281 since by definition all terms except the ones weighted by the spoilers vanish , we can write as @xmath282 in order to let the sum over @xmath126 vanish in the case @xmath283 ( i.e. , to exploit ) , we must guarantee @xmath284 for all @xmath280 , i.e. , @xmath285 . in case @xmath286 the sum over @xmath126 is empty and thus it is sufficient to guarantee @xmath287 . due to the definition of @xmath288 $ ] we have @xmath289 $ ] in both cases and thus @xmath290 and the @xmath124 , @xmath43 , are roots of multiplicity @xmath291 $ ] of @xmath104 and thus it can be factored as in . the bound on the degrees of the @xmath292 follows from a comparison of the involved polynomial degrees . just as the re - encoding prefactors @xmath110 , @xmath107 , from section [ sec : reenc ] , the _ sierpinski prefactors _ @xmath293}$ ] , @xmath276 , are fixed a - priori and do not depend on the received vector @xmath32 . [ ex : sier3 ] we have already seen with the help of lemma [ lemma : zerocolumnexists ] that for the setting of example [ ex : sier1 ] any result of the polynomial step of the gsa contains a univariate constituent polynomial @xmath294 with factorization @xmath205 , where @xmath206 and @xmath207 . from example [ ex : sier2 ] we have @xmath295 , i.e. , theorem [ thm : sier_factors ] additionally guarantees factorizations of @xmath296 , @xmath297 , and @xmath298 . they are @xmath299 with @xmath300 , @xmath301 , and @xmath302 , respectively , because @xmath303=g[6]=g[7]=8 $ ] , @xmath304 , @xmath305 , and @xmath306 . the situation is depicted in fig . [ fig : sier_ex ] . for the zero columns with resolvable spoilers @xmath148 , @xmath264 , @xmath267 , and @xmath270 the summands @xmath260 from lemma [ lemma : zerocolumnsspoilers ] are enclosed by dashed rectangles and the sets of ( resolvable ) spoilers @xmath307 , @xmath308 , are enclosed by solid rectangles . the greatest spoilers @xmath288 $ ] are the lowermost entries within the solid rectangles . we have seen in sections [ sec : reenc ] and [ sec : sierpinski ] that some of the univariate constituent polynomials of the gsa interpolation step ( problem [ prob : interpolation ] ) have certain prefactors that are fixed a - priori . we will show in this section how this knowledge can be exploited in order to simplify solving the interpolation step . more precisely , we show that the associated linear system of equations in @xmath309 unknowns can be reduced to a linear system of smaller size . so far , we have treated re - encoding and sierpinski prefactors separately . this is not practical , one obviously wishes to exploit both types of prefactors jointly whenever possible . since re - encoding and sierpinski prefactors of a given univariate constituent polynomial @xmath104 , @xmath211 are in general not coprime , we have to calculate the _ combined prefactors _ in the following manner . [ thm : combined_factors ] let @xmath14 be a grs code , @xmath70 with @xmath71 , and @xmath105 such that the gsa can correct at most @xmath106 errors . let further @xmath21 , @xmath22 with @xmath23}}\leq\varepsilon_0 $ ] and @xmath24 . when the gsa is applied to @xmath99 $ ] it yields a bivariate result polynomial @xmath34 $ ] whose univariate constituent polynomials @xmath104 can be factored as @xmath310{ll } w_\nu(x)p_\mathcal{j}(x)^{r-\nu}p_{\mathcal{i}\setminus\mathcal{j}}(x)^{r - g[\nu ] } & \nu < r , \nu\in\mathcal{r}_0^{(\ell)}\\ u_\nu(x)p_\mathcal{j}(x)^{r-\nu } & \nu < r,\nu\not\in\mathcal{r}_0^{(\ell ) } \end{array}\right .. \]]the degrees of the @xmath311 and @xmath111 are bounded by @xmath312\leq { \ensuremath{d_{q_{\nu}}}}-k(r-\nu)-(n - k)(r - g[\nu]){\ensuremath{\triangleq}}{\ensuremath{d_{w_{\nu}}}}\ ] ] and @xmath109\leq { \ensuremath{d_{u_{\nu}}}}$ ] , respectively . the second case is a repetition of theorem [ thm : reenc_factors ] and there is nothing left to prove . as for the first case , we have @xmath70 . this allows to write the sierpinski prefactors from theorem [ thm : sier_factors ] as @xmath293}=p_\mathcal{j}(x)^{r - g[\nu]}p_{\mathcal{i}\setminus\mathcal{j}}(x)^{r - g[\nu]}$ ] . since @xmath313}$ ] is a spoiler , we have @xmath288\leq \nu$ ] . this allows to write the re - encoding prefactors from theorem [ thm : reenc_factors ] as @xmath314}p_\mathcal{j}(x)^{\nu - g[\nu]}$ ] . the factor @xmath315}$ ] is common to both factorizations ( sierpinski and re - encoding ) and must be counted not more than once in a combined factorization . hence , we obtain @xmath316}p_\mathcal{j}(x)^{\nu - g[\nu]}}_{=p_\mathcal{j}^{r-\nu}}p_{\mathcal{i}\setminus\mathcal{j}}(x)^{r - g[\nu]},\ ] ] which proves the first case . the degree constraints follow directly from the degree constraints in theorems [ thm : reenc_factors ] and [ thm : sier_factors ] . in the following , it will be convenient to have the sets @xmath317 [ ex : sier4 ] for the setting of example [ ex : sier1 ] , a solution of the gsa interpolation step problem [ prob : interpolation ] for the projected received vector @xmath99 $ ] with @xmath318 could be @xmath319 the re - encoding prefactors for @xmath320 according to theorem [ thm : reenc_factors ] are @xmath321 , where @xmath322 we have @xmath323 and from example [ ex : sier2 ] we have @xmath324 , @xmath325 , @xmath326 , and , finally , @xmath295 . @xmath327 since @xmath148 is a zero column . with @xmath303=g[6]=g[7]=g[8]=8 $ ] we obtain the combined prefactors @xmath328 , @xmath329 , @xmath330 , and @xmath331 of @xmath296 , @xmath297 , @xmath298 , and @xmath294 , respectively . re - encoding- , sierpinski- and combined prefactors for the code under consideration are shown in table [ tab : sier_example ] . the table contains upper bounds on the degrees of the quotient polynomials ( after dividing by the prefactors ) for each of the factorizations . re - encoding and combined prefactors for this example give @xmath332 , while sierpinski prefactors give @xmath333 . [ 1 ] > + m#1 @xmath127 & bound @xmath334 & re - encoding factorization & bound @xmath335 & sierpinski factorization & bound @xmath336 & combined factorization & bound @xmath335 , @xmath337 + 0 & 199 & @xmath338 & 39 & & 199 & @xmath338 & 39 + 1 & 184 & @xmath339 & 40 & & 184 & @xmath339 & 40 + 2 & 169 & @xmath340 & 41 & & 169 & @xmath340 & 41 + 3 & 154 & @xmath341 & 42 & & 154 & @xmath341 & 42 + 4 & 139 & @xmath342 & 43 & & 139 & @xmath342 & 43 + 5 & 124 & @xmath343 & 44 & @xmath344 & 72 & @xmath345 & 24 + 6 & 109 & @xmath346 & 45 & @xmath347 & 57 & @xmath348 & 25 + 7 & 94 & @xmath349 & 46 & @xmath350 & 42 & @xmath351 & 26 + 8 & 79 & @xmath352 & 47 & @xmath353 & 27 & @xmath354 & 27 + 9 & 64 & @xmath355 & 48 & & 64 & @xmath355 & 48 + 10 & 49 & & 49 & & 49 & & 49 + 11 & 34 & & 34 & & 34 & & 34 + 12 & 19 & & 19 & & 19 & & 19 + 13 & 4 & & 4 & & 4 & & 4 + @xmath356 & 1421 & & 541 & & 1213 & & 461 + as noted before , gsa interpolation ( problem [ prob : interpolation ] ) amounts to finding the solution of a linear system of equations . this can be done naively using gaussian elimination . several faster methods @xcite have been developed , all of which exploit the _ structure of the involved coefficient matrix_. we conjecture that all these methods can be applied to a reduced linear system of equations , which can be obtained using re - encoding- , sierpinski- , or combined prefactors . the key idea here is to exploit the a - priori known _ structure of the solutions _ , which follows from the a - priori known prefactors . this can be accomplished using the following lemma . [ lemma : lincomb_solution ] for an arbitrary field @xmath357 let @xmath358 and @xmath359 be coefficient matrix and vector of constant terms of a linear system of equations @xmath360 that has at least one solution @xmath361 . let @xmath362 be a linear combination of the solution variables @xmath363 , i.e. , @xmath364 where @xmath365 , @xmath366 . then @xmath367 is the coefficient matrix of a linear system @xmath368 that has a solution @xmath369 . we can augment the linear system @xmath360 in order to obtain a linear system @xmath370 with @xmath371 and @xmath372 that has a solution @xmath373 . the last row of this system can be used to annihilate the column vector @xmath374 , which results in @xmath375 this allows to split the system into linear combination and a part @xmath368 that is independent of @xmath376 . now let us consider the factorization of a univariate constituent polynomial @xmath104 , @xmath377 , into a prefactor ( re - encoding , sierpinski , or combined ) and the corresponding quotient polynomial ( @xmath111 , @xmath292 , or @xmath311 ) . in order to prescind from the actual type of factorization let us denote the prefactor ( whatever type it is ) by @xmath378 } f_{\nu , \mu}x^\mu$ ] and the corresponding quotient polynomial by @xmath379 } g_{\nu , \mu}x^\mu$ ] with @xmath380\leq{\ensuremath{d_{g_{\nu}}}}$ ] . this gives @xmath381 for @xmath377 with coefficients @xmath382 where we implicitly used that the @xmath186th coefficient of a polynomial with @xmath383 or @xmath384 the degree of the polynomial is zero . in order to simplify the following description , let us agree on trivial prefactors @xmath385 for all @xmath104 , @xmath386 . in these cases , the quotient polynomials are @xmath387 and @xmath388 . note that the constant term @xmath389 of any prefactor is nonzero due to . this allows us to write @xmath390 which shows that @xmath391 is a linear combination of the @xmath392 , @xmath393 , and @xmath394 . what we will do in the following is to exploit for @xmath395 in order to obtain a solvable linear system whose solution comprises the coefficients of @xmath396 ( _ preparation _ ) and then to exploit for @xmath397 in order to dispose of the redundant columns of the coefficient matrix ( _ reduction _ ) . both steps preparation and reduction are based on applying lemma [ lemma : lincomb_solution ] with certain parameters . for simplicity , let us consider @xmath398 . we can apply the lemma with @xmath399 , and @xmath400 in order to exchange solution variable @xmath401 for @xmath402 . after that , we can apply the lemma again with @xmath399 , @xmath403 , and @xmath404 in order to exchange @xmath405 for @xmath406 . the process can be repeated until @xmath407 is replaced by @xmath408 , which closes the preparation step for @xmath409 . the reduction step is performed by applying lemma [ lemma : lincomb_solution ] with @xmath410 in order to exchange @xmath411 by @xmath412 . obviously , this allows to delete the last column from the coefficient matrix . the process is repeated with @xmath413 in order to exchange @xmath414 by @xmath412 and ( after deletion of the last column ) repeated again until @xmath415 is exchanged by @xmath412 ( allowing to delete the last column ) , which happens for @xmath416 preparation and reduction are then executed for all remaining @xmath127 , i.e. , @xmath417 . algorithm [ alg : reduction ] cumulates the whole process . as a result , the original coefficient matrix associated with , whose @xmath309 columns are associated with @xmath418 is converted into a reduced coefficient matrix , whose @xmath419 columns are associated with @xmath420 the quotient polynomials @xmath396 , @xmath377 , as well as the @xmath104 , @xmath386 , can be directly read from any solution vector of the reduced linear system . the remaining @xmath104 , @xmath377 , can be reconstructed from the corresponding quotient polynomials and the prefactors @xmath421 . this allows to set up the bivariate polynomial @xmath422 as a solution to the gsa interpolation step . note that lemma [ lemma : lincomb_solution ] is formulated such that the leading column of the coefficient matrix @xmath374 and the first solution variable @xmath376 is removed and a new column @xmath423 and a new solution variable is appended . thus , repeated application of the lemma processes the coefficient matrix from left to right , which makes algorithm [ alg : reduction ] particularly easy to understand and at the same time saves a few indices in this part of the paper . a rather technical analysis of the algorithm shows that columns @xmath424 associated with @xmath394 , @xmath377 and @xmath425 , are replaced by @xmath426 and associated with @xmath391 , while columns associated with @xmath394 , @xmath377 and @xmath427 are deleted . columns associated with @xmath394 , @xmath386 , are simply carried over from the original coefficient matrix . this allows the following reformulation of the gsa interpolation step : [ prob : interpolation_reduced ] given a received vector @xmath32 with prefactors @xmath421 , @xmath377 , find a nonzero bivariate polynomial @xmath428\ ] ] such that @xmath429\leq{\ensuremath{d_{g_{\nu}}}}$ ] and @xmath430\leq{\ensuremath{d_{q_{\nu}}}}$ ] and @xmath431 } g_{\nu , \mu}\\ + \sum_{\substack{\nu\in\mathcal{f}^c\\\nu\geq t}}\binom{\nu}{t}z^{\nu - t } \sum_{\mu = s}^{{\ensuremath{d_{q_{\nu } } } } } \underbrace{\binom{\mu}{s}x^{\mu - s}}_{{\ensuremath{\triangleq}}\mathrm{lut}[s , i , \nu , \mu ] } q_{\nu , \mu } \bigg|_{(x , z)=\left(\alpha^{-i } , y_i\right)}=0 . \end{gathered}\ ] ] note that in case of re - encoding or combined prefactors with respect to @xmath70 , @xmath71 , we can replace @xmath432 by @xmath85 in . in that case , the input vector @xmath32 must be the new received vector after the re - encoding projection , i.e. , @xmath433}}$ ] with the actual received vector @xmath434 . the coefficient matrix associated with the reduced interpolation problem can be set up using algorithm [ alg : gsa_matrix_reduced ] . we emphasize that the sums in with summation index @xmath435 are independent of the received vector @xmath32 and thus their addends can be pre - calculated and stored in an @xmath436 lookup table @xmath437 $ ] . hence , the overlapping double summations that occur for @xmath377 have no negative effect on the complexity of setting up the coefficient matrix . as mentioned before , a solution @xmath40 of the gsa interpolation step ( problem [ prob : interpolation ] ) can be recovered from a solution @xmath438 of the reduced gsa interpolation step ( problem [ prob : interpolation_reduced ] ) using the prefactors . @xmath40 can then be used as input to the gsa factorization step ( problem [ prob : factorization ] ) in order to complete the decoding . a special case of the re - encoding projection that makes the reconstruction of @xmath40 particularly simple and hardware - friendly has been introduced in @xcite . a _ reduced _ gsa factorization step that could operate directly on @xmath438 in order to construct the result list was proposed in @xcite . the focus of this paper is on interpolation , which is why we do not delve into the details of factorization . [ ex : matrix1 ] for the setting of example [ ex : sier1 ] , table [ tab : sier_example ] delivers a sum of univariate constituent degrees of @xmath439 , which means that the number of unknowns in the original gsa interpolation step is @xmath440 . in contrast to this , we obtain @xmath441 unknowns for the reduced gsa interpolation step presented in this section if only sierpinski prefactors are used , @xmath442 unknowns if only re - encoding prefactors are used , and a mere @xmath443 unknowns if both types of prefactors are combined . the reduced coefficient matrices for the latter three cases can be set up using algorithm [ alg : gsa_matrix_reduced ] and the following inputs : 1 . sierpinski : @xmath444 , @xmath333 , @xmath445 and @xmath446}$ ] as in theorem [ thm : sier_factors ] . re - encoding : @xmath318 , @xmath332 , @xmath447 and @xmath448 as in theorem [ thm : reenc_factors ] . 3 . combined : @xmath318 , @xmath332 , either @xmath449 , @xmath450}$ ] or @xmath447 , @xmath448 as in theorem [ thm : combined_factors ] . [ ex : matrix2 ] it turns out that sierpinski prefactors work particularly well for the two conventional rs codes @xmath451 and @xmath452 considered by ktter and vardy in @xcite . the gsa for @xmath453 can correct up to @xmath454 errors with multiplicity @xmath455 and list size @xmath456 . this requires solving a linear system in @xmath457 unknowns . sierpinski prefactors alone reduce the system to @xmath458 unknowns , re - encoding alone to @xmath459 unknowns . when both techniques are combined , the number of unknowns shrinks to @xmath460 . a practically probably more relevant example is the gsa with multiplicity @xmath461 and list size @xmath462 for @xmath463 , it can correct up to @xmath464 errors , which is four errors beyond half its minimum distance . the associated linear system has @xmath465 unknowns , which can be diminished to @xmath466 unknowns using sierpinski prefactors alone , @xmath467 unknowns using re - encoding alone , and a mere @xmath468 unknowns when both techniques are combined . this is only slightly more than the @xmath469 unknowns that are required to decode up to @xmath470 errors with multiplicity @xmath471 and list size @xmath472 using re - encoding prefactors . hence , the newly introduced combined prefactors ( based on sierpinski prefactors ) allow to correct two additional errors at the comparatively low cost of having @xmath473 additional unknowns . we introduced the concept of a - priori known prefactors in the gsa interpolation step and showed how the well - known re - encoding projection can be interpreted in this framework . our main contribution is section [ sec : sierpinski ] , where we introduced the new type of sierpinski prefactors , which exist for a fairly wide range of code and gsa parameters . as opposing to re - encoding prefactors , sierpinski prefactors do not require a modification of the received vector since they are based on a simple property of the ground field . in general , prefactors allow to reduce the linear system of equations that is associated with the gsa interpolation step , i.e. , they allow to reconstruct a solution of the original big system from the solution of a reduced smaller system . the reduction of the linear system for re - encoding prefactors , sierpinski prefactors , and the ( practically most relevant ) combination of both has been described in section [ sec : factors ] . we stress that solving the reduced linear system using gaussian elimination is straightforward with complexity cubic in @xmath1 . however , it should not be too hard to adapt techniques like those from @xcite for solving the reduced system with _ quadratic _ complexity in @xmath1 since the structure of the original coefficient matrix is only marginally obstructed by the simple linear operator associated with the matrix reduction algorithm [ alg : reduction ] . in similar manner , it should be possible to adapt ktter interpolation @xcite such that arbitrary prefactors can be exploited . this would be a generalization of results from @xcite . we leave these problems for future work . another way to generalize our results would be to consider varying multiplicities for the received symbols , e.g. , in the context of algebraic soft - decision decoding with the ktter vardy algorithm @xcite . note that lines [ algline : gsa_matrix_reduced : betainit ] , [ algline : gsa_matrix_reduced : deltaloop ] , and [ algline : gsa_matrix_reduced : betaincr ] in algorithm [ alg : gsa_matrix_reduced ] can be deleted if @xmath474 in line [ algline : gsa_matrix_reduced : setval ] is replaced by @xmath437 $ ] as in . the algorithm admits a number of improvements that we ignored for the sake of slender pseudo code , e.g. , the binomial coefficient @xmath135 can be calculated directly at the beginning of the @xmath127-loop in line [ algline : gsa_matrix_reduced : nuloop ] and the rest of the loop can be skipped in case it is zero modulo @xmath133 . solving the linear system with a reduced coefficient matrix obtained by first calculating the full coefficient matrix with algorithm [ alg : gsa_matrix ] and then reducing it with algorithm [ alg : reduction ] results in the same solution ( space ) as directly solving the linear system with reduced coefficient matrix as obtained from algorithm [ alg : gsa_matrix_reduced ] . @xmath475 zero matrix over @xmath10 with @xmath476 rows and @xmath477 columns @xmath478 ( ) @xmath479 from @xmath120 to @xmath480 ( ) @xmath126 from @xmath120 to @xmath481 ( ) @xmath482 from @xmath120 to @xmath483 @xmath484 ( ) @xmath127 from @xmath126 to @xmath47 ( ) @xmath435 from @xmath479 to @xmath485 @xmath486 @xmath487 @xmath488 ( ) @xmath127 from @xmath120 to @xmath47 calculate @xmath421 and @xmath485 ( ) @xmath489 , @xmath490 @xmath475 zero matrix over @xmath10 with @xmath491 rows and @xmath492 columns @xmath478 ( ) @xmath479 from @xmath120 to @xmath480 ( ) @xmath126 from @xmath120 to @xmath481 @xmath493 ( ) @xmath482 from @xmath120 to @xmath483 ( ) @xmath494 @xmath495 @xmath496 ( ) @xmath127 from @xmath126 to @xmath47 @xmath497 ( ) @xmath435 from @xmath120 to @xmath485 ( ) @xmath498 from @xmath120 to @xmath499 @xmath500 @xmath501 @xmath502 @xmath503 the author is grateful to frank r. kschischang for stimulating discussions and valuable comments on the manuscript . r. koetter , j. ma , a. vardy , and a. ahmed , `` efficient interpolation and factorization in algebraic soft - decision decoding of reed solomon codes , '' in _ international symposium on information theory . isit 2003_.1em plus 0.5em minus 0.4emieee , jun . 2003 , p. 365 , doi : http://dx.doi.org/10.1109/isit.2003.1228381[10.1109/isit.2003.1228381 ] . j. massey , `` shift - register synthesis and bch decoding , '' _ information theory , ieee transactions on _ , vol . 15 , no . 1 , 122127 , jan . 1969 , doi : http://dx.doi.org/10.1109/tit.1969.1054260[10.1109/tit.1969.1054260 ] . y. sugiyama , m. kasahara , s. hirasawa , and t. namekawa , `` a method for solving key equation for decoding goppa codes , '' _ information and control _ , vol . 27 , no . 1 , pp . 8799 , jan . 1975 , doi : http://dx.doi.org/10.1016/s0019-9958(75)90090-x[10.1016/s0019-9958(75)90090-x ] . u. k. sorger , `` a new reed solomon code decoding algorithm based on newton s interpolation , '' _ ieee transactions on information theory , _ , vol . 39 , no . 2 , 358365 , mar . 1993 , doi : http://dx.doi.org/10.1109/18.212267[10.1109/18.212267 ] . gemmell and m. sudan , `` highly resilient correctors for polynomials , '' _ information processing letters _ , vol . 43 , no . 4 , pp . 169174 , sep . 1992 , doi : http://dx.doi.org/10.1016/0020-0190(92)90195-2[10.1016/0020-0190(92)90195-2 ] . m. sudan , `` decoding of reed solomon codes beyond the error - correction bound , '' _ journal of complexity _ , vol . 13 , no . 1 , 180193 , mar . 1997 , doi : http://dx.doi.org/10.1006/jcom.1997.0439[10.1006/jcom.1997.0439 ] . v. guruswami and m. sudan , `` improved decoding of reed solomon and algebraic - geometry codes , '' _ ieee transactions on information theory _ 45 , no . 6 , pp . 17571767 , sep . 1999 , doi : http://dx.doi.org/10.1109/18.782097[10.1109/18.782097 ] . v. olshevsky and m. a. shokrollahi , _ a displacement approach to decoding algebraic codes _ , ser . contemporary mathematics ( american mathematical society).1em plus 0.5em minus 0.4emams / siam , 2003 , vol . 265292 . a. zeh , c. gentner , and d. augot , `` an interpolation procedure for list decoding reed-solomon codes based on generalized key equations , '' _ ieee transactions on information theory _ , 57 , no . 9 , pp . 59465959 , 2011 , doi : http://dx.doi.org/10.1109/tit.2011.2162160[10.1109/tit.2011.2162160 ] . m. alekhnovich , `` linear diophantine equations over polynomials and soft decoding of reed solomon codes , '' in _ proceedings of the 43rd symposium on foundations of computer science _ , ser . focs 02.1em plus 0.5em minus 0.4emwashington , dc , usa : ieee computer society , 2002 , pp . 439448 . p. v. trifonov , `` efficient interpolation in the guruswami-sudan algorithm , '' _ ieee transactions on information theory _ , vol . 56 , no . 9 , pp . 43414349 , sep . 2010 , doi : http://dx.doi.org/10.1109/tit.2010.2053901[10.1109/tit.2010.2053901 ] . w. j. gross , f. r. kschischang , r. koetter , and p. g. gulak , `` towards a vlsi architecture for interpolation - based soft - decision reed solomon decoders , '' _ the journal of vlsi signal processing _ , vol . 39 , no . 1 , 93111 , jan . 2005 , doi : http://dx.doi.org/10.1023/b:vlsi.0000047274.68702.8d[10.1023/b:vlsi.0000047274.68702.8d ] . r. koetter , j. ma , and a. vardy , `` the re - encoding transformation in algebraic list - decoding of reed solomon codes , '' _ ieee transactions on information theory _ , vol . 57 , no . 2 , 633647 , feb . 2011 , doi : http://dx.doi.org/10.1109/tit.2010.2096034[10.1109/tit.2010.2096034 ] . i. s. reed and g. solomon , `` polynomial codes over certain finite fields , '' _ journal of the society for industrial and applied mathematics _ , vol . 8 , no . 2 , pp . 300304 , 1960 , doi : http://dx.doi.org/10.1137/0108018[10.1137/0108018 ] . p. delsarte , `` on subfield subcodes of modified reed solomon codes ( corresp . ) , '' _ ieee transactions on information theory , _ , 21 , no . 5 , pp . 575576 , sep . 1975 , doi : http://dx.doi.org/10.1109/tit.1975.1055435[10.1109/tit.1975.1055435 ] . y. wu , `` new list decoding algorithms for reed-solomon and bch codes , '' _ ieee transactions on information theory _ , vol . 54 , no . 8 , pp . 36113630 , aug . 2008 , doi : http://dx.doi.org/10.1109/tit.2008.926355[10.1109/tit.2008.926355 ] . h. hasse , `` theorie der hheren differentiale in einem algebraischen funktionenkrper mit vollkommenem konstantenkrper bei beliebiger charakteristik . '' _ j. reine angew . 175 , pp . 5054 , 1936 . a. zeh , c. gentner , and m. bossert , `` efficient list - decoding of reed solomon codes with the fundamental iterative algorithm , '' in _ information theory workshop 2009 . itw 2009_.1em plus 0.5em minus 0.4emieee , oct . 2009 , pp . 130134 , doi : http://dx.doi.org/10.1109/itw.2009.5351241[10.1109/itw.2009.5351241 ] . r. m. roth and g. ruckenstein , `` efficient decoding of reed solomon codes beyond half the minimum distance , '' _ ieee transactions on information theory _ , vol . 46 , no . 1 , 246257 , jan . 2000 , doi : http://dx.doi.org/10.1109/18.817522[10.1109/18.817522 ] . j. justesen and t. hholdt , _ a course in error - correcting codes_.1em plus 0.5em minus 0.4emzrich : european mathematical society , feb . [ online ] . available : http://www.worldcat.org/isbn/3037190019 c. senger , `` the periodicity transform in algebraic decoding of reed solomon codes , '' in _ communication , control , and computing ( allerton ) , 2012 50th annual allerton conference on_.1em plus 0.5em minus 0.4em ieee , oct . 2012 , pp . 168175 , doi : http://dx.doi.org/10.1109/allerton.2012.6483214[10.1109/allerton.2012.6483214 ] . a. ahmed , r. koetter , and n. r. shanbhag , `` vlsi architectures for soft - decision decoding of reed solomon codes , '' _ ieee transactions on information theory _ , vol . 57 , no . 2 , 648667 , feb . 2011 , doi : http://dx.doi.org/10.1109/tit.2010.2095210[10.1109/tit.2010.2095210 ] . r. koetter and a. vardy , `` algebraic soft - decision decoding of reed solomon codes , '' _ ieee transactions on information theory _ , vol . 49 , no . 11 , pp . 28092825 , 2003 , doi : http://dx.doi.org/10.1109/tit.2003.819332[10.1109/tit.2003.819332 ] .

to aid in the physical interpretation of planetary radii constrained through observations of transiting planets , or eventually direct detections , we compute model radii of pure hydrogen - helium , water , rock , and iron planets , along with various mixtures . masses ranging from 0.01 earth masses to 10 jupiter masses at orbital distances of 0.02 to 10 au are considered . for hydrogen - helium rich planets , our models are the first to couple planetary evolution to stellar irradiation over a wide range of orbital separations ( 0.02 to 10 au ) through a non - gray radiative - convective equilibrium atmosphere model . stellar irradiation retards the contraction of giant planets , but its effect is not a simple function of the irradiation level : a planet at 1 au contracts as slowly as a planet at 0.1 au . we confirm the assertion of guillot that very old giant planets under modest stellar irradiation ( like that received by jupiter and saturn ) develop isothermal atmospheric radiative zones once the planet s intrinsic flux drops to a small fraction of the incident flux . for hydrogen - helium planets , we consider cores up to 90% of the total planet mass , comparable to those of uranus and neptune . if `` hot neptunes '' have maintained their original masses and are not remnants of more massive planets , radii of.30 - 0.45 are expected . water planets are% larger than rocky planets , independent of mass . finally , we provide tables of planetary radii at various ages and compositions , and for ice - rock - iron planets we fit our results to analytic functions , which will allow for quick composition estimates , given masses and radii , or mass estimates , given only planetary radii . these results will assist in the interpretation of observations for both the current transiting planet surveys as well as upcoming space missions , including corot and kepler .

astro-ph0612671

we are still in the early days of a revolution in the field of planetary sciences that was triggered by the discovery of planets around other stars . exoplanets now number over 200 , with masses as small as @xmath05 - 7 @xmath3 @xcite . comparative planetology , which once included only our solar system s planets and moons , now includes sub - neptune to super - jupiter - mass planets in other solar systems . currently the most important class of exoplanets are those that transit the disk of their parent stars , allowing for a determination of planetary radii . the 14 confirmed transiting planets observed to date are all more massive than saturn , have orbital periods of only a few days , and orbit stars bright enough such that radial velocities can be determined , allowing for a calculation of planetary masses and bulk densities ( see * ? ? ? . a planetary mass and radius allows us a window into planetary composition @xcite . the 14 transiting planets are all gas giants ( see @xcite for calculations of structure and contraction of many planets ) although one planet , hd 149026b , appears to be @xmath02/3 heavy elements by mass @xcite . understanding how the transiting planet mass - radius relations change as a function of orbital distance , stellar mass , stellar metallicity , or uv flux , will provide insight into the fundamentals of planetary formation , migration , and evolution . tentatively , some work in this general direction is beginning @xcite . the transit method of planet detection is biased towards finding planets that orbit relatively close to their parent stars . this means that radial velocity followup will be possible for some planets as the stellar `` wobble '' signal is larger for shorter period orbits . however , for transiting planets that are low - mass , or that orbit very distant stars , stellar radial velocity measurements may not be possible . for planets at larger orbital distances , radial velocity observations may take years . therefore , for the foreseeable future a measurement of planetary radii will be our only window into the structure of these planets . estimates of masses will still be important , however , for useful constraints on planet formation theories . this will have to involve some degree of presumption regarding the composition of planets . orbital distances may give some clues as to a likely composition , but our experience over the past decade with pegasi planets ( or `` hot jupiters '' ) has shown us the danger of assuming certain types of planets can not exist at unexpected orbital distances . here we compute planetary radii as function of mass , composition , and stellar irradiation to help in these future planetary mass and composition estimates . in 2 we give a brief overview of the corot and kepler missions . in 3 we describe the scope of the calculations and our aims in this study . later , 4 describes our methods while 5 and 6 give our results for ice - rock - iron planets and gaseous planets , respectively . finally , 7 is our conclusions and suggestions for future work . the french / european corot mission , set to launch in december 2006 , and american kepler mission , set to launch in november 2008 , will revolutionize the study of exoplanets . corot will monitor 12,000 stars in each of 5 different fields , each for 150 continuous days @xcite . planets as small as 2 @xmath4 should be detectable around solar - type stars @xcite . the mission lifetime is expected to be at least 2.5 years . the kepler mission will continuously monitor one patch of sky , monitoring over 100,000 main sequence stars @xcite . the expected mission lifetime is at least 4 years . detection of sub - earth size planets is the mission s goal , with detection of planets with radii as small at 1 mercury radius possible around m stars . with these missions , perhaps hundreds of planets will be discovered with masses ranging from sub - mercury to many times that of jupiter . of course while planets close to their parent stars will preferentially be found , due to their shorter orbital periods and greater likelihood to transit , planetary transits will be detected at all orbital separations . in general , the detection of three successive transit will be necessary for a confirmed detection , which will limit confirmed planetary - radius objects to @xmath01.5 au . it is important to remember that corot and kepler will not determine planetary masses , only radii , so until followup radial velocity or astrometry work is done , planetary radii will be our only window into the composition of these new planets . in the next few years technology will allow the detection of transiting planets that range from many jupiter masses ( 1 @xmath5= 317.89 @xmath3 ) , to perhaps as small as 0.01 @xmath3 . interpreting these observations will require planetary radii to be understood over more than five orders of magnitude in mass . discoveries to date have been surprising , including transiting planets that are larger than standard models can explain ( such as hd 209458b and hat - p-1b ) , saturn - mass planets that have four neptune - masses worth of heavy elements ( such as hd 149026b ) , and amongst those that do not transit , neptune - mass planets that are quite hot and perhaps not ice - rich ( such as hd 68930b ) , and @xmath05 - 10 @xmath3 planets that some are calling `` super - earths '' ( such as gliese 876d ) . therefore , we think it is useful to take as broad a view as possible . we make few assumptions regarding composition , and we calculate radii of pure hydrogen - helium planets as small as 10 @xmath3 , water and rock dominated planets up to 1000 @xmath3(@xmath03 @xmath5 ) , along with many compositions in between . since we have the theoretical tools at our disposal , excluding some compositions a priori is unnecessary at this point . we use state - of - the - art equations of state ( eos ) for iron ( sesame 2140 , * ? ? ? * ) , olivine ( for generic `` rock '' ) ( aneos , * ? ? ? * ) , water ( aneos , * ? ? ? * ) , helium @xcite , and hydrogen @xcite . although detailed models for the thermal evolution of super terrestrial planets have been published @xcite , issues discussed in such detailed models , such as lithospheric thickness and interior temperature structure have only a small effect on planetary radii and are unobservable in the forseeable future . here we are interested in the most readily observed quantity : planetary radii . even for precisely determined light curves , planetary radii deteminations are somewhat imprecise owing to uncertainties in parent star radii , which can often reach 10% . for hydrogen / helium planets , irradiation from a parent star is a significant energy source that strongly affects evolution and contraction , and must be accounted for . we accurately incorporate stellar heating into our evolution ( contraction ) calculations through self - consistent non - gray model atmospheres under irradiation from 0.02 to 10 au . this is the first investigation of the evolution and contraction of hydrogen / helium planets under irradiation over such a wide range of orbital separations . other authors have previously computed radii for some subset of the compositions here . @xcite is perhaps the best known , in which they calculated the radius of zero - temperature spheres of h , he , c , fe , and mg for masses from 0.3 @xmath3 to 10 m@xmath6 . their thomas - fermi - dirac eoss are accurate at pressures where some degree of ionization takes place but less so for the lower pressures of our solar system s terrestrial planets . similary , @xcite calculated radii from 1 to 1000 @xmath3 for cold and warm h , h / he , ice , and rock planets . @xcite have recently investigated the radii of earth - like and mercury - like planets up to 10 @xmath3 . @xcite have calculated planetary radii of giant planets with and without cores . @xcite also investigated the radii of 51 peg b - like planets , with compositions ranging from h / he to rock . more recently @xcite have calculated radii of h / he planets with and without cores , at various orbital separations , but these models lack realistic atmospheric boundary conditions and equations of state for core materials . our results for water , rock , and iron bodies do not differ substantially from these other works , but we compute accurate radii over a wider range of masses , and also include mixed compositions . our calculations for the radii of h / he rich planets are an important improvement over other works , as we explicitly and accurately include irradiation from a parent star . in addition , we include the effects on planetary radii across a wide range of core masses . first we will look in a bit more detail into our equations of state before elaborating on the construction of the planet models . the total pressure , ( @xmath7 ) , can be expressed as : @xmath8 where @xmath9 is the pressure at zero - temperature and @xmath10 is the thermal pressure . for the degenerate interiors of massive planets , thermal effects are quite small for high atomic number species . for instance , zero - temperature equations of state for rock are thought to be accurate to within @xmath01 - 2% for use in uranus and neptune interior models @xcite , so we make no thermal corrections for rock and iron . for water , thermal pressure can be important at the @xmath010% level . @xcite find a relation for the @xmath11 of water that is suitable for the interior pressure - temperature ( _ p t _ ) profiles of uranus and neptune . this relation , @xmath12 where @xmath11 is in mbar , @xmath13 is in g @xmath14 , and @xmath15 in k , is relevant for `` typical planetary interior conditions . '' for @xmath16 g @xmath14 and @xmath17 k , @xmath18 mbar , or 10% of the total pressure , @xmath7 . for planets composed of any fraction of water , we assume that interior temperatures within the ice follow the uranus / neptune adiabat of @xcite , which reaches @xmath0550 k at 1 kar and 4000 k at 1 mbar , and add the thermal pressure correction of equation ( [ ice ] ) at every @xmath9 . this is our eos of `` warm ice . '' for the time being we will ignore the additional complexity of including the eos of methane and ammonia , solar system `` ices '' that are not as abundant as water . these molecules condense at colder temperatures and it is not clear how abundant these species may be in transiting planets , which are preferentially found relatively close to their parent stars . in we show our equations of state for cold ice , warm ice , rock , and iron . for hydrogen / helium envelopes , we compute internal adiabats with a helium mass fraction @xmath19 and do not include heavy elements . a description of this detailed h / he eos can be found in @xcite . the structure of spherically symmetric planets in hydrostatic equilibrium follow the relations set out below . equations ( [ dm1 ] ) and ( [ dm2 ] ) define mass conservation and hydrostatic equilibrium , respectively . equation ( [ dm3 ] ) defines energy conservation , which is employed in our evolution calculations of planets with a h / he envelope . @xmath20 @xmath21 @xmath22 here @xmath23 is the radius of a mass shell , @xmath24 mass of a given shell , @xmath13 the local mass density , @xmath7 the pressure , @xmath25 the gravitational constant , @xmath26 the planet s intrinsic luminosity , @xmath15 the temperature , @xmath27 the specific entropy , and @xmath28 the time . for planets composed only of water , rock , or iron , we do not utilize equation ( [ dm3 ] ) , as we assume a constant radius with age . given the small thermal component of the pressure for these materials , and the expected uncertainty in radius measurements , this assumption is valid . for planets where hydrogen and helium make up an appreciable mass fraction , following the thermal evolution and contraction of these planets is essential . we note that in these models we do not include additional interior energy sources such as tidal dissapation . this may be important for the hot jupiters . we also neglect helium phase separation , which will add @xmath01000 km in radius to cold giant planets at gyr ages @xcite . recall that for the planet with the most precisely determined radius hd 209458b , the 1@xmath29 radius uncertainty is still 1.9% ( @xmath01800 km ) , due to uncertainties in the stellar parameters @xcite . our evolution code for the calculation of the cooling and contraction of adiabatic giant planets is well - tested . it has been used to produce evolutionary models of jupiter and saturn @xcite , cool extrasolar giant planets @xcite , hot jupiters @xcite , and it is described in detail in @xcite and fortney04 . for all of these planets , it is the radiative atmosphere that serves as the bottleneck for cooling above the adiabatic h / he envelope . this is accounted for with our fully non - gray , self - consistent model atmosphere grids . the importance of using detailed atmosphere models for evolutionary calculations of hot jupiters is discussed in @xcite and @xcite . below the h / he envelope we assume that heavy elements are found within a distinct core . to model this core we use the eos of a 50/50 by mass ice / rock mix using the aneos zero - temperature water and olivine eoss @xcite . the compositions of the cores of jupiter and saturn are not known , and given that we model planets that likely formed at a variety of orbital distances , in which different ratios of ice / rock could be accumulated , this simple choice is a reasonable one . we ignore the heat content of the core on the thermal evolution of the planets . this is often done for evolutionary models of jupiter and saturn @xcite , as the error involved is small compared to other unknowns . please see @xcite for additional discussion on this point . we also neglect the `` transit radius '' effect : the apparent radius of a transiting planet is the radius where the slant optical depth through the planet s atmosphere reaches unity . the corresponding atmospheric pressure can vary across many orders of magnitude , depending on the wavelength @xcite . @xcite and @xcite have estimated this effect to be @xmath010% and 5% respectively , for hd 209458b , compared to some reference radius , such as the radiative - convective boundary or the 1 bar level . the _ hst _ light curve of @xcite and @xcite was obtained in a narrow wavelength band that overlaps the strong sodium d line absorptions at 589 nm . across a broad visible wavelength band , for most planets , the transit radius effect would likely be only a few percent . for the models presented here the radii correspond to a pressure of 1 bar . based on models from @xcite , an extension of @xmath05 - 6 atmosphere scale heights from the 1 bar level is needed to reach the optical transit radius . giant planets have been shown to be fully convective , or nearly so , beneath their thin radiative atmospheres ( for a review , see * ? ? ? the convection is thought to be quite efficient , and hence it is the radiative planetary atmosphere that serves as the bottleneck for escaping radiation and controls the cooling and contraction of the interior @xcite . as giant planet atmospheres have a number of atomic and molecular absorbers , including water , ammonia , methane , sodium , and potassium , these atmospheres are far from blackbodies @xcite . a model atmosphere grid , which serves as the upper boundary condition in these evolution calculations , relates the specific entropy ( @xmath27 ) of the planet s internal adiabat and atmospheric surface gravity ( @xmath30 ) to the planet s effective temperature ( @xmath31 ) . while @xmath27 and @xmath30 are calculated from the planetary structure , @xmath31 can only be accurately determined from a non - gray planetary atmosphere code . to compute the boundary condition for the evolution we compute self - consistent radiative - convective equilibrium atmospheric structure models on a large grid of gravities , intrinsic effective temperatures , and incident fluxes . each model computes an atmospheric temperature structure , accounting for deposition and re - radiation of incident starlight and convective transport and emission of internal thermal energy . the model , based on one developed for titan @xcite , was originally applied to the study of giant planet thermal structure by @xcite and has also been employed to study the atmospheres of the hot jupiters @xcite as well as brown dwarfs @xcite , albeit without incident radiation . the radiative transfer methods , chemical equilibrium calculations , and molecular and atomic opacities are summarized in the above publications as well as r. s. freedman & k. lodders , in prep . we assume solar metallicity atmospheres @xcite , and while the effect of condensation on atmospheric composition is included in the chemical equilibrium calculation @xcite we neglect the opacity of clouds . @xcite , @xcite , @xcite , and @xcite have previously discussed the importance of properly incorporating incident stellar flux into the hot jupiter atmosphere grids that serve as the upper boundary condition for the evolution of these planets , but this is also necessary at greater orbital separations for old and relatively low - mass planets . these planets have comparatively little internal energy , and their intrinsic effective temperature ( @xmath32 ) quickly falls below its equilibrium temperature ( @xmath33 ) , which is set entirely by absorption of stellar flux . here , by definition : @xmath34 when @xmath32 is small , incident stellar flux dominates over intrinsic flux and a deep atmospheric radiative zone grows , similar to a highly irradiated hot jupiter . this was previously discussed by j. b. pollack in the early 1990s and mentioned in @xcite , but we believe this is the first time this effect has been explicitly shown with detailed model atmospheres . shows model atmospheres computed at 0.1 au ( highly irradiated ) and 9.5 ( modestly irradiated ) . in both cases once @xmath35 , an isothermal region connects the deep interior adiabat to the upper atmosphere , whose structure is governed only by absorption of stellar flux ( see * ? ? ? note that the very low @xmath32 values in for this saturn - like planet would only occur after several hubble times of evolution . for all models we calculate a planet - wide average _ profile that is representative of the planet as a whole . in practice this means that the incident stellar flux is diluted by a factor of 1/4 ( see * ? ? ? * ) . the common approximation described in @xcite for including stellar flux into an atmosphere grid computed for _ isolated _ model atmospheres is only valid when @xmath36 . following the @xcite prescription to small @xmath32 leads to planetary radii that reach an asymptotic value governed by their bond albedo @xcite , and hence overestimates the radii of old , cold planets . we have computed model atmosphere grids across a range of surface gravities , for @xmath32 from 50 - 2000 k ( and as low at 10 k at low gravity ) , including the proper solar insolation for orbital distances of 0.02 , 0.045 , 0.1 , 1.0 , and 9.5 au ( which is saturn s orbital distance ) . extrapolation off the grid was performed as well . this generally only affected highly irradiated core - free low - mass planets . ] at gyr ages , we find bond albedos of @xmath00.05 - 0.1 at distances less than 0.1 au , and higher values of @xmath00.3 - 0.4 from 1 - 10 au , where cooler temperatures prevail and sodium and potassium , which absorb strongly in the optical , have condensed into clouds below the visible atmosphere @xcite . we have elected to ignore cloud opacity here , since considerable uncertainties remain concerning their effect on the atmospheric structure and albedos of egps . we will pursue this area in more detail in a later paper that focuses on the evolution of jupiter and saturn . a separate important issue is the opacity in the deep atmosphere at pressures near 1 kbar . while the temperature structure of highly irradiated atmospheres near @xmath371 kbar is of great importance for understanding giant planet thermal evolution @xcite , the opacities at these pressures remain highly uncertain @xcite . shows _ p t _ profiles computed for a jupiter - like planet from 0.02 to 10 au from the sun . the surface gravity @xmath30 is 25 m s@xmath38 and @xmath39 k in all models . these profiles are meant to roughly illustrate the atmospheres of jupiter - like planets at 4.5 gyr . @xmath39 k is very close to jupiter s current value @xcite , and we find that in our cooling calculations that the model planets reach a @xmath32 of @xmath0102 - 110 k at 4.5 gyr , which is only weakly dependent on stellar irradiation . a deep external radiative zone is found in the most highly irradiated models . for the planets at @xmath400.05 au convection does not begin until @xmath41 kbar . from 0.1 to 2 au the deep internal adiabat for all models begins at 300 + bar , but there is a 2nd , detached convective zone at pressures close to 1 bar . this detached convective zone grows at stellar distance increases , and by 3 au the convective zones have merged . only when these convective zones merge is the interior adiabat cooler as a function of orbital distance . the models from 0.1 to 2 au have essentially the same internal adiabat , meaning the planets would have the same radius at a given mass . as we will see , a striking consequence of this effect is that stellar irradiation at 2 au has approximately the same effect on retarding cooling and contraction as at 0.1 au , even though the incident fluxes vary by a factor of 400 ! it seems likely that planets with masses within an order of magnitude of the earth s mass will be composed primarily of more refractory species , like the planetary ices , rocks , and iron . within our solar system , objects of similar radius can differ by over a factor of 3 in mass , due to compositional differences . a planet with the radius of mercury , which is potentially detectable with kepler , could indicate a mass of 0.055 @xmath3 , like mercury itself , or a mass of 1/3 this value , like callisto , which has a radius that differs by only 30 km . with our equations of state , we are able to explore the radii of objects with any possible combination of ice , rock , and iron . in order to keep this task manageable , we have limited our calculations to several illustrative compositions . these include pure ice and ice / rock mixtures , which could be described as `` water worlds '' or ocean planets . " such objects in our solar system , like the icy satellites of the outer planets , generally have small masses . however , @xcite and @xcite have pointed out water - rich objects could reach many earth masses ( perhaps as failed giant planet cores ) and migrate inward to smaller orbital distances . we also consider planets composed of pure rock , rock and iron mixtures , and pure iron , more similar to our own terrestrial planets . the ice / rock and rock / iron mixtures are computed for 75/25 , 50/50 , and 25/75 percentages by mass , with ice always overlaying rock , and rock always overlaying iron . our results are shown in . since we make few assumptions regarding what is a reasonable planet , we have computed radii from masses of 0.01 to 1000 @xmath3 . for all compositions , the radii initially grow as @xmath42 , but at larger masses , compression effects become important . as a greater fraction of the electrons become pressure ionized , the materials begin to behave more like a fermi gas , and there is a flattening of the mass - radius curves near 1000 @xmath3 . eventually the radii shrink as mass increases , with radii falling with @xmath43 ( see * ? ? ? * ) . at the top left of we also show the size of various levels of uncertainty in planetary mass , as a percentages of a given mass , from 10 to 200% . for instance , if one could determine the mass of a 1 @xmath3 planet to within 50% , even a radius determination accurate to within 0.25 @xmath4 would lead to considerable ambiguity concerning composition , ranging from 50/50 ice / rock to pure iron . the shallow slope of the mass - radius curves below a few @xmath3 makes accurate mass determinations especially important for understanding composition . in we give the mass and radius for a subset of these planets . we note that from 1 - 10 @xmath3 we find excellent agreement between our models and the more detailed `` super - earth '' models of @xcite . on we have also plotted , in open circles , the masses and radii of solar system planets and moons . these planets can be used to validate our methods . for instance , detailed models of the earth s interior indicate that the earth is approximately 33% iron by mass with a core - mantle boundary at 3480 km @xcite . this composition is readily recovered from , where earth plots between the 25% and 50% iron curves , but closer to 25% . our simple earth model , with a iron / rock boundary at 3480 km yields a planetary radius within 100 km ( 1.5% smaller ) of the actual earth . given that our model lacks thermal corrections to eoss that are found in detailed earth models and that we ignore lower density species such as sulfur that are likely mixed with iron into the earth s core , we regard this agreement as excellent , and entirely sufficient with regard to the expected radii uncertainties as measured by transit surveys . elsewhere in the solar system , one can see that we recover ice / rock or rock / iron ratios of other bodies , which are derived by more complex models . a brief overview of the structure of the terrestrial planets and icy moons is given in @xcite . earth s moon is composed almost entirely of rock , with a very small iron core of radius @xmath44 km . here , the moon ( the leftmost circle ) plots on top of the line for pure rock . mercury is calculated to be @xmath060% iron by mass , and with our models mercury falls between the 50/50 ( rock / iron ) and 25/75 curves , but again , closer to 50/50 , which shows excellent agreement . titan is calculated to be composed of @xmath035% ices , and again we find excellent agreement , as titan falls between the 50/50 ( ice / rock ) and 25/75 curves , slightly closer to 25/75 . uranus and neptune can not be composed purely of ice , although ice likely makes up the bulk of their masses . in addition to we have also fit the radii of these planets to analytic functions that are quadratic in @xmath45 and linear in composition . these are equation ( [ fit1 ] ) , for ice / rock planets , and ( [ fit2 ] ) , for rock / iron , shown below : @xmath46 @xmath47 here @xmath48 is in @xmath4 and @xmath49 is in @xmath3 , while @xmath50 is the ice mass fraction ( 1.0 for pure ice and 0.0 for pure rock ) and @xmath51 is the rock mass fraction ( 1.0 for pure rock and 0.0 for pure iron ) . the fits were performed for planetary masses from 0.01 to 100 @xmath3 . for ice / rock , equation ( [ fit1 ] ) is on average accurate to within 2.5% . for rock / iron , equation ( [ fit2 ] ) is accurate to 1.5% . accuracy near 1 @xmath3 for both equations is better than 0.5% , although deviations can reach 10% for ice / rock at @xmath00.01 @xmath3 . if a mass and radius can be determined for a given planet these equations will allow for a quick and reliable composition estimate . however , since radial velocity or astrometric followup for these small planets will be extremely difficult and time consuming , radii may have to suffice as a proxy for mass for some time . given an assumed composition , such as earth - like , " one could assign masses to terrestrial - sized transiting planets . the distribution of planetary masses vs. orbital distance and stellar type could then be compared with planet formation models . planets around the mass of uranus and neptune ( @xmath52 @xmath3 ) to objects as large at 75 @xmath5 can be described by the same cooling theory . in general , planets with larger cores will have smaller radii , and planets closer to their parent stars will have larger radii at a given age then planets at larger orbital distances . in we plot the contraction of planets from 0.1 to 3 @xmath5 , as a function of age at various orbital distances . we also show the effect of a core of 25 @xmath3 , the approximate mass of heavy elements within jupiter and saturn @xcite . the first feature to notice is that , independent of mass , the spread in radii between 0.045 and 1 au is quite small compared to the large radii at 0.02 au and small radii at 9.5 au . as expected , the effect of the 25 @xmath3 core diminishes with increased planet mass , as the core becomes a relatively smaller fraction of the planet s mass . the radii at early ages it quite large , especially for the low - mass planets under intense stellar irradiation . these hot low - gravity planets are potentially susceptible to evaporation @xcite . as discussed in detail by @xcite , the physical properties of giant planets at young ages are quite uncertain . the models presented here do not include a formation mechanism and are arbitrarily large and hot at very young ages . @xcite have found that in their implementation of the core - accretion mechanism of giant planet formation @xcite giant planets form with radii than can be several tenths of a jupiter radius smaller than one computes with an arbitrarily large and hot start . these differences may last @xmath010 myr for a 1 @xmath5 object to hundreds of millions of years for planets of several jupiter masses . therefore , determining the radius of young giant planets , has the potential to elucidate their post - formation structure , and give us clues to their formation mechanism . in addition , a young hot jupiter would place constraints on planetary migration times @xcite . radii as a function of orbital distance , at 4.5 gyr , are shown in . for the planets from 0.3 ( saturn s mass ) to 3 @xmath5 , the radius curves are nearly flat between 0.1 and 1 au . these models predict that irradiation effects will remain important for planets @xmath53 au from their parent stars . _ planets with the same composition , from 0.1 to @xmath02 au , should have nearly the same radii . _ this is a consequence of the atmospheric temperature structures shown in . planets with cores are smaller , but there radii curves also flatten from 0.1 to @xmath01 au . we can also consider the effects of extremely large and small core masses . one view of our full range of 4.5 gyr - old planets is shown in where radii vs. mass is plotted at 5 orbital separations from 0.02 au to 9.5 au . coreless planets and planets with cores that are 10% , 50% , and 90% of their _ total planet mass _ are shown . note that these models have a core mass mass that is a constant _ percentage _ of planetary mass , meaning that a 50% core mass for a 10 @xmath3 planet is 5 @xmath3 , and for a 5 @xmath5 planet it is 800 @xmath3 . this plot is meant to show extremes , but is still quite illustrative . several real planets are shown as well . in uranus and neptune lie very close to the 90% heavy elements curves , as expected from more detailed models @xcite , although these planets likely contain more ice than rock . the radii of `` hot neptunes , '' should be similar to that of uranus and neptune , _ if _ hot neptunes do not suffer the effects of significant mass loss and are @xmath010% h / he , like uranus and neptune . @xcite and @xcite have shown that if these planets are remnants of much larger original planets , their radii may well exceed 1 @xmath5 . looking at the solar system s gas giants , it is clear that for jupiter because its mean radius is 2.2% smaller than its equatorial radius at 1 bar , 71492 km , which has become the standard `` jupiter radius . '' we note that planetary oblateness will be extremely difficult to determine from transit light curves @xcite . ] at 5.2 au , 10% heavy elements by mass is a reasonable estimate as it falls between the 1 au and 9.5 au 10% heavy element curves . saturn , at 9.5 au , is clearly more enhanced in heavy elements that jupiter , and from this plot perhaps 20% heavy elements would be estimated . given that one of the first ten transiting planets found has a core mass of @xmath060 - 80 @xmath3 , and it orbits a star with a metallicity 2.3@xmath54 that of the sun , it is probably realistic to expect the occasional planet around metal rich stars with core masses of @xmath0100 @xmath3 or more . we now turn to models computed with a constant core mass ( 10 , 25 , 50 , and 100 @xmath3 ) as function of age , mass , and stellar irradiation . and [ fig : rcores]_b _ are dense plots that show the radii of planets at two ages , 500 myr and 4.5 gyr . the contraction of planets by @xmath00.1 - 0.2 @xmath1 across this factor of @xmath010 in age is clear . planets at 0.045 au ( blue ) , 0.1 au ( orange ) , and 1.0 au ( red ) are very similar in radius at every mass , so their mass - radius curves trace similar path , while planets at 0.02 au ( magenta ) are always substantially larger , and planets at 9.5 au ( green ) are always substantially smaller . in tables [ t300 ] , [ t1000 ] , and [ t4500 ] contain our calculations of the radii of these planets at ages of 300 myr , 1 gyr , and 4.5 gyr . looking at radii at 4.5 gyr , we find generally good agreement with the models of @xcite for planets @xmath550.69 @xmath5 ( within 0.03 @xmath1 ) , but at lower masses we predict significantly smaller planets with differences than can approach 0.4 @xmath1 for coreless models at 0.02 au . it is likely that in the very low - mass core - free , highly irradiated corner of phase space that evolution models are perhaps most uncertain . evaporation effects may be important here as well . extending our calculations to parent stars other than the sun should be done with some care . @xcite have shown that planetary bond albedos are a function of the spectral type of the primary star . a planet around a later - type star , with a spectrum peaking closer to the infrared , will have a greater fraction of its flux absorbed by a planetary atmosphere , meaning a lower planetary bond albedo . although this must surely be accounted for eventually , given current uncertainties in chemistry and cloud formation in these planets , which leads directly to uncertainties in their atmospheric absorption and scattering properties , for now we will ignore this effect . however , the incident stellar fluxes of the currently known transiting planets already differ by a factor of _ _ 18 _ _ , so the differing magnitudes of incident radiation can not be ignored . if a planet s orbital distance from its parent star is @xmath56 , then the distance from the sun that the planet would have to be to receive this same flux , @xmath57 , is given by : @xmath58 or , if one uses a mass - radius relation such as , @xmath59 where @xmath60 may typically be @xmath03.5 across the h - r diagram , then @xmath61 for instance , hd 209458b , at 0.046 au from its g0v parent star ( @xmath62 @xmath63 ) , would receive that same level of irradiation at 0.036 au from the sun . shows the mass and radius of three transiting planets . the planet with the lowest mass , hd 149026b , receives an incident flux equivalent to that of a planet at 0.026 au from the sun . the figure clearly suggests that the planet has a core of perhaps 70 @xmath3 , which is confirmed by more detailed models @xcite . the middle diamond is tres-1 , which receives insolation equal to what is received from the sun at 0.059 au . using these models , one would estimate that the planet has a core of perhaps 25 @xmath3 , very similar to the bulk abundance of heavy elements in jupiter and saturn @xcite . hd 209458b is anomalous on this plot and must have an additional interior energy source . however , a 0.1 @xmath5 planet of pure h / he at 0.02 au ( if such a planet exists ) with a similar radius would have the correct radius . it should also be pointed out that all the hot jupiters could have additional energy sources of similar magnitude . as discussed in @xcite and @xcite , this would require planets such as tres-1 and hd 149026b to have even larger cores . with a common energy source , differences in the radii of giant planets would then be attributable to differing masses of heavy elements within their interiors . this idea is strengthened by the recent discovery of hot jupiters with large radii , such as hat - p-1b @xcite and wasp-1b @xcite , which shows that inflated radii are common features of these irradiated planets . in this paper we have taken a broad look at the radii of planets at nearly all possible masses and orbital separations . these calculations serve as a baseline for comparisons with transiting planets currently being discovered from the ground and those that will soon be discovered from space . since our main effort here is to calculate radii over a very large phase space , there are surely particular corners of space where these calculations could be improved , as we outline below . if in the coming years it becomes clear under what conditions planets may have an additional interior energy source , and how this changes as a function of orbital separation , incident stellar flux , and stellar type , this could be included in a new grid of calculations . if evaporation effects are observed for numerous transiting planets , across a wide range of planetary surface gravities and orbital separations , then the effects of appreciable mass loss could be added to these evolution calculations . the work of @xcite and @xcite is moving in this direction . ice - rich terrestrial planets close to their parent stars could have steam atmospheres @xcite which , depending on the atmospheric scale height , could lead to larger radii at a given mass for these objects . if the cooling of uranus and neptune are eventually better understood , then the additional energy due to the cooling of their heavy element interiors could be added to our evolution models of these lower - mass egps . including the eos of other forms of rock , or other astrophysical ices , such at ch@xmath64 and nh@xmath65 , would be useful when more detailed models are eventually needed . one could also envision more exotic planets , such as those that formed in solar systems where c / o@xmath661 , which could lead to carbon dominated terrestrial planets " @xcite , which would likely have radii intermediate between pure rock and ice ( zapolsky & salpeter , 1969 ; m. kuchner , personal communication ) . as there is in the solar system , there will always be ambiguity in the bulk composition of exoplanets , especially those that do not posses substantial low density gaseous envelopes . for instance , since a mixture of h / he gas and rocks has an eos that is similar to that of ice , this leads to considerable uncertainty in the composition of uranus and neptune . for a terrestrial - mass `` ocean planet , '' perhaps the best that we can expect will be be an understanding of a planet s ice / rock ratio , with a given uncertainty based on the mass and radius measurements , along with the knowledge that a the uncertainty could be even larger if the planet has a substantial iron core . since iron and silicates condense at similar temperatures @xcite , it is possible that these species will condense out in ratios similar to that found in the earth , although subsequent collisions ( e.g. , as experienced by the moon and probably mercury ) could alter this ratio , especially for smaller bodies . experience will tell us how to best classify newly discovered terrestrial - type planets , but at this point we would advocate composition classes based on simple ratios of ice / rock and rock / iron . our analytic fits reproduce the behavior of our models and will allow for mass estimates prior to radial velocity or astrometric follow up work . a clear prediction from our models for giant planets is that irradiation effects on planetary radii are not a simple function of stellar insolation . at a given planetary mass and composition , we predict a flattening in planetary radius as a function of orbital distance from @xmath00.1 to @xmath01 - 2 au . however , if unknown additional energy sources are still in play at these orbital distances , then this effect may be more difficult to see . given the recent flurry of transit detections it is likely that we are now just seeing the tip of the iceberg of transiting planets . even before detections from corot and kepler are announced it seems likely that we will have a steady stream of new planets to challenge our understanding . we thank kevin zahnle for a helpful discussion at the start of this project . jjf acknowledges the support of a spitzer fellowship from nasa and nsf grant ast-0607489 and msm from the nasa origins and planetary atmospheres programs . , g. a. , noyes , r. w. , kovacs , g. , latham , d. w. , sasselov , d. d. , torres , g. , fischer , d. a. , stefanik , r. p. , sato , b. , johnson , j. a. , pal , a. , marcy , g. w. , butler , r. p. , esquerdo , g. a. , stanek , k. z. , lazar , j. , papp , i. , sari , p. , & sipocz , b. 2006 , apj in press , astro - ph/0609369 , j .- p . , bennett , d. p. , fouqu , p. , williams , a. , dominik , m. , jorgensen , u. g. , kubas , d. , cassan , a. , coutures , c. , greenhill , j. , hill , k. , menzies , j. , sackett , p. d. , albrow , m. , brillant , s. , caldwell , j. a. r. , calitz , j. j. , cook , k. h. , corrales , e. , desort , m. , dieters , s. , dominis , d. , donatowicz , j. , hoffman , m. , kane , s. , marquette , j .- b . , martin , r. , meintjes , p. , pollard , k. , sahu , k. , vinter , c. , wambsganss , j. , woller , k. , horne , k. , steele , i. , bramich , d. m. , burgdorf , m. , snodgrass , c. , bode , m. , udalski , a. , szymaski , m. k. , kubiak , m. , wic eckowski , t. , pietrzyski , g. , soszyski , i. , szewczyk , o. , wyrzykowski , . , paczyski , b. , abe , f. , bond , i. a. , britton , t. r. , gilmore , a. c. , hearnshaw , j. b. , itow , y. , kamiya , k. , kilmartin , p. m. , korpela , a. v. , masuda , k. , matsubara , y. , motomura , m. , muraki , y. , nakamura , s. , okada , c. , ohnishi , k. , rattenbury , n. j. , sako , t. , sato , s. , sasaki , m. , sekiguchi , t. , sullivan , d. j. , tristram , p. j. , yock , p. c. m. , & yoshioka , t. 2006 , , 439 , 437 , a. c. , bouchy , f. , hbrard , g. , maxted , p. , pollacco , d. , pont , f. , skillen , i. , smalley , b. , street , r. a. , west , r. g. , wilson , d. m. , aigrain , s. , christian , d. j. , clarkson , w. i. , enoch , b. , evans , a. , fitzsimmons , a. , fleenor , m. , gillon , m. , haswell , c. a. , hebb , l. , hellier , c. , hodgkin , s. t. , horne , k. , irwin , j. , kane , s. r. , keenan , f. p. , loeillet , b. , lister , t. a. , mayor , m. , moutou , c. , norton , a. j. , osborne , j. , parley , n. , queloz , d. , ryans , r. , triaud , a. h. m. j. , udry , s. , & wheatley , p. j. 2007 , , 375 , 951 , b. , fischer , d. a. , henry , g. w. , laughlin , g. , butler , r. p. , marcy , g. w. , vogt , s. s. , bodenheimer , p. , ida , s. , toyota , e. , wolf , a. , valenti , j. a. , boyd , l. j. , johnson , j. a. , wright , j. t. , ammons , m. , robinson , s. , strader , j. , mccarthy , c. , tah , k. l. , & minniti , d. 2005 , , 633 , 465 ccccccccccc ice & 0.38 & 0.55 & 0.79 & 1.12 & 1.55 & 2.12 & 2.87 & 3.74 & 4.68 & 5.43 + 50/50 & 0.33 & 0.48 & 0.69 & 0.97 & 1.36 & 1.85 & 2.48 & 3.23 & 4.03 & 4.67 + rock & 0.25 & 0.37 & 0.54 & 0.77 & 1.08 & 1.48 & 1.97 & 2.54 & 3.14 & 3.64 + 67/33 ( earth - like ) & 0.24 & 0.34 & 0.50 & 0.71 & 1.00 & 1.36 & 1.80 & 2.31 & 2.84 & 3.29 + 50/50 & 0.23 & 0.33 & 0.48 & 0.68 & 0.95 & 1.30 & 1.71 & 2.19 & 2.69 & 3.12 + iron & 0.19 & 0.27 & 0.39 & 0.55 & 0.77 & 1.04 & 1.36 & 1.72 & 2.09 & 2.42 + [ tiri ] 0.02 & 0 & * & 2.326 & 1.883 & 1.656 & 1.455 & 1.378 & 1.342 & 1.327 & 1.308 & 1.311 & 1.315 & 1.284 + 0.02 & 10 & 1.102 & 1.388 & 1.465 & 1.422 & 1.349 & 1.325 & 1.311 & 1.306 & 1.295 & 1.304 & 1.310 & 1.281 + 0.02 & 25 & - & 0.493 & 0.945 & 1.133 & 1.220 & 1.253 & 1.267 & 1.275 & 1.276 & 1.294 & 1.304 & 1.277 + 0.02 & 50 & - & - & - & 0.801 & 1.030 & 1.144 & 1.193 & 1.226 & 1.245 & 1.276 & 1.292 & 1.270 + 0.02 & 100 & - & - & - & - & 0.669 & 0.939 & 1.055 & 1.128 & 1.187 & 1.242 & 1.270 & 1.256 + 0.045 & 0 & 2.795 & 1.522 & 1.345 & 1.255 & 1.240 & 1.228 & 1.212 & 1.206 & 1.199 & 1.210 & 1.203 & 1.170 + 0.045 & 10 & 0.801 & 1.012 & 1.091 & 1.124 & 1.168 & 1.185 & 1.185 & 1.188 & 1.188 & 1.204 & 1.199 & 1.168 + 0.045 & 25 & - & 0.447 & 0.793 & 0.968 & 1.071 & 1.124 & 1.147 & 1.161 & 1.173 & 1.195 & 1.193 & 1.164 + 0.045 & 50 & - & - & - & 0.719 & 0.921 & 1.033 & 1.084 & 1.119 & 1.148 & 1.179 & 1.183 & 1.157 + 0.045 & 100 & - & - & - & - & 0.627 & 0.863 & 0.968 & 1.036 & 1.101 & 1.148 & 1.163 & 1.146 + 0.1 & 0 & 1.595 & 1.395 & 1.270 & 1.197 & 1.202 & 1.198 & 1.187 & 1.182 & 1.178 & 1.189 & 1.178 & 1.144 + 0.1 & 10 & 0.755 & 0.956 & 1.035 & 1.084 & 1.134 & 1.157 & 1.160 & 1.164 & 1.168 & 1.183 & 1.174 & 1.142 + 0.1 & 25 & - & 0.438 & 0.767 & 0.938 & 1.042 & 1.099 & 1.123 & 1.138 & 1.153 & 1.174 & 1.169 & 1.138 + 0.1 & 50 & - & - & - & 0.702 & 0.899 & 1.011 & 1.063 & 1.098 & 1.129 & 1.158 & 1.159 & 1.132 + 0.1 & 100 & - & - & - & - & 0.618 & 0.847 & 0.950 & 1.018 & 1.084 & 1.128 & 1.140 & 1.121 + 1.0 & 0 & 1.504 & 1.325 & 1.222 & 1.169 & 1.182 & 1.182 & 1.173 & 1.169 & 1.168 & 1.179 & 1.169 & 1.136 + 1.0 & 10 & 0.727 & 0.921 & 1.004 & 1.063 & 1.116 & 1.141 & 1.146 & 1.152 & 1.158 & 1.173 & 1.165 & 1.134 + 1.0 & 25 & - & 0.433 & 0.754 & 0.923 & 1.027 & 1.085 & 1.110 & 1.127 & 1.143 & 1.164 & 1.159 & 1.130 + 1.0 & 50 & - & - & - & 0.693 & 0.888 & 0.999 & 1.051 & 1.087 & 1.120 & 1.149 & 1.149 & 1.124 + 1.0 & 100 & - & - & - & - & 0.613 & 0.839 & 0.941 & 1.009 & 1.075 & 1.119 & 1.131 & 1.113 + 9.5 & 0 & 0.929 & 0.951 & 0.983 & 1.020 & 1.070 & 1.106 & 1.127 & 1.146 & 1.167 & 1.169 & 1.156 & 1.130 + 9.5 & 10 & 0.565 & 0.733 & 0.847 & 0.939 & 1.016 & 1.072 & 1.104 & 1.131 & 1.157 & 1.163 & 1.152 & 1.127 + 9.5 & 25 & - & 0.394 & 0.664 & 0.826 & 0.942 & 1.024 & 1.073 & 1.146 & 1.142 & 1.153 & 1.146 & 1.124 + 9.5 & 50 & - & - & - & 0.635 & 0.823 & 0.951 & 1.020 & 1.072 & 1.119 & 1.137 & 1.137 & 1.118 + 9.5 & 100 & - & - & - & - & 0.587 & 0.810 & 0.920 & 0.999 & 1.072 & 1.107 & 1.119 & 1.107 + [ t300 ] 0.02 & 0 & * & 1.770 & 1.539 & 1.387 & 1.309 & 1.281 & 1.258 & 1.248 & 1.235 & 1.244 & 1.240 & 1.199 + 0.02 & 10 & 0.909 & 1.150 & 1.221 & 1.211 & 1.228 & 1.234 & 1.229 & 1.229 & 1.224 & 1.237 & 1.235 & 1.197 + 0.02 & 25 & - & 0.461 & 0.838 & 1.022 & 1.121 & 1.169 & 1.189 & 1.200 & 1.206 & 1.228 & 1.229 & 1.192 + 0.02 & 50 & - & - & - & 0.746 & 0.958 & 1.072 & 1.122 & 1.156 & 1.180 & 1.211 & 1.218 & 1.186 + 0.02 & 100 & - & - & - & - & 0.640 & 0.888 & 0.997 & 1.068 & 1.130 & 1.179 & 1.198 & 1.173 + 0.045 & 0 & 1.490 & 1.271 & 1.183 & 1.144 & 1.163 & 1.167 & 1.160 & 1.157 & 1.156 & 1.164 & 1.149 & 1.107 + 0.045 & 10 & 0.698 & 0.888 & 0.975 & 1.043 & 1.099 & 1.127 & 1.134 & 1.140 & 1.147 & 1.158 & 1.145 & 1.105 + 0.045 & 25 & - & 0.426 & 0.739 & 0.908 & 1.012 & 1.072 & 1.099 & 1.115 & 1.132 & 1.149 & 1.140 & 1.101 + 0.045 & 50 & - & - & - & 0.684 & 0.877 & 0.988 & 1.041 & 1.077 & 1.109 & 1.134 & 1.130 & 1.095 + 0.045 & 100 & - & - & - & - & 0.607 & 0.831 & 0.932 & 0.999 & 1.065 & 1.105 & 1.111 & 1.084 + 0.1 & 0 & 1.298 & 1.197 & 1.127 & 1.105 & 1.133 & 1.143 & 1.139 & 1.138 & 1.139 & 1.147 & 1.130 & 1.087 + 0.1 & 10 & 0.665 & 0.847 & 0.934 & 1.012 & 1.072 & 1.105 & 1.114 & 1.122 & 1.130 & 1.141 & 1.126 & 1.085 + 0.1 & 25 & - & 0.420 & 0.719 & 0.883 & 0.989 & 1.051 & 1.080 & 1.097 & 1.116 & 1.132 & 1.121 & 1.081 + 0.1 & 50 & - & - & - & 0.670 & 0.859 & 0.970 & 1.023 & 1.059 & 1.094 & 1.117 & 1.111 & 1.076 + 0.1 & 100 & - & - & - & - & 0.600 & 0.818 & 0.918 & 0.984 & 1.050 & 1.088 & 1.093 & 1.065 + 1.0 & 0 & 1.229 & 1.148 & 1.095 & 1.086 & 1.118 & 1.130 & 1.128 & 1.127 & 1.130 & 1.137 & 1.121 & 1.079 + 1.0 & 10 & 0.646 & 0.823 & 0.915 & 0.996 & 1.058 & 1.092 & 1.103 & 1.111 & 1.121 & 1.131 & 1.117 & 1.077 + 1.0 & 25 & - & 0.416 & 0.709 & 0.871 & 0.977 & 1.040 & 1.069 & 1.087 & 1.107 & 1.123 & 1.112 & 1.073 + 1.0 & 50 & - & - & - & 0.663 & 0.850 & 0.961 & 1.014 & 1.050 & 1.085 & 1.108 & 1.102 & 1.068 + 1.0 & 100 & - & - & - & - & 0.595 & 0.811 & 0.910 & 0.976 & 1.042 & 1.080 & 1.085 & 1.057 + 9.5 & 0 & 0.857 & 0.877 & 0.910 & 0.955 & 1.003 & 1.044 & 1.068 & 1.089 & 1.113 & 1.119 & 1.109 & 1.074 + 9.5 & 10 & 0.532 & 0.683 & 0.791 & 0.882 & 0.955 & 1.013 & 1.047 & 1.075 & 1.104 & 1.113 & 1.105 & 1.072 + 9.5 & 25 & - & 0.386 & 0.631 & 0.780 & 0.888 & 0.970 & 1.018 & 1.089 & 1.090 & 1.105 & 1.100 & 1.069 + 9.5 & 50 & - & - & - & 0.610 & 0.784 & 0.904 & 0.970 & 1.021 & 1.068 & 1.090 & 1.091 & 1.063 + 9.5 & 100 & - & - & - & - & 0.570 & 0.775 & 0.878 & 0.954 & 1.026 & 1.063 & 1.074 & 1.053 + [ t1000 ] 0.02 & 0 & * & 1.355 & 1.252 & 1.183 & 1.190 & 1.189 & 1.179 & 1.174 & 1.170 & 1.178 & 1.164 & 1.118 + 0.02 & 10 & 0.726 & 0.934 & 1.019 & 1.072 & 1.123 & 1.148 & 1.153 & 1.157 & 1.160 & 1.172 & 1.160 & 1.116 + 0.02 & 25 & - & 0.430 & 0.756 & 0.928 & 1.032 & 1.091 & 1.116 & 1.131 & 1.145 & 1.163 & 1.155 & 1.112 + 0.02 & 50 & - & - & - & 0.695 & 0.891 & 1.004 & 1.056 & 1.091 & 1.121 & 1.148 & 1.144 & 1.106 + 0.02 & 100 & - & - & - & - & 0.613 & 0.841 & 0.944 & 1.011 & 1.076 & 1.118 & 1.125 & 1.095 + 0.045 & 0 & 1.103 & 1.065 & 1.038 & 1.049 & 1.086 & 1.105 & 1.107 & 1.108 & 1.113 & 1.118 & 1.099 & 1.053 + 0.045 & 10 & 0.599 & 0.775 & 0.878 & 0.964 & 1.029 & 1.069 & 1.083 & 1.092 & 1.104 & 1.112 & 1.095 & 1.050 + 0.045 & 25 & - & 0.403 & 0.686 & 0.846 & 0.952 & 1.019 & 1.050 & 1.069 & 1.090 & 1.104 & 1.090 & 1.047 + 0.045 & 50 & - & - & - & 0.648 & 0.831 & 0.942 & 0.996 & 1.033 & 1.068 & 1.090 & 1.081 & 1.042 + 0.045 & 100 & - & - & - & - & 0.587 & 0.798 & 0.896 & 0.961 & 1.026 & 1.062 & 1.063 & 1.032 + 0.1 & 0 & 1.068 & 1.027 & 1.005 & 1.024 & 1.062 & 1.085 & 1.090 & 1.092 & 1.099 & 1.104 & 1.084 & 1.038 + 0.1 & 10 & 0.592 & 0.755 & 0.858 & 0.942 & 1.008 & 1.051 & 1.067 & 1.077 & 1.090 & 1.098 & 1.080 & 1.036 + 0.1 & 25 & - & 0.404 & 0.675 & 0.829 & 0.934 & 1.002 & 1.034 & 1.054 & 1.077 & 1.090 & 1.075 & 1.033 + 0.1 & 50 & - & - & - & 0.639 & 0.817 & 0.928 & 0.982 & 1.019 & 1.055 & 1.076 & 1.066 & 1.027 + 0.1 & 100 & - & - & - & - & 0.582 & 0.788 & 0.884 & 0.949 & 1.014 & 1.049 & 1.049 & 1.018 + 1.0 & 0 & 1.014 & 0.993 & 0.983 & 1.011 & 1.050 & 1.074 & 1.081 & 1.084 & 1.091 & 1.096 & 1.075 & 1.030 + 1.0 & 10 & 0.576 & 0.738 & 0.845 & 0.931 & 0.997 & 1.041 & 1.058 & 1.068 & 1.082 & 1.090 & 1.072 & 1.028 + 1.0 & 25 & - & 0.400 & 0.666 & 0.820 & 0.924 & 0.993 & 1.026 & 1.046 & 1.069 & 1.082 & 1.067 & 1.025 + 1.0 & 50 & - & - & - & 0.633 & 0.810 & 0.920 & 0.974 & 1.011 & 1.048 & 1.068 & 1.058 & 1.020 + 1.0 & 100 & - & - & - & - & 0.578 & 0.782 & 0.878 & 0.942 & 1.007 & 1.041 & 1.041 & 1.010 + 9.5 & 0 & 0.798 & 0.827 & 0.866 & 0.913 & 0.957 & 0.994 & 1.019 & 1.037 & 1.056 & 1.062 & 1.055 & 1.023 + 9.5 & 10 & 0.508 & 0.653 & 0.759 & 0.844 & 0.911 & 0.966 & 0.999 & 1.024 & 1.048 & 1.057 & 1.052 & 1.021 + 9.5 & 25 & - & 0.378 & 0.611 & 0.750 & 0.849 & 0.926 & 0.972 & 1.037 & 1.035 & 1.050 & 1.047 & 1.018 + 9.5 & 50 & - & - & - & 0.594 & 0.754 & 0.865 & 0.926 & 0.973 & 1.015 & 1.037 & 1.039 & 1.013 + 9.5 & 100 & - & - & - & - & 0.558 & 0.746 & 0.842 & 0.911 & 0.976 & 1.012 & 1.023 & 1.004 + [ t4500 ]

the paramagnetic phase of heavy fermion systems is investigated , using a non - perturbative local moment approach to the asymmetric periodic anderson model within the framework of dynamical mean field theory . the natural focus is on the strong coupling kondo - lattice regime wherein single - particle spectra , scattering rates , d.c . transport and optics are found to exhibit scaling in terms of a single underlying low - energy coherence scale . dynamics / transport on all relevant ()-scales are encompassed , from the low - energy behaviour characteristic of the lattice coherent fermi liquid , through incoherent effective single - impurity physics likewise found to arise in the universal scaling regime , to non - universal high - energy scales ; and which description in turn enables viable quantitative comparison to experiment .

cond-mat0505148

lanthanide based heavy fermion ( hf ) metals constitute a major , long studied class of correlated electron materials @xcite . their behaviour is quite distinct from conventional clean metals , the basic physics being driven by strong spin - flip scattering from essentially localised @xmath3-levels , generating the large effective mass and attendant low - energy scale indicative of strong interactions . the low - temperature ( @xmath4 ) state is a lattice - coherent fermi liquid with well defined quasiparticles and coherently screened @xmath3-spins , crossing over with increasing @xmath4 to essentially incoherent screening via independent kondo scattering , before attaining characteristic clean metallic behaviour . physical properties of hf are in consequence typically ` anomalous ' : e.g. the resistivity @xmath5 shows a strong non - monotonic @xmath4-dependence , while optics often exhibit rich structure from the microwave to the near infrared , and pronounced evolution on low temperature scales @xcite . theoretical treatments of hf centre on the periodic anderson model ( pam ) , in which a non - interacting conduction band hybridizes locally with a correlated @xmath3-level in each unit cell of the lattice ; or on its strong coupling limit , the kondo lattice model . the absence of exact results ( save for some in one dimension , see e.g. @xcite ) has long spurred the search for suitable approximation schemes . one such framework , which has had a major impact in recent years , is provided by dynamical mean field theory ( dmft , for reviews see @xcite ) . formally exact in the large - dimensional limit , the self - energy within dmft becomes momentum independent and hence spatially local , but still retains full temporal dynamics ; such that all lattice models map onto an effective single - impurity model with a self - consistently determined host @xcite . that raises an immediate question , easier asked than answered : to what extent are the properties of real hf materials captured within a dmft approach to the pam ? to answer this clearly requires direct quantitative comparsion of theory to experiment . and a prerequisite to that in turn is a method to solve the pam which dmft does not _ per se _ provide . the latter has of course been studied extensively using a wide variety of techniques . full scale numerical methods include the numerical renormalization group ( nrg ) @xcite , quantum monte carlo @xcite and exact diagonalization @xcite , while theoretical approaches encompass finite - order perturbation theory in the interaction @xmath6 @xcite , iterated perturbation theory @xcite , the lattice non - crossing approximation @xcite and the average @xmath7-matrix approximation @xcite , large-@xmath8 mean - field theory / slave bosons @xcite , the gutzwiller variational approach @xcite and the recently developed local moment approach @xcite . all of these methods naturally have their own virtues . but most possess significant , well known limitations @xcite , be it the general inability of perturbative approaches ( and in practice quantum monto carlo ) to handle strong interactions ; failure to recover fermi liquid behaviour at low - energies as arises in nca - based approaches , restriction to the lowest - energy fermi liquid behaviour as in large-@xmath8/slave boson mean - field theories , finite - size effects limiting exact diagonalization , and so on . to enable viable comparison to experiment requires an approach that can adequately handle all experimentally relevant energy and/or temperature scales in the strongly correlated hf regime of primary interest ; and indeed ideally also across the full spectrum of interaction strengths , such that intermediate valence and related behaviour can likewise be treated . one such is employed here , the local moment approach ( lma ) @xcite . via study of the generic asymmetric pam , our essential aims are ( i ) to provide a many - body description of dynamical and transport properties of paramagnetic hf , notably single - particle dynamics , d.c . transport and optical conductivities ; as considered here . ( ii ) to make direct quantitative comparison with experiment . that is taken up in the following paper where comparison to transport / optical properties of @xmath9 , @xmath10 , @xmath11 and @xmath12 is made . some remarks on the lma are apposite at this point since the paper will focus mainly on results obtained using the approach , with minimal technical details . intrinsically non - perturbative and as such capable of handling strong interactions , the lma @xcite introduces the physically intuitive notion of local moments @xcite from the outset . this leads directly to a ` two - self - energy ' description in which , post mean - field level , the key correlated spin - flip dynamics is readily captured ; corresponding in physical terms to dynamical tunneling between initially degenerate local moment configurations , which lifts the erstwhile spin degeneracy and restores the local singlet symmetry characteristic of a fermi liquid state . as with all techniques for lattice models within dmft , the lma originated in study of the single - impurity anderson model ( aim ) @xcite , where results for dynamics are known to give good agreement with nrg calculations @xcite , and for static magnetic properties with known exact results @xcite the approach has recently been developed to encompass the anderson lattice ( pam ) ; initially for the particle - hole symmetric limit @xcite appropriate to the kondo insulating sub - class of heavy electron materials , where for all interaction strengths the system is an ` insulating fermi liquid ' that evolves continuously from its simple non - interacting limit of a hybridization - gap insulator @xcite . from this a rich description of transport and optical properties of kondo insulators arises @xcite , particularly in strong coupling where the system is characterized by an exponentially small indirect gap scale @xmath13 , such that dynamics / transport exhibit scaling as functions of @xmath14 . exploiting that scaling enables direct comparison to experiment with minimal use of ` bare ' material / model parameters ; and in particular for three classic kondo insulators @xmath15 , @xmath16 and @xmath17 , leads to what we regard as excellent agreement between theory and experiment on essentially all relevant energy and temperature scales @xcite . the particle - hole symmetric pam is of course special , confined as it is to the case of kondo insulators . most recently the lma has been non - trivially extended to handle the generic asymmetric pam @xcite and hence hf metals ( with the insulating symmetric limit recovered simply as a particular case ) . single - particle dynamics at @xmath18 were considered in @xcite , with a natural emphasis on the strongly correlated kondo lattice regime of localised @xmath3-electrons but general conduction ( ` @xmath19 ' ) band filling @xmath20 . the problem was found to be characterized by a _ single _ low - energy coherence scale @xmath1 the precise counterpart of the insulating indirect gap scale @xmath21 , and likewise exponentially small in strong coupling in terms of which dynamics exhibit one - parameter universal scaling as a function of @xmath22 , independently of either the interaction strength or local @xmath23 hybridization . with increasing @xmath24 dynamics cross over from the low - energy quasiparticle behaviour required by and symptomatic of the coherent fermi liquid state , to essentially incoherent single - impurity kondo scaling physics at high-@xmath24 but still in the @xmath24-scaling regime and as such incompatible @xcite with a two - scale ` exhaustion ' scenario @xcite . in this paper we extend the work of @xcite to finite temperature , thereby enabling access to d.c . transport and optics . our primary focus is again the strongly correlated hf regime and attendant issues of scaling / universality ( that play a key role in comparing to experiment ) , the paper being organised as follows . the model and a bare bones description of background theory is introduced in section 2 , together with preliminary consideration of transport / optics . results for the thermal evolution of single - particle dynamics and scattering rates , and the connection between the two , are given in section 3 . the d.c . resistivity is considered in section 4 , with particular emphasis in this context on the crossover from the low-@xmath25 coherent fermi liquid to the high-@xmath26 incoherent regime , and explicit connection to single - impurity scaling behaviour . optical conductivities on all relevant @xmath27- and @xmath4-scales are investigated in section 5 ; and the paper concludes with a brief summary . the hamiltonian for the pam is given by @xmath28 : = _ c_i , c^_i c^_i & -&t_(i , j),c^_i c^_i + _ i,(_f+ f^_i - f^_i- ) f^_if^_i + & + & v_i , ( f^_i c^_i + ) [ 2.1 ] the first two terms represent the uncorrelated conduction ( @xmath19 ) band , + @xmath29 ( @xmath30 ) ; with @xmath19-orbital site energies @xmath31 and nearest - neighbour hopping matrix element @xmath32 , rescaled as @xmath33 in the large dimensional limit where the coordination number @xmath34 @xcite . the third term describes the correlated @xmath3-levels , @xmath35 , with site energies @xmath36 and on site coulomb repulsion @xmath6 ; while the final term @xmath37 hybridizes the @xmath19- and @xmath3-levels locally via the matrix element @xmath38 , rendering the otherwise localised @xmath3-electrons itinerant . the model is thus characterized by four independent dimensionless parameters , @xmath39 and @xmath40 ( @xmath41 sets the scale for the width of the free conduction band and is taken as the basic unit of energy , @xmath42 ) . an equivalent and somewhat more convenient set of ` bare'/material parameters ( with @xmath43 ) is @xmath44 and @xmath45 , where @xmath46 . this parameter space is large , and as such encompasses a wide range of physical behaviour for the paramagnetic phases we consider . the system is of course generically metallic , with non - integral @xmath3-level and @xmath19-band occupancies ( @xmath47 and @xmath48 respectively ) . that in turn extends from the trivial case of weakly correlated , perturbative behaviour , through intermediate valence to the strongly correlated heavy fermion ( hf ) regime . it is naturally the latter , characterized by a low - energy coherence scale @xmath49 , that is of primary interest . the hf ( or kondo lattice ) regime corresponds to essentially localised @xmath3-electrons , @xmath50 , but with arbitrary conduction band filling @xmath51 , the latter being controlled by @xmath52 ( which determines the centre of gravity of the free ( @xmath53 ) conduction band relative to the fermi level ) . it arises when @xmath54 , for @xmath55 and @xmath56 ( whence @xmath57 ) ; where @xmath58 , with @xmath59 the free conduction electron density of states as specified below and @xmath60 the fermi level . the heavy fermion regime forms our main focus here ; intermediate valence behaviour will be discussed in an experimental context in the following paper . the exception to the above behaviour arises when @xmath61 . here the system is generically a kondo insulator ( see _ eg @xcite ) , with an indirect gap in both its @xmath18 single - particle spectrum _ and _ optical conductivity @xcite ; the canonical example being the particle - hole symmetric pam with @xmath62 and @xmath63 , where @xmath64 for all @xmath6 . just like its metallic counterpart arising for @xmath65 , the kondo insulator is however a fermi liquid , evolving continuously with increasing interaction strength from its non - interacting limit ( in this case a ` hybridization gap insulator ' @xcite ) . as such , the kondo insulating state is obtained simply as a particular limit of the underlying theory . _ a knowledge of local single - particle dynamics and their thermal evolution is well known to be sufficient within dmft @xcite to determine transport properties ( see section 2.2 below ) . our initial focus is thus on the local retarded green functions @xmath66 and likewise @xmath67 for the @xmath19-levels , with corresponding spectra @xmath68 ( and @xmath69 or @xmath3 ) . some brief comments on the free conduction band are first required ( @xmath53 in equation ( [ 2.1 ] ) , where the @xmath19- and @xmath3-subsystems decouple ) ; specified by the local propagator @xmath70 with corresponding density of states ( dos ) @xmath59 . this is given by g_0^c ( ) & = & ( ^+-_c ) [ 2.2a ] + & = & [ 2.2b ] with @xmath71 , where for arbitrary complex @xmath72 h(z ) = ^+_- d [ 2.3 ] denotes the hilbert transform with respect to @xmath73 ; such that from equation ( [ 2.2a ] ) , @xmath74 corresponds simply to a rigid shift of @xmath75 by @xmath52 . equation ( [ 2.2b ] ) defines the feenberg self - energy @xmath76 @xcite as used below , with @xmath77 $ ] alone ( since @xmath78 from equations ( 2.2 ) ) . the free conduction band is thus determined by the non - interacting dos @xmath79 which , modulo the rigid @xmath52-shift , reflects the underlying host bandstructure , @xmath80 . while the formalism below holds for an arbitrary @xmath73 , explicit results will later be given for the hypercubic lattice ( hcl ) , for which within dmft @xcite @xmath81 is an unbounded gaussian ; and the bethe lattice ( bl ) , with compact spectrum @xmath82 for @xmath83 @xcite . the hcl will in fact be the primary case , because the bloch states characteristic of it ultimately underlie the lattice coherence inherent to low - temperature metallic hf behaviour . the major simplifying feature of dmft is that the self - energy becomes momentum - independent and hence site - diagonal @xcite ; and since we are interested in the homogeneous paramagnetic phase , the local green functions @xmath84 ( @xmath85 ) are also site - independent . straightforward application of feenberg renormalized perturbation theory @xcite , then gives the @xmath86 as g^c()&=&[2.4a ] + g^f()&= & [ 2.4b ] + & = & [ 2.4c ] where @xmath87 is the retarded @xmath3-electron self - energy ( @xmath88 such that @xmath89 ) . in equations ( 2.4 ) , @xmath90 is the feenberg self - energy for the fully interacting case , with @xmath91 $ ] the _ same _ functional of @xmath92 as it is of @xmath93 in the @xmath53 limit . in consequence , @xmath92 is given using equations ( [ 2.4a ] ) , ( 2.2 ) , ( [ 2.3 ] ) as g^c()=h ( ) [ 2.5a ] where @xmath94 ( @xmath95 ) is given by ( ; t)=^+-_c- . [ 2.5b ] let us first point up the physical interpretation of equations ( 2.4 ) , ( 2.5 ) . @xmath92 is a _ local _ propagator , and as such familiarly expressed as @xmath96 ; with the @xmath97-resolved conduction electron propagator @xmath98^{-1}$ ] and the usual conduction electron self - energy @xmath99 thus defined . since @xmath80 , it follows directly that g^c ( ) = ^+_-d _ 0()g^c ( ; ) g^c ( ; ) _ [ 2.6 ] with ( @xmath100 ) @xmath101^{-1}$ ] . but equation ( [ 2.6 ] ) is precisely the form equation ( [ 2.5a ] ) ( with equation ( [ 2.3 ] ) for @xmath102 ) , showing that g^c ( ; ) = [ ( ; t)-]^-1 [ 2.7 ] with @xmath94 related to the conduction electron self - energy by ( ; t)=^+-_c - _ c(;t ) ; [ 2.8 ] and hence ( via equation ( [ 2.5b ] ) ) that @xmath103^{-1}$ ] in terms of the @xmath3-electron self - energy alone ( because the @xmath3-levels alone are correlated ) . for an arbitrary conduction band ( specified by @xmath104 ) equations ( 2.4 ) , ( 2.5 ) are central ; for given the self - energy @xmath87 , and hence @xmath94 from equation ( [ 2.5b ] ) , @xmath92 follows directly from the hilbert transform equation ( [ 2.5a ] ) , and @xmath105 in turn from equation ( [ 2.4c ] ) . that statement hides however the truly difficult part of the problem : obtaining the self - energy @xmath87 . this is not merely a calculational issue , e.g. the need to solve the problem iteratively and self - consistently ( any credible approximation to @xmath87 will in general be a functional of self - consistent propagators ) . it reflects by contrast the longstanding problem of obtaining an approximate @xmath87 that , ideally : ( i ) handles non - perturbatively the full range of interaction strengths , from weak coupling ( itself accessible by perturbation theory or simple variants thereof @xcite ) all the way to the strongly correlated kondo lattice regime that is dominated by spin - fluctuation physics and typified by an exponentially small coherence scale @xmath1 . ( ii ) respects the asymptotic dictates of fermi liquid behaviour on the lowest energy ( @xmath27 ) and/or @xmath4 scales on the order of @xmath106 itself yet can also handle the _ full _ @xmath27 and/or @xmath4 range ; including the non - trivial dynamics that arise on energy scales up to many multiples of @xmath1 yet which remain universal ( and the existence of which we find to dominate transport and optics ) , as well as the non - universal energy scales prescribed by the bare material parameters of the problem . the success of any theory naturally hinges on the inherent approximation to @xmath87 . in this paper we employ the local moment approach ( lma ) @xcite , for it is known to satisfy the above desiderata and to our knowledge is currently the only theory that does . it is based on an underlying two - self - energy description a natural consequence of the mean - field approach from which it starts , and from which the conventional single self - energy @xmath107 follows together with the concept of _ symmetry restoration _ that is central to the lma generally @xcite . full details of the lma for the pam , including discussion of its physical basis and content , are given in @xcite . in particular the generic asymmetric pam ( as considered here ) is detailed in @xcite for @xmath18 ; and extension of it to finite-@xmath4 , required to consider transport and optics , follows the approach of @xcite where the particle - hole symmetric pam appropriate to the case of kondo insulators was considered . for that reason further discussion of the approach is omitted here . the reader is instead directed to @xcite on the pam , from which appropriate results will be used when required ; and to @xcite for anderson impurity models _ per se _ where details of the lma , including its stengths and limitations in relation to other approaches , are fully discussed . as mentioned above , a knowledge of single - particle dynamics is sufficient within dmft to determine @xmath108 transport properties @xcite . this arises because the strict absence of vertex corrections in the skeleton expansion for the current - current correlation function means only the lowest - order conductivity bubble survives @xcite , and a formal result for it is thus readily obtained . denoting the trace of the conductivity tensor by @xmath109 ( @xmath110 of which , denoted by @xmath111 , provides an approximation to the isotropic conductivity of a 3-dimensional system ) , this may be cast in the form ( ; t)=_0 f(;t ) [ 2.9 ] with @xmath112 merely an overall scale factor ( @xmath113 is the lattice constant and @xmath114 is typically of order @xmath115 ) . the dimensionless dynamical conductivity @xmath116 naturally depends on the lattice type , and for a bloch decomposable lattice such as the hcl is given ( with @xmath43 ) by @xcite f_hcl(;t)=^_- d_1d^c(;_1 ) d^c(;_1 + ) _ [ 2.10 ] where @xmath117^{-1}$ ] is the fermi function . here ( as in equation ( [ 2.6 ] ) ) , the notation @xmath118 denotes an average with respect to the non - interacting conduction band dos @xmath73 ; and the spectral density @xmath119 with @xmath120^{-1}$ ] from equation ( [ 2.7 ] ) . physically , @xmath121 ( @xmath122 ) represents the @xmath27-dependent conduction electron scattering rate ( inverse scattering time ) arising from electron interactions , @xmath123 ( @xmath124 from equation ( [ 2.8 ] ) ) . it is given using equation ( [ 2.5b ] ) by = _ i(;t ) = [ 2.11 ] in terms of the @xmath3-electron self - energy ; a knowledge of which thus determines the scattering rates ( considered explicitly in section 3.1 ) , and in consequence the dynamical conductivity equation ( [ 2.10 ] ) ( noting that @xmath125 ^ 2 + [ \gamma_i(\om;t)]^2)$ ] ) . results for @xmath126 obtained using the lma will be considered in sections 4,5 . here we simply point out an exact result , not apparently well known , for the weight of the drude peak in the @xmath18 conductivity . at @xmath18 , scattering at the fermi level is absent since the system is a fermi liquid , i.e. @xmath127 and hence @xmath128 . the leading low - frequency behaviour of @xmath129 is given by ^r_f(;0 ) ~^r_f(0;0 ) - ( -1)[2.12 ] where @xmath130^{-1}$ ] is the usual quasiparticle weight / inverse mass renormalization ; hence ( from equation ( [ 2.5b ] ) ) @xmath131 , where ^*_f = _ f + ^r_f(0;0 ) [ 2.13 ] is the renormalized @xmath3-level energy . a straightforward evaluation of equation ( [ 2.10 ] ) for @xmath18 and @xmath132 then shows that @xmath133 contains a @xmath134 drude ` peak ' ( as it must , reflecting the total absence of fermi level scattering and a vanishing @xmath18 resistivity ) . denoted by @xmath135 , it is given explicitly by f_drude(;0 ) = ( ) _ 0(-_c+ ) [ 2.14a ] or equivalently f_drude(;0 ) = ( ) _ 0(-_c+ ) [ 2.14b ] where @xmath136 and _ l = zv^2 . [ 2.15 ] equations ( 2.14 ) are exact , and bear comment . in the trivial limit @xmath53 where ( equation(2.1 ) ) the @xmath3-levels decouple from the conduction band , the total drude weight is naturally @xmath137 , the free conduction band dos at the fermi level ( recall @xmath138 ) . for any @xmath139 , the luttinger integral theorem requires ( n_c+n_f ) = ^-_c+1/_f^*_- _ 0 ( ) d + ( -_f^ * ) [ 2.16 ] ( with @xmath140 merely the unit step function ) . this again is an exact result , proven in @xcite . it holds for _ any _ interaction @xmath6 , reflecting the adiabatic continuity to the non - interacting limit that is intrinsic to a fermi liquid ; and shows in general that ( any ) fixed total filling @xmath141 determines @xmath142 entering equations ( 2.14 ) . of particular interest is of course the strongly correlated hf regime , where @xmath143 . here @xmath1 in equation ( [ 2.15 ] ) ( @xmath144 with @xmath145 ) is the coherence scale : exponentially small in strong coupling ( because @xmath146 is ) , it is the _ single _ low - energy scale in terms of which all properties of the system exhibit universal scaling ( as shown in @xcite and pursued below ) . in the hf regime , @xmath20 itself is moreover given ( see @xcite ) by n_c = ^-_c_-_0 ( ) d[2.17a ] showing that @xmath147 and @xmath20 are in essence synonymous , @xmath148 being determined by @xmath147 alone . conjoining this with equation ( [ 2.16 ] ) gives n_f = ^-_c+1/_f^*_-_c _ 0 ( ) d + ( -_f^ * ) [ 2.17b ] so as @xmath143 ( the hf regime ) , @xmath149 is _ also _ determined by @xmath147 alone , and is typically of order unity . it is this that determines @xmath150 entering equation ( [ 2.14b ] ) for @xmath135 ; showing in turn that the net drude weight is itself @xmath151 , and hence exponentially diminished compared to the free conduction band limit . we add that kondo insulators , arising generically for @xmath152 as mentioned earlier , are also encompassed by the above . using @xmath153 , the luttinger theorem equation ( [ 2.16 ] ) shows that @xmath152 arises either for @xmath154 ( for an unbounded @xmath104 ) or for @xmath142 outside the band edges of a compact @xmath104 ; such that in either case the drude weight in equations ( 2.14 ) vanishes , symptomatic of the vanishing @xmath18 d.c . conductivity characteristic of the kondo insulating state . our focus above has naturally been on the canonical case of a bloch decomposable lattice . for a bethe lattice by contrast , @xmath116 is given @xcite by ( _ cf _ equation ( [ 2.10 ] ) ) f_bl(;t)=^_- d_1d^c(_1 ) d^c(_1 + ) [ 2.18 ] where @xmath155 ( @xmath156 ) is the local conduction band spectrum . in particular the d.c . conductivity at @xmath18 follows as @xmath157 ^ 2 $ ] ; which , using @xmath158 together with equations ( [ 2.2a ] ) , ( 2.5 ) , ( [ 2.12 ] ) , is given by f_bl(0;0 ) = [ _ 0(-_c+)]^2 . [ 2.19 ] in contrast to equations ( 2.14 ) there is thus no drude @xmath134-peak and the @xmath18 d.c . resistivity is in general finite , reflecting of course that the underlying one - particle states of the bl are not coherent bloch states . hence , aside from the case of kondo insulators where the bl ( like the hcl ) does capture the vanishing @xmath18 d.c . conductivity and indirect - gapped optics characteristic of the insulator @xcite , the ` joint density of states ' type formula equation ( [ 2.19 ] ) should not be taken seriously when considering transport / optics of real materials on sufficiently low @xmath4 and/or @xmath27 scales ( as discussed further in section 4 ) . we turn now to lma results for single - particle dynamics at finite-@xmath4 . our natural focus will be the strong coupling kondo lattice regime ( where @xmath143 ) , characterized by the low - energy lattice scale @xmath159 . this scale is of course a complicated function of the bare / material parameters , @xmath160 ( detailed lma results for it are given in @xcite , and nrg results in @xcite ) . that dependence is however a subsidiary issue in comparison to the fact that , because @xmath1 becomes exponentially small in strong coupling , physical properties exhibit scaling in terms of it ; i.e. depend universally on @xmath161 , independently of the interaction strength . universality in strong coupling single - particle dynamics at @xmath18 has been considered in @xcite for the generic pam ; the essential findings of which are first reprised for use below . ( i ) both the @xmath19-electron spectrum @xmath155 ( @xmath162 with @xmath43 ) and the @xmath3-electron spectrum @xmath163 ( with @xmath164 introduced in section 2 ) , exhibit universal scaling as a function of @xmath22 in a manner that is _ independent _ of both the interaction strength @xmath6 _ and _ hybridization matrix element @xmath38 . ( ii ) that scaling depends in general only on @xmath147 ( or equivalently the conduction band filling @xmath20 , see equation ( [ 2.17a ] ) ) which embodies the conduction band asymmetry ; and on @xmath165 reflecting the @xmath3-level asymmetry . more specifically , ( iii ) in the coherent fermi liquid regime arising for @xmath166 , the @xmath3- scaling spectra depend only on @xmath147 and are in fact independent of @xmath45 as well as @xmath6 and @xmath38 . in this low-@xmath24 regime the scaling spectra amount in essence to the quasiparticle behaviour ( equations ( 3.11 ) of @xcite ) required by the asymptotic dictates of low - energy fermi liquid theory . ( iv ) for @xmath167 by contrast the @xmath3- scaling spectra depend on the @xmath3-level asymmetry @xmath45 ( albeit rather weakly ) , but are now independent of @xmath147 and indeed also of the lattice type ; and the spectrum contains a long , logarithmically slowly decaying spectral tail . ( v ) the latter behaviour , which sets in progressively for @xmath168 , reflects in turn the crossover to incoherent effective single - impurity physics that one expects to arise for sufficiently high @xmath27 ( and/or @xmath4 ) : for @xmath167 the _ scaling form _ of the @xmath3-spectrum is found to be precisely that of an anderson impurity model ( aim ) . with increasing @xmath24 , dynamics thus cross over from the low - energy quasiparticle behaviour symptomatic of the lattice coherent fermi liquid state to single - impurity kondo scaling physics at high @xmath24 ( and that this crossover occurs in a single @xmath161 scaling regime is thus incompatible with the occurrence of ` two - scale exhaustion ' @xcite as explained in @xcite ) . figure [ fig1 ] summarises representative results for @xmath18 scaling dynamics ( irrelevant non - universal energy scales such as @xmath6 , @xmath41 ( @xmath169 ) or @xmath170 are of course projected out in scaling spectra @xcite ) . the main figure shows @xmath23 scaling spectra for the hcl as functions of the scaled frequency @xmath171 , for @xmath172 with @xmath173 ( dashed , and @xmath174 ) and @xmath175 ( solid , with @xmath176 ) . the @xmath177 example corresponds to the particle - hole ( p - h ) symmetric kondo insulator , whose spectra are thus gapped at the fermi level @xmath178 ( with @xmath159 here corresponding to the insulating gap scale @xcite ) . for the asymmetric conduction band @xmath179 by contrast , the gap ( which is well developed in strong coupling @xcite ) moves above the fermi level ; and a sharp lattice - kondo resonance symptomatic of the hf metal , straddling the fermi level and of width @xmath180 , takes its place in the @xmath3-spectra . the inset shows the @xmath3-spectra on a much larger @xmath24 scale ; displaying the @xmath31-independence of the slow logarithmic tails @xcite and reflecting the crossover to effective single - impurity behaviour ( which we emphasise arises whether the system is a hf metal or a kondo insulator ) . at finite temperatures , what one expects for the strong coupling scaling spectra is clear : they should now depend universally on @xmath181 _ and _ @xmath182 . that this arises correctly within the present lma is shown in figure [ fig2 ] . for a fixed @xmath183 , the @xmath3- and @xmath19-spectra are shown for progressively increasing interaction strengths @xmath184 and @xmath185 with @xmath186 ; for @xmath187 and @xmath172 ( corresponding results for the p - h symmetric limit have been obtained in @xcite ) . the inset shows the @xmath3-spectra on an absolute scale ( _ vs _ @xmath188 ) , where the exponential reduction of the @xmath1-scale with increasing @xmath6 is clearly seen from the change in the width of the resonance . the main figures by contrast show the spectra as functions of @xmath24 , from which the @xmath6-independent scaling collapse is evident ; repeating the calculations with different @xmath189 likewise shows the scaling to be independent of @xmath38 . this behaviour is not of course confined to the chosen @xmath26 , and figure [ fig3 ] shows the resultant lma scaling spectra for a range of @xmath26 ( again for the representative @xmath190 ) . figures [ fig3 ] and [ fig2 ] show clearly the thermal broadening and ultimate collapse of the @xmath3-resonance with increasing @xmath26 ; which is naturally accompanied by a redistribution of spectral weight leading to infilling of the ( @xmath191 ) spectral gap seen in figure [ fig1 ] for @xmath18 . in fact by @xmath192 this gap is already obliterated , and the lattice kondo resonance also significantly eroded . this behaviour is typical of the metallic hf state . by contrast , corresponding results for the p - h symmetric kondo insulator ( @xmath193 ) are shown in figure 4 of @xcite . in that case the insulating gap at the fermi level fills up with increasing temperature , and the fermi level @xmath194 in particular increases monotonically with increasing temperature ; in contrast to the the asymmetric hf spectra shown above where @xmath195 diminishes with @xmath26 . two further points regarding figure [ fig3 ] should be noted . first , the thermal evolution of the @xmath3- and @xmath19-spectra differ somewhat in terms of the persistence of a pseudogap the @xmath3-spectrum shows no sign of the gap by @xmath196 , while a weak pseudogap structure persists in the @xmath19-spectrum up to @xmath197 ; this reflects the rapid spread of spectral weight caused by the meltdown of the sharp resonance in the @xmath3-spectra , of which there is no counterpart in the @xmath19-spectra . second , the inset to figure [ fig3 ] shows the @xmath3-spectra on an enlarged frequency scale out to @xmath198 , from which it is seen that the high frequency behaviour of the finite-@xmath26 scaling spectra coincide with that for @xmath18 . this is physically natural , since one expects the dominant influence of temperature to be confined to frequencies @xmath199 . the corollary of course is that non - universal frequencies are affected only on non - universal , and thus in general physically irrelevant , temperature scales ( as shown in figure 5 of @xcite for the p - h symmetric case ) . we consider now the scattering rates @xmath200 that underlie the evolution of the conductivity , and are given explicitly in terms of the @xmath3-electron self - energy by equation ( [ 2.11 ] ) . since the system is a fermi liquid with @xmath201 , at @xmath18 there is of course no scattering at the fermi level , @xmath202 . the low frequency behaviour of the @xmath18 scattering rate can be understood qualitatively by using the low-@xmath27 expansion of @xmath203 ( equation ( [ 2.12 ] ) ) and simply neglecting the imaginary part @xmath204 ; leading to ^-1(;t=0 ) ( - _ f^ * ) [ 3.1 ] with @xmath205 the renormalized level . restoring the small but strictly non - vanishing @xmath206 naturally implies a narrow resonance centred on @xmath207 instead of a pure @xmath208-function . at finite temperature , we likewise expect the scattering rate to increase from zero in the neighbourhood of the fermi level , reflecting the finite-@xmath4 contribution to @xmath209 ; and that this will simultaneously lead to further , thermal broadening of the resonance at @xmath150 . the above picture is corroborated by lma results as shown in figure [ fig4 ] , displaying the @xmath22 dependence of @xmath200 ( in units of @xmath42 ) arising in strong coupling for @xmath187 and @xmath172 , for a range of temperatures @xmath210 between @xmath211 and @xmath212 . in this case the renormalized level is found to be @xmath213 , precisely where @xmath214 has a narrow resonance . with increasing temperature the resonance is indeed seen to broaden and decrease in intensity ; and we reiterate that this occurs for temperatures @xmath4 set by the scale @xmath1 the sole low - energy scale characteristic of the problem in strong coupling . excepting the lowest @xmath26 we also note that scattering rates in the vicinity of the fermi level are on the order of @xmath215 of the bandwidth @xmath41 , values some two or so orders of magnitude higher than for conventional clean metals ( and indicative of the higher d.c . resistivities that are typical of heavy fermion materials @xcite ) . neither is this behaviour confined to a narrow @xmath26 regime since even for @xmath216 the scattering rates decay very slowly with @xmath26 ; the fermi level scattering rate for example is readily shown to decay as @xmath217 . the scattering rates are also related to the @xmath3-electron scaling spectra considered above . for the kondo insulating p - h symmetric pam , it was shown in @xcite that the dimensionless scattering rate defined as = _ i(;t ) [ 3.2 ] coincides asymptotically with the @xmath3-spectral function , specifically ~_0 d^f ( ) [ 3.3 ] in the regime @xmath218 for any @xmath26 ( the spectral ` tails ' ) , and for all @xmath219 for sufficiently large @xmath216 . equation ( [ 3.3 ] ) is in fact readily shown to be quite general , and not dependent on p - h symmetry . that it holds for hf metals embodied in the asymmetric pam is illustrated in figure [ fig5 ] , where for @xmath187 and @xmath172 the strong coupling @xmath220 and @xmath221 _ vs _ @xmath24 are compared , for @xmath222 in the left panel and @xmath183 and @xmath223 in the right panel . the high - frequency behaviour of the scaling spectrum @xmath221 is itself known , being given ( here for @xmath224 explicitly ) by @xcite _ 0 d^f ( ) ( + ) [ 3.4 ] with @xmath113 a pure constant @xmath225 . these slowly decaying logarithmic tails are evident in figure [ fig5 ] , and as mentioned in section 3 embody the connection to effective incoherent single - impurity physics on high energy scales . they are independent of the interaction @xmath6 , local hybridization @xmath38 , underlying conduction band asymmetry @xmath31 , and even of the lattice type ; depending , albeit weakly , only on the @xmath3-level asymmetry @xcite . the above discussion of scattering rates leads naturally to consideration of transport ; beginning with the d.c . limit where ( section 2.2 ) the static conductivity @xmath226 , with @xmath116 given for the hypercubic lattice by equation ( [ 2.10 ] ) . in the strong coupling regime we expect static transport to exhibit universal scaling in terms of @xmath25 , and our aim here is to understand its thermal evolution across the full @xmath26 range . transport on non - universal temperature scales @xmath227 ( @xmath228 ) or @xmath229 , will be discussed briefly at the end of the section . for the p - h symmetric kondo insulator , lma results for the @xmath26-dependence of the scaling resistivity have been considered in @xcite ( in this case @xmath230 is equivalently the insulating gap scale ` @xmath13 ' ) . the @xmath18 resistivity is naturally infinite reflecting the gapped ground state , the scaling resistivity @xmath231 has an activated form @xmath232 for @xmath233 ( with @xmath234 a pure constant @xmath235 and hence a ` transport gap ' of @xmath236 ) ; and @xmath5 decreases monotonically with increasing @xmath26 , tending asymptotically to incoherent single - impurity scaling behaviour ( @xcite and figures [ fig7],[fig8 ] below ) . for the general case of heavy fermion metals the situation is of course quite different , and what one expects in qualitative terms well known @xcite . the @xmath18 resistivity vanishes , reflecting the absence of fermi level scattering and the underlying coherence generic to any bloch decomposable lattice . with increasing temperature @xmath5 increases ( initially as @xmath237 for @xmath233 @xcite ) , passes through a maximum at @xmath238 a classic signature of hf compounds @xcite and decreases thereafter in the strong coupling , kondo lattice regime of interest . figure [ fig6 ] shows lma results for @xmath5 _ vs _ @xmath26 for fixed @xmath239 , and with increasing interaction @xmath240 and @xmath185 for @xmath186 . the scaling collapse is clearly evident : while the low - energy scale @xmath241 itself diminishes exponentially on increasing @xmath6 , universal scaling of @xmath5 as a function of @xmath25 indeed arises in strong coupling , independent of interaction strength ( and likewise readily shown to be @xmath38-independent on repeating the calculations varying @xmath189 ) . this leads us first to comment briefly on the issue of ` the coherence scale ' , characterising the crossover from low - temperature lattice coherent behaviour to high - temperature effective single - impurity behaviour . experimentally , many such identifications of the low - energy scale are commonly employed . some groups use @xmath238 at which @xmath5 peaks , others identify the scale via the inflection points ( @xmath242 , marked by arrows in figure [ fig6 ] ) , via the leading @xmath237 behaviour of @xmath5 at low-@xmath4 , or via the onset of the ` log - linear ' regime @xcite ( shown in inset ( a ) to figure [ fig6 ] and seen in many experimental systems @xcite ) ; the inverse of the @xmath243 paramagnetic susceptibility , or the width of the lattice kondo resonance , are other possibilities . this leads to what at first sight might seem a plethora of low - energy scales . the key point however is that , because physical properties in strong coupling scale universally in terms of _ one _ low - energy scale , all the above definitions of ` the coherence scale ' are fundamentally equivalent : all are proportional to @xmath1 , and hence to each other in figure [ fig6 ] for example , the inflection points in @xmath5 lie at @xmath244 and @xmath245 , and the peak maximum at @xmath246 . as for single - particle dynamics and scattering rates considered in section 3 , the @xmath26-dependent scaling resistivity is independent of @xmath6 or @xmath38 ( as above ) but depends in general on @xmath147 ( reflecting the conduction band asymmetry and determining @xmath20 via equation ( [ 2.17a ] ) ) and @xmath45 ( reflecting the @xmath3-level asymmetry ) . to consider this figure [ fig7 ] shows the resultant scaling resistivities @xmath5 _ vs _ @xmath26 for @xmath172 , and a range of different @xmath147 @xmath247 and @xmath248 , corresponding respectively to conduction band fillings @xmath249 and @xmath250 . the @xmath177 example is the kondo insulator @xcite , with its characteristic diverging @xmath5 as @xmath251 . the others are all hf metals , and exhibit the same qualitative behaviour for all @xmath147 a positive slope for @xmath252 , going through the maximum and then decreasing monotonically for @xmath253 ; the coherence peak itself increasing monotonically with @xmath147 , albeit slowly such that @xmath254 for the @xmath147-range shown . qualitatively similar behaviour is found on varying the @xmath3-level asymmetry @xmath45 for fixed conduction band asymmetry embodied in @xmath147 , although quantitatively this effect is appreciably less . the significant @xmath147-dependence of @xmath255 seen in figure [ fig7 ] for @xmath256 is intuitively natural : the strong coupling kondo lattice regime corresponds to @xmath257 , but with variable conduction band filling ( @xmath20 ) controlled by @xmath147 ( equation ( [ 2.17a ] ) ) ; and on decreasing @xmath20 ( increasing @xmath147 ) one expects the static conductivity to diminish and hence an increased @xmath5 , as found . to understand the @xmath147-dependence , and in turn to enable connection to incoherent effective single - impurity behaviour at high-@xmath26 , we first consider an approximate evaluation of @xmath258 ( equation ( [ 2.10 ] ) ) ; in which the energy dependence of the free conduction band dos @xmath259 is neglected , @xmath260 being replaced by its fermi level value . employing this ` flat band ' approximation in equation ( [ 2.10 ] ) ( where it enters via the @xmath261 average ) leads to f_hcl(0;t ) [ _ 0(-_c)]^2 ^+_-d ( ; t ) [ _ 0(-_c)]^2 [ 4.1 ] expressed as a physically intuitive thermal average of the dimensionless scattering time @xmath262 ( equation ( [ 3.2 ] ) ) . for kondo insulators this approximation is qualitatively inadequate at low-@xmath26 @xcite , but as illustrated in figure [ fig7 ] ( inset ) it is entirely respectable for the hf metals and in particular recovers precisely the high-@xmath26 asymptotics of @xmath5 . as shown in section 3.1 , the large @xmath24 and/or @xmath26 dependence of the reduced scattering rate @xmath220 coincides with the @xmath3-spectral function @xmath221 ( equation ( [ 3.3 ] ) ) ; and in section 3 ( see also @xcite ) the latter were shown to have common spectral tails , independently of @xmath147 . this suggests that the primary effect of @xmath147 seen in figure [ fig7 ] for @xmath255 is contained in the @xmath263 ^ 2 $ ] of equation ( [ 4.1 ] ) . that this is so is seen in figure [ fig8 ] where the results of figure [ fig7 ] are now shown as @xmath264 _ vs _ @xmath26 , where ^(t)=. [ 4.2 ] for @xmath265 or so in practice , @xmath264 is seen in particular to be _ independent _ of the conduction band filling embodied in @xmath147 ; including we note the kondo insulator , whose ` high ' temperature resistivity is thus seen to be that of a regular heavy fermion metal . indeed as readily demonstrated , and evident in part from the above discussion , the behaviour seen in figure [ fig8 ] is barely dependent on the details ( @xmath266-dependence ) of the host bandstructure embodied in @xmath104 . [ h ] the obvious final question here concerns the high-@xmath26 form of @xmath264 for the pam . to that end we consider the anderson single - impurity model ( aim ) , with @xmath267 denoting as usual the change of resistivity due to addition of the impurity to the non - interacting host , and @xmath268 . this is given by @xcite ( _ _ cf__equations ( [ 4.1]),([4.2 ] ) ) = ^+_-d _ imp(;t ) [ 4.3 ] with the impurity scattering rate @xmath269 ; where @xmath270 is the impurity spectral function such that @xmath271 follows from the friedel sum rule @xcite in the singly occupied , strong coupling kondo regime of the aim . the lma scaling resistivity @xmath272 _ vs _ @xmath26 is also shown in figure [ fig8 ] , where @xmath273 and @xmath274 is the aim kondo scale ( with @xmath275 the impurity quasiparticle weight ) . from this it is seen that the high-@xmath26 scaling behaviour of @xmath264 for the pam is precisely that of the aim ; in particular the leading @xmath216 behaviour of the lma @xmath264 is readily shown analytically to be given by @xmath276 , which is exact in the kondo limit of the impurity model @xcite . this reflects again the crossover in the strong coupling pam from low - temperature lattice coherent behaviour to incoherent effective single - impurity scaling physics , here in the context of d.c . transport . as for its counterpart in the case of single - particle dynamics @xcite , we point out ( a ) that since this connection is established from scaling considerations it is entirely independent of how the scales @xmath1 and @xmath277 for the two distinct models ( pam and aim ) depend on the underlying bare / material parameters of the respective problems ; and ( b ) the fact that it arises in the @xmath278 scaling regime precludes a two - scale description of the crossover from lattice - coherent to incoherent effective single - impurity physics . our focus above has naturally been on the strong coupling , kondo lattice regime . we now look briefly at d.c . transport on non - universal scales . what one expects here is that when the temperature is a not insignificant fraction of the hybridization @xmath170 or bandwidth scale @xmath279 , kondo screening will be washed out , and hence @xmath5 should cross over from the logarithmically decreasing single - impurity form at @xmath280 ( figures [ fig7 ] and [ fig8 ] ) to conventional metallic behaviour @xmath281 at non - universal temperatures ; and thus as such must go through a minimum . that this indeed happens can be seen in the inset to figure [ fig8 ] where we show @xmath282 _ vs _ @xmath25 for @xmath187 , @xmath283 and @xmath284 ( solid line ) and @xmath285 ( point - dash ) . for the lower @xmath6 example , a minimum is seen at @xmath286 , which corresponds in ` absolute ' units ( @xmath41 ) to a temperature @xmath287 an appreciable fraction of the hybridization @xmath288 . the corresponding minimum does of course exist for the higher @xmath6 , but is pushed beyond @xmath289 ( and to concomitantly lower vales of @xmath5 ) ; and @xmath5 in this case lies on the universal scaling curve throughout the @xmath26-range shown in figure [ fig8 ] . a final point is worth noting here . for @xmath290 , the @xmath5 _ vs _ @xmath25 for the two @xmath6 s shown in figure [ fig8 ] ( inset ) are in essence coincident ; each lies on the universal scaling curve . what distinguishes different interaction strengths is of course the location of the minimum , occurring as it does on non - universal temperature scales . no real hf material is however in the universal scaling regime ` for ever ' with increasing @xmath4 the scaling regime will be exited sooner or later . and the temperature for which the experimental @xmath5 is a minimum ( once phonon contributions have been subtracted out ) can provide valuable information on the interaction strength , as we shall see in action in the following paper . we have considered almost exclusively the hypercubic lattice , for the obvious reason that its one - particle bloch states ultimately underlie the low - temperature lattice coherence of the interacting problem . for the bethe lattice , the strong coupling scaling resistivity @xmath291 ( with @xmath292 from equation ( [ 2.18 ] ) ) is shown _ vs _ @xmath26 in inset ( b ) to figure [ fig6 ] , for @xmath293 . in contrast to its counterpart for the hcl shown in the main figure , the @xmath222 resistivity is non - vanishing ( given by equation ( [ 2.19 ] ) ) , reflecting the absence of coherent bloch states for the bl . further , the _ high_-@xmath26 asymptote of the bl @xmath5 in the scaling regime is likewise non - zero ; being given by the @xmath18 value of the free ( @xmath53 ) conduction band resistivity , namely @xmath294 ^ 2 = \pi^2/[4(1-\epsilon_c^2)]$ ] as marked by a cross in figure [ fig6 ] inset ( and arising for the same physical reasons discussed for kondo insulators in @xcite ) . the qualitative contrast between @xmath5 for the canonically bloch decomposable hcl , and that for the bl , illustrates why the latter more specifically the associated ` joint density of states ' type formula equation ( [ 2.18 ] ) for @xmath295 that is not uncommonly employed in the literature gives a poor caricature of d.c . transport for hf metals in which the lattice coherence is of central importance . we turn now to the optical conductivity @xmath296 ( with @xmath126 given by equation ( [ 2.10 ] ) ) . in the strong coupling kondo lattice regime @xmath126 is of course independent of @xmath6 and @xmath189 , and a universal function of @xmath22 and @xmath25 for fixed @xmath147 and @xmath45 . lma results for @xmath126 are shown in figure [ fig9 ] , for @xmath179 and @xmath224 . the right panel shows the thermal evolution of the optical conductivity ( on a linear @xmath24-scale ) for temperatures @xmath297 and @xmath223 ; while the left panel ( on a log - log scale ) shows the behaviour for a lower range of temperatures up to @xmath298 . the latter in particular illustrates the thermal evolution of the optical drude peak , which at @xmath299 consists of an @xmath300 @xmath208-function given by equations ( 2.14 ) ( with net weight @xmath180 in strong coupling ) . on increasing @xmath26 from @xmath211 the drude peak naturally broadens , and is well fit by a lorentzian up to its half - width or so , after which it decays more slowly in @xmath24 . at the lowest @xmath26 shown the drude peak is well separated from the ` optical edge ' in @xmath126 seen at @xmath301 ( although we add that @xmath126 is strictly non - zero for all @xmath24 ) , and with increasing @xmath26 is seen to persist as an essentially separate entity up to @xmath302 or so ; after which it is progressively destroyed as expected , merging into an optical pseudogap in the neighbourhood of @xmath303 , which is reasonably well filled up by @xmath304 and all but gone by @xmath305 ( see figure [ fig9 ] , right panel ) . similar behaviour is naturally found on varying @xmath147 and/or @xmath45 . figure [ fig10 ] shows in particular the influence of @xmath147 ( varying conduction band filling ) on the optical pseudogap for a fixed temperature @xmath306 , from which it is seen that the pseudogap becomes shallower with increasing @xmath147 . the above behaviour should be compared to the p - h symmetric kondo insulator ( ki ) @xmath307 considered in @xcite . in that case the @xmath18 optical conductivity is characterized by an indirect gap @xmath308 , and there is of course no @xmath18 drude peak . instead a drude - like peak in the optical conductivity actually builds up on initially increasing @xmath26 from zero ( see figure 15 of @xcite ) , before being thermally broadened and subsumed into the optical pseudogap . for @xmath309 or so the low - frequency optics of the ki are thus very different from those of the hf metal , as expected . but for @xmath310 the optical behaviour of the two is qualitatively similar as shown by comparison of figure [ fig9 ] ( right panel ) and its counterpart for the ki , figure 14 of @xcite . this too is physically natural , since the infilling of the indirect optical gap on temperature scales @xmath192 means that the ki behaves to all intents and purposes as a hf metal ; as seen also in figures [ fig7 ] or [ fig8 ] for the static transport . a second point should be emphasised here , obvious though it is from the preceding discussion : whether for hf metals or kondo insulators , it is the low - energy scale @xmath159 that sets the intrinsic scale for both the @xmath27-dependence of the low - energy optical conductivity and its thermal evolution . and in strong coupling that scale is wholly distinct from the optical _ direct _ gap , @xmath311 . the latter arises at its simplest in the commonly employed renormalized band picture ( see e.g. @xcite ) , as the minimum direct gap for which optical transitions are allowed . in this effective single - particle description the imaginary part of the @xmath3-electron self - energy and hence all scattering is neglected entirely , and the corresponding real part @xmath312 is replaced by its leading low-@xmath27 behaviour equation ( [ 2.12 ] ) ( as also inherent to a slave boson mean - field approximation @xcite ) . the two branches of the renormalized bandstructure , denoted by @xmath313 with @xmath314 , then follow from the zeros of @xmath315^{-1 } = [ \gamma(\om)-\epsilon_{\mathbf k}]$ ] ( see equation ( [ 2.7 ] ) ) with the approximate @xmath316 @xmath317^{-1}$ ] from equation ( [ 2.5b ] ) ; and the resultant @xmath266-dependent direct gap @xmath318 $ ] is given by @xmath319^{1/2}$ ] with @xmath320 the usual renormalized level . the minimum direct gap , @xmath311 , occurs for @xmath321 ( @xmath322 in strong coupling ) and is thus _ [ 5.1 ] the corresponding result for the optical conductivity @xmath126 is readily determined from equation ( [ 2.10 ] ) . denoted by @xmath323 it is given for @xmath18 ( and all @xmath324 ) by f_o(;0 ) = [ 5.2 ] with @xmath325 the free conduction band dos and @xmath326 the unit step function ; and is thus non - zero only for frequencies @xmath327 _ above _ the direct gap ( which result is also readily shown to hold for _ all _ temperatures ) . two points should be noted here . first that the low - energy scale @xmath159 intrinsic to hfs or kis is qualitatively distinct from the direct gap @xmath311 . in fact since @xmath328 it follows that in strong coupling where the quasiparticle weight @xmath146 and hence @xmath1 becomes exponentially small , optics on the direct gap scale do not even lie in the @xmath329 scaling regime ; although neither do they occur on truly non - universal scales ( because @xmath330 ) and in that sense belong to the ` low - frequency ' optical spectrum . second , we emphasise the inherent naivet of interpreting optics in terms of renormalized single - particle interband transitions : it is scattering due to electron interactions that generates _ all _ the optical density below the direct gap scale . failure to include such , as in a renormalized band picture and regardless of how sophisticated the underlying band structure employed in practice inevitably leads to a qualitatively inadequate description of optics ( as illustrated explicitly in figure [ fig11 ] below ) . neither is this situation ameliorated in materials application by the introduction of _ ad hoc _ @xmath27-dependent broadening factors , for that simply avoids the basic underlying physics . lma results for optics on all frequency scales are given in figure [ fig11 ] , for @xmath190 ; where @xmath126 is shown _ vs _ @xmath22 on a log scale spanning five orders of magnitude , for the same range of temperatures @xmath25 employed in figure [ fig9 ] ( right panel ) . to encompass all @xmath27 including non - universal energies , the bare parameters @xmath6 and @xmath189 must of course be specified , @xmath331 and @xmath332 here being chosen for illustration ; although note that the optical conductivity as a function of @xmath22 remains ` universal ' ( independent of @xmath6 or @xmath189 ) up to large but finite values of @xmath24 determined by the particular @xmath6 and @xmath189 chosen , in this example @xmath333 ( a directly analogous situation for the resistivity @xmath5 is shown in the inset to figure [ fig8 ] ) . the inset to figure [ fig11 ] shows the renormalized bandstructure @xmath334 _ vs _ the free conduction band energy @xmath266 ; determined as above from solution of @xmath335 ( with @xmath336 the full @xmath337 ) . this enables the notional direct gap to be determined , @xmath338 here well separated from the low - energy coherence scale @xmath1 in strong coupling and indeed seen to occur for @xmath339 . the essential points from figure [ fig11 ] are clear . as expected and well known ( see e.g. @xcite ) , significant optical absorption occurs in the vicinity of the direct gap ; strongly broadened to low - energies due to electron interactions as above , and all but ` dead ' on non - universal energy scales ( e.g. the hybridization @xmath340 for the chosen bare parameters ) . regarding the thermal evolution of the optical conductivity note also that temperatures on the order of a few multiples of the coherence scale @xmath1 which control the thermal evolution of the low - energy optics have essentially no effect on frequencies of the order of the direct gap , reflecting the clean separation between @xmath1 and @xmath311 characteristic of strong coupling . as a corollary the direct gap should be thermally eroded only for @xmath341 ; as indeed seen in figure [ fig12 ] where ( for the same parameters as figure [ fig11 ] ) the thermal evolution of @xmath126 is shown for temperatures up to @xmath342 . significant thermal erosion sets in by about @xmath343 or so , and is well developed by the highest temperature shown . the clear scale separation between @xmath1 and @xmath311 will not however be captured properly if one is restricted to relatively low interactions and high temperatures as e.g. in quantum monte carlo @xcite , or from theories in which the quasiparticle weight @xmath146 is algebraically rather than exponentially small in the interaction strength , such as iterated perturbation theory @xcite . we have considered here the periodic anderson lattice , the canonical model for understanding heavy fermion metals , kondo insulators , intermediate valence and related materials . optical conductivities , d.c . transport and single - particle dynamics of the paramagnetic phase have been investigated , using the local moment approach within a dmft framework . for obvious physical reasons our main focus has been the strongly correlated kondo lattice regime , where we find the problem to be characterised by a single , exponentially small coherence scale @xmath1 ; in terms of which the frequency and temperature dependence of physical properties scale being universally dependent on @xmath344 and/or @xmath25 regardless of the interaction or hybridization strengths . all relevant energy / temperature scales are handled by the theory , from the low - energy coherent fermi liquid domain out to large ( and in the strict scaling limit arbitrarily large ) multiples of @xmath1 where incoherent many - body scattering dominates the physics ; followed by the crossover out of the scaling regime to non - universal , high energy / temperature scales dictated by ` bare ' model / material parameters . and while our emphasis has been on strong correlations we add that all interaction strengths from weak to strong coupling are encompassed by the lma @xcite , such that intermediate valence behaviour in particular can also be addressed . the first question posed in the introduction nonetheless remains : to what extent does the model , and our theory for it , capture experiment ? we turn to that in the following paper where direct comparison of theory and experiment is made for three heavy fermion materials and a classic intermediate valence compound . we are grateful to the epsrc for supporting this research .

we construct a complete five dimensional quark - lepton symmetric model , with all fields propagating in the bulk . the extra dimension forms an orbifold with the zero mode fermions corresponding to standard model quarks localised at one fixed point . zero modes corresponding to left(right)-chiral leptons are localised at ( near ) the other fixed point . this localisation pattern is motivated by the symmetries of the model . shifting the right - handed neutrinos and charged leptons slightly from the fixed point provides a new mechanism for understanding the absence of relations of the type or in quark - lepton symmetric models . flavour changing neutral currents resulting from kaluza klein gluon exchange , which typically arise in the quark sector of split fermion models , are suppressed due to the localisation of quarks at one point . the separation of quarks and leptons in the compact extra dimension also acts to suppress the proton decay rate . this permits the extra dimension to be much larger than that obtained in a previous construct , with the bound tev obtained .

hep-ph0611164

it is known that the evident differences between quarks and leptons may be no more than the low energy manifestation of a more symmetric underlying theory . in particular it has been shown that one may construct four dimensional models which are invariant under a discrete quark - lepton ( ql ) symmetry , whereby one interchanges all quarks and leptons in the lagrangian @xcite . this idea requires the introduction of leptonic colour , an @xmath4 gauge group acting on a set of generalised leptons , and thus predicts new gauge bosons and fermions . recently the notion of a ql symmetry has been investigated in five dimensions @xcite . in that work the scalar and gauge fields of the ql symmetric model were assumed to propagate in the bulk , whilst all quark and lepton fields were confined to a brane . the extra dimension was assumed to form an @xmath0 orbifold , the construction of which reduced the five dimensional ql symmetric gauge group to one of its subgroups . the model required an order @xmath5 gev cut - off to suppress the proton decay rate and the extra dimension was taken to be of order @xmath6 gev . in the present work we construct a complete five dimensional ql symmetric model , with all fields assumed to propagate in a five dimensional spacetime . placing fermions in the bulk provides new ways of addressing the problems encountered in the previous construct . provided that the standard model ( sm ) quarks and leptons are localised at ( or near ) different ends of the compact extra dimension it may be as large as @xmath7 tev , permitting the observation of higher dimensional physics at future colliders . a common problem encountered in ql symmetric models is the presence of tree level mass relations of the type @xmath1 or @xmath2 . the higher dimensional framework provides a new way to understand the absence of such mass relations . the mass relations may be removed from the effective four dimensional theory if zero mode fermions corresponding to sm quarks and leptons have different profiles along the extra dimension . in our model this occurs as follows . all zero mode fermions corresponding to sm quarks are localised at one orbifold fixed point , with the varying widths of the fifth dimensional quark wavefunctions determining the degree of wavefunction overlap between left- and right - chiral sm quarks . this in turn determines the size of the effective four dimensional quark yukawa couplings with the sm higgs scalar . leptons are localised at the opposite end of the extra dimension , with @xmath8 doublet leptons localised at the fixed point and right - chiral neutrinos and charged leptons shifted slightly into the bulk . thus the fifth dimensional wavefunction overlaps in the lepton sector tend to be reduced relative to that of the quark sector , motivating the flavour differences observed between quarks and leptons . by shifting the right - chiral neutrinos further into the bulk than the right - chiral charged leptons , neutrino dirac masses may be suppressed below the electroweak scale . the symmetries of the model motivate the localisation pattern of sm fermions outlined above . the localisation of quarks at one point in the extra dimension allows one to evade the flavour changing neutral current bounds on the size of the extra dimension which arise generically in the quark sector of split fermion models @xcite . these bounds may be quite severe , requiring @xmath9 to be as large as @xmath10 @xcite . the flavour changing neutral current bounds which arise in the lepton sector are much weaker and permit the extra dimension to be relatively large . we note that the concept of leptonic colour has recently been generalised in @xcite and studied within the context of unified theories in @xcite . split fermions have also been employed to remove quark - lepton mass relations in a different context in @xcite . the layout of this paper is as follows . in section [ sec : bulk_5d_ql_gauge_sector ] the gauge and scalar content of the model are discussed . section [ sec : bulk_5d_ql_fermion_parities ] introduces fermions and we determine the orbifold parity assignments necessary to obtain a realistic zero mode fermion spectrum . the yukawa sector of the model is introduced in section [ sec : bulk_5d_ql_yukawas ] whilst the issue of proton decay is addressed in section [ sec : bulk_5d_ql_p_decay ] . fermion masses are considered in section [ sec : bulk_5d_ql_fermion_mass ] , with the neutrino sector considered separately in section [ bulk_ql_nu_mass ] . the size of the extra dimension is discussed in section [ sec : rev_bulk_ql_bounding_r ] whilst section [ sec : bulk_5d_ql_sterile_cosmology ] addresses the lifetime and cosmological bounds on the mass of the lightest sm singlet neutrino . a discussion of the expected scale for new physics and some experimental signatures is provided in section [ sec : bulk_ql_bounding_scales ] and the paper concludes in section [ sec : bulk_5d_ql_conc ] . the ql symmetric gauge group is @xmath11 where @xmath12 is the lepton colour group and @xmath13 is the usual colour group for quarks . the action of the group @xmath8 on the zero mode fermion spectrum will be identified with the usual weak group of the sm and @xmath14 , where @xmath15 is the sm hypercharge . under the discrete ql symmetry the gauge fields transform as : @xmath16 where @xmath17 are the @xmath18 gauge bosons , @xmath19 are the weak bosons and @xmath20 is the @xmath21 gauge boson . the five dimensional lorentz index takes the values @xmath22 , where @xmath23 is the @xmath24 dimensional lorentz index . the additional spatial dimension is taken as the orbifold @xmath0 , whose coordinate is labelled as @xmath25 . the construction of the orbifold proceeds via the identification @xmath26 under @xmath27 and @xmath28 under the @xmath29 symmetry , where @xmath30 . the physical region in @xmath25 is given by the interval @xmath31 $ ] . the lagrangian is required to be invariant under the discrete @xmath32 symmetry , whose action in the space gauge fields is defined as follows : @xmath33 where @xmath34 denotes a generic gauge field . we take @xmath35 and @xmath36 to be trivial for the @xmath13 , @xmath8 and @xmath21 gauge bosons . for the @xmath12 gauge bosons we choose @xmath37 and @xmath38 , which will reduce the leptonic colour symmetry to @xmath39 in the four dimensional effective theory . we write the five dimensional @xmath12 gauge bosons as @xmath40 and find their @xmath32 parities to be @xmath41 one may expand the gauge bosons as a fourier series in the compact extra dimension , with the @xmath32 parities constraining the series as usual . one has @xmath42 where @xmath43 represents a generic field . thus the four dimensional charge 1/2 bosons @xmath44 , @xmath45 , @xmath46 and @xmath47 do not possess zero modes , with the @xmath48th mode possessing a mass of @xmath49 . the fields @xmath50 , @xmath51 , @xmath52 and @xmath53 all have zero modes , with the higher modes possessing a mass of @xmath54 . after compactification the zero mode gauge group is @xmath55 and the next stage of symmetry breaking requires @xmath56 which shall be achieved by the usual higgs mechanism . the scalar content necessary to break @xmath57 is @xmath58 with the action of the discrete ql symmetry given by , @xmath59 where @xmath60 and @xmath61 is the two dimensional anti - symmetric tensor . the @xmath62 parities of @xmath63 are trivial whilst for @xmath64 we have @xmath65 with @xmath37 and @xmath38 . for @xmath66 we take @xmath67 with @xmath35 and @xmath36 trivial . under the symmetry reduction @xmath68 one has @xmath69 where @xmath70 and @xmath71 have the @xmath32 parities @xmath72 the zero mode for @xmath73 may develop a vev to break the gauge symmetry as follows : @xmath74 at this stage the hypercharge generator may be identified as @xmath75 where @xmath76 is a diagonal generator of @xmath12 . the final stage of symmetry breaking occurs when the neutral component of @xmath63 develops the vev @xmath77 to give @xmath78 assuming @xmath79 , where @xmath80 [ @xmath81 is the @xmath8 [ @xmath21 ] coupling constant and @xmath83 denotes the common @xmath12 and @xmath13 coupling constant , one may write the neutral gauge boson mass eigenvalues as @xcite @xmath84 note that the zero modes possess the same masses as the neutral gauge bosons in the minimal four dimensional ql symmetric model @xcite . these zero modes couple to fermions in exactly the same way as the neutral gauge bosons in the minimal ql symmetric model , making the phenomenology of these sates identical to that of the neutral gauge bosons studied in @xcite . the zero modes consist of the massless photon , the @xmath85 boson with mass of order @xmath77 , the electroweak scale , and an additional neutral boson @xmath86 with mass of order @xmath87 , the @xmath88 symmetry breaking scale . the phenomenological bound of @xmath89 gev obtained in @xcite also applies to the zero mode @xmath86 boson in the present model . thus we obtain a lower bound on the @xmath88 symmetry breaking scale of @xmath90 tev , which is low enough to permit observation of the @xmath86 boson at the lhc . the kaluza - klein ( kk ) tower of charged 1/2 bosons possess the mass @xmath91 where the zero mode is absent . we shall discuss the bounds on @xmath92 and the @xmath15 boson mass scales in section [ sec : rev_bulk_ql_bounding_r ] . the mass of the @xmath34 bosons are given by @xmath93 with the zero mode corresponding to the usual @xmath34 boson . whilst the five dimensional theory is vector - like and necessarily free from anomalies , we shall be introducing four dimensional chirality via the orbifold boundary conditions @xcite . the effective four dimensional gauge theory is free from anomalies if all anomalies cancel amongst the zero mode fermions @xcite . after compactification the gauge group is given in equation ( [ bulk_5d_ql_compact_gauge_group ] ) . to cancel anomalies involving @xmath94 factors one requires both the sm leptons and extra coloured leptons to appear at the zero mode level . this may be achieved by doubling the fermion content of the ql symmetric model , in analogy with the doubling required in five dimensional left - right symmetric models @xcite . we denote the fermions by @xmath95 where the quantum numbers under @xmath57 are shown and @xmath96 labels distinct fermion multiplets . the action of the ql symmetry on the fermions is @xmath97 the transformation rules for a fermion field under the @xmath98 discrete symmetries are @xmath99 where @xmath100 denotes a generic fermion field and the signs may be chosen independently for the distinct discrete symmetries . the matrices @xmath35 and @xmath36 are identical to the ones used for the gauge sector in the last section , with the only non - trivial matrix being @xmath36 acting in leptonic colour space . we demand that the lepton fields acquire the @xmath98 quantum numbers : @xmath101 here the numerical superscripts @xmath102 label the different lepton colours and the numerical subscripts @xmath103 label the different five dimensional fields . as @xmath13 is not broken by the orbifold compactification the three quark colours all possess the same @xmath104 parities . suppressing the quark colour index , we enforce the following orbifold parities for quarks : @xmath105 only fermion fields with the orbifold parities @xmath106 appear at the zero mode level . we identify the zero modes of the fields @xmath107 , @xmath108 , @xmath109 , @xmath110 and @xmath111 with the sm fields @xmath112 , @xmath113 , @xmath114 , @xmath115 and @xmath116 respectively . the zero modes of the fields @xmath117 , @xmath118 and @xmath119 are the usual exotic leptons found in ql symmetric models ( known as liptons in the literature @xcite ) . the liptons form doublets under the remnant leptonic colour symmetry @xmath120 and are confined into two particle bound states . these states form one of the key signatures of ql symmetric models , leading to an exotic spectrum of particles which may decay into sm fields by creation of electroweak gauge bosons @xcite . we shall see below that the zero mode liptons acquire mass at the symmetry breaking scale @xmath87 and may be observed at the lhc . note that the zero mode liptons appear in different multiplets to the sm leptons . consequently the lightest liptons will not couple directly to sm leptons via @xmath15-boson exchange . this is one of the major phenomenological differences between the current model and that constructed in @xcite . we explore the implications of this difference below . the zero mode of @xmath121 is an additional neutrino which is sterile under the electroweak gauge group . this state will play the role of the usual sm singlet neutrino and we shall label it as @xmath122 . we shall discuss the mass of this state in section [ bulk_ql_nu_mass ] and issues relating to its cosmological evolution in section [ sec : bulk_5d_ql_sterile_cosmology ] . the yukawa lagrangian must remain invariant under both gauge and orbifold parity transformations . it is convenient to split the five dimensional yukawa lagrangian into two portions , with the non - electroweak portion given by @xmath123+h_2[l_2^tc_5^{-1}l_2\chi + q_2^tc_5^{-1}q_2\chi']+\right.\nonumber\\ & & \left.h_1'[n_1^{ct}c_5^{-1}e_1^c\chi^\dagger + d_1^{ct}c_5^{-1}u_1^c\chi'^\dagger]+h_2'[n_2^{ct}c_5^{-1}e_2^c\chi^\dagger + d_2^{ct}c_5^{-1}u_2^c\chi'^\dagger]+\mathrm{h.c.}\right\}.\label{bulk_5d_ql_non - ew_lagrangian}\end{aligned}\ ] ] here the @xmath124 s are yukawa coupling matrices in flavour space , @xmath125 is the five dimensional charge conjugation matrix and @xmath126 is the cut - off . contraction over @xmath18 colour indices with the three dimensional anti - symmetric tensor @xmath127 is implied in ( [ bulk_5d_ql_non - ew_lagrangian ] ) . when the zero mode of @xmath64 develops a vev the @xmath128 and @xmath129 terms in ( [ bulk_5d_ql_non - ew_lagrangian ] ) generate an order @xmath87 mass for the liptons . all sm fields remain massless at this stage of symmetry breaking . the electroweak portion of the yukawa lagrangian is @xmath130+\lambda_2[e_2^c c_5^{-1}l_2\phi^\dagger + u_2^c c_5^{-1}q_2\tilde{\phi}^{\dagger}]+\right.\nonumber\\ & & \left.\lambda_1'[n_1^c c_5^{-1}l_1\tilde{\phi}^{\dagger}+d_1^c c_5^{-1}q_1\phi^{\dagger}]+\lambda_2'[n_2^c c_5^{-1}l_2\tilde{\phi}^{\dagger } + d_2^c c_5^{-1}q_2\phi^{\dagger}]+\mathrm{h.c.}\right\},\label{bulk_5d_ql_ew_yukawa_lagrangian}\end{aligned}\ ] ] where the @xmath131 s are flavour space yukawa coupling matrices . this lagrangian generates fermion dirac mass terms when @xmath132 develops a vev , with the sm fermions acquiring mass through the @xmath133 , @xmath134 terms . if the sm fermions have uniform profiles along the extra dimension , then the troublesome tree level mass relations @xmath1 and @xmath135 arise however , if one is able to generate different five dimensional wavefunction profiles for quarks and leptons , the troublesome tree level relations will not persist in the effective four dimensional theory . we shall have more to say on this matter in section [ sec : bulk_5d_ql_fermion_mass ] , but it will prove useful to consider the stability of the proton within our model first . at the five dimensional level , the lowest dimension non - renormalizable operator which leads to proton decay in our model is @xmath136 where @xmath137 , @xmath138 , @xmath139 are quark colour indices , @xmath140 is the lepton colour index and @xmath141 is a dimensionless coupling constant . if the sm quarks and leptons have uniform profiles across the extra dimension , proton decay occurs in the effective four dimensional theory via the operator @xmath142 where @xmath143 and @xmath112 denote four dimensional quark and lepton fields respectively . current experimental bounds require the lifetime of the proton to be in excess of @xmath144 years @xcite . with @xmath145 tev and @xmath146 one requires @xmath147 gev to suppress the proton decay rate . it is possible to reduce the scale of the cut - off by localising quarks and leptons at different points in the extra dimension @xcite . if quarks and leptons are separated in the extra dimension , the diminished overlap of the quark and lepton wavefunctions in the fifth dimension serves to reduce the proton decay rate . to suppress the proton decay rate we shall localise quarks and leptons at different fixed points . to illustrate our idea it will suffice to consider an @xmath148 orbifold . let us add a gauge singlet , bulk scalar field @xmath149 which transforms as @xmath150 under the @xmath27 symmetry and has the potential @xmath151 where @xmath152 ( @xmath153 ) is a dimensionless ( dimensionful ) constant . the minimum of the potential clashes with the orbifold boundary conditions ( [ bulk_ql_sigma_bcs ] ) and results in the vacuum profile @xcite ( see also @xcite ) , @xmath154\tanh[\xi y],\end{aligned}\ ] ] where @xmath155 and we have introduced @xmath156 . the points @xmath157 are identified under the orbifold construction . one can simplify the analysis when @xmath158 by treating the vev profile of @xmath149 as a step function @xcite , @xmath159 where , @xmath160 in five dimensions , gauge invariance permits the yukawa couplings @xmath161 for all fermion fields , where @xmath162 is a yukawa coupling constant . the shape of zero mode fermions in the extra dimension , when @xmath163 , is subsequently given by @xcite @xmath164 or via the replacement @xmath165 if , demonstrating the localisation of zero mode fermions at different fixed points on the @xmath148 orbifold , depending on the sign of the coupling constant @xmath162 . we wish to suppress the proton decay rate by localising quarks and leptons at different fixed points . let us take @xmath149 to be odd under the discrete ql symmetry @xcite . the resulting yukawa lagrangian is , @xmath166 + h_{u_i}[u^{c2}_i - e^{c2}_{i } ] \right.\nonumber\\ & & \left.\mkern25mu + h_{q_i}[q^{2}_i - l^{2}_{i}]\right\}\sigma,\label{orb_ql_bulk_localizer_lagrangian_p_decay}\end{aligned}\ ] ] where @xmath167 , @xmath168 and @xmath169 are yukawa coupling constants . we use an obvious notation with @xmath170 and we have suppressed family indices . if one takes all yukawa couplings to be greater than zero , quarks and leptons are automatically localised at different fixed points . after integrating over the extra dimension , the operator ( [ orb_ql_bulk_p_decay_operator_5d ] ) produces a proton decay inducing operator in the four dimensional theory . the approximate form of this operator is @xmath171 where @xmath172 and @xmath173 are dimensionless constants and @xmath143 and @xmath112 are four dimensional quark and lepton fields respectively . for @xmath126 of order @xmath174 tev and an order tev ql symmetry breaking scale , the lower bound on the lifetime of the proton requires @xmath175 if one takes @xmath173 to be order unity . we have seen that one is able to understand the long lifetime of the proton if quarks and leptons are localised at different fixed points . in section [ sec : bulk_5d_ql_yukawas ] we noted that despite the yukawa coupling relations induced by the ql symmetry in equation ( [ bulk_5d_ql_ew_yukawa_lagrangian ] ) , one may understand the absence of relations of the type @xmath1 in the effective four dimensional theory if quarks and leptons have different wavefunction profiles in the extra dimension . inspection of equations ( [ orb_ql_bulk_zero_mode_fermi_profile ] ) and ( [ orb_ql_bulk_localizer_lagrangian_p_decay ] ) reveals that the methods employed to separate quarks and leptons in the extra dimension induce identical fifth dimensional wavefunction profiles for fermions related by the ql symmetry . upon integrating over the fifth dimension the troublesome tree level mass relationships persist , despite quarks and leptons being localised at different fixed points . we could in principle add a second sm higgs doublet to the model and generate enough parameter freedom to remove the unwanted mass relations @xcite . however this will not allow us to understand the lightness of the known neutrinos relative to the electroweak scale . in this section we shall apply a purely higher dimensional mechanism to remove the undesirable tree level mass relations . the full construction of a theory of flavour is beyond the scope of the present work . we shall sketch our ideas in what follows and to this end it is helpful to recall some key results obtained in previous studies involving split fermions . the work of arkani - hamed and schmaltz ( as ) @xcite demonstrated that four dimensional flavour could be addressed in terms of fifth dimensional wavefunction overlaps . as spatially separated the left- and right - chiral fermion fields in an extra dimension , with the size of the spatial separation influencing the degree of wavefunction overlap . the amount of overlap then determined the size of the effective four dimensional yukawa couplings to the sm higgs . however one need not separate the left- and right - chiral fields to address flavour in terms of fifth dimensional wavefunction overlaps . indeed one can localise all the fermions at one point in the extra dimension , provided the left- and right - chiral fermions have fifth dimensional wavefunctions with different widths @xcite . we shall employ each of these mechanisms to realize flavour . in particular we shall localise all quark fields at one point in the extra dimension , with the different widths of the left- and right - chiral quark fields determining the effective four dimensional yukawa couplings in the quark sector . for the leptons we shall separate the left- and right - chiral fields in the extra dimension , thus motivating the observed flavour differences between the quark and lepton sectors . as separated fermion fields by introducing distinct bulk dirac mass terms for the different bulk fermions . in an orbifold theory , however , the fermion transformations necessary to introduce four dimensional chirality preclude bulk dirac mass terms . one may localise fermions at non - fixed points in an orbifold theory by introducing a second localising scalar @xcite . this works as follows @xcite . with one localising scalar the chiral zero mode of a fermion @xmath100 is localised at one of the orbifold fixed points . the point of localisation is determined by the sign of the product @xmath176 as mentioned already in section [ sec : bulk_5d_ql_p_decay ] , which amounts to the sign of @xmath162 when @xmath177 . if the field @xmath100 couples to a second bulk scalar with an opposite sign yukawa coupling , the second scalar tends to localise the zero mode at the opposite fixed point . when one scalar is dominant the fermion is found localised at the fixed point preferred by that scalar . in general , however , an interplay between the two scalars results in a compromise which sees the fermion localised in the bulk . more technical details may be found in @xcite . let us now investigate this idea in the ql symmetric framework , assuming an @xmath148 orbifold again for simplicity . we introduce a second bulk scalar , @xmath178 , which transforms as @xmath179 under the orbifold @xmath27 symmetry and transforms trivially under the ql symmetry . the yukawa lagrangian for @xmath178 is , @xmath180 + f_{u_i}[u^{c2}_i + e^{c2}_{i } ] \right.\nonumber\\ & & \left.\mkern25mu + f_{q_i}[q^{2}_i + l^{2}_{i}]\right\}\sigma.\label{orb_ql_bulk_localizer2_lagrangian_p_decay}\end{aligned}\ ] ] we achieved quark - lepton separation in the last section by demanding @xmath181 let us further demand that @xmath182 so that all quark yukawa couplings to @xmath149 and @xmath178 are positive . consider first the effects of the second scalar @xmath178 on the quark sector . inspection of ( [ orb_ql_bulk_localizer_lagrangian_p_decay ] ) and ( [ orb_ql_bulk_localizer2_lagrangian_p_decay ] ) reveals that both @xmath149 and @xmath178 will attempt to localise the quarks at the same fixed point . thus for all positive values of the yukawa couplings , which we generically denote as @xmath162 and @xmath183 , the quark fields will be localised at one fixed point . note that the sm fields @xmath143 , @xmath184 and @xmath185 will each have , in general , different profiles in the extra dimension , allowing one to reproduce the necessary quark flavour spectrum along the lines of @xcite . let us now consider the lepton sector . this sector is more complicated because @xmath149 and @xmath178 attempt to localise a given chiral zero mode lepton at different fixed points . we have shown in the last section that the long lifetime of the proton may be understood if quarks and leptons are localised at different fixed points . to retain this result we shall require @xmath149 to dominate over @xmath178 for the sm fields @xmath112 and @xmath186 . the situation with @xmath187 is not as clear . we shall discuss neutrino masses further in the next section , but for now let us estimate the degree of separation required between the field @xmath112 and the field @xmath186 to obtain an order mev electron mass . it was shown in @xcite that , in the decoupling limit of the two localising scalars , one may arrange zero mode fermions to be localised in the bulk of an orbifold theory with fifth dimensional profiles of the form @xmath188 here @xmath189 is a normalization factor , @xmath190 denotes the vev of one of the localising scalars and @xmath191 is the location of the wavefunction maxima . we show in section [ sec : bulk_ql_bounding_scales ] that we expect @xmath192 tev and use this lower bound in what follows . to leading order the dimensionless constant @xmath193 depends on the largest fermion - bulk scalar yukawa coupling and the dimensionless scalar quartic self - coupling @xmath194 . note that naive dimensional analysis permits values of @xmath195 if the underlying theory has strong couplings at the cut - off @xcite . however we shall be able to obtain acceptable suppression of the electroweak scale with significantly lower @xmath193 values @xcite . the charged lepton mass terms arise from the couplings @xmath196 which lead to dirac masses of the form @xmath197 where @xmath198 is the electron wavefunction in the extra dimension which takes the form of @xmath199 in equation ( [ bulk_ql_perez_gaussian ] ) . the localisation of the zero mode of the field @xmath112 is assumed to be dominated by the scalar @xmath149 such that @xmath200 has the exponential form of equation ( [ orb_ql_bulk_zero_mode_fermi_profile ] ) . taking @xmath201 in ( [ bulk_5d_ql_split_dirac_mass ] ) reveals that a charged lepton dirac mass of order mev is obtained if @xmath186 is localised a distance of @xmath202 from the fixed point at which @xmath112 is localised . thus one is not required to shift the zero modes corresponding to the sm right - chiral charged leptons far to obtain a realistic charged lepton spectrum . this slight shift from the fixed point should not produce an observable enhancement of the proton decay rate . neutrinos acquire dirac masses at the electroweak symmetry breaking scale via the coupling @xmath203 we wish to suppress the electroweak contribution to the neutrino mass by separating @xmath187 and @xmath204 in the extra dimension . given that neutrino masses are required to be less than an ev it is clear that one must localise the sm neutrinos away from the quarks to suppress the proton decay mode @xmath205 . we shall shift @xmath187 into the bulk to suppress the electroweak contribution to the neutrino mass , however the acceptable proximity of @xmath187 to the quarks is not as clear . we must consider the mass scale of this field to determine its acceptable proximity to the quarks , and this requires a consideration of the non - renormalizable contributions to the neutrino mass sector . consider the non - renormalizable operators which induce majorana mass for the neutrinos in our model . the operator @xmath206 written in terms of five dimensional fields , with @xmath207 a dimensionless constant , has dimension ten and in the effective four dimensional theory induces the operator @xmath208 which has dimension seven . when @xmath209 and @xmath132 develop vev s , @xmath210 produces majorana masses for the sm neutrinos @xmath204 . these masses take the form @xmath211 we require @xmath212 ev so that this mass is in the right range to accommodate the atmospheric and solar neutrino oscillation data . the neutrinos @xmath187 also acquire mass at the non - renormalizable level . at the five dimensional level , one has the operator @xmath213 where @xmath214 is a dimensionless coupling constant . this produces a majorana mass for @xmath187 in the effective four dimensional theory , @xmath215 we intend to reduce the effective dirac mass between @xmath204 and @xmath187 in the four dimensional theory below the electroweak scale by separating @xmath112 and @xmath187 in the extra dimension . to estimate the degree of suppression required of the effective dirac mass we shall take @xmath216 to be @xmath217 ( we shall see in section [ sec : bulk_5d_ql_sterile_cosmology ] that masses in this range are compatible with our framework ) . consider the complete neutrino mass matrix in the majorana basis for one generation , @xmath218 where @xmath219 is the effective four dimensional dirac mass . under the usual see - saw hierarchy @xmath220 the mass eigenvalues take the approximate form @xmath221 the seesaw contribution to the light neutrino mass eigenvalue , @xmath222 , is required to be order ev or less to ensure that @xmath223 is not too heavy . hence one requires @xmath224 kev for @xmath225 gev to ensure that @xmath226 ev . this requires @xmath187 to be localised a distance @xmath227 from the boundary at which @xmath112 is localised . we show in section [ sec : bulk_5d_ql_sterile_cosmology ] that @xmath216 is required to be at least of order gev . thus proton decays into @xmath187 are kinematically forbidden and localising @xmath187 further into the bulk does not effect the stability of the proton , a result which holds even if @xmath187 is localised at the ` quark end ' of the extra dimension . we now consider the bounds on @xmath92 , the compactification scale . in section [ sec : bulk_5d_ql_gauge_sector ] it was shown that the mass of the lightest @xmath15 boson has a contribution inversely proportional to @xmath92 ( the usual kk contribution ) . in 4d @xcite and 5d @xcite ql models explored previously , bounds on the @xmath15 boson mass have been obtained by considering the decay @xmath228 , which proceeds radiatively by the creation of virtual @xmath15 bosons . this results in an approximate bound of @xmath229 tev . in @xcite this led to a bound on @xmath92 , however the current model requires a re - evaluation of this bound . we note from section [ sec : bulk_5d_ql_fermion_parities ] that the liptons which form @xmath12 multiplets with sm leptons do not possess zero modes . vertices in the effective 4d lagrangian which couple sm leptons to liptons via @xmath15 boson exchange will generally take the form @xmath230 where @xmath15 , @xmath231 and @xmath232 denote @xmath15 boson , lipton and sm lepton operators respectively and @xmath233 is a numerical factor representing the wavefunction overlap of @xmath15 , @xmath232 and @xmath231 in the extra dimension . the lightest lipton that couples to a sm lepton @xmath232 via @xmath15 exchange is one of the @xmath234 kk liptons in the same @xmath12 multiplet as @xmath232 . however the bulk scalars employed to localise chiral zero mode fermions in sections [ sec : bulk_5d_ql_p_decay ] and [ sec : bulk_5d_ql_fermion_mass ] also alter the mass and 5d profiles of @xmath235 kk modes . when one localising scalar is employed it has been shown that the low lying kk modes ( with @xmath236 ) also become localised , albeit with broader 5d wavefunctions @xcite . furthermore the odd and even kk modes tend to be localised at opposite fixed points on an @xmath148 model . given that we localise sm leptons at one boundary of the extra dimension , the @xmath234 kk liptons will be found at the other boundary . thus @xmath237 for @xmath234 liptons , rendering the associated vertex vanishingly small . consequently the radiative decay @xmath228 will occur only via the coupling of sm charged leptons to even @xmath48 kk mode liptons , with the dominant contribution arising from the @xmath238 mode . it is difficult to evaluate @xmath233 exactly for an @xmath238 lipton @xmath231 in our model , given the two localising scalars employed . this would require a determination of the fermion 5d wavefunction profiles for a two localising scalar scenario , which is beyond the scope of the present study . to approximate @xmath233 we use the one bulk scalar results of @xcite where analytic expressions for the kk tower 5d wavefunctions were obtained . we find that the overlap is very small for a range of parameter values , with @xmath239 for a zero mode lepton @xmath232 and an @xmath238 lipton @xmath231 . assuming similar values for our model we find that the decay @xmath228 is highly suppressed due to the factor of @xmath240 in the rate . the associated bound on @xmath241 , and thus @xmath92 , is exceptionally weak . a stronger bound on @xmath92 may be obtained by considering flavour changing neutral currents ( fcncs ) . it is a generic feature of models involving split fermions that fcncs arise @xcite . zero mode gauge bosons possess uniform profiles in the extra dimension and thus couple uniformly to localised fermions . however the @xmath235 kk mode gauge bosons are not uniform along the extra dimension and thus , in general , couple to different localised fermions with different strengths , resulting in fcncs . the most stringent bounds from fcncs in split fermion models arise in the quark sector , where exchange of kk mode gluons can lead to bounds of @xmath242 tev @xcite . we have localised all quarks at one fixed point , aiming to realize quark flavour through varying width 5d quark wavefunctions . this significantly reduces the bounds from fcncs in the quark sector to @xmath243 tev @xcite . in the lepton sector we have localised all left - chiral sm leptons at one fixed point , whilst the sm right - chiral leptons are localised near the fixed point , with different points of localisation for different fields . thus only gauge bosons which couple to right - chiral fields , namely the hypercharge and @xmath244 gauge bosons , will have markedly different couplings to different leptons and thus give rise to fcncs in the lepton sector . by considering the contribution of the kk tower for the electrically neutral @xmath8 gauge boson to @xmath228 a bound of @xmath245 tev has previously been obtained @xcite . in our model the kk tower for the @xmath86 gauge bosons will contribute to @xmath228 and we expect this bound to apply , even though kk @xmath8 gauge bosons will not mediate this decay . thus we require @xmath245 tev to avoid trouble with fcncs in the model . we note that if all right - chiral sm leptons were localised at one point the lower bound of @xmath243 tev would apply , though this would have to be arbitrarily imposed on the model . we have seen in section [ sec : bulk_5d_ql_fermion_mass ] that the neutrinos will acquire majorana masses at the non - renormalizable level . the cosmological density of the neutrino @xmath187 must be considered to ensure that this state does not spoil any of the successful features of the standard cosmological model . the neutrinos @xmath187 will remain in thermal equilibrium in the early universe . this results from annihilation s involving @xmath86 bosons , namely : @xmath246 where @xmath232 denotes a sm lepton . at some temperature the state @xmath187 will freeze out and the remaining cosmological abundance will contribute to the energy density of the universe . the perturbations to the usual scenario induced by @xmath187 depend on its mass and lifetime . it is known from 4d models that right - chiral neutrinos may decay via photon emission or via @xmath120 glueball emission . before we enter into the specifics , let us summarise the findings of this section for readers not interested in the more technical details . below we show that in our 5d model the neutrino decay @xmath247 is expected to dominate the decays involving glueball emission . known cosmological bounds between the mass of heavy neutrinos and the lifetime for radiative decays then require @xmath248 gev . this result is obtained under the assumption that the neutrino dirac mass @xmath219 is suppressed to kev energies by localising @xmath187 in the bulk . however the lightest @xmath187 effectively becomes stable if @xmath219 is suppressed by localising @xmath187 at the quark end of the extra dimension . in this case the usual bound for stable massive neutrinos applies and one requires @xmath249 gev . let us now consider these decay mechanisms in more detail . consider the decay @xmath250 , where @xmath204 denotes a sm neutrino . in ql models this occurs radiatively via the creation of a @xmath15 boson and a lipton ( see figure [ fig : neutrinodecay ] ) . in quark - lepton symmetric models.,title="fig:",scaledwidth=33.0% ] in quark - lepton symmetric models.,title="fig:",scaledwidth=33.0% ] the rate for these decays can be obtained by modifying existing results for similar neutrino decays . in particular reference @xcite studied radiative neutrino decays of the type @xmath251 when exotic leptons and gauge bosons exist . equation ( 2 ) of that work identifies the rate for @xmath251 to be @xmath252 where @xmath253 is the fermi constant , @xmath137 is the fine structure constant , @xmath254 is a numerical vertex factor and @xmath255 ( @xmath256 ) is the mass of an exotic lepton ( muon neutrino ) . this expression may be applied to the decay @xmath257 in the 5d ql model by making the following replacements @xmath258 here ( [ goldman_replacement_alpha ] ) is required as the @xmath15 bosons and liptons carry electric charge @xmath259 . the factor of @xmath260 is due to the enhanced coupling of the @xmath235 kk mode @xmath15 boson and @xmath261 is the equivalent of the fermi constant for the @xmath15 bosons , @xmath262 the angle @xmath263 represents the mixing between a lipton which couples to @xmath187 and a lipton which couples to @xmath204 . this mixing is induced by electroweak symmetry breaking and vanishes in the limit @xmath264 . the angle takes the approximate form @xmath265 where @xmath266 is the 5d wavefunction overlap of the @xmath187 lipton with the @xmath204 lipton . we shall comment further on @xmath266 shortly . the low lying @xmath235 kk mode liptons will have mass of order @xmath190 , the bulk scalar symmetry breaking scale @xcite . note that the interplay of the @xmath267 factor in the numerator of ( [ goldman_bound ] ) and the @xmath268 dependence of @xmath269 renders the rate for @xmath257 independent of @xmath190 . we expect the @xmath238 kk lipton to be the lightest state to contribute significantly to the decay @xmath257 . the @xmath234 state is not localised at the same fixed point as the @xmath270 chiral modes @xcite and thus its contribution to the decay will be suppressed by the @xmath266 factor . the factors @xmath271 and @xmath272 in equation ( [ goldman_replacement_k_1_2_g_y ] ) represent the wave function overlap between @xmath270 chiral mode neutrinos , the @xmath238 lipton which couples to this neutrino and the @xmath234 @xmath15 gauge boson . they take the form @xmath273 where @xmath274 and @xmath200 are zero mode sm singlet neutrino and electroweak lepton doublet 5d wavefunctions respectively . @xmath275 represents an @xmath238 lipton 5d wavefunction and the @xmath15 boson fifth dimensional profile is given by the cosine factor . the absence of analytic expressions for the 5d fermion kk tower profiles in two bulk scalar models and the dependence on free parameters makes it difficult to determine these factors exactly . we approximate these using the results obtained for the one localising scalar case in @xcite and find that typical values are @xmath276 . to estimate @xmath266 we consider the case where @xmath187 and @xmath204 are localised by one bulk scalar , but the potential trapping @xmath187 in the bulk has a minimum shifted away from the fixed point . noting that @xmath187 is shifted some distance away from the fixed point such that the effective dirac mass coupling @xmath187 and @xmath204 in the 4d theory is @xmath219 and using the analytic @xmath238 kk mode wavefunctions of @xcite we find that typically @xmath277 . the dirac mass coupling the @xmath238 liptons is expected to be larger than the associated @xmath270 mode dirac mass @xmath219 as the 5d wavefunctions for the higher kk modes generally have broader wavefunctions in the extra dimension than the zero modes . putting all this together we find that @xmath278 taking @xmath279 kev and @xmath7 tev gives @xmath280 ( @xmath281 ) for @xmath282mev ( 1 gev ) . such long lived neutrinos would still be decaying and would contribute to the diffuse photon background . studies of the diffuse photon background require heavy neutrinos to satisfy the bound @xmath283 if their lifetime exceeds @xmath284 @xcite . thus one requires @xmath285 ( @xmath286 ) for @xmath282mev ( 1 gev ) , which is not satisfied by the radiative @xmath187 decays . note that increasing @xmath216 decreases the radiative lifetime of @xmath187 . for @xmath287 gev one finds that @xmath288 . if the lifetime of a heavy neutrino lies in the range @xmath289 , where @xmath290 ( @xmath291 ) is the time of recombination ( age of the universe ) , a different relationship between the neutrinos mass and lifetime must be satisfied . this relationship takes the form @xcite @xmath292 using ( [ bulk_ql_radiative_nu_r_lifetime ] ) gives @xmath293 where all masses should be given in gev . using the lower bound for @xmath9 and @xmath294 kev ( @xmath295 kev ) gives @xmath296 gev ( 40 gev ) so that an order @xmath297 gev @xmath187 is disallowed . for masses @xmath298 gev the bound ( [ heavy_nu_bound_bulk_ql_2 ] ) may be satisfied and thus consideration of the radiative decay of @xmath187 into light neutrinos in a cosmological context leads to the bound @xmath299 gev . as mentioned already , the neutrino @xmath187 may also decay via emission of an @xmath120 glueball in ql symmetric models . before concluding that cosmological considerations demand @xmath299 gev we must further investigate this decay mode . the interaction eigenstate right - chiral neutrino is an @xmath300 singlet and as such does nt couple directly to the @xmath300 gluons ( henceforth referred to as ` stickons ' , adopting the notation of @xcite ) . however the physical mass eigenstate @xmath187 does couple to the stickons . recall that @xmath300 is predicted to be an unbroken symmetry and is thus expected to be confining . stickon exchange leads to bound state fermions formed by the liptons and the @xmath15 bosons . interestingly these bound state fermions possess the same quantum numbers as the sm leptons @xcite . consequently they mix with the known leptons and the resulting physical leptons contain small admixtures of states which couple to the stickons . of importance to us is the mixing of @xmath187 with the sm singlet bound state fermions ( henceforth bound state neutrinos ) . it has been shown that the neutrino - bound state neutrino mixing leads to heavy neutrino decays of the type @xmath301 , where @xmath302 is an @xmath300 glueball ( henceforth a stickball ) . the stickballs have a mass set by the @xmath300 strong interaction scale @xmath303 , which has been estimated in @xcite to be of order 10 mev . the lightest stickball is expected to be a lorentz scalar @xmath304 and the neutrino @xmath187 may decay by emitting a real stickball if @xmath305 . if @xmath306 the stickball must be virtual and the decay mode is @xmath307 . the mixing between @xmath187 and the bound state neutrinos was quantified in @xcite in terms of two mixing angles @xmath308 . it was found that a maximal value of @xmath309 is expected for 4d ql models . the rate for @xmath310 then depends on @xmath311 . in the 5d theory additional factors arise . again , because @xmath15 boson exchange only couples @xmath187 to @xmath235 kk mode liptons a factor of @xmath312 must be included . also the study of @xcite assumed dirac neutrinos . we have majorana neutrinos and the decay rate for stickball emission will depend on the mixing between @xmath204 and @xmath122 , which takes the standard seesaw form of @xmath313 in our model . putting this together we find that the replacement @xmath314 is required to utilise the results of the 4d study . in @xcite it was found that the lifetime for a 17 kev neutrino to decay by stickball emission could be @xmath315 . employing the replacement ( [ bulk_ql_glueball_decay_replacement ] ) one finds that the lifetime for stickball emission is @xmath316 for a heavy neutrino with mass @xmath282 mev in the 5d model . the lifetime is inversely proportional to @xmath216 so that larger mass values decrease the lifetime . however the dominant decay mode will be @xmath317 for @xmath318 mev , so that the bound of @xmath299 gev still applies . interestingly both the stickball and photon decay modes for @xmath187 depend on the dirac mass coupling @xmath204 and @xmath187 . for @xmath319 these decays do not occur and @xmath187 becomes stable . in this limit it is important to ensure that the density of @xmath187 particles which remain at the freeze out temperature of @xmath320 does not overclose the universe . this leads to the well known lee - weinberg bound of @xmath321 gev for a massive majorana neutrino . this is much weaker than the bound obtained when considering the radiative decay . in the present model the limit @xmath319 corresponds to the limit in which @xmath187 is localised at the opposite boundary to @xmath204 . we note that this limit does not lead to rapid proton decays of the type @xmath322 as kinematic considerations of any neutrino which satisfies the lee - weinberg bound will preclude proton decay via this channel . given that this is the lowest bound on the mass of @xmath187 we will consider this scenario in what remains . this setup has the added advantage of allowing for an improved understanding of the hierarchy between quark and lepton masses in a ql symmetric framework , a matter which is currently under investigation @xcite . having obtained the lower bound on the mass of @xmath187 in the preceding section we may now consider the bounds on the remaining scales in the theory . the ql symmetry breaking scale must satisfy the bound @xmath323 tev , due to leptonic annihilations involving the @xmath86 bosons . if we take @xmath324 tev in accordance with the lower bound obtained by considering kk contributions to fcncs , then the lower bound on @xmath216 and the upper bound of 1 ev on @xmath325 translate into lower bounds on @xmath87 and @xmath126 . using ( [ bulk_ql_sm_neutrin_mass ] ) and ( [ bulk_ql_sterile_nu_mass ] ) gives : @xmath326 so that the lower bound on @xmath216 translates into the bounds @xmath327 for @xmath324 tev . furthermore the demand that the proton does not decay too rapidly translates into a bound on @xmath190 , which we find to be @xmath328 tev . this enables us to summarise the mass scales for the exotic fields in our construction and some of the associated phenomenology . with @xmath324 tev and @xmath329 in accordance with the lower bounds , we find the zero mode @xmath86 boson mass to be @xmath330 tev . the heaviest liptons acquire an order @xmath87 mass and all liptons will appear at energies @xmath331 . these liptons do not couple directly to the known leptons via @xmath15 exchange . they appear in different @xmath12 multiplets and the @xmath32 parities preclude a direct coupling of the type @xmath332 . this is a major distinction between this model and previous ql symmetric constructs . assuming order one yukawa couplings the lightest liptons will possess masses less than @xmath87 . the lower bound of @xmath333 tev permits these liptons to appear at tev energies and thus these states may be observed at the lhc . although they do not couple directly to leptons via @xmath15 boson exchange they will couple with the known fermions through electroweak interactions and through @xmath86 exchange . thus the @xmath270 liptons may appear at the lhc via interactions of the type @xmath334 etc . a key signature for this construct would be the appearance of liptons at the lhc . one could then discriminate this model from other ql symmetric models by studying the coupling of the liptons to leptons at the proposed ilc . the liptons of this model do not couple to leptons via @xmath15 boson exchange and thus only electroweak and @xmath86 interactions would couple @xmath335 pairs to the liptons . we remind the reader that the unbroken @xmath300 symmetry confines liptons into two particle bound states . having the lightest liptons in different @xmath12 multiplets to the sm leptons does not alter the stability of the lightest lipton bound states . these decay via the electromagnetic or weak interactions . as the electroweak bosons have uniform profiles in the fifth dimension their couplings to the liptons are the same as the 4d case . thus the lifetimes of the lightest confined liptons are the same as the 4d case and these bound states present no cosmological concern . the liptons which do couple to the sm leptons are @xmath235 members of kk towers . these states possess mass of order @xmath336 tev . their couplings to sm leptons via @xmath15 exchange are highly suppressed due to the localisation methods employed in this work , thus the associated phenomenology ( like the rare decay @xmath228 ) is also suppressed . whilst new physics may appear at tev energies in this model , it is not until energies of order @xmath337 tev that the higher dimensional nature of the theory reveals itself . at these scales the kk excitations for the neutral gauge bosons @xmath85 and @xmath86 , the photon , the gluons the @xmath34 bosons appear . the @xmath15 bosons also appear at this scale , though at these energies they will only manifest themselves in precision experiments through couplings to the other gauge bosons . we have constructed a complete five dimensional ql symmetric model . this model differs from a previous five dimensional ql symmetric model @xcite in that all fermions are assumed to propagate in the bulk . placing fermions in the bulk provides the following advantages advantages over the earlier framework . namely : + @xmath338 the longevity of the proton is readily understood by localising quarks and leptons at ( or near ) different fixed points in the extra dimension . + @xmath338 the extra dimension may be as large as @xmath92=1/(30 tev ) allowing the phenomenology associated with the kk towers of the gauge sector to be observed at future colliders . + @xmath338 the higher dimensional framework allows one to understand the absence of mass relations of the type @xmath1 or @xmath2 in a ql symmetric framework due to the different profiles of quark and lepton wavefunctions in the extra dimension . + @xmath338 the five dimensional model permits a purely higher dimensional mechanism whereby one may suppress the neutrino mass scale relative to the electroweak scale by spatially separating left- and right - chiral neutrino fields . the bounds on the mass of @xmath187 depend on its localisation point . we find that @xmath248 gev is required if @xmath187 is localised near @xmath204 , whilst @xmath187 can be of order gev if it is localised at the ` quark end ' of the extra dimension . the model as it stands has some features which are introduced in a somewhat arbitrary fashion and it would be pleasing to uncover deeper reasons for their implementation . in particular it would be satisfying to understand why the five dimensional fermion @xmath339 couples more strongly to the bulk scalar @xmath149 than do @xmath340 and @xmath341 . it would also be pleasing to discover a connection between the fermions that undergo @xmath4 symmetry breaking and the fermions which couple with opposite sign yukawa coupling constants to the bulk scalars @xmath149 and @xmath178 . in four dimensional ql symmetric models one starts with two sets of fermions which are indistinguishable at high energies . at low energies leptons are defined as those fermions which experience @xmath4 symmetry breaking . in our five dimensional model there are two independent features which distinguish quarks from leptons . leptons are defined to be those fermions which experience @xmath4 symmetry breaking _ and _ couple to @xmath178 and @xmath149 with opposite signs . it would be pleasing to develop a mechanism which ensures that the fermions which undergo @xmath4 symmetry breaking _ must _ also couple to the two bulk scalars with opposite signs . we note also that some interesting steps towards understanding flavour in a five dimensional left - right symmetric model have been made in @xcite . an intriguing direction for further study would be to attempt to combine the methods employed in @xcite with those of the present study . the authors thank ray volkas , alison demaria and damien george for discussions and a. perez - lorenzana for a helpful correspondence . this work was supported in part by the australian research council . 999 r. foot and h. lew , phys . d * 41 * ( 1990 ) 3502 . r. foot and h. lew , phys . d * 42 * ( 1990 ) 945 . r. foot and h. lew , mod . lett . a * 5 * , 1345 ( 1990 ) . r. foot , h. lew and r. r. volkas , phys . d * 44 * , 1531 ( 1991 ) . y. levin and r. r. volkas , phys . d * 48 * , 5342 ( 1993 ) [ arxiv : hep - ph/9308256 ] . d. s. shaw and r. r. volkas , phys . rev . d * 51 * , 6490 ( 1995 ) [ arxiv : hep - ph/9410350 ] . r. foot and r. r. volkas , phys . b * 358 * , 318 ( 1995 ) [ arxiv : hep - ph/9505331 ] . k. l. mcdonald and b. h. j. mckellar , phys . d * 74 * , 056005 ( 2006 ) [ arxiv : hep - ph/0609110 ] . b. lillie , jhep * 0312 * , 030 ( 2003 ) [ arxiv : hep - ph/0308091 ] . a. delgado , a. pomarol and m. quiros , jhep * 0001 * , 030 ( 2000 ) [ arxiv : hep - ph/9911252 ] . r. foot and r. r. volkas , phys . b * 645 * , 345 ( 2007 ) [ arxiv : hep - ph/0607047 ] . g. c. joshi and r. r. volkas , phys . d * 45 * , 1711 ( 1992 ) . k. s. babu , e. ma and s. willenbrock , phys . d * 69 * , 051301 ( 2004 ) [ arxiv : hep - ph/0307380 ] . s. l. chen and e. ma , mod . a * 19 * , 1267 ( 2004 ) [ arxiv : hep - ph/0403105 ] . a. demaria , c. i. low and r. r. volkas , phys . d * 72 * , 075007 ( 2005 ) [ arxiv : hep - ph/0508160 ] . a. demaria , c. i. low and r. r. volkas , phys . d * 74 * , 033005 ( 2006 ) [ arxiv : hep - ph/0603152 ] . a. demaria and k. l. mcdonald , arxiv : hep - ph/0610346 . p. q. hung , nucl . b * 720 * , 89 ( 2005 ) [ arxiv : hep - ph/0412262 ] . h. georgi , a. k. grant and g. hailu , phys . d * 63 * , 064027 ( 2001 ) [ arxiv : hep - ph/0007350 ] . n. arkani - hamed , a. g. cohen and h. georgi , phys . b * 516 * , 395 ( 2001 ) [ arxiv : hep - th/0103135 ] . r. n. mohapatra and a. perez - lorenzana , phys . d * 66 * , 035005 ( 2002 ) [ arxiv : hep - ph/0205347 ] . g. perez , phys . d * 67 * , 013009 ( 2003 ) [ arxiv : hep - ph/0208102 ] . nath and p. f. perez , arxiv : hep - ph/0601023 . n. arkani - hamed and m. schmaltz , phys . d * 61 * , 033005 ( 2000 ) [ arxiv : hep - ph/9903417 ] . d. e. kaplan and t. m. p. tait , jhep * 0111 * , 051 ( 2001 ) [ arxiv : hep - ph/0110126 ] . a. coulthurst , k. l. mcdonald and b. h. j. mckellar , arxiv : hep - ph/0610345 . y. grossman and g. perez , phys . d * 67 * , 015011 ( 2003 ) [ arxiv : hep - ph/0210053 ] . y. grossman and g. perez , pramana * 62 * , 733 ( 2004 ) [ arxiv : hep - ph/0303243 ] . z. chacko , m. a. luty and e. ponton , jhep * 0007 * , 036 ( 2000 ) [ arxiv : hep - ph/9909248 ] . p. q. hung and n. k. tran , phys . d * 69 * , 064003 ( 2004 ) [ arxiv : hep - ph/0309115 ] . t. goldman and g. j. stephenson , phys . d * 16 * , 2256 ( 1977 ) . e. w. kolb and m. s. turner , the early universe , front . phys . 69 ( 1990 ) 1 . r. foot , h. lew and r. r. volkas , int . j. mod . phys . a * 8 * ( 1993 ) 983 . a. coulthurst and k. l. mcdonald , in preparation . p. q. hung , phys . d * 67 * ( 2003 ) 095011 [ arxiv : hep - ph/0210131 ] .

after a brief review of various mappings of fermion pairs to bosons , we rigorously derive a general approach . following the methods of marumori and otsuka , arima , and iachello , our approach begins with mapping states and constructs boson representations that preserve fermion matrix elements . in several cases these representations factor into finite , hermitian boson images times a projection or norm operator that embodies the pauli principle . we pay particular attention to truncated boson spaces , and describe general methods for constructing hermitian and approximately finite boson image hamiltonians , including effective operator theory to account for excluded states . this method is akin to that of otsuka , arima , and iachello introduced in connection with the interacting boson model , but is more rigorous , general , and systematic .

nucl-th9410012

the original attempt at bosonization of the nuclear many - fermion system was motivated by collective particle - hole modes in nuclei@xcite . since that time the interacting boson model @xcite ( ibm ) has been phenomenologically very successful in explaining low energy nuclear spectroscopy for heavy nuclei . the bosons in this model are thought to represent monopole ( j=0 ) , quadrupole ( j=2 ) , and sometimes hexadecapole ( j=4 ) correlated pairs of valence nucleons in the shell model . the ibm hamiltonian is hermitian , usually has at most two - boson interactions , and conserves boson number , reflecting the particle - particle , rather than particle - hole , nature of the underlying fermion pairs . while one can numerically diagonalize the general ibm hamiltonian , one of the strengths of ibm is the existence of algebraic limits corresponding to the subgroups su(3 ) , u(5 ) , or o(6 ) , with analytic expressions for excitation bands and transition strengths , which encompass an enormous amount of nuclear data . the microscopic reasons for the success of such a simple model are elusive . otsuka , arima , and iachello , along with talmi , have used a mapping of the shell model hamiltonian to the ibm hamiltonian @xcite based on the seniority model @xcite , but these attempts have not done well for well - deformed nuclei @xcite . for this reason we have revisited boson mappings to see if we can understand the success of the ibm starting from the shell model . we will begin by sketching out various historic approaches to boson mappings @xcite . we then specificially follow marumori @xcite and otsuka et al . @xcite ( oai ) in our mapping procedure which maps fermion states into boson states and construct boson operators that reproduce fermion matrix elements . we give the boson representation of the hamiltonian and review the result that , in the full boson fock space , it factorizes into a boson image , which is the same as the belyaev - zelevinskii hamiltonian @xcite in fact in the full space all mappings yield the same results times a normalization operator which projects out the spurious states . however , since our goal is to understand the ibm , which only deals with a few of the enormous degrees of freedom of the shell model , we go on to discuss boson images in truncated spaces . this , we shall see , gives rigorous insight into the oai mapping and shows how to systematically extend it . the fundamental goal is to solve the many - fermion schrdinger equation @xmath0 and find transition matrix elements between eigenstates , @xmath1 . for fermion many - body ( shell - model ) basis states one often uses slater determinants , antisymmetrized products of single - fermion wavefunctions which we can write using fock creation operators : @xmath2 for @xmath3 fermions . for an even number of fermions one can instead construct states from products of fermion pairs , @xmath4 if the number of fermion is fixed at @xmath3 then @xmath5 runs from 1 to @xmath6 . as the fermion fock space may be so large as to make direct solution intractable , the idea of a boson mapping is to replace the fermion operators with boson operators , _ using only a minimal number of boson degrees of freedom _ , that approximate the spectrum and transition matrix elements of the original fermion problem . there are two approaches to boson mappings which we now review . the first approach , epitomized in nuclear physics by belyaev and zelevinskii ( bz ) @xcite , is to map fermion operators to boson operators so as to preserve the original algebra . specifically , consider a space with @xmath7 single - fermion states ; @xmath8 signify fermion creation and annihilation operators . the set of all bilinear fermion operators , @xmath9 , form the lie algebra of @xmath10 , as embodied by the commutation relations @xmath11 = 0 \\ % \left [ a_i a_j , a^\dagger_k a^\dagger_l \right ] = \delta_{il } \delta_{jk } + \delta_{ik } a^\dagger_l a_j + \delta_{jl } a^\dagger_k a_i - ( i \leftrightarrow j ) \\ \left [ a_i a_j , a_k^\dagger a_l \right ] = \delta_{jk } a_i a_l - ( i \leftrightarrow j ) \\ \left [ a^\dagger_i a_j , a^\dagger_k a_l \right ] = \delta_{jk } a^\dagger_i a_l -\delta_{il } a^\dagger_k a_j\end{aligned}\ ] ] at this point it is convenient to introduce collective fermion pair operators @xmath12 we always choose the @xmath13 matrices @xmath14 to be antisymmetric so as to preserve the underlying fermion statistics , thus eliminating the need later on to distinguish between ` ideal ' and ` physical ' bosons . we also assume the following normalization and completeness relations for the matrices : @xmath15 generic one- and two - body fermion operators we represent by @xmath16 , @xmath17 where @xmath18 ; from such operators one can construct a fermion hamiltonian @xmath19 . now one has the following commutation relations : @xmath20 = \left [ \hat{a}^\dagger_\alpha , \hat{a}^\dagger_\beta \right ] = 0 ; \label{commute1 } \\ \left [ \hat{a}_\alpha , \hat{a}^\dagger_\beta \right ] = \delta_{\alpha \beta } -2 \sum_{ij } \left ( { \bf a}^\dagger_\beta { \bf a}_\alpha \right ) _ { ij } a^\dagger_i a_j ; \\ \left [ \hat{a}_\alpha , \hat{t } \right ] = 2\sum_\beta { \rm tr \ , } \left ( { \bf a}_\alpha { \bf t } { \bf a}^\dagger_\beta \right ) \hat{a}_\beta \\ \left [ \hat{t}_1 , \hat{t}_2 \right ] = \sum_{ij } \left ( { \bf t}_1{\bf t}_2 -{\bf t}_2{\bf t}_1 \right)_{ij } a^\dagger_i a_j \label{commutelast}\end{aligned}\ ] ] the method of belyaev and zelevinskii is to find boson images of the bifermion operators , @xmath21 where @xmath22 are boson creation and annihilation operators , @xmath23 = \delta_{\alpha \beta}$ ] , with the coefficients @xmath24 chosen so that the images @xmath25 have the same commutation relations as in ( [ commute1])-([commutelast ] ) . because the algebra is exactly matched , if one builds boson states in exact analogy to the fermion states then the full boson fock space is not spanned and one does not have nonphysical or spurious states . in the full boson fock space , that is , no truncation of the boson degrees of freedom , the image of one body operators is finite and given quite simply by @xmath26 . since in the full space any fermion hamiltonian can be written in terms of one - body operators , the boson image of a finite fermion hamiltonian will be finite in the full space . the states that one must use then are built from the boson representations of the fermion pairs given in ( [ bzpair ] ) which will not just be products of bosons but will include exchange terms . for example , for two bosons and using ( [ bzpair ] ) , @xmath27 these exchange terms are due to the antisymmetry . we shall take care of such exchange effects by introducing a norm operator in the boson space . for _ truncated spaces _ , however , blind application of the boson representations of the pair operators ( [ bzpair ] ) will produce states outside the truncated space . therefore , if these states are omitted , then the hamiltonian needs to be renormalized to account for these omissions , which in general leads to an infinite bz hamiltonian . marshalek @xcite points out that there exist mappings that are both finite and hermitian , but these in general require projection operators to eliminate spurious states . we will regain this result later on in this paper . a variant of belyaev - zelevinskii that also preserves commutation relations is the dyson mapping @xcite : @xmath28 the operators are then clearly finite ; on the other hand they are just as clearly non - hermitian . from a computational viewpoint non - hermiticity is only a minor barrier , but it is an obstacle to an understanding of the microscopic origin of hermitian ibm hamiltonians . furthermore , unless the truncated pairs constitute a subalgebra , under truncation , renormalization will produce an infinite expansion , although it may be possible that this expansion may be more convergent than the bz expansion . the second major approach , pioneered by marumori @xcite , is to map fermion states and construct boson representation operators that preserve matrix elements . the original work of marumori , however , focused on particle - hole excitations and so the number of pairs and consequently bosons were not fixed . however , this method can be applied to particle - particle pairs as well . marumori constructs the norm matrix @xmath29 and then the usui operator @xmath30 where bosons states are constructed in strict analogy to the fermion states , @xmath31 then the marumori expansion of any fermion operator is @xmath32 we shall show that , in the full space , these boson representation operators factor into a finite boson image times a norm operator . furthermore , the boson image of the hamiltonian is the same as the bz image , and hence , the two methods are equivalent in the full space . otsuka , arima , and iachello ( oai ) , along with talmi @xcite , investigated the microscopic origins of the interacting boson model through boson mappings . although they also mapped states , they differed from marumori in some key details . first of all , they built states built on a fixed number of particle - particle , not particle - hole , pairs . in addition , the space was truncated to include only one monopole ( @xmath33 ) and quadrupole ( @xmath34 ) pair . these states were orthogonalized based on seniority . that is , they constructed , for @xmath35 fermions , low - seniority basis states of @xmath36 and @xmath37 fermion pairs , @xmath38 , and then orthonormalized the states such that the zero - seniority state is mapped to itself , and states of higher seniority @xmath39 were orthogonalized against states of lower seniority , @xmath40 then oai calculated the matrix elements @xmath41 for @xmath42 and obtained the coefficients for their one plus two - boson hamiltonian . these coefficients have an implicit @xmath43-dependence ( and for large @xmath43 and arbitrary systems such matrix elements are not trivial to calculate , especially in analytic form ! ) and thus a many - body dependence . at first sight this is not entirely unreasonable as it is well known the ibm parameters change substantially as a function of the number of bosons , even within a major ` shell ' . nonetheless the oai mapping has three drawbacks . the first is that it is not clear how to systematically calculate many - body contributions beyond that contained in the oai prescription , whereas the method we shall describe is fully and rigorously systematic . the second is that the oai prescription can induce many - body effects where none are needed . this point will be illustrated in section [ approximages ] . thirdly , only the @xmath44 space is exactly mapped , but very deformed systems will involve large @xmath45 . in fact , for an axial rotor limit , the average number of d - bosons in the ground state band is 2/3 the total number of bosons . correcting this by also mapping matrix elements with @xmath46 will involve many - body terms . as an alternative to oai , skouras , van isacker , and nagarajan @xcite proposed a `` democratic '' mapping where the orthogonalization is based on eigenvectors of the norm matrix rather than seniority . in what follows we present a unified state - mapping method and obtain four strong results . first we derive matrix elements of the fermion operators in the pair basis ( 2 ) . second , we give general expressions for fermion matrix elements via boson representations . third , we show how in several cases one can have exact , finite , and hermitian boson images of fermion operators . finally , we show how to extend the oai and democratic mappings in a systematic and rigorous fashion , and illustrate how the choice of orthogonalization can affect the many - body dependence of the boson images . the starting point of any state - mapping method is the calculation of matrix elements of fermion operators between states constructed from fermion pairs of the form ( [ statedefn ] ) , including the overlap : @xmath47 , @xmath48 , @xmath49 , and so on . these matrix elements are much more difficult to compute than the corresponding matrix elements between slater determinants . as we shall show , however , full and careful attention paid to the problem of calculation matrix elements can yield powerful results . silvestre - brac and piepenbring @xcite , laboriously using commutation relations , derived a wick theorem for fermion pairs . rowe , song and chen @xcite using ` vector coherent states ' ( we would say fermion - pair coherent states ) found matrix elements between pair - condensate wavefunctions , states of the form @xmath50 using a theorem by lang et al . @xcite , we have generalized the method of rowe , song and chen and recovered ( actually discovered independently ) the expressions of silvestre - brac and piepenbring . to derive the generalized wick s theorem , we use the following theorem ( its proof , which requires fermion coherent states and integration over grassmann variables , is found in appendix a of @xcite ) : let @xmath51 be an operator of the form @xmath52 where each @xmath53 is of the form @xmath54.\ ] ] again , the matrices @xmath55 , @xmath56 are antisymmetric ( @xmath57 is not , in general ) , and are of dimension @xmath58 . we introduce the @xmath59 matrix representation of @xmath51 in the basis of the single - particle fock operators @xmath60 . then : @xmath61 the conclusion of the theorem is the vacuum expectation value of @xmath51 : @xmath62 appropriate derivatives of ( [ vev ] ) bring down pair creation and annihilation operators , leading to the desired matrix elements . we do this in some detail for the overlap matrix elements @xmath63 for the states given in ( 2 ) ; then we simply give the results for matrix elements of one- and two - body operators which are found in the same way . if one begins with @xmath64 where @xmath65 are defined in ( 7 ) , then the overlap is @xmath66 in applying the theorem the matrix algebra is straightforward ; keeping only terms linear in @xmath67 etc . one finds that for the overlap @xmath68 , and @xmath69 next , insert @xmath70 into ( [ vev ] ) , use @xmath71 and expand the logarithm to arrive at the generating function @xmath72 finally , one applies ( [ diffu ] ) . the result is constructed from objects of the form @xmath73 , which we term a _ k - contraction _ ( related to the contractions of @xmath74 phonons in @xcite ) and represent with the notation @xmath75 $ ] ( from the order it is clear that the @xmath76 s represent final states and the @xmath77 s initial states ) . then @xmath63 is a finite sum of products of @xmath74-contractions ; each term in the sum is a product of @xmath78 1-contractions , @xmath79 2-contractions , , and @xmath80 @xmath43-contractions , with the @xmath81 taking all possible values under the constraint that @xmath82 . for example , there will be terms consisting of @xmath43 1-contractions , terms consisting of @xmath83 1-contractions and one 2-contraction , and so on , up to terms consisting of 1 @xmath43-contraction . the full expression is @xmath84 \label{norm}\ ] ] where @xmath85 means the sum is over all permutations of @xmath86 and @xmath87 . the coefficients @xmath88 are given by @xmath89 \ , c_2[\alpha_3 , \beta_3 ; \alpha_4 , \beta_4 ] \ , \ldots c_2[\alpha_{m_2 -1 } , \beta_{m_2 - 1 } ; \alpha_{m_2 } , \beta_{m_2 } ] \ , } _ { m_2 } \\ \times \underbrace { c_3[\alpha'_1 , \beta'_1 ; \alpha'_2 , \beta'_2 ; \alpha'_3 , \beta'_3 ] \ldots c_3[\alpha'_{m_3 - 2 } , \beta'_{m_3 - 2 } ; \alpha'_{m_3 - 1 } , \beta'_{m_3 - 1 } ; \alpha'_{m_3 } , \beta'_{m_3 } ] } _ { m_3 } \times \ldots \nonumber\end{aligned}\ ] ] the organization of ( [ norm ] ) is such as to ease the interpretation in bosons in the next section . the coefficients @xmath90 are found through the expansion of ( [ vev_trace ] ) : @xmath91 where @xmath92 and @xmath93 , and @xmath94 ( hence the normalization chosen above ) . note that we have set @xmath95 and that the @xmath96 have no explicit dependent on @xmath43 . to illustrate , consider a single @xmath97-shell fermion space , with @xmath98 , and further consider only the fermion pair with total @xmath99 : @xmath100_0 $ ] ; the matrix @xmath101 is antidiagonal , @xmath102 . then the @xmath74-contraction @xmath103 . following the above definitions , @xmath104 @xmath105 and @xmath106 then from equation ( [ norm ] ) , including the sum over all permuations which gives a factor @xmath107 , one obtains for example the norm of the state with 5 @xmath108 pairs : @xmath109 which agrees with the general result from the commutation relation @xmath110 = 1 - \hat{n}/ \omega $ ] ( @xmath111 is the fermion number operator ) , that is , the norm is @xmath112 . in the same way as for the overlap one can derive the matrix elements for one- and two - body operators @xmath113 ( and for general @xmath3-body operator , if so desired ) : @xmath114 w_{\ell -k}^0 % \prod_{m = \ell+1}^n c_k [ \alpha_k , \beta_k ] .\ ] ] for a one - body operator @xmath115 we define @xmath116 ; then @xmath117 .\ ] ] for a two - body operator @xmath118 @xmath119 c_{k -l + 1 } [ \alpha_1,\nu ; \alpha_{l+1 } , \beta_{l+1 } ; \ldots ; \alpha_k , \beta_k ] \right . -2 ( \delta_{l,1 } - 1 ) c_{k+1 } [ \mu , \beta_1 ; \alpha_2 , \beta_2 ; \ldots ; \alpha_l , \beta_l ; \alpha_1,\nu ; \alpha_{l+1 } , \beta_{l+1 } ; \ldots ; \alpha_k , \beta_k ] \right ) .\end{aligned}\ ] ] returning to our illustration , first consider the the number operator @xmath111 , which is represented by just the unit matrix ; in this case , again restricting ourselves to @xmath108 pairs , @xmath120 . similarly for the pairing interaction @xmath121 , @xmath122 the reader is invited to check that these coefficients reproduce the correct matrix elements . therefore , given any two states constructed from fermion pairs and the matrices representing those constituent pairs , the above formulas give exactly the overlap and the matrix element for one- and two - body operators . for a pair - condensate wavefunction , the matrix elements can be found quickly through recursion @xcite . throughout this paper we will discuss the effect of truncating the boson fock space , that is taking a restricted number of boson species from which to construct states and operators , on boson mappings . but which boson species should we keep ? in regards to this question we merely wish to comment that rowe , song and chen give a variational principle @xcite which seems useful in this regard , and is probably related in some approximation to the hartree - fock - bogoliubov states that otsuka and yoshinaga @xcite use in their mapping of deformed nuclei . we now want to translate the fermion matrix elements into boson space . we take the simple mapping of fermion states into boson states @xmath123 where the @xmath124 are boson creation operators . we construct boson operators that preserve matrix elements , introducing boson operators @xmath125 , @xmath126 , and most importantly the _ norm operator _ @xmath127 such that @xmath128 , @xmath129 and @xmath130 . we term @xmath125,@xmath126 the boson _ representations _ of the fermion operators @xmath115 , @xmath118 . the boson basis @xmath131 is an orthogonal basis . the fermion norm operator in the boson space will be given by @xmath132 where the coefficients are given by ( [ norm ] ) . because the matrices @xmath133 are orthogonal ( see ( 8) ) , the one - contraction is simply @xmath134 = \delta_{\alpha_{k } , \beta_{k}}$ ] . using the fact that @xmath135 , the number operator , we find the ` linked - cluster ' ( la kishimoto and tamura@xcite although with differences ) expansion of the representations to be of the form @xmath136 and similarly for @xmath137 . in the norm operator the @xmath138-body terms express the fact that the fermion - pair operators do not have exactly bosonic commutation relations , and act to enforce the pauli principle . in the example of a single @xmath97-shell given in the previous section , using the coefficients ( [ w_coef ] ) and with the mapping @xmath139 , the purely @xmath140-boson part of the norm is @xmath141 which again yields the correct matrix elements . the norm operator can be conveniently and compactly expressed in terms of bosons by using the fermion norm matrix ( [ diffu ] ) in terms of the fermion generating function ( [ vev_trace ] ) . taking derivatives with respect to the @xmath142 in ( [ diffu ] ) is like contracting @xmath143 with a @xmath144 , and with respect to @xmath145 is like contracting a @xmath146 with a @xmath147 . hence the norm matrix is just the generating function ( [ vev_trace ] ) with @xmath148 and @xmath149 . however , since these bosons do not commute with one another , we must take the normal order : @xmath150 where the colons ` : ' refer to normal - ordering of the boson operators , and @xmath151 is the @xmath74th - order casimir of @xmath152 , with @xmath153 ( the trace is over the matrices and not the boson fock space ) . this norm operator , which takes into account the exchange terms in the bz expansion of a fermion pair given in ( [ bzpair ] ) , is found in ref . @xcite . similarly and this is a new result we have not seen elsewhere in the literature the representations @xmath154 can also be written in compact form : @xmath155 b^\dagger_\sigma b_\tau \hat { \cal n}_b \colon \label{onebodyimage } \\ \hat{\cal v}_b & = & \sum_{\mu , \nu } \left \langle \mu \left | v \right | \nu \right \rangle \sum_{\sigma , \tau } \colon \left \ { { \rm tr \ , } \left [ { \bf a}_\sigma { \bf a}^\dagger_\mu { \bf g } \right ] { \rm tr \ , } \left [ { \bf a}_\nu { \bf a}^\dagger_\tau { \bf g } \right ] \right . \hspace*{1.00 in } & + & \left.4 \,{\rm tr \ , } \left [ { \bf a}_\sigma { \bf a}^\dagger_\mu { \bf p g a}_\mu { \bf a}^\dagger_\tau { \bf g } \right ] \right \ } b^\dagger_\sigma b_\tau \hat { \cal n}_b \colon , \label{twobodyimage}\end{aligned}\ ] ] where @xmath156 . these compact forms are useful for formal manipulation . furthermore they have the powerful property of exactly expressing the fermion matrix elements under _ any _ truncation , a fact not previously appreciated in the literature even for the norm operator @xcite . by this we mean the following : suppose we truncate our fermion fock space to states constructed from a restricted set of pairs @xmath157 . such a truncation need _ not _ correspond to any subalgebra . then the representations in the corresponding truncated boson space , which still exactly reproduce the fermion matrix elements and which we denote by @xmath158_t$ ] etc . , are the same as those given above , retaining only the ` allowed ' bosons with unrenormalized coefficients . for example @xmath159_t = \colon \exp \left(-\frac{1}{2 } \sum^{\infty}_{k=2 } \frac{(-1)^k}{k } \left[\hat{c}_k\right]_t\right):\ ] ] where @xmath160_t = 2 \colon { \rm tr \ , } ( \left [ { \bf p}\right ] _ t ) ^k \colon \ , , \ , \left [ { \bf p}\right ] _ t = \sum_{\bar{\sigma}\bar{\tau } } b^{\dagger}_{\bar{\sigma } } b_{\bar{\tau } } { \bf a}_{\bar{\sigma } } { \bf a}^{\dagger}_{\bar{\tau}}.\ ] ] this invariance of the coefficients under truncation will not hold true for the boson _ images _ introduced below . with the boson representations of fermion operators in hand , one can express the fermion schrdinger equation ( [ schrodinger1 ] ) with @xmath161 as a generalized boson eigenvalue equation , @xmath162 here @xmath163 is the boson representation of the fermion hamiltonian . every physical fermion eigenstate in ( [ schrodinger1 ] ) has a corresponding eigenstate , with the same eigenvalue , in ( [ schrodinger2 ] ) . because the space of states constructed from pairs of fermions is overcomplete , there also exist spurious boson states that do not correspond to unique physical fermion states . these spurious states will have zero eigenvalues and so can be identified . in general the boson representations given in ( [ compactnorm ] ) , ( [ onebodyimage ] ) and ( [ twobodyimage ] ) do not have good convergence properties , so that simple termination of the series such as ( [ normexpand ] ) in @xmath138-body terms is impossible and use of the generalized eigenvalue equation ( [ schrodinger2 ] ) , as written , is problematic . instead we `` divide out '' the norm operator to obtain the _ boson image _ , i.e. schematically , @xmath164 that this is reasonable is suggested by the explicit forms of ( [ onebodyimage ] ) and ( [ twobodyimage ] ) . the hope of course is that @xmath165 is finite or nearly so , so that a 1 + 2-body fermion hamiltonian is mapped to an image @xmath166 with the @xmath138-body terms , @xmath167 , zero or greatly suppressed . we now discuss how to `` divide out '' the norm . it turns out that for a number of cases the image of the hamiltonian is exactly finite . in particular , for the full boson fock space the representations factor in a simple way : @xmath168 and @xmath169 , where the factored operators @xmath170 , which we term the boson images of @xmath171 , commute with the norm operator and have simple forms : @xmath172 @xmath173\ ] ] the proof of the factorization and commutation requires use of the identities @xmath174 ( @xmath175 is the transpose of @xmath176 ) which in turn are proved using the completeness relation ( [ completeness ] ) . the image hamiltonian @xmath177 is the one determined by bz if one decomposes the hamiltonian into multipole - multipole form and then maps these multipole operators . as discussed earlier , these bz multipole operators are finite in the full space . this result , and its relation to other mappings , was noted by marshalek @xcite thus any boson representation of a hamiltonian factorizes : @xmath178 in the full space . since the norm operator is a function of the @xmath152 casimir operators it commutes with the boson images of fermion operators , and one can simultaneously diagonalize both @xmath163 and @xmath127 . then eqn . ( [ schrodinger2 ] ) becomes @xmath179 where @xmath180 for the physical states , but @xmath181 for the spurious states is no longer necessarily zero . the boson hamiltonian @xmath182 is by construction hermitian and , if one starts with at most only two - body interactions between fermions , has at most two - body boson interactions . all physical eigenstates of the original fermion hamiltonian will have counterparts in ( [ schrodinger3 ] ) . it should be clear that transition amplitudes between physical eigenstates will be preserved . spurious states will also exist but , since the norm operator @xmath127 commutes with the boson image hamiltonian @xmath182 , the physical eigenstates and the spurious states will not admix . also the spurious states can be identified because , while they will no longer have zero energy eigenvalues , they have eigenvalue zero with respect to the norm operator . the boson schrdinger equation ( [ schrodinger3 ] ) , though finite , is not of much use as the boson fock space is much larger than the original fermion fock space , and we still must truncate the boson fock space . although the representations remain exact under truncation , the factorization into the image does not persist in general : @xmath183_t \neq \left [ \hat{\cal n}_b \right ] _ t \left [ \hat{h}_b \right ] _ again for example consider the pairing interaction , even in multiple @xmath97-shells , and a truncation to just @xmath140-bosons ; then @xmath184_t = s^{\dagger } s + { 1\over 2 \omega^2 } s^{\dagger } s^{\dagger } s s$ ] whereas the appropriate image is actually @xmath185 . this was recognized by marshalek @xcite . ( an alternate formulation @xcite does not require the complete fock space , but mixes physical and spurious states and so always requires a projection operator . ) if the truncated set @xmath186 represents a closed subalgebra , that is , if the truncated set of fermion pairs are closed under double commutations : @xmath187 = \delta _ { \bar{\alpha } , \bar{\beta } } - t_{\bar{\alpha } \bar{\beta } } , \\ \left [ t_{\bar{\alpha } \bar{\beta } } , a^\dagger_{\bar{\gamma } } \right ] = \sum_{\bar{\sigma } } \gamma _ { \bar{\alpha } \bar{\beta } ; \bar{\gamma}}^{\bar{\sigma } } a^\dagger_{\bar{\sigma}};\end{aligned}\ ] ] then a factorization @xcite @xmath188_t = \left [ \hat{\cal n}_b \right ] _ t \hat{h}_d = \hat{h}_d^\dagger \left [ \hat{\cal n}_b \right ] _ t \label{dysonfactor}\ ] ] _ does _ exist , with @xmath189 at most two - body , but not necessarily hermitian : @xmath190 we term this a generalized _ dyson _ image @xcite . although @xmath191 , if @xmath189 is non - hermitian , that is @xmath192_t , \hat{h}_d \right ] \neq 0 $ ] , then unfortunately @xmath193 . in the full space , of course , all definitions of boson images coincide and yield the same result . we have found conditions under which @xmath189 is additionally hermitian and commutes with the truncated norm operator @xmath194_t$ ] . consider a partition of the single fermion states labeled by @xmath195 , where the dimension of each subspace is @xmath196 , @xmath197 so that @xmath198 . we denote the amplitudes for the truncated space as @xmath199 and assume they can be factored , @xmath200 , with @xmath201 and @xmath202 , where @xmath203 ( symmetric ) or @xmath204 ( antisymmetric ) . furthermore we assume the completeness relation ( [ completeness ] ) , which was crucial for proving that @xmath178 @xcite , is valid for the truncated space ; i.e. , @xmath205 . \label{newcomplete}\ ] ] the norm operator in the truncated space then becomes @xmath206_t = \colon { \rm exp } \sum_{k = 2 } \left(\frac{-1}{\omega_a}\right)^{k-1 } \frac{1}{k } { \rm tr } ( \bar{\bf p}^k ) \colon , \ ] ] where @xmath207 so that @xmath208_t = \left(\frac{1}{2 \omega_{a}}\right)\bar{\bf p}$ ] . in this case the boson image of a one - body operator is the truncation of the boson image in the full space , @xmath209_t = \left [ \hat{\cal n}_b \right ] _ t \left [ \hat { t}_b \right ] _ t\ ] ] @xmath210_t = 2\sum _ { \bar{\sigma } , \bar{\tau } } { \rm tr \ , } \left ( { \bf a}_{\bar{\sigma } } { \bf t } { \bf a}^\dagger_{\bar{\tau } } \right ) b^\dagger_{\bar{\sigma } } b_{\bar{\tau}}.\ ] ] the representation of a two - body interaction can be factored into a boson image times the truncated norm , @xmath211_t = \left [ \hat{\cal n}_b \right ] _ t \hat{v}_d ; \label{dysonimage2}\ ] ] however , @xmath212 , while finite ( 1 + 2-body ) , is not simply related to @xmath213_t$ ] as is the case for one - body operators . if one writes @xmath214 then matrix elements of the two - boson interaction are @xmath215 \nonumber \\ - \omega_a ( 2\omega_a + ( -1)^p ) { \rm tr}_c ( { \bf a}_{\nu } { \bf k}^{\dagger } \bar{\bf a}^\dagger_{\bar{\tau } } \bar{\bf a}_{\bar{\sigma}^{\prime } } \bar{\bf a}^\dagger_{\bar{\tau}^{\prime } } ) \delta_{\bar{\sigma } , \mu } \rbrace , \label{biglongeqn}\end{aligned}\ ] ] where @xmath216 are traces only in the @xmath217- and @xmath218- spaces , respectively . upon inspection one sees the image ( [ biglongeqn ] ) is not constrained to be hermitian . consider the additional condition between the matrix elements of the interaction : @xmath219 where the factor @xmath220 is the number of pairs in the excluded subspace . while condition ( [ hermcondition ] ) looks complicated there are interactions that satisfy it ; for example , two - body interactions constructed from one - body operators @xmath221 where @xmath222 $ ] . when ( [ hermcondition ] ) is satisfied then @xmath212 is hermitian and although @xmath223_t$ ] they are simply related : @xmath224 with @xmath225 renormalizing the two - boson part of @xmath226_t$ ] by a factor which ranges from unity ( full space ) to 2 for a very small subspace . not all interactions satisfy ( [ hermcondition ] ) ; for example , the pairing interaction never does except in the full space . for the pairing interaction @xmath227 , and @xmath228 , and the image ( [ dysonimage2 ] ) @xmath229 becomes ( remembering @xmath198 ) @xmath230 + \sum_{\bar{\tau } \bar{\tau}^\prime \neq 0 , \bar{\sigma } } { \rm tr}\ ; ( \bar{{\bf a}}_{\bar{\sigma } } \bar{{\bf a}}^\dagger_{\bar{\tau } } \bar{{\bf a}}_0 \bar{{\bf a}}_{\bar{\tau}^\prime}^\dagger ) b^\dagger_{\bar{\sigma } } b^\dagger_0 b_{\bar{\tau } } b_{\bar{\tau}^\prime } \right \rbrace , \label{pairingimage}\ ] ] where @xmath231 is the total number of bosons , @xmath232 , and @xmath233 . the second term in ( [ pairingimage ] ) is not hermitian but can be transformed away by a similarity transformation @xcite , leaving the first term as a finite hermitian image which gives the correct eigenvalues for all n. the so(8 ) and sp(6 ) models @xcite belong to a class of models which have a subspace for which ( [ newcomplete ] ) is valid and interactions which satisfy ( [ hermcondition ] ) . in these models the shell model orbitals have a definite angular momentum @xmath234 and are partitioned into a pseudo orbital angular momentum @xmath235 and pseudospin @xmath236 , @xmath237 . the amplitudes are then given as products of clebsch - gordon coefficients , @xmath238 where @xmath239 and @xmath240 are the total pseudo orbital angular momentum and pseudospin respectively of the pair of nucleons . for the so(8 ) model @xmath241 and one considers the subspace of pairs with @xmath242 @xmath243 , @xmath244 ; in the sp(6 ) model @xmath245 and one considers the subspace with @xmath246 @xmath247 , @xmath248 . the complicated conditions ( [ hermcondition ] ) hold true for important cases , such as the quadrupole - quadrupole and other multipole - multipole interactions in the so(8 ) and sp(6 ) models ( that is , interactions of the generic form @xmath249 in the notation of @xcite ) where therefore have hermitian dyson images . not all interactions in these models have hermitian dyson images . for example , pairing in any model ( see ( [ pairingimage ] ) ) and , in the so(8 ) model , the combination @xmath250 , where @xmath251 , @xmath252 , which is the so(7 ) limit . it so happens that these particular cases nonetheless can be brought into finite , hermitian form as discussed in the next section . the most general image hamiltonian one can define is @xmath253_t^{-1/2 } \left [ \hat{{\cal h \rm}}_b \right ] _ t \left [ \tilde{\cal n}_b \right ] _ t^{-1/2 } { \cal u}^\dagger , \label{himage}\ ] ] which is manifestly hermitian for any truncation scheme and any interaction , with @xmath254 a unitary operator . ( because the norm is a singular operator it can not be inverted . instead @xmath255_t^{-1/2}$ ] is calculated from the norm only in the physical subspace , with the zero eigenvalues which annihilate the spurious states retained . then @xmath165 does not mix physical and spurious states . ) if @xmath256 , this is the democratic mapping @xcite . again , for the full space @xmath257 = 0 $ ] and hence @xmath258 . this prescription is , we argue , useful for a practical derivation of boson image hamiltonians . ignoring for the moment the unitary transformation @xmath254 , consider the expansion ( [ htruncate ] ) of @xmath165 . the operators @xmath259_t$ ] and @xmath255_t^{-1/2}$ ] have similar expansions , and by multiplying out ( [ himage ] ) one sees immediately that the coefficient @xmath260 depends only on up to @xmath138-body terms in @xmath261_t$ ] and @xmath255_t^{-1/2}$ ] , derived from @xmath262-fermion matrix elements which are tractable for @xmath138 small . ideally @xmath165 would have at most two - body terms , and our success in finding finite images in the previous section gives us hope that the high - order many - body terms may be small ; at any rate the convergence can be calculated and checked term - by - term . specifically , consider the convergence of the series ( [ htruncate ] ) as a function of @xmath138 . a rough estimate is that , for an @xmath43-boson fock space , one can truncate to the @xmath138-body terms if for @xmath263 , @xmath264 is sufficiently small compared to @xmath265 ; the strictest condition is to require @xmath266 . although we have given in section iv analytic expressions for the coefficients of the boson representations , in practice one only needs the fermion matrix elements , given in section iii , for the norm and the hamiltonian or other operators . the coefficients of the image @xmath165 are then found by numerical induction . suppose one has the coefficients of @xmath165 up to @xmath138-boson terms . one takes the matrix elements of @xmath255_t^{-1/2}$]note that in calculating @xmath255_t^{-1/2}$ ] , one _ first _ truncates and _ then _ calculates the inverse - square - root ; the two operations do not commmute!and of the hamiltonian or transition operator in the @xmath267 fermion space , and multiply out those matrices as in ( [ himage ] ) , yielding the matrix elements of @xmath165 in the @xmath268-boson space . the coefficients @xmath269 of @xmath165 in ( [ htruncate ] ) are then uniquely determined , up to the freedom embodied in the unitary transform @xmath254 . the hermitian image @xmath165 , defined in ( [ himage ] ) , is related to the dyson image @xmath189 , defined in ( [ dysonfactor ] ) , by a similarity transformation @xmath270_t^{1/2}$ ] , @xmath271 the similarity transformation @xmath272 orthogonalizes the fermion states @xmath273 inasmuch @xmath274 in the physical space ( and @xmath275 in the spurious space ) . this is akin to gram - schmidt orthogonalization and the freedom to choose @xmath254 , and @xmath272 , corresponds to the freedom one has in ordering the vectors in the gram - schmidt procedure . the oai and democratic mappings are just two particular choices out of many ; the former orders the states by seniority whereas the latter takes @xmath276 . we can use this freedom in the choice of @xmath254 to our advantage , which we illustrate in the so(8 ) model @xcite , truncating to the space of @xmath277 pairs ( see eqn . ( [ so8pairdefn ] ) ) , for which one has only an @xmath278 pair ( mapped to an @xmath140 boson ) and a quintuplet of @xmath279 pairs ( mapped to @xmath280 bosons , @xmath281 ) . to second order , the norm is @xmath282_l \cdot \left [ \tilde{d } \otimes \tilde{d } \right ] _ l \nonumber \\ - { \sqrt{5 } \over 2 \omega } \left \ { s^\dagger s^\dagger \left [ \tilde{d } \otimes \tilde{d } \right ] _ 0 + \left [ d^\dagger \otimes d^\dagger \right ] _ 0 s s \right \ } -{2 \over \omega } \hat{n}_d \hat{n}_s.\end{aligned}\ ] ] we pay particular attention to three interactions which correspond to algebraic limits : the pure pairing interaction @xmath283 , the quadrupole - quadrupole interaction @xmath284 (= @xmath285 in the notation of @xcite ) , which can be written in terms of so(6 ) casimir operators , and the linear combination of pairing and quadrupole @xmath286 which can be written in terms of so(7 ) casimirs . as discussed in the previous section , the dyson image of @xmath284 is hermitian and finite , and hence @xmath287 with @xmath288 : @xmath289_0 + \left [ d^\dagger \otimes d^\dagger \right ] _ 0 s s \right \ } + 8 \hat{n}_d \hat{n}_s . \label{qqimage}\ ] ] the dyson images of the pairing and so(7 ) interactions are finite but non - hermitian : @xmath290_0 -2 \hat{n}_d \hat{n}_s , \\ \left ( v^{7 } \right ) _ d = ( \omega+5 ) s^\dagger s + d^\dagger \cdot \tilde{d } - s^\dagger s^\dagger s s + \sqrt{5 } \left [ d^\dagger \otimes d^\dagger \right ] _ 0 s s.\end{aligned}\ ] ] we arrived at these images by computing the fermion matrix elements using the methods of @xcite and writing the norm and the representations of the interactions as matrices , which in second order are at most @xmath291 , and then directly multiplied the matrices @xmath292 . we went to third order to confirm the dyson images are finite . using these same matrices we could also calculate @xmath293 , taking @xmath288 , which yields again ( [ qqimage ] ) for @xmath284 ; for @xmath294 and @xmath295 the images are then hermitian but with nonzero third - order , and presumably higher - order , terms . we then found @xmath254 s @xmath296 for both the pairing and so(7 ) cases ( but not the same @xmath254 ) such that their respective hermitian images @xmath165 are finite ; the one for pairing is unsurprisingly the oai prescription , while that for so(7 ) is the opposite , orthogonalizing states of low seniority against states of higher seniority . these finite , hermitian images are : @xmath297 note that these general hermitian images are not simply related to the truncation of the full image through an overall renormalization , as is the hermitian dyson image ( [ hermdysonimage ] ) . for the pairing interaction , the coefficient of the @xmath298 term changes sign , while off - diagonal terms such as @xmath299 , which exist in the image in the full space , vanish , and in some cases like so(8 ) other diagonal terms such as @xmath300 can also vanish ( depending on the model space ) . if one uses an `` inappropriate '' transform @xmath301 it can induce an unneeded and unwanted many - body dependence . this principle we illustrate in the so(8 ) model , by deriving up to second order the hermitian image of @xmath295 but using @xmath302 : @xmath303_0 \cdot \left [ \tilde{d } \otimes \tilde{d } \right ] _ 0 \nonumber \\ - { 2 \sqrt{5 } \over 1 - { 1 \over \omega } } \sqrt { \left ( 1 -{2 \over \omega } \right ) \left ( 1 + { 4 \over \omega } \right ) } \left \ { s^\dagger s^\dagger \left [ \tilde{d } \otimes \tilde{d } \right ] _ 0 + \left [ d^\dagger \otimes d^\dagger \right ] _ 0 s s \right \ } \label{v7oai}\end{aligned}\ ] ] plus higher order terms which we drop ; this is equivalent to the standard oai procedure @xcite computed in the 2-boson space . in figure 1 we display the spectra of ( [ v7exact ] ) on the left , which is the exact so(7 ) result , and ( [ v7oai ] ) on the right , taking @xmath304 and @xmath305 . the distortions in the right - hand spectrum from the exact result , such as the overall energy shift and the large perturbation in the third band , indicates the importance of the missing many - body terms . in oai @xcite these many - body terms would appear implicitly in the @xmath43-dependence of the coefficients for the two - body terms . from the existence of ( [ v7exact ] ) we see that this strong oai @xmath43-dependence is , for this case at least , a needless complication . therefore it is possible that some of the @xmath43-dependence of oai is induced by their choice of orthogonalization and could be minimized with a different choice . we are currently exploring how to exploit this freedom to best effect . so far we have considered the mapping of a hamiltonian in a truncated fermion space to a boson space . that truncated fermion space , however , may be inadequate for reproducing the spectrum of the full space . consider a single @xmath97-shell fermion space . if the interaction is dominated by pairing , then truncation to just @xmath306 ( @xmath307 ) pairs is reasonable@xcite ; but for other interactions , particularly the quadrupole - quadrupole interaction , the @xmath307-space is inadequate@xcite . to rectify these shortcomings one must introduce an effective interaction theory for boson mappings . an important issue is at what stage to introduce effective operators . for example , one could start from an effective fermion hamiltonian . because the fermion - pair basis is non - orthogonal , however , computation of the effective fermion interaction would be tricky ; furthermore , the starting hamiltonian would have a number dependence which would be difficult to separate from the number dependence induced by the subsequent boson mapping . at the other extreme , sakamoto and kishimoto @xcite start from a boson image and use perturbation theory to account for excluded states . they do not rigorously derive their effective interaction and it is not clear that they properly account for the exchange terms , etc . , included in the norm operator . a better , intermediate approach , is to calculate the boson image in a larger space say @xmath308and then renormalize the pure @xmath309 interaction so as to account for the effect of the @xmath310-boson @xcite . ideally , one should start with a sufficient number of species of bosons so as to exactly span the fermion fock space , and then further truncate the space and renormalize . the number of species required will depend on the number of bosons ( or pairs ) @xmath43 , however , and we know of no prescription for determining this set of bosons ( except for @xmath311 when all are necessary ) . we present a rigorous and general approach to effective interactions . following the usual feshbach @xcite derivation , we partition the _ boson _ fock space using @xmath312 to project out the allowed space and @xmath313 its compliment , with @xmath314 , @xmath315 , @xmath316 and @xmath317 . then the truncated representations are simply @xmath318_t = p { \cal h}_b p , \\ \left [ { \cal n}_b \right ] _ t = p { \cal n}_b p,\end{aligned}\ ] ] and @xmath319 . then the generalized eigenvalue equation in the full space becomes @xmath320_t^{\rm eff}(e_\lambda ) \left | \phi_\lambda \right ) _ t = e_\lambda \left [ \hat { \cal n}_b \right]_t^{\rm eff}(e_\lambda ) \left | \phi_\lambda \right ) _ t , \nonumber\ ] ] with @xmath321_t^{\rm eff}(e ) = p \hat{\cal h}_b p + p \hat{\cal h}_b q { 1 \over q ( e\hat{\cal n}_b- \hat{\cal h}_b ) q } q \hat{\cal h}_b p \nonumber \\ - p \hat{\cal h}_b q { e \over q ( e\hat{\cal n}_b - \hat{\cal h}_b ) q } q \hat{\cal n } _ b p - p \hat{\cal n}_b q { e \over q ( e\hat{\cal n}_b - \hat{\cal h}_b ) q } q \hat{\cal h } _ b p \nonumber \\ + p \hat{\cal n}_b q { e^2 \over q ( e \hat{\cal n}_b - \hat{\cal h}_b ) q } q \hat{\cal n}_b p + e{\cal a}(e ) , \label{effective } \\ \nonumber \\ \left [ \hat{\cal n}_b \right ] _ t^{\rm eff}(e ) = p \hat{\cal n}_b p + { \cal a}(e ) . \nonumber\end{aligned}\ ] ] one can also in principle construct energy - independent , but non - hermitian , effective representations @xcite . there is some ambiguity in the definition of the effective representations as denoted by @xmath322 . for example , one could define @xmath323_t^{\rm eff}$ ] to be simply @xmath324 with the remaining terms in ( [ effective ] ) absorbed into the definition of @xmath325_t^{\rm eff}(e)$ ] , and in principle @xmath322 could be anything at all . now consider boson images , where one divides out the norm operator . we suggest that effective operator theory may be more efficient when applied to representations rather than images , by which we mean that the corrections are smaller . suppose one started with the image hamiltonian in the full space , @xmath326 as defined previously , and from that constructed an effective image in the usual way , @xmath327_t + \delta h_b(e).\ ] ] now compare that with the effective image constructed from effective representations , @xmath328_t^{\rm eff}(e ) \right ) ^{-1/2 } \left [ { \cal h}_b \right ] _ t^{\rm eff } \left ( \left [ \tilde{\cal n}_b \right ] _ t^{\rm eff } ( e)\right ) ^{-1/2 } = \hat{h}_b + \delta h_b(e)\ ] ] ( leaving aside the issue of the choice of an overall unitary transformation @xmath254 ) . now @xmath329_t$ ] . which approach is better ? in those cases , such as so(8 ) and sp(6 ) , where , the @xmath312-space decouples completely from the @xmath313 space , @xmath330 but @xmath331 can not be zero . hence the corrections @xmath332 from using effective representations can be smaller than the corrections @xmath331 determined from performing effective operator theory directly on the image . we now would like to speculate on the possible use of the ambiguity operator @xmath322 . in effective operator theory the eigenstates are no longer orthogonal because of the truncation of the model space ; this is expressed by the fact that one uses either energy - dependent or non - hermitian effective interactions . we propose that this non - orthogonality could also be embedded in the choice of @xmath322 , so that the similarity transform on the basis is now @xmath333_t + { \cal a}(e)}\left | \phi_{\bar{\alpha } } \right ) $ ] . the ambiguity operator @xmath322 could be chosen so as to minimize the energy dependence of the final boson image . although the similarity transformation is now energy - dependent , this would not show up in the calculation of the spectrum , but only in the calculation of effective transition operators . these speculations need to be explored in greater detail . in order to investigate rigorous foundations for the phenomenological interacting boson model , we have presented a rigorous microscopic mapping of fermion pairs to bosons , paying special attention to exact mapping of matrix elements , hermiticity , truncation of the model space , and many - body terms . first we presented new , general and compact forms for boson _ representations _ that preserve fermion matrix elements . we then considered the boson _ image _ hamiltonian which results from `` dividing out '' the norm from the representation ; in the full boson fock space the image is always finite and hermitian ; in addition we discussed several analytic cases for truncated spaces where the image is also finite and hermitian . next , we gave a prescription which is a generalization of both the oai and democratic mappings ; in the most general case for truncated spaces the hermitian image hamiltonian may not be finite but we have demonstrated there is some freedom in the mapping that one could possibly exploit to minimize the many - body terms . this freedom , which manifests itself in a similarity transformation that orders the orthogonalization of the underlying fermion basis , depends on the hamiltonian . finally , we discussed effective operator theory for boson mappings . 999 s.t . belyaev and v.g . zelevinskii , _ nucl . * 39 * ( 1962 ) 582 . t. marumori , m. yamamura , and a. tokunaga , _ prog . theor . * 31 * ( 1964 ) 1009 ; t. marumori , m. yamamura , a. tokunaga , and a.takeda , _ prog . * 32 * ( 1964 ) 726 . f. iachello and a. arima , _ the interacting boson model _ ( cambridge university press , 1987 ) . t. otsuka , a. arima , f. iachello and i. talmi , _ phys . lett._*76b * ( 1978 ) 139 . t. otsuka , a. arima , and f. iachello , _ nucl . phys . _ * a309 * ( 1978 ) 1 . i. talmi , _ simple models of complex nuclei _ , ( harwood academic publishers , 1994 ) . t. otsuka , _ nucl . phys . _ * a368 * ( 1981 ) 244 . a. klein and e. r. marshalek , _ rev . phys . _ * 63 * ( 1991 ) 375 . p. ring and p. schuck , _ the nuclear many - body problem _ ( springer - verlag , 1980 ) . e. r. marshalek , _ phys . c _ * 38 * ( 1988 ) 2961 . dyson _ phys . _ * 102 * ( 1956 ) 1217 . l. d. skouras , p. van isacker , and m. a. nagarajan , _ nucl . phys . _ * a516 * ( 1990 ) 255 . b. silvestre - brac and r. piepenbring _ phys . c _ * 26 * ( 1982 ) 2640 . rowe , t. song and h. chen , _ phys . c _ * 44*(1991 ) r598 . lang , c.w . johnson , s.e . koonin , and w.e . ormand , _ phys . c _ * 48 * ( 1993 ) 1518 . t. otsuka and n. yoshinaga , _ phys . lett . _ * 168b * ( 1986 ) 1 . t. kishimoto and t. tamura , _ phys . c _ * 27 * ( 1983 ) 341 . h. sakamoto and t. kishimoto , _ nucl . _ * a486 * ( 1988 ) 1 . j. dobaczewski , h.b . geyer , and f.j.w . hahne , _ phys . c _ * 44 * ( 1991 ) 1030 . ginocchio and c.w . johnson , to be published in `` frontiers of nuclear structure physics '' , t. otsuka , ed . ( world scientific , 1994 ) . ginocchio and i. talmi , _ nucl . phys . _ * a337 * ( 1980 ) 431 . ginocchio , _ ann . * 126 * ( 1980 ) 234 . halse , l. jaqua and b. r. barrett _ phys . c _ * 40 * ( 1989 ) 968 . t. otsuka and j. n. ginocchio , _ phys . _ * 55 * ( 1985 ) 276 . p. navratil and h. b. geyer , _ nucl . * a556 * ( 1993 ) , 165 . o. scholten in _ computational nuclear physics 1 _ , k. langanke , j.a . maruhn and s.e . koonin , eds . , ( springer - verlag 1991 ) chapter 5 .

we report on the discovery of cepheids in the spiral galaxy 2841 , based on observations made with the wide field and planetary camera 2 on board the _ hubble space telescope_. 2841 was observed over 12 epochs using the f555w filter , and over 5 epochs using the f814w filter . photometry was performed using the daophot / allframe package . we discovered a total of variables , including high - quality cepheids with periods ranging from 15 to 40 days . period - luminosity relations in the v and i bands , based on the high - quality cepheids , yield a reddening - corrected distance modulus of mag , which corresponds to a distance of mpc . our distance is based on an assumed lmc distance modulus of mag ( kpc ) and a metallicity dependence of the cepheid p - l relation of mag / dex .

astro-ph0105491

this paper presents the discovery of @xmath8 cepheids and a distance determination to 2841 . 2841 is an isolated , flocculent spiral galaxy , classified as sb(r ) : i @xcite or sb i @xcite , and located at @xmath9 9@xmath10 22@xmath11 03@xmath12 , @xmath13 + 50@xmath14 58@xmath15 36@xmath16 ( j2000.0 ) . it exhibits a heliocentric redshift of @xmath17 km / s , and has a major axis diameter of @xmath18 . @xcite places it in the northwestern - most association of the leo spur ( group 15 + 10 ) . the inclination angle of 2841 is @xmath19 @xcite , making it a suitable calibrator of the tully - fisher relation . furthermore , the large observed 21-cm line - width of 2841 ( @xmath20 km / s , * ? ? ? * ) significantly extends the range in rotation velocity over which the tully - fisher relation has been calibrated using cepheid distances . this topic will be covered by @xcite , which will also address the implications of the 2841 cepheid distance regarding the predictions from the mond ( modified newtonian dynamics ) theory . 2841 has been the host of four supernovae in the last century : 1912a , 1957a , 1972r and 1999by . the last one is a well - observed type ia sn which has been classified as a fast decliner ( hence sub - luminous ) . thus , it is of special importance for the calibration of the type ia sne distance indicator . this topic will be addressed by @xcite . 2 describes the observations and preliminary reductions of the images . 3 contains the details of photometry and calibration of the instrumental magnitudes . 4 presents the sample of variables and their properties . 5 describes the fiducial cepheid period - luminosity relations used in our analysis , derives a distance modulus , and lists the sources of uncertainty . 6 discusses some implications of our distance measurement , and 7 presents our conclusions . 2841 was observed by the wide field and planetary camera 2 ( wfpc2 ) @xcite on board hst on twelve epochs between 2000 february 29 and april 19 . the observations were performed using the f555w ( approximately johnson v ) and the f814w ( approximately kron - cousins i ) filters . all twelve epochs contained a pair of cr - split images in the f555w band , while five epochs contained an additional pair of cr - split images in the f814w . each individual image had an exposure time of 1100s . all observations were made with the telescope guiding in fine lock with a stability of approximately 3 mas . the gain and readout noise were 7 e@xmath21/dn and 7 e@xmath21 , respectively . the ccd was operated at a temperature of 88@xmath22c for all observations . a log of the observations is presented in table 1 . the sampling strategy followed in these observations was similar to the one employed by the _ _ key project on the extragalactic distance scale @xcite , it followed a power - law distribution in time , which provides an optimum sampling of the light curves of cepheids in the period range between 10 and 50 days . the images were calibrated using the pipeline processing at the space telescope institute ( stsci ) . the full reduction procedure @xcite consisted of : a correction for a / d conversion errors , the subtraction of a bias level for each chip , the subtraction of a superbias frame , the subtraction of a dark frame , a correction for shutter shading , and a division by a flat field . furthermore , the images were corrected for vignetting and geometrical distortions in the wfpc2 optics ( the latter correction was done using files kindly provided by j. holtzman ) . lastly , the images were multiplied by four and converted to two - byte integers , to reduce disk usage and allow image compression . this conversion led to an effective readout noise of 4 dn and a gain of 1.75e@xmath23/dn . figure 1 displays a ground - based image of 2841 , obtained at the whipple observatory 1.2-m telescope , with a super - imposed mosaic of the wfpc2 field of view . the mosaic is shown in greater detail in figure 2 . photometry of 2841 was performed using the daophot / allframe package . @xcite contain comprehensive overviews of the package and the reduction strategy , which we briefly summarize here . as in the case of previous allframe reductions of wfpc2 data , we used external point - spread functions ( psfs ) , determined by p.b.s . for each chip and filter combination from high @xmath24 , uncrowded observations of galactic globular cluster fields . these external psfs take into account the variation in the psf across the field of view of each wfpc2 chip . the photometric reduction started with a preliminary detection of sources in each individual image . daomatch and daomaster were used to derive offsets and rotations between frames , relative to the first f555w frame . in the case of these observations , the shifts were @xmath25 pixel for 31 frames , and @xmath26 pixels for the other two . once the transformations were established , montage was used to create a medianed image , free of cosmic rays and chip defects . the find routine in daophot was used to detect the sources present in that image , and their positions were refined by running allstar . the resulting list of objects was used by allframe to perform the photometry of each object in every frame . the conversion from psf to standard @xcite 0.5@xmath27-radius aperture magnitudes involved the determination of growth curves . first , we selected bright and isolated stars ( hereafter , `` secondary standards '' ) that were located in areas with low and slowly - varying backgrounds . next , all other objects were removed from the frames and aperture photometry was carried out at a variety of radii . daogrow @xcite was used to derive growth curves and to obtain @xmath28-radius aperture magnitudes for the secondary standards in each frame . lastly , we used collect to calculate the `` aperture correction '' coefficient for each frame from the difference between psf and aperture magnitudes of the secondary standards . the conversion from @xmath29 and @xmath30 magnitudes to johnson v and kron - cousins i bandpasses followed the precepts of @xcite : @xmath31\ ( \vi)\\ \nonumber & & + 0.027 [ \pm0.002]\ ( \vi)^2+z_{i , f555w}\\[6pt ] i&=&f814w-0.062 [ \pm0.009]\ ( \vi)\\ \nonumber & & + 0.025 [ \pm0.002]\ ( \vi)^2+z_{i , f814w}\end{aligned}\ ] ] where @xmath29 and @xmath30 equal @xmath32 and the color terms come from table 7 of @xcite . the zeropoint terms ( @xmath33 ) , which are listed in table 2 , were determined by p.b.s . and reflect evolution in the understanding of the wfpc2 calibration since the publication of @xcite ( cf . table 5 of that paper ) . for those epochs without f814w data , the transformation of the @xmath29 magnitudes into the standard v bandpass was carried out using the mean @xmath34 color of each object . additionally , the magnitudes of each object were corrected for the effects of charge transfer inefficiency following the prescription of @xcite . in order to facilitate the comparison of our magnitude system with future studies , we list in table 3 the mean v and i magnitudes for the secondary standards present in each chip . the initial search for variables was carried out using trial , a program that calculated the value of the @xmath35 statistic for each star , based on the v - band data . the @xmath35 statistic is a robust index of variability developed by @xcite , and it is designed so that objects with constant magnitudes yield values near zero . the mean value and _ r.m.s . _ deviation of this index for all stars present in our database was @xmath36 . we flagged @xmath37 objects with @xmath38 as possible variables , and extracted their individual light curves from the photometry database for further analysis using a suite of programs developed by the direct team @xcite . we calculated periods for the candidate variables using a modified version of the lafler - kinman technique @xcite ; we restricted the range of possible periods to the range between 10 and 50 days . next , template cepheid light curves ( derived by * ? ? ? * ) were simultaneously fitted to the phased v- and i - band photometry of each candidate . the use of templates allows a robust classification of a candidate variable as a cepheid , refines the period of the variable , and yields reliable mean magnitudes through numerical integration of the best - fit template . if the candidate object was classified as a cepheid , an iterative @xmath39 rejection algorithm was applied to the data to reject anomalous photometric measurements , most of them associated with cosmic - ray hits , and the light curve was analyzed again . about 5% of the individual data points were rejected in this manner . the automated classification software rejected @xmath40 candidates and classified @xmath8 objects as cepheids . an additional @xmath41 objects are clearly variables , and most likely cepheids , but their periods are longer that our observing window and we can not obtain accurate values for their mean magnitudes . table 4 lists the following properties of the @xmath0 variables which passed the analysis : identification number ( c01-c@xmath8 in order of descending period , for the cepheids , and v01-v0@xmath41 in order of increasing r.a . , for the other variables ) ; chip where the object is located ; @xmath42 coordinates ( based on the coordinate system of the first v band image ) ; j2000.0 coordinates ; best - fit period and uncertainty ( cepheids only ) ; v and i mean magnitudes , derived from the best - fit template light curves ; and a selection flag ( see 5.2 for details ) . figure 3 contains chip - wide images of each of the wfpc2 ccds , indicating the locations of the variables . finding charts for each of the stars are displayed in figure 4 ; the charts encompass a 50 by 50 pixel area around each variable ( i.e. , @xmath43 for the pc and @xmath44 for the wf ) . the individual v and i band photometric measurements are listed in tables 5 and 6 , respectively , and shown as phase - magnitude plots ( for the cepheids ) and as time series ( for the other variables ) in figure 5 . that figure also shows the v and i best - fit template light curves for each cepheid , which were used to determine their mean magnitudes . figure 6 shows a color - magnitude diagram of the @xmath45 stars detected in our images . as expected , the cepheids occupy a region of the diagram that is consistent with the location of the instability strip . figure 7 shows the observed differential luminosity functions for the v and i data , which indicate that our completeness limits are v@xmath46 mag and i@xmath47 mag . additionally , the i band luminosity function does not exhibit the signature associated with the tip of the red giant branch , yielding a lower limit for the distance to 2841 of @xmath48 mag or @xmath49 mpc the method used to derive v- and i - band apparent distance moduli is the same one that was used by the _ hst _ key project on the extragalactic distance scale , which is described in detail by @xcite . it relies on fiducial period - luminosity relations of lmc cepheids , corrected for reddening and scaled based on an assumed true distance modulus of @xmath50 = 18.50@xmath510.10 mag . the p - l relations used are those from @xcite , based on a sample of @xmath52 ogle cepheids with periods ranging from 2.5 to 28 days : -6pt @xmath53 -4.218(\pm0.02 ) , \\ m_i&=&-2.962(\pm0.02)\left[\log{p } -1\right ] -4.904(\pm0.04).\end{aligned}\ ] ] these relations can be combined to obtain an equation for the reddening - corrected distance modulus of a cepheid @xcite : -6pt @xmath54 + 5.899(\pm0.01),\ ] ] where @xmath55 is the reddening - free wesenheit magnitude @xcite . before determining distance moduli from equations 3 - 5 , we must impose a cut in period at the short end to avoid an incompleteness bias ( see appendix a of * ? ? ? the completeness limits determined at the end of 4 imply that this cut must be applied around @xmath56 d , resulting in the rejection of cepheids c01-c05 . we also rejected cepheids c07-c09 due to their rather poor quality i band light curves . thus , the final sample of cepheids used for distance determination consists of @xmath1 variables , identified with an asterisk in the last column of table 4 . figures 8 and 9 show the v and i band p - l relations of the final sample of cepheids . by fitting the p - l relations described in equations ( 3 ) and ( 4 ) , we obtain apparent v and i band distance moduli of @xmath57 mag and @xmath58 mag ( internal errors only ) , and a mean color excess of @xmath59 mag . the application of equation ( 5 ) results in an extinction - corrected distance modulus of @xmath60 mag ( internal errors only ) . the measured value of @xmath61 must be corrected for the metallicity dependence of the cepheid p - l relation , using the following equation : -6pt @xmath62_{n2841 } - [ o / h]_{lmc } \big ) , \ ] ] where @xmath63 is the metallicity dependence index for the v and i filters , and @xmath64_{gal } = \log ( o / h)_{gal } / \log ( o / h)_\odot$ ] . we adopt @xmath65 mag / dex @xcite , which represents the range of measurements found in the literature @xcite . the adopted @xmath64 $ ] value for the lmc is @xmath66 dex @xcite . the @xmath64 $ ] value for our field in 2841 can be calculated using @xmath67 \lambda\lambda 3727)/h\beta$ ] and @xmath68 \lambda\lambda 4959,5007)/h\beta$ ] line flux ratios listed in table 3 of @xcite for four hii regions located at galactocentric radii similar to those spanned by our cepheids . the sum of these two line flux ratios ( known as @xmath69 ) can be turned into a value of @xmath64 $ ] by using the appropriate equation from @xcite . we find @xmath64_{n2841}= 0.4\pm0.15 $ ] dex , which reflects the scatter in the abundances of the four hii regions ( @xmath70 ) and the uncertainty in the calibration of the metallicity-@xmath69 relation for high abundances . in conclusion , equation ( 6 ) results in an overall correction of @xmath71 mag to our distance modulus , making it @xmath72 mag . this uncertainty includes several sources of error , which we list in the following sub - section . the error budget for our distance determination is listed in table 7 , and explained in greater detail here . the internal error due to the dispersion in individual values of @xmath61 is a small contribution to the total error , which is heavily dominated by external sources of error . these are , in decreasing order of importance : the metallicity dependence of the cepheid p - l relation , the distance to the lmc , the absolute calibration of wfpc2 , contamination of the cepheid magnitudes due to unresolved blends , and the uncertainty in the reddening law of 2841 . _ i. metallicity correction _ : as stated in 5.3 , the uncertainty in the adopted value of the metallicity correction index @xmath63 is @xmath73 mag / dex , and the uncertainty in the difference in abundances between the lmc and 2841 is @xmath74 dex . the combination of these two uncertainties results in a contribution to the error budget of @xmath75 mag . _ lmc distance _ : the uncertainty in the distance to the lmc has been recently summarized by @xcite . we have adopted @xmath76 mag in order to maintain consistency with the cepheid distances presented in that paper . however , different distance indicators yield widely different ( and non - overlapping ) estimates of the lmc distance modulus , ranging from 18.3 to 18.65 mag ( see table 13 of * ? ? ? blending _ : an additional source of uncertainty is the possible contamination of our cepheid magnitudes by nearby stars , physically unrelated to the variables , but located at distances smaller than the resolution provided by the telescope and detector . @xcite have studied this contamination by comparing ground - based and hst images of long - period ( @xmath77 ) cepheids in m33 . they conclude that the median value of this effect on hst data for galaxies at 1015 mpc is @xmath78 in v and @xmath79 in i. if these numbers were to apply to our cepheids , our @xmath80 magnitudes would be biased by @xmath81 , or @xmath82 mag . we take this value to represent a @xmath83 uncertainty due to the possible effects of blending . _ iv . wfpc2 calibration _ : @xcite have estimated the size of the uncertainty related to the absolute photometric calibration of wfpc2 to be @xmath84 mag ; this includes the uncertainties in the characterization of the zeropoints , the color transformations , the camera gain ratios , and the effects of cte . _ v. reddening law _ : the reddening - free wesenheit magnitudes used in equation ( 5 ) were calculated using an assumed value of total - to - selective extinction @xmath85 . if the interstellar medium of 2841 were to follow a different extinction law than the ism of the milky way , the adoption of the standard value of @xmath86 would introduce an additional systematic error to our distance estimate . this issue was addressed in a recent study by @xcite , who obtained nicmos @xmath87 band photometry of extragalactic cepheids in a dozen galaxies . they found that the extinction corrections determined from either v and i data or from v and @xmath87 data , using the standard extinction law , are consistent with each other . thus , the assumption of a standard reddening law is a valid one , and does not introduce a significant source of error . the aforementioned sources of error ( i - iv ) can be combined in quadrature to arrive at a total external uncertainty in our metallicity - corrected true distance modulus of @xmath88 mag . the addition of the internal error term ( @xmath89 mag ) increases the total uncertainty to @xmath90 mag , which we simplify as @xmath91 mag . 2841 is one of eight galaxies with cepheid distances associated with the leo cloud & spur , out of a total of 65 members listed in @xcite . table 8 lists some properties of those eight galaxies . all distances ( except for the one derived in this paper ) are from @xcite . the mean velocity and dispersion of the previously observed galaxies is @xmath92 km / s , while the average distance ( excluding the nearby ngc@xmath73621 ) is @xmath93 mpc . the small velocity dispersion for this relatively spread out structure is somewhat surprising , but consistent with bulk motion of the leo cloud . kinematically , at a redshift of @xmath94 km / s , 2841 is clearly a member of the leo spur , and our derived distance of @xmath3 mpc places it on the far side of this structure . figure 10 shows the bvri surface brightness profiles of 2841 calculated from ccd images obtained at the fred l. whipple observatory 1.2-m telescope . the procedure we followed to obtain these profiles is identical to the one described in @xcite . we find a @xmath95 isophotal radius of @xmath96 , which corresponds to a linear isophotal radius of @xmath97 kpc , given the distance of @xmath3 mpc derived in the previous section . thus , 2841 is an impressively large disk galaxy , about twice the size of m31 , a galaxy which shares many of its properties . furthermore , the values of extinction - corrected effective radius , @xmath98 kpc , and extinction - corrected effective surface brightness , @xmath99 , place 2841 in a relatively unique position in the surface brightness radius plane ( see figure 2 of * ? ? ? we have discovered a total of @xmath8 cepheids and @xmath41 other variables in the sb spiral galaxy 2841 , a member of the leo spur . we have applied fiducial ogle lmc v- and i - band cepheid period - luminosity relations to a sub - sample of @xmath1 cepheids . we adopted an lmc distance of @xmath61(lmc)@xmath100 mag , and a metallicity dependence of @xmath65 mag / dex . based on these assumptions , we obtain a distance modulus of @xmath2 mag , corresponding to a distance of @xmath3 mpc . our distance estimate is consistent with an association of this galaxy to the leo spur . the mean reddening of the cepheid sample is @xmath101 mag . a companion paper by @xcite will address the impact of the cepheid distance to 2841 on the absolute calibration of the tully - fisher relation , as well as the implication of our distance for the mond theory . a second companion paper by @xcite will combine our cepheid distance with observations of 1999by to further the absolute calibration of the type ia sne distance indicator . we would like to thank the referee , john graham , for his helpful comments , and the stsci and nasa support staff that have made these observations possible . this work has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , caltech , under contract with the national aeronautics and space administration . llll 2000 feb 29 & 2451603.35 & @xmath102 & f555w + & 2451603.41 & @xmath102 & f814w + 2000 mar 02 & 2451605.29 & @xmath102 & f555w + 2000 mar 04 & 2451607.24 & @xmath102 & f555w + 2000 mar 06 & 2451609.66 & @xmath102 & f555w + & 2451609.72 & @xmath102 & f814w + 2000 mar 09 & 2451612.54 & @xmath102 & f555w + 2000 mar 12 & 2451615.49 & @xmath102 & f555w + 2000 mar 16 & 2451619.38 & @xmath102 & f555w + & 2451619.44 & @xmath102 & f814w + 2000 mar 20 & 2451623.60 & @xmath102 & f555w + 2000 mar 26 & 2451629.30 & @xmath102 & f555w + & 2451629.36 & @xmath102 & f814w + 2000 apr 02 & 2451636.27 & @xmath102 & f555w + 2000 apr 09 & 2451643.24 & @xmath102 & f555w + & 2451643.31 & @xmath102 & f814w + 2000 apr 19 & 2451653.56 & @xmath102 & f555w + lcrrrrcc s01 & 1 & 132.2 & 467.7 & 09:22:10.649 & 50:56:12.66 & @xmath111 & @xmath112 + s02 & 1 & 141.4 & 530.3 & 09:22:10.725 & 50:56:09.88 & @xmath113 & @xmath114 + s03 & 1 & 238.3 & 600.8 & 09:22:11.223 & 50:56:07.16 & @xmath115 & @xmath116 + s04 & 1 & 540.2 & 144.5 & 09:22:12.433 & 50:56:29.20 & @xmath117 & @xmath118 + s05 & 1 & 508.3 & 685.9 & 09:22:12.556 & 50:56:04.64 & @xmath119 & @xmath120 + s06 & 1 & 591.2 & 687.4 & 09:22:12.952 & 50:56:04.99 & @xmath121 & @xmath122 + s07 & 1 & 659.5 & 98.5 & 09:22:12.976 & 50:56:31.81 & @xmath123 & @xmath124 + s08 & 1 & 715.8 & 262.0 & 09:22:13.329 & 50:56:24.75 & @xmath125 & @xmath126 + s09 & 2 & 117.0 & 744.0 & 09:22:02.603 & 50:56:16.07 & @xmath127 & @xmath128 + s10 & 2 & 390.5 & 543.7 & 09:22:05.010 & 50:55:51.50 & @xmath129 & @xmath130 + s11 & 2 & 272.8 & 410.5 & 09:22:06.260 & 50:56:04.66 & @xmath131 & @xmath132 + s12 & 2 & 193.1 & 396.0 & 09:22:06.315 & 50:56:12.69 & @xmath133 & @xmath134 + s13 & 2 & 68.9 & 371.7 & 09:22:06.418 & 50:56:25.17 & @xmath135 & @xmath136 + s14 & 2 & 464.0 & 355.9 & 09:22:07.063 & 50:55:46.39 & @xmath137 & @xmath138 + s15 & 2 & 349.2 & 308.1 & 09:22:07.424 & 50:55:58.29 & @xmath131 & @xmath139 + s16 & 2 & 189.2 & 271.2 & 09:22:07.615 & 50:56:14.49 & @xmath140 & @xmath141 + s17 & 2 & 208.3 & 181.5 & 09:22:08.574 & 50:56:13.62 & @xmath142 & @xmath143 + s18 & 2 & 225.7 & 160.3 & 09:22:08.816 & 50:56:12.14 & @xmath144 & @xmath145 + s19 & 2 & 332.0 & 64.2 & 09:22:09.944 & 50:56:02.75 & @xmath146 & @xmath147 + s20 & 3 & 730.9 & 495.3 & 09:22:02.202 & 50:57:07.63 & @xmath148 & @xmath149 + s21 & 3 & 723.7 & 457.7 & 09:22:02.319 & 50:57:04.01 & @xmath150 & @xmath151 + s22 & 3 & 705.2 & 400.4 & 09:22:02.576 & 50:56:58.57 & @xmath152 & @xmath153 + s23 & 3 & 716.0 & 214.2 & 09:22:02.682 & 50:56:40.13 & @xmath154 & @xmath155 + s24 & 3 & 689.1 & 373.8 & 09:22:02.774 & 50:56:56.12 & @xmath156 & @xmath157 + s25 & 3 & 702.8 & 156.2 & 09:22:02.888 & 50:56:34.59 & @xmath158 & @xmath147 + s26 & 3 & 631.8 & 364.0 & 09:22:03.382 & 50:56:55.77 & @xmath159 & @xmath160 + s27 & 3 & 634.1 & 284.6 & 09:22:03.450 & 50:56:47.92 & @xmath161 & @xmath162 + s28 & 3 & 607.5 & 139.7 & 09:22:03.897 & 50:56:33.96 & @xmath163 & @xmath164 + s29 & 3 & 509.6 & 207.4 & 09:22:04.840 & 50:56:41.64 & @xmath165 & @xmath141 + s30 & 3 & 473.9 & 177.8 & 09:22:05.247 & 50:56:39.12 & @xmath166 & @xmath167 + s31 & 3 & 462.7 & 101.3 & 09:22:05.452 & 50:56:31.73 & @xmath168 & @xmath169 + s32 & 3 & 434.2 & 248.7 & 09:22:05.580 & 50:56:46.53 & @xmath170 & @xmath171 + s33 & 3 & 425.0 & 225.3 & 09:22:05.704 & 50:56:44.32 & @xmath172 & @xmath173 + s34 & 3 & 385.3 & 100.9 & 09:22:06.261 & 50:56:32.52 & @xmath174 & @xmath175 + s35 & 3 & 360.1 & 82.5 & 09:22:06.545 & 50:56:31.00 & @xmath176 & @xmath177 + s36 & 3 & 282.6 & 110.9 & 09:22:07.322 & 50:56:34.63 & @xmath178 & @xmath171 + s37 & 3 & 280.5 & 93.6 & 09:22:07.363 & 50:56:32.96 & @xmath154 & @xmath179 + s38 & 3 & 249.9 & 78.8 & 09:22:07.698 & 50:56:31.85 & @xmath180 & @xmath181 + s39 & 3 & 220.4 & 141.7 & 09:22:07.936 & 50:56:38.34 & @xmath182 & @xmath183 + s40 & 3 & 163.9 & 162.4 & 09:22:08.500 & 50:56:41.01 & @xmath184 & @xmath185 + s41 & 4 & 85.5 & 262.1 & 09:22:12.214 & 50:56:37.22 & @xmath130 & @xmath186 + s42 & 4 & 588.9 & 323.0 & 09:22:12.322 & 50:57:27.53 & @xmath137 & @xmath187 + s43 & 4 & 296.7 & 300.6 & 09:22:12.392 & 50:56:58.40 & @xmath188 & @xmath189 + s44 & 4 & 132.5 & 309.9 & 09:22:12.663 & 50:56:42.30 & @xmath190 & @xmath191 + s45 & 4 & 202.5 & 371.4 & 09:22:13.233 & 50:56:49.80 & @xmath192 & @xmath193 + s46 & 4 & 502.3 & 421.4 & 09:22:13.443 & 50:57:19.95 & @xmath194 & @xmath195 + s47 & 4 & 154.1 & 389.7 & 09:22:13.476 & 50:56:45.21 & @xmath196 & @xmath197 + s48 & 4 & 257.2 & 420.9 & 09:22:13.695 & 50:56:55.69 & @xmath198 & @xmath199 + s49 & 4 & 229.0 & 439.2 & 09:22:13.916 & 50:56:53.09 & @xmath200 & @xmath201 + s50 & 4 & 684.1 & 510.6 & 09:22:14.185 & 50:57:38.74 & @xmath125 & @xmath202 + s51 & 4 & 738.3 & 543.7 & 09:22:14.473 & 50:57:44.37 & @xmath125 & @xmath203 + s52 & 4 & 595.7 & 535.8 & 09:22:14.543 & 50:57:30.30 & @xmath204 & @xmath205 + s53 & 4 & 507.6 & 545.1 & 09:22:14.733 & 50:57:21.69 & @xmath206 & @xmath207 + s54 & 4 & 595.3 & 562.4 & 09:22:14.821 & 50:57:30.52 & @xmath208 & @xmath134 + s55 & 4 & 722.2 & 577.1 & 09:22:14.838 & 50:57:43.11 & @xmath209 & @xmath210 + s56 & 4 & 752.0 & 587.9 & 09:22:14.919 & 50:57:46.12 & @xmath161 & @xmath211 + s57 & 4 & 157.4 & 539.7 & 09:22:15.042 & 50:56:47.04 & @xmath212 & @xmath213 + s58 & 4 & 131.8 & 652.5 & 09:22:16.243 & 50:56:45.68 & @xmath214 & @xmath215 + s59 & 4 & 662.6 & 737.1 & 09:22:16.565 & 50:57:38.77 & @xmath216 & @xmath217 + s60 & 4 & 481.2 & 718.5 & 09:22:16.569 & 50:57:20.77 & @xmath218 & @xmath219 + s61 & 4 & 459.2 & 747.9 & 09:22:16.897 & 50:57:18.89 & @xmath220 & @xmath221 + lcrrrrcccc c01 & 4 & 51.7 & 387.2 & 09:22:13.556 & + 50:56:35.13 & @xmath222 & @xmath223 & @xmath224 & + c02 & 3 & 263.2 & 194.9 & 09:22:07.430 & + 50:56:43.10 & @xmath225 & @xmath226 & @xmath227 & + c03 & 4 & 660.6 & 112.5 & 09:22:10.055 & + 50:57:32.45 & @xmath228 & @xmath229 & @xmath230 & + c04 & 4 & 284.2 & 325.8 & 09:22:12.669 & + 50:56:57.42 & @xmath231 & @xmath232 & @xmath233 & + c05 & 3 & 52.3 & 355.3 & 09:22:09.441 & + 50:57:01.23 & @xmath234 & @xmath235 & @xmath236 & + c06 & 2 & 111.0 & 118.1 & 09:22:09.112 & + 50:56:23.87 & @xmath237 & @xmath238 & @xmath239 & * + c07 & 3 & 438.1 & 165.2 & 09:22:05.636 & + 50:56:38.26 & @xmath240 & @xmath226 & @xmath241 & + c08 & 2 & 129.0 & 102.3 & 09:22:09.298 & + 50:56:22.28 & @xmath242 & @xmath243 & @xmath244 & + c09 & 4 & 378.8 & 311.1 & 09:22:12.416 & + 50:57:06.63 & @xmath245 & @xmath226 & @xmath246 & + c10 & 3 & 762.2 & 719.8 & 09:22:01.632 & + 50:57:29.27 & @xmath247 & @xmath248 & @xmath249 & * + c11 & 3 & 675.6 & 715.4 & 09:22:02.530 & + 50:57:29.84 & @xmath250 & @xmath251 & @xmath252 & * + c12 & 4 & 437.0 & 292.6 & 09:22:12.161 & + 50:57:12.21 & @xmath253 & @xmath254 & @xmath255 & * + c13 & 1 & 382.6 & 145.9 & 09:22:11.681 & + 50:56:28.38 & @xmath256 & @xmath257 & @xmath258 & * + c14 & 4 & 59.4 & 290.3 & 09:22:12.536 & + 50:56:34.94 & @xmath259 & @xmath260 & @xmath261 & * + c15 & 4 & 251.0 & 293.2 & 09:22:12.363 & + 50:56:53.81 & @xmath262 & @xmath263 & @xmath264 & * + c16 & 3 & 666.2 & 528.2 & 09:22:02.836 & + 50:57:11.58 & @xmath265 & @xmath266 & @xmath267 & * + c17 & 4 & 626.0 & 338.8 & 09:22:12.449 & + 50:57:31.34 & @xmath268 & @xmath269 & @xmath270 & * + c18 & 3 & 328.2 & 66.4 & 09:22:06.896 & + 50:56:29.77 & @xmath271 & @xmath272 & @xmath273 & * + c19 & 4 & 334.8 & 66.9 & 09:22:09.914 & + 50:56:59.89 & @xmath274 & @xmath275 & @xmath276 & * + c20 & 2 & 253.8 & 75.2 & 09:22:09.735 & + 50:56:10.33 & @xmath277 & @xmath266 & @xmath278 & * + c21 & 4 & 551.9 & 191.4 & 09:22:10.985 & + 50:57:22.56 & @xmath279 & @xmath280 & @xmath281 & * + c22 & 1 & 689.7 & 435.7 & 09:22:13.295 & + 50:56:16.80 & @xmath282 & @xmath152 & @xmath283 & * + c23 & 3 & 487.1 & 227.6 & 09:22:05.051 & + 50:56:43.88 & @xmath284 & @xmath285 & @xmath149 & * + c24 & 4 & 88.9 & 281.4 & 09:22:12.412 & + 50:56:37.74 & @xmath286 & @xmath121 & @xmath138 & * + c25 & 2 & 291.2 & 93.9 & 09:22:09.587 & + 50:56:06.44 & @xmath287 & @xmath288 & @xmath289 & * + c26 & 4 & 532.1 & 311.9 & 09:22:12.264 & + 50:57:21.81 & @xmath290 & @xmath180 & @xmath207 & * + v01 & 2 & 218.2 & 384.2 & 09:22:06.469 & + 50:56:10.35 & @xmath291 & @xmath292 & @xmath293 & + v02 & 3 & 349.3 & 157.6 & 09:22:06.573 & + 50:56:38.48 & @xmath291 & @xmath294 & @xmath295 & + v03 & 1 & 460.2 & 188.6 & 09:22:12.073 & + 50:56:26.84 & @xmath291 & @xmath296 & @xmath297 & + lccccc & c01 & c02 & c03 & c04 & c05 + 603.36 & @xmath298 & @xmath299 & @xmath300 & @xmath301 & @xmath112 + 603.38 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & @xmath306 + 605.30 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 + 605.32 & @xmath312 & @xmath313 & @xmath314 & @xmath315 & @xmath316 + 607.25 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 + 607.26 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 + 609.67 & @xmath327 & @xmath328 & @xmath329 & @xmath330 & @xmath331 + 609.68 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 + 612.55 & @xmath337 & @xmath317 & @xmath338 & @xmath317 & @xmath339 + 612.56 & @xmath340 & @xmath341 & @xmath342 & @xmath343 & @xmath344 + 615.50 & @xmath345 & @xmath346 & @xmath347 & @xmath348 & @xmath349 + 615.51 & @xmath350 & @xmath317 & @xmath351 & @xmath352 & @xmath353 + 619.39 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & @xmath358 + 619.40 & @xmath359 & @xmath360 & @xmath361 & @xmath362 & @xmath363 + 623.61 & @xmath364 & @xmath365 & @xmath366 & @xmath367 & @xmath368 + 623.62 & @xmath369 & @xmath370 & @xmath371 & @xmath372 & @xmath373 + 629.31 & @xmath374 & @xmath317 & @xmath375 & @xmath376 & @xmath377 + 629.32 & @xmath378 & @xmath379 & @xmath380 & @xmath381 & @xmath382 + 636.28 & @xmath383 & @xmath384 & @xmath385 & @xmath386 & @xmath387 + 636.29 & @xmath388 & @xmath317 & @xmath389 & @xmath390 & @xmath391 + 643.25 & @xmath392 & @xmath393 & @xmath394 & @xmath395 & @xmath396 + 643.27 & @xmath397 & @xmath398 & @xmath399 & @xmath400 & @xmath401 + 653.57 & @xmath402 & @xmath403 & @xmath404 & @xmath405 & @xmath406 + 653.59 & @xmath407 & @xmath408 & @xmath409 & @xmath410 & @xmath411 + & c06 & c07 & c08 & c09 & c10 + 603.36 & @xmath412 & @xmath413 & @xmath414 & @xmath317 & @xmath415 + 603.38 & @xmath416 & @xmath417 & @xmath418 & @xmath419 & @xmath420 + 605.30 & @xmath421 & @xmath422 & @xmath423 & @xmath424 & @xmath425 + 605.32 & @xmath426 & @xmath427 & @xmath428 & @xmath429 & @xmath430 + 607.25 & @xmath431 & @xmath432 & @xmath433 & @xmath434 & @xmath435 + 607.26 & @xmath436 & @xmath413 & @xmath437 & @xmath438 & @xmath439 + 609.67 & @xmath440 & @xmath441 & @xmath442 & @xmath443 & @xmath444 + 609.68 & @xmath445 & @xmath446 & @xmath317 & @xmath447 & @xmath448 + 612.55 & @xmath449 & @xmath450 & @xmath451 & @xmath452 & @xmath453 + 612.56 & @xmath454 & @xmath455 & @xmath456 & @xmath457 & @xmath458 + 615.50 & @xmath459 & @xmath460 & @xmath461 & @xmath462 & @xmath314 + 615.51 & @xmath463 & @xmath464 & @xmath465 & @xmath466 & @xmath467 + 619.39 & @xmath468 & @xmath469 & @xmath470 & @xmath471 & @xmath472 + 619.40 & @xmath473 & @xmath474 & @xmath475 & @xmath476 & @xmath477 + 623.61 & @xmath478 & @xmath479 & @xmath480 & @xmath481 & @xmath482 + 623.62 & @xmath483 & @xmath484 & @xmath485 & @xmath486 & @xmath487 + 629.31 & @xmath488 & @xmath489 & @xmath490 & @xmath491 & @xmath492 + 629.32 & @xmath493 & @xmath494 & @xmath495 & @xmath496 & @xmath497 + 636.28 & @xmath498 & @xmath499 & @xmath500 & @xmath501 & @xmath502 + 636.29 & @xmath503 & @xmath504 & @xmath505 & @xmath506 & @xmath507 + 643.25 & @xmath318 & @xmath508 & @xmath509 & @xmath510 & @xmath464 + 643.27 & @xmath511 & @xmath512 & @xmath513 & @xmath484 & @xmath425 + 653.57 & @xmath514 & @xmath515 & @xmath516 & @xmath517 & @xmath473 + 653.59 & @xmath317 & @xmath518 & @xmath317 & @xmath519 & @xmath317 + & c11 & c12 & c13 & c14 & c15 + 603.36 & @xmath317 & @xmath520 & @xmath521 & @xmath522 & @xmath523 + 603.38 & @xmath524 & @xmath525 & @xmath526 & @xmath527 & @xmath528 + 605.30 & @xmath529 & @xmath530 & @xmath531 & @xmath532 & @xmath533 + 605.32 & @xmath317 & @xmath534 & @xmath535 & @xmath536 & @xmath537 + 607.25 & @xmath538 & @xmath539 & @xmath330 & @xmath540 & @xmath541 + 607.26 & @xmath542 & @xmath543 & @xmath544 & @xmath545 & @xmath541 + 609.67 & @xmath546 & @xmath547 & @xmath548 & @xmath549 & @xmath550 + 609.68 & @xmath551 & @xmath552 & @xmath553 & @xmath554 & @xmath555 + 612.55 & @xmath556 & @xmath557 & @xmath558 & @xmath559 & @xmath522 + 612.56 & @xmath560 & @xmath561 & @xmath562 & @xmath563 & @xmath564 + 615.50 & @xmath565 & @xmath566 & @xmath567 & @xmath317 & @xmath568 + 615.51 & @xmath317 & @xmath534 & @xmath569 & @xmath570 & @xmath571 + 619.39 & @xmath572 & @xmath573 & @xmath317 & @xmath574 & @xmath317 + 619.40 & @xmath317 & @xmath575 & @xmath317 & @xmath317 & @xmath576 + 623.61 & @xmath577 & @xmath578 & @xmath579 & @xmath580 & @xmath581 + 623.62 & @xmath582 & @xmath583 & @xmath584 & @xmath585 & @xmath586 + 629.31 & @xmath587 & @xmath588 & @xmath425 & @xmath589 & @xmath590 + 629.32 & @xmath355 & @xmath591 & @xmath592 & @xmath593 & @xmath594 + 636.28 & @xmath595 & @xmath596 & @xmath597 & @xmath598 & @xmath599 + 636.29 & @xmath600 & @xmath601 & @xmath602 & @xmath603 & @xmath604 + 643.25 & @xmath605 & @xmath606 & @xmath607 & @xmath608 & @xmath609 + 643.27 & @xmath610 & @xmath611 & @xmath612 & @xmath613 & @xmath614 + 653.57 & @xmath615 & @xmath616 & @xmath617 & @xmath618 & @xmath619 + 653.59 & @xmath620 & @xmath621 & @xmath622 & @xmath623 & @xmath624 + & c16 & c17 & c18 & c19 & c20 + 603.36 & @xmath625 & @xmath379 & @xmath626 & @xmath627 & @xmath628 + 603.38 & @xmath629 & @xmath630 & @xmath631 & @xmath632 & @xmath633 + 605.30 & @xmath634 & @xmath635 & @xmath416 & @xmath636 & @xmath637 + 605.32 & @xmath638 & @xmath639 & @xmath640 & @xmath641 & @xmath642 + 607.25 & @xmath643 & @xmath644 & @xmath645 & @xmath646 & @xmath647 + 607.26 & @xmath648 & @xmath649 & @xmath650 & @xmath651 & @xmath652 + 609.67 & @xmath653 & @xmath654 & @xmath366 & @xmath317 & @xmath655 + 609.68 & @xmath656 & @xmath657 & @xmath658 & @xmath659 & @xmath536 + 612.55 & @xmath660 & @xmath661 & @xmath392 & @xmath662 & @xmath663 + 612.56 & @xmath664 & @xmath665 & @xmath666 & @xmath667 & @xmath668 + 615.50 & @xmath669 & @xmath317 & @xmath670 & @xmath317 & @xmath671 + 615.51 & @xmath672 & @xmath673 & @xmath674 & @xmath675 & @xmath667 + 619.39 & @xmath676 & @xmath677 & @xmath678 & @xmath679 & @xmath680 + 619.40 & @xmath681 & @xmath682 & @xmath683 & @xmath684 & @xmath685 + 623.61 & @xmath385 & @xmath686 & @xmath687 & @xmath688 & @xmath456 + 623.62 & @xmath689 & @xmath317 & @xmath690 & @xmath691 & @xmath692 + 629.31 & @xmath693 & @xmath694 & @xmath695 & @xmath692 & @xmath696 + 629.32 & @xmath697 & @xmath698 & @xmath699 & @xmath356 & @xmath317 + 636.28 & @xmath700 & @xmath653 & @xmath701 & @xmath317 & @xmath404 + 636.29 & @xmath702 & @xmath703 & @xmath704 & @xmath686 & @xmath705 + 643.25 & @xmath317 & @xmath389 & @xmath706 & @xmath707 & @xmath708 + 643.27 & @xmath317 & @xmath709 & @xmath710 & @xmath711 & @xmath478 + 653.57 & @xmath712 & @xmath713 & @xmath714 & @xmath715 & @xmath716 + 653.59 & @xmath717 & @xmath718 & @xmath317 & @xmath719 & @xmath317 + & c21 & c22 & c23 & c24 & c25 + 603.36 & @xmath720 & @xmath721 & @xmath722 & @xmath723 & @xmath724 + 603.38 & @xmath725 & @xmath726 & @xmath727 & @xmath728 & @xmath729 + 605.30 & @xmath730 & @xmath731 & @xmath732 & @xmath733 & @xmath734 + 605.32 & @xmath735 & @xmath736 & @xmath317 & @xmath737 & @xmath738 + 607.25 & @xmath739 & @xmath153 & @xmath740 & @xmath741 & @xmath742 + 607.26 & @xmath743 & @xmath744 & @xmath745 & @xmath746 & @xmath747 + 609.67 & @xmath748 & @xmath552 & @xmath749 & @xmath750 & @xmath317 + 609.68 & @xmath751 & @xmath414 & @xmath317 & @xmath746 & @xmath752 + 612.55 & @xmath753 & @xmath754 & @xmath755 & @xmath504 & @xmath756 + 612.56 & @xmath757 & @xmath758 & @xmath759 & @xmath760 & @xmath498 + 615.50 & @xmath761 & @xmath502 & @xmath762 & @xmath763 & @xmath764 + 615.51 & @xmath765 & @xmath766 & @xmath767 & @xmath768 & @xmath414 + 619.39 & @xmath317 & @xmath317 & @xmath769 & @xmath770 & @xmath771 + 619.40 & @xmath678 & @xmath350 & @xmath498 & @xmath772 & @xmath773 + 623.61 & @xmath774 & @xmath775 & @xmath776 & @xmath777 & @xmath778 + 623.62 & @xmath779 & @xmath780 & @xmath781 & @xmath782 & @xmath783 + 629.31 & @xmath784 & @xmath317 & @xmath785 & @xmath786 & @xmath730 + 629.32 & @xmath787 & @xmath636 & @xmath788 & @xmath789 & @xmath736 + 636.28 & @xmath790 & @xmath791 & @xmath792 & @xmath624 & @xmath793 + 636.29 & @xmath794 & @xmath795 & @xmath317 & @xmath796 & @xmath797 + 643.25 & @xmath798 & @xmath799 & @xmath800 & @xmath801 & @xmath802 + 643.27 & @xmath803 & @xmath720 & @xmath804 & @xmath805 & @xmath806 + 653.57 & @xmath807 & @xmath808 & @xmath317 & @xmath809 & @xmath810 + 653.59 & @xmath317 & @xmath811 & @xmath812 & @xmath813 & @xmath317 + & c26 & v01 & v02 & v03 & + 603.36 & @xmath814 & @xmath815 & @xmath816 & @xmath817 & + 603.38 & @xmath317 & @xmath818 & @xmath819 & @xmath820 & + 605.30 & @xmath821 & @xmath822 & @xmath823 & @xmath824 & + 605.32 & @xmath825 & @xmath822 & @xmath803 & @xmath826 & + 607.25 & @xmath317 & @xmath317 & @xmath827 & @xmath828 & + 607.26 & @xmath829 & @xmath830 & @xmath831 & @xmath493 & + 609.67 & @xmath832 & @xmath833 & @xmath834 & @xmath835 & + 609.68 & @xmath797 & @xmath264 & @xmath836 & @xmath547 & + 612.55 & @xmath294 & @xmath837 & @xmath838 & @xmath839 & + 612.56 & @xmath840 & @xmath317 & @xmath841 & @xmath842 & + 615.50 & @xmath843 & @xmath844 & @xmath845 & @xmath846 & + 615.51 & @xmath847 & @xmath848 & @xmath849 & @xmath850 & + 619.39 & @xmath292 & @xmath851 & @xmath852 & @xmath317 & + 619.40 & @xmath853 & @xmath155 & @xmath854 & @xmath855 & + 623.61 & @xmath678 & @xmath856 & @xmath857 & @xmath858 & + 623.62 & @xmath859 & @xmath860 & @xmath861 & @xmath317 & + 629.31 & @xmath862 & @xmath863 & @xmath864 & @xmath865 & + 629.32 & @xmath866 & @xmath867 & @xmath317 & @xmath387 & + 636.28 & @xmath868 & @xmath869 & @xmath230 & @xmath509 & + 636.29 & @xmath870 & @xmath871 & @xmath872 & @xmath858 & + 643.25 & @xmath873 & @xmath874 & @xmath616 & @xmath875 & + 643.27 & @xmath876 & @xmath877 & @xmath878 & @xmath879 & + 653.57 & @xmath818 & @xmath880 & @xmath881 & @xmath882 & + 653.59 & @xmath317 & @xmath317 & @xmath883 & @xmath884 & + lccccc & c01 & c02 & c03 & c04 & c05 + 603.42 & @xmath885 & @xmath886 & @xmath887 & @xmath888 & @xmath889 + 603.43 & @xmath890 & @xmath891 & @xmath892 & @xmath893 & @xmath894 + 609.73 & @xmath895 & @xmath896 & @xmath897 & @xmath898 & @xmath899 + 609.74 & @xmath900 & @xmath901 & @xmath902 & @xmath903 & @xmath904 + 619.45 & @xmath905 & @xmath906 & @xmath907 & @xmath908 & @xmath909 + 619.47 & @xmath317 & @xmath317 & @xmath910 & @xmath911 & @xmath912 + 629.37 & @xmath913 & @xmath914 & @xmath915 & @xmath916 & @xmath917 + 629.39 & @xmath918 & @xmath919 & @xmath317 & @xmath920 & @xmath921 + 643.32 & @xmath922 & @xmath923 & @xmath924 & @xmath925 & @xmath926 + 643.33 & @xmath927 & @xmath928 & @xmath929 & @xmath930 & @xmath931 + & c06 & c07 & c08 & c09 & c10 + 603.42 & @xmath932 & @xmath933 & @xmath934 & @xmath935 & @xmath936 + 603.43 & @xmath937 & @xmath317 & @xmath938 & @xmath939 & @xmath940 + 609.73 & @xmath941 & @xmath942 & @xmath943 & @xmath944 & @xmath945 + 609.74 & @xmath946 & @xmath947 & @xmath948 & @xmath949 & @xmath950 + 619.45 & @xmath951 & @xmath952 & @xmath953 & @xmath954 & @xmath955 + 619.47 & @xmath956 & @xmath957 & @xmath958 & @xmath959 & @xmath960 + 629.37 & @xmath961 & @xmath962 & @xmath963 & @xmath964 & @xmath965 + 629.39 & @xmath966 & @xmath967 & @xmath968 & @xmath969 & @xmath970 + 643.32 & @xmath971 & @xmath972 & @xmath973 & @xmath974 & @xmath975 + 643.33 & @xmath976 & @xmath977 & @xmath978 & @xmath979 & @xmath317 + & c11 & c12 & c13 & c14 & c15 + 603.42 & @xmath317 & @xmath980 & @xmath981 & @xmath982 & @xmath919 + 603.43 & @xmath317 & @xmath983 & @xmath984 & @xmath985 & @xmath986 + 609.73 & @xmath987 & @xmath317 & @xmath988 & @xmath989 & @xmath966 + 609.74 & @xmath990 & @xmath991 & @xmath992 & @xmath993 & @xmath994 + 619.45 & @xmath317 & @xmath995 & @xmath317 & @xmath317 & @xmath996 + 619.47 & @xmath317 & @xmath997 & @xmath317 & @xmath317 & @xmath998 + 629.37 & @xmath999 & @xmath1000 & @xmath1001 & @xmath317 & @xmath1002 + 629.39 & @xmath1003 & @xmath1004 & @xmath317 & @xmath1005 & @xmath1006 + 643.32 & @xmath1007 & @xmath1008 & @xmath317 & @xmath1009 & @xmath895 + 643.33 & @xmath1010 & @xmath1011 & @xmath317 & @xmath317 & @xmath1012 + & c16 & c17 & c18 & c19 & c20 + 603.42 & @xmath1013 & @xmath1014 & @xmath1015 & @xmath1016 & @xmath1017 + 603.43 & @xmath1018 & @xmath1019 & @xmath1020 & @xmath957 & @xmath1021 + 609.73 & @xmath1022 & @xmath1023 & @xmath1024 & @xmath1025 & @xmath1026 + 609.74 & @xmath1027 & @xmath1028 & @xmath1029 & @xmath1030 & @xmath587 + 619.45 & @xmath1031 & @xmath1032 & @xmath1033 & @xmath1034 & @xmath1035 + 619.47 & @xmath1036 & @xmath1037 & @xmath1038 & @xmath1039 & @xmath317 + 629.37 & @xmath317 & @xmath1040 & @xmath1041 & @xmath1042 & @xmath1043 + 629.39 & @xmath1044 & @xmath1045 & @xmath1046 & @xmath1047 & @xmath1048 + 643.32 & @xmath1049 & @xmath1050 & @xmath1051 & @xmath1052 & @xmath317 + 643.33 & @xmath1053 & @xmath1054 & @xmath994 & @xmath1055 & @xmath411 + & c21 & c22 & c23 & c24 & c25 + 603.42 & @xmath317 & @xmath1056 & @xmath1057 & @xmath1058 & @xmath317 + 603.43 & @xmath1059 & @xmath1060 & @xmath1061 & @xmath1062 & @xmath1063 + 609.73 & @xmath1064 & @xmath1065 & @xmath1066 & @xmath1067 & @xmath1068 + 609.74 & @xmath1069 & @xmath1070 & @xmath1071 & @xmath1072 & @xmath941 + 619.45 & @xmath1073 & @xmath1074 & @xmath1075 & @xmath1076 & @xmath1077 + 619.47 & @xmath1078 & @xmath1079 & @xmath1080 & @xmath901 & @xmath1081 + 629.37 & @xmath1082 & @xmath1083 & @xmath958 & @xmath1084 & @xmath1085 + 629.39 & @xmath1086 & @xmath1087 & @xmath1088 & @xmath1089 & @xmath1090 + 643.32 & @xmath1091 & @xmath317 & @xmath1092 & @xmath1093 & @xmath1094 + 643.33 & @xmath1095 & @xmath1096 & @xmath1097 & @xmath1098 & @xmath1099 + & c26 & v01 & v02 & v03 & + 603.42 & @xmath1100 & @xmath1101 & @xmath1102 & @xmath1103 & + 603.43 & @xmath1104 & @xmath1105 & @xmath295 & @xmath1106 & + 609.73 & @xmath1107 & @xmath1108 & @xmath1109 & @xmath1110 & + 609.74 & @xmath1111 & @xmath1112 & @xmath1113 & @xmath1114 & + 619.45 & @xmath1115 & @xmath1116 & @xmath1117 & @xmath1118 & + 619.47 & @xmath1119 & @xmath1120 & @xmath1121 & @xmath1122 & + 629.37 & @xmath1123 & @xmath1124 & @xmath1125 & @xmath1126 & + 629.39 & @xmath1127 & @xmath1121 & @xmath1128 & @xmath1129 & + 643.32 & @xmath1130 & @xmath295 & @xmath1131 & @xmath317 & + 643.33 & @xmath1132 & @xmath1133 & @xmath1134 & @xmath1135 & + lcllcccccc ngc@xmath72541 & 11.2 & 08:14:40.2 & + 49:03:42 & @xmath1139 & 40.5 & -22.9 & 7.8 & 6.7 & -4.4 + * ngc@xmath72841 * & 14.1 & 09:22:02.7 & + 50:58:36 & @xmath94 & 49.6 & -16.0 & 8.7 & 10.3 & -3.9 + ngc@xmath73198 & 13.8 & 10:19:54.9 & + 45:32:59 & @xmath1140 & 60.6 & -13.3 & 6.6 & 11.7 & -3.2 + ngc@xmath73319 & 13.3 & 10:39:08.8 & + 41:41:16 & @xmath1141 & 66.0 & -12.8 & 5.3 & 11.8 & -2.9 + ngc@xmath73351 & 10.0 & 10:43:57.8 & + 11:42:14 & @xmath1142 & 94.1 & -27.1 & -0.6 & 8.9 & -4.6 + ngc@xmath73368 & 10.5 & 10:46:45.7 & + 11:49:12 & @xmath1143 & 94.3 & -26.4 & -0.7 & 9.4 & -4.7 + ngc@xmath73621 & 6.6 & 11:18:16.0 & 32:48:42 & @xmath1144 & 145.7 & -28.5 & -4.8 & 3.3 & -3.2 + ngc@xmath73627 & 10.1 & 11:20;15.0 & + 12:59:30 & @xmath1145 & 96.6 & -18.4 & -1.1 & 9.5 & -3.2 +

using kadeishvili s formulas with appropriate signs , we show that the classical cobar construction from coalgebras to algebras can be enhanced to a functor from hopf algebras to algebras ( for a certain choice of operad ) , which , unlike its classical counterpart , is not strictly adjoint , but homotopically equivalent to the left adjoint of the enhanced bar construction studied by gerstenhaber voronov and fresse .

1309.2820

the purpose of this article is to prove a precise homotopical relationship between the category of @xmath4 algebras and the category of hopf algebras . by @xmath4 we mean the @xmath1 suboperad of the @xmath5 operad of mcclure smith @xcite , called the sequence operad , or the surjection operad . classical results from @xcite ( theorems 3.6 , 4.4 , and 4.5 ) show that the cobar and bar constructions define an adjunction @xmath6 between coalgebras and algebras such that the unit and counit maps are homology equivalences . voronov @xcite ( section 3.2 ) , showed that the bar construction induces a functor @xmath7 . this has been generalized by fresse @xcite ( theorem 5.d , see also @xcite for specific formulas mod @xmath8 ) who showed that the bar construction induces functors @xmath9 , where @xmath10 is the barratt eccles @xmath11 operad studied by berger fresse @xcite , which has the property that @xmath12 is the associative operad ( thus the usual bar construction is a special case ) , and there are equivalences @xmath13 compatible with the respective inclusions @xmath14 and @xmath15 . in the other direction , kadeishvili @xcite showed that the cobar construction induces a functor @xmath16 . this has also been generalized by fresse in unpublished work showing that the cobar construction induces a functor @xmath17 , which for @xmath18 reduces via the map @xmath19 to kadeishvili s functor . from all of this a natural question arises : to what extent can one recover the @xmath20 algebra @xmath21 from @xmath22 ? @xmath23 and @xmath21 are equivalent as @xmath24 algebras . the conjecture is a classical theorem for @xmath25 , and remains open for @xmath26 . the main result of this paper is a proof of this conjecture in the case @xmath18 . in this case the @xmath27 structure reduces via the map @xmath19 to an @xmath4 structure and so the statement is as follows . ( this theorem appears in the main text as corollary [ equivthm ] . ) [ introthm ] if @xmath21 is an @xmath4 algebra , then @xmath28 and @xmath21 are equivialent as @xmath4 algebras . instead of proving the theorem directly , we study the strict left adjoint of @xmath29 , denoted @xmath30 , and prove that there is a natural equivalence of functors @xmath31 . as we will show in section [ tildecobar ] , the theorem then follows from classical results . the theorem above of theoretical interest , as there does not exist a model category structure on @xmath32 , but according to mandell @xcite(section 13 , see also fresse @xcite ( part iii section 12 ) ) @xmath4 algebras do possess a _ semi_-model structure . thus , in the language of chachlski scherer @xcite ( chapter i section 5 ) , we have a right semi - model approximation ( a generalization of a quillen equivalence ) @xmath33 and therefore , there are mapping spaces , homotopy colimits , etc , in the category of hopf algebras . the original purpose of this work was to enable the study of @xmath4 algebras @xmath21 in terms of relatively simpler hopf algebras @xmath22 . mandell s theorem @xcite ( called the main theorem in the paper ) says that the @xmath34-algebra structure on the cochains @xmath35 of a nice space @xmath36 is enough to determine the homotopy type of @xmath36 . since @xmath34 is filtered by suboperads @xmath37 one can naturally ask : what homotopy information about @xmath36 remains if we just consider these `` simpler '' structures ? our eventual goal is to prove that for spaces that are sufficiently connected and relatively low dimensional , @xmath35 is equivalent as an @xmath4-algebra to a strictly commutative algebra . the idea is to study instead the hopf algebra @xmath38 and show that it is equivalent as a hopf algebra to a strictly commutative hopf algebra . this connects to work of anick @xcite , who showed that for certain @xmath39 algebras @xmath21 , @xmath22 is equivalent in a weak sense to a commutative hopf algebra of the form @xmath40 , the dual of the universal enveloping algebra of a lie algebra . we prove in the final section that @xmath41 is equivalent as an @xmath4 algebra to @xmath42 , the chevalley eilenberg cochain complex . this shows that if @xmath22 is equivalent as a hopf algebra to some @xmath40 , then @xmath21 is equivalent as an @xmath4 algebra to a commutative algebra . we will show in a future paper that for spaces @xmath36 as above , @xmath38 is equivalent as a hopf algebra to @xmath43 , thus @xmath35 is equivalent as an @xmath4-algebra to @xmath44 which is a strictly commutative algebra . we now briefly outline the structure of the paper . we enhance the classical bar and cobar constructions to functors @xmath7 , and @xmath16 , respectively , in sections [ barhopf ] and [ cobar ] . we then study the strict left adjoint of @xmath29 , called the @xmath4 cobar construction , @xmath45 and show that there is a natural equivalence of functors @xmath31 ; this occupies most of section [ tildecobar ] . finally , in section [ chevalley ] , we study the @xmath4 algebra @xmath41 . i would like to thank benot fresse for sharing his work on the cobar construction during a visit in lille , and for many helpful comments and references . i would also to thank kathryn hess for many helpful suggestions on the writing and organization of the paper . finally , i would like to thank mike mandell for providing the original problem and references that gave rise to this work , guiding its development as my advisor , and for many helpful suggestions both mathmatical and stylistic . in this section we establish conventions and sketch definitions for the notions used in this paper . let @xmath46 denote a commutative ring . our convention is that chain complexes @xmath47 will have differential of degree @xmath48 . sometimes , we use upper grading " @xmath49 which should be interpreted as notation for @xmath50 . we also recall the standard symmetric monoidal structure on the category of chain complexes . in particular , the tensor product @xmath51 , and the internal hom @xmath52 . we use @xmath53 to denote the linear dual chain complex @xmath54 . given a chain complex @xmath47 we define its suspension @xmath55 to be the chain complex @xmath56\otimes c$ ] , where @xmath56 $ ] is the chain complex given by @xmath46 concentrated in degree @xmath57 . similarly we define the desuspension of a chain complex @xmath47 to be @xmath58\otimes c$ ] . the signs are determined by the koszul rule . we also define a map of chain complexes @xmath59 to be a _ weak equivalence _ if it induces an isomorphism on homology . we use @xmath60 and @xmath61 for simplicial ( or singular ) chains and cochains , respectively , with coefficients in @xmath46 . if the coefficients are @xmath62 , we use @xmath63 and @xmath35 . let @xmath21 be a chain complex . we say that @xmath21 is an _ associative algebra _ provided it is equipped with maps of chain complexes @xmath64 , @xmath65 , and @xmath66 called multiplication , unit and augmentation respectively . the multiplication map is associative , the unit map satisfies unit conditions , and the augmentation is a homomorphism of the multiplication and unit maps from @xmath21 to @xmath46 . for brevity , we will usually use the term _ algebra _ for associative algebras , and we will specify if we mean some other kind of algebra . we can write the condition that @xmath67 is a map of chain complexes element - wise : for @xmath68 we have @xmath69 , this is called the _ leibniz rule_. a map of algebras @xmath70 is a map of chain complexes that preserves the multiplication , unit and augmentation maps . there is an obvious dual to the structure of an algebra that is important in this work . let @xmath47 be a chain complex . we say that @xmath47 is a _ coassociative coalgebra _ provided that it is equipped with maps of chain complexes @xmath71 , @xmath72 , and @xmath73 , called comultiplication , counit , and coaugmentation , respectively . the comultiplication map is required to be coassociative , the counit map satisfies counit conditions , and the augmentation is a homomorphism of the comultiplication and counit maps from @xmath47 to @xmath46 . a map of coalgebras is a map of chain complexes that preserves the comultiplication , counit and coaugmentation maps . again , we will usually use the term _ coalgebra _ for coassociative coalgebras . a coalgebra @xmath47 is called _ conilpotent _ if , for all @xmath74 , there exists @xmath75 such that for @xmath76 , @xmath77 . we make the following assumption throughout the paper . all coalgebras are conilpotent . finally , if we combine an algebra and a coalgebra structure coherently , we obtain the structure of a hopf algebra . a _ hopf algebra _ @xmath78 is simultaneously a ( coassociative ) coalgebra and an ( associative ) algebra such that the multiplication map @xmath79 is a map of coalgebras , or equivalently , the comultiplication map @xmath80 is a map of algebras . the corresponding commutative diagram is as follows . @xmath81^m \ar[d]_{\delta\otimes \delta } & & h \ar[dd]^\delta\\ h\otimes h\otimes h \otimes h \ar[d]_{1\otimes \tau \otimes 1 } & & \\ h\otimes h\otimes h\otimes h \ar[rr]_{m\otimes m}&&h\otimes h}\ ] ] where @xmath82 is the twist map . note that the counit is an augmentation of the underlying algebra of @xmath78 , and the unit is a coaugmentation of the underlying coalgebra . a map of hopf algebras is simultaneously a map of algebras and coalgebras . for an algebra @xmath21 , with augmentation @xmath83 , we define @xmath84 and call this the _ augmentation ideal_. for a coalgebra @xmath47 , with coaugmentation @xmath85 , we define @xmath86 and call this the _ coaugmentation coideal_. we find that @xmath47 splits as a chain complex , @xmath87 . thus , we can define the reduced diagonal @xmath88 by the composition @xmath89^\delta & c\otimes c \ar[r ] & { \overline}c \otimes { \overline}c } \ ] ] we will sometimes view @xmath90 as a submodule of @xmath91 , which will be clear from the context . we will systematically use sumless notation in this paper . in particular , the ordinary diagonal will be denoted @xmath92 , and the reduced diagonal will be denoted @xmath93 . a _ symmetric sequence of chain complexes _ @xmath94 is a collection of chain complexes @xmath95 , @xmath96 , such that each @xmath95 is equipped with a @xmath97 action , where @xmath97 denotes the symmetric group on @xmath98 letters . a map of symmetric sequences of chain complexes is a collection of equivariant maps on each component complex . an _ operad _ is a @xmath99 complex @xmath100 together with composition maps @xmath101 @xmath102and a unit map @xmath103 satisfying appropriate associativity , equivariance , and unit conditions . we also use the notation @xmath104 for @xmath105 where @xmath106 , and @xmath107 is in the @xmath98th spot . a map of operads is a map of symmetric sequences of chain complexes that preserves the composition and unit maps . we will insist for the purposes of this work that the unit map is an isomorphism , and also that @xmath108 . let @xmath100 be an operad . a chain complex @xmath21 is called a @xmath100-algebra if it is equipped with evaluation maps @xmath109 satisfying associativity , equivariance and unit conditions . a map of @xmath100-algebras is a map of chain complexes that preserves the evaluation maps . a subcomplex @xmath110 of a @xmath100-algebra is called an _ ideal _ if @xmath111 whenever some @xmath112 . in this case the quotient @xmath100-algebra is well - defined and denoted by @xmath113 . a chain complex @xmath47 is called a @xmath100-coalgebra if it is equipped with coevaluation maps @xmath114 satisfying the appropriate dual conditions . a map of @xmath100-coalgebras is a map of chain complexes that preserves the coevaluation maps . define the tensor product of symmetric sequences of chain complexes @xmath94 and @xmath115 , by @xmath116 . if @xmath100 and @xmath117 are operads , then @xmath118 inherits the structure of an operad . an operad @xmath119 is called a _ hopf operad _ if it is equipped with a coassociative map of operads @xmath120 . if @xmath21 is an @xmath119-algebra , then @xmath121 inherits the structure of an @xmath119-algebra . if @xmath21 is also a ( coassociative ) coalgebra , then we say that @xmath21 is an _ @xmath119 hopf algebra _ provided that the comultiplication @xmath122 is a map of @xmath119 algebras . similarly , if @xmath47 is an @xmath119-coalgebra , then @xmath91 inherits the structure of an @xmath119-coalgebra . suppose that @xmath47 is also an algebra . we say that @xmath47 is an _ @xmath119 hopf coalgebra _ provided that the multiplication map @xmath123 is a map of @xmath119-coalgebras . the associative operad @xmath124 given by @xmath125 $ ] is a hopf operad whose algebras are associative algebras , and whose coalgebras are coassociative coalgebras . furthermore , an @xmath124 hopf algebra ( or @xmath124 hopf coalgebra ) is exactly a hopf algebra . the commutative operad @xmath126 given by @xmath127 with trivial @xmath97 action , is a hopf operad whose algebras are commutative algebras , and whose coalgebras are cocommutative coalgebras . since we insist that all operads @xmath100 satisfy @xmath108 and @xmath128 , we find that there is a canonical map @xmath129 of operads . therefore , @xmath46 itself is a @xmath100 algebra and a @xmath100 coalgebra for any operad @xmath100 . thus , it makes sense to speak of augmentations @xmath130 ( coaugmentations @xmath131 ) for @xmath100-algebras ( @xmath100-coalgebras ) @xmath36 . we will assume in general that all @xmath100-algebras are augmented , and all @xmath100-coalgebras are coaugmented . this is equivalent to working nonunitally , however for many purposes it is more convenient to have the unit or counit available . for clarity , we state this assumption separately . all @xmath100-algebras are augmented and all @xmath100-coalgebras are coaugmented . in particular , ( co)associative ( co)algebras are ( co)augmented . let @xmath100 be an operad , and let @xmath36 be a chain complex . we define a @xmath100 algebra @xmath132 as follows . as a complex , @xmath133 , with operad action given by @xmath134 we can interpret this as giving @xmath135 the structure of a monad on the category of chain complexes , and the algebras over @xmath135 are exactly the same as the @xmath100 algebras . note that because of our assumptions on @xmath100 , i.e. @xmath108 , @xmath132 has a natural unit and augmentation . note further that @xmath136 is a free @xmath100 algebra on the generating chain complex @xmath36 . an operad @xmath100 is called an _ @xmath5 operad _ if the map @xmath129 is a weak equivalence , and each @xmath137 is @xmath97 projective and concentrated in nonnegative degrees . thus , @xmath126 itself is not an @xmath5 operad , if the ring @xmath46 does not have enough inverses . an operad @xmath100 is called an _ @xmath20 operad _ if each @xmath137 is @xmath97 projective and concentrated in nonnegative degrees , and @xmath100 is weakly equivalent to @xmath138 as an operad , where @xmath139 is the topological operad of little @xmath140-cubes as defined in @xcite . in @xcite , smith constructed an @xmath5 operad @xmath34 called the sequence operad , and showed that it acts naturally on simplicial chains ( as a coalgebra ) , and simplicial cochains ( as an algebra ) . smith also defined a complexity filtration @xmath141 such that @xmath142 is an @xmath20 operad , and @xmath143 is the associative operad . the elements of @xmath144 can be represented by formal linear combinations of non - degenerate surjections @xmath145 , where non - degenerate means that @xmath146 , for all @xmath147 . we will often write elements of this operad as ordered lists of integers , for @xmath148 the corresponding list is denoted @xmath149 . the complexity filtration can be expressed in terms of the number of times any given pair of integers switches order in the list . for example , @xmath150 has complexity @xmath151 and thus lives in @xmath39 . in @xcite , berger fresse constructed another @xmath5 operad @xmath152 called the barratt eccles operad , and an operad map @xmath153 , which is therefore a weak equivalence . the operad @xmath152 is given by simplicial chains on a simplicial operad given in arity @xmath98 by @xmath154 the standard contractible @xmath97-complex . the operad structure comes from the simplicial operad structure on the discrete operad @xmath124 . fresse also define a complexity filtration , coming from a filtration on the simplicial level , @xmath155 such that the map @xmath153 restricts to maps @xmath156 , which are weak equivalences of operads . furthermore , @xmath12 is also the associative operad , so the map @xmath157 is an isomorphism . in addition , since each @xmath24 is given by chains on a simplicial operad , each @xmath24 is a hopf operad . the signs in @xmath152 are thus implicitly determined by the signs in the eilenberg - zilber map @xmath158 , and the signs in the chain differential . since the morphism @xmath153 is surjective , this implicitly determines the operad structure and the signs on @xmath34 . we should note that these signs for @xmath34 differ from the signs given by mcclure smith in @xcite . we use these signs because the signs on @xmath152 are natural , and we would like the @xmath152 and @xmath34 algebra structures to be directly compatible . in @xcite ( section 2.2 ) , the authors give explicit formulas for the action of @xmath34 , with this choice of signs , on chains ( as a coalgebra ) , and cochains ( as an algebra ) . note that we have arranged our conventions so that , for finite - type complexes @xmath36 , @xmath36 is a @xmath100-algebra if and only if @xmath159 is a @xmath100-coalgebra . similarly , for hopf operads @xmath119 , @xmath36 is an @xmath119 hopf algebra if and only if @xmath159 is an @xmath119 hopf coalgebra . we will occasionally switch between these dual points of view without explicitly mentioning it . the purpose of this section is to define and give first properties of the hopf algebra structure on @xmath22 , the bar construction of an @xmath4 algebra @xmath21 . the results of this section are due to gerstenhaber voronov @xcite . let @xmath21 denote an algebra . we define the bar construction as follows : @xmath160 with comultiplication given by the tensor comultiplication , differential defined by @xmath161 ) & = -\sum_{i=1}^k ( -1)^{m_i } [ a_1 | \cdots |da_i|\cdots |a_k ] \\ & + \sum_{i=2}^k ( -1)^{m_i } [ a_1|\cdots|a_{i-1 } a_{i}|\cdots |a_k],\end{aligned}\ ] ] where @xmath162 , and obvious coaugmentation @xmath163 . this gives @xmath22 the structure of a coalgebra . let @xmath47 be a coalgebra and let @xmath21 be an algebra . an @xmath46-linear map @xmath164 of degree @xmath48 is called a twisting morphism if @xmath165 where @xmath64 is the multiplication map on @xmath21 and @xmath166 is the reduced diagonal . it is a theorem of @xcite ( iii , prop 3.5 ) that coalgebra maps @xmath167 are in one to one correspondence with twisting morphisms @xmath168 . in particular , the projection @xmath169 is the twisting morphism corresponding to the identity @xmath170 . given a twisting morphism @xmath168 one constructs a morphism of coalgebras @xmath167 by iterating diagonals . for ease of exposition , we recall that an @xmath4 algebra is an algebra @xmath21 over the operad @xmath4 with signs as given in @xcite ( section 1.2 ) , and briefly outlined in section [ conv ] . we note that since the operad @xmath4 has known generators and relations , we can describe the structure of an @xmath4 algebra as an algebra together with _ brace operations _ given by @xmath171 satisfying appropriate associativity , differential , and mutiplication formulas . we observe that the @xmath4 operation @xmath172 corresponds to the @xmath140-fold product : @xmath173 . the relations are determined by computing @xmath174 @xmath175 and @xmath176 using the explicit formulas given in @xcite . note that these signs differ from those given in @xcite and in @xcite . we will give the associativity identity without explicit signs because , with the exception of @xmath177 , the signs are not critical . [ assoc ] @xmath178 where the sum runs over all sequences @xmath179 . [ diff ] @xmath180 where @xmath181 . [ mult ] @xmath182 where @xmath183 . recall also that we require algebras over operads to be augmented , and thus all brace operations vanish if one of the inputs is the unit . suppose that @xmath21 is a commutative algebra . define @xmath184 for all @xmath185 . then , @xmath21 is an @xmath4 algebra . we now recall a theorem of @xcite that the structure of an @xmath4 algebra on @xmath21 is equivalent to the structure of a hopf algebra on @xmath22 with a left non - decreasing multiplication ( term to be defined below ) . the structure of a hopf algebra on the coalgebra @xmath22 is determined by a map of differential coalgebras @xmath186 such that @xmath187 is associative . to say that @xmath187 is left non - decreasing means that the restriction of @xmath187 to @xmath188 lands in @xmath189 . the structure of an @xmath4 algebra on a is equivalent to an associative unital left non - decreasing map of coalgebras @xmath190 . the multiplication is uniquely determined by the associated twisting morphism @xmath191 . by the left non - decreasing condition , this is uniquely determined by the restrictions @xmath192 . in turn , these uniquely determine operations corresponding to the braces @xmath193 one then needs to verify that the twisting morphism formula is equivalent to identities [ diff ] and [ mult ] on the associated braces , and that the multiplication is associative if and only if identity [ assoc ] holds on the braces . the following is a commutative diagram of functors @xmath194^b\ar[d ] & { \mathit{hopfalg}}\ar[d]\\ { \mathcal{s}_{1}\mathit{alg } } \ar[r]^b & { \mathit{coalg}},}\ ] ] where the vertical functors are the forgetful ones . we will now study @xmath195 as an @xmath4-algebra for what is essentially the smallest nontrivial space : @xmath196 . we will see that already in this case we have interesting structure . let @xmath197 considered as a commutative @xmath4 algebra ( with trivial braces ) . the following example was explained to the author by m. mandell , and is an unpublished result of mandell and n. kuhn . the @xmath4-algebra @xmath198 is not equivalent as an @xmath4 algebra to a commutative algebra , and in particular ( in fact , equivalently ) , @xmath198 is not equivalent to @xmath21 . the hopf algebra @xmath199 is given by numerical polynomials @xmath200 $ ] where @xmath201 has degree zero , i.e. , the polynomials @xmath202 $ ] such that @xmath203 for all @xmath204 , which is generated by @xmath205-linear combinations of the following polynomials : @xmath206 the hopf algebra @xmath22 is given by the free divided power algebra on one generator @xmath207 $ ] given by @xmath205-linear combinations of @xmath208 . the proof of kuhn mandell used sophisticated methods , we will now give a new proof using only straightforward computations . we use the simplicial model @xmath209/\partial \delta[1]$ ] . let @xmath210 denote the dual of the @xmath57-simplex given by the quotient map @xmath211 \to s^1 $ ] . in @xcite , berger fresse ( see @xcite ( fact 3.3.4 ) ) computed explicitly that , for @xmath212 we have @xmath213 unless @xmath214 ( up to permutations ) , in which case @xmath215 . in brace notation , @xmath216 . now consider the commutative algebra @xmath217 , as an @xmath4 algebra ( with trivial braces ) . the algebra @xmath21 is the only commutative algebra up to weak equivalence with cohomology @xmath21 , which is easy to see if the ground ring contains @xmath218 ; a longer argument , which we omit , proves the general case . we can distinguish @xmath21 from @xmath198 as @xmath4 algebras using extra structure given by the steenrod operation @xmath219)$ ] where @xmath220 , which we will now explain . let @xmath36 be an @xmath4 algebra over @xmath205 . let @xmath221 be a cycle with @xmath222 , then define @xmath223 . this defines a function @xmath224 , natural in @xmath36 , which will not necessarily be a homomorphism ( unless , for example , @xmath36 is an @xmath39 algebra ) . if @xmath36 is commutative , then @xmath225 for all @xmath140 . on the other hand , we have computed @xmath226 ) = [ x ] \neq 0 $ ] in @xmath227 . thus , @xmath198 is not equivalent as an @xmath4 algebra to a commutative algebra , or equivalently , it is not formal as an @xmath4 algebra . now , we compute the hopf algebra @xmath199 . we introduce the notation @xmath228 $ ] in the hopf algebra @xmath199 . one can compute : @xmath229,\end{aligned}\ ] ] where the sum is over all sequences @xmath230 , and @xmath231 for all @xmath98 . there are two cases in which we get nonzero terms : when @xmath232 , in which case we get a @xmath233 term , and when @xmath234 , in which case we get a @xmath235 term . by inspecting these cases , we get @xmath140 terms of the first type and @xmath236 terms of the second type . one can compute that the sign is always @xmath237 . one can also compute that @xmath238 , and so : @xmath239 now , if we use the hopf algebra embedding @xmath240 , we see that @xmath241 is a polynomial algebra on a primitive @xmath242 , and that the embedding @xmath243 maps @xmath244 therefore , we see that @xmath245 . here , we use a well - known fact from algebra that the polynomials @xmath246 generate all numerical polynomials over @xmath205 . from this computation it is clear that in @xmath22 we have the same linear structure , with generators denoted by @xmath247 . the multiplicative relations are given by @xmath248 thus , via the embedding @xmath249 , we find that @xmath250 , with @xmath251 . classically , the cobar construction is a functor @xmath0 , which is left adjoint to the bar construction . in @xcite ( p. 36 ) , adams observes that over @xmath252 the cobar construction of a hopf algebra @xmath47 , denoted @xmath253 , can be equipped with a @xmath254 product satisfying @xmath255 in @xcite , kadeishvili proves that this product can be extended to the structure of an @xmath4 algebra on @xmath253 , with @xmath256 . the purpose of this section is to work out appropriate signs to show that @xmath253 , where @xmath47 is a hopf algebra , has the structure of an @xmath4 algebra over any ground ring @xmath46 . we will then characterize maps @xmath257 of hopf algebras , and in particular observe that the unit map of the adjunction @xmath258 is a map of hopf algebras . let @xmath47 denote a coalgebra . we define the cobar construction as follows : @xmath259 with tensor multiplication , differential determined by @xmath260 ) = -[dc ] + ( -1)^{|c^{(1)}| } [ { \overline}c^{(1 ) } | { \overline}c^{(2)}]\ ] ] and the leibniz rule , and the obvious augmentation @xmath261 . this gives @xmath253 the structure of an algebra . restricting a morphism @xmath262 of algebras @xmath263 gives a twisting morphism @xmath168 ( note the degree shift in the inclusion @xmath264 ) . this correspondence is a bijection . thus , the functor @xmath0 is adjoint to the functor @xmath265 . observe that if @xmath47 and @xmath21 are finite type , then @xmath266 and @xmath267 . we now proceed to the goal of this section . we will assume that @xmath47 is a hopf algebra , and construct braces such that @xmath253 is an @xmath4 algebra . the formula for the first brace is : @xmath268\{[y_1 | \cdots | y_q]\ } = ( -1)^\alpha [ x^{(1)}y_1 | \cdots | x^{(q)}y_q]\ ] ] where @xmath269 is the @xmath107th iterated diagonal , and @xmath270|)\right ] + |x| - 1.\ ] ] first , we need to verify identity [ diff ] for the first brace , i.e. , we need to verify the following formula . @xmath271\{[y_1 | \cdots | y_q]\ } ) & = -[x|y_1|\cdots|y_q ] + ( -1)^{|[x]|(\sum |[y_i]|)}[y_1|\cdots|y_q|x ] + \\ & - d[x]\{[y_1|\cdots|y_q]\ } + ( -1)^{|x|}[x]\{d[y_1|\cdots|y_q]\}.\end{aligned}\ ] ] we verify the formula in the case that @xmath272 are all primitive elements with the zero differential . the general case requires more involved bookkeeping , but poses no additional problems by our assumptions , the formula reduces to the following @xmath271\{[y_1 | \cdots | y_q]\ } ) & = -[x|y_1|\cdots|y_q ] + ( -1)^{|[x]|(\sum |[y_i]|)}[y_1|\cdots|y_q|x].\end{aligned}\ ] ] by definition of the first brace , @xmath271\{[y_1 | \cdots | y_q]\ } ) & = ( -1)^\alpha \sum_{i=1}^q ( -1)^{m_i}[x^{(1)}y_1 | \cdots| d([x^{(i)}y_i])| \cdots | x^{(q)}y_q],\\\end{aligned}\ ] ] where @xmath273 is determined by the koszul rule . since @xmath274 is primitive , @xmath268\{[y_1 | \cdots | y_q]\}= ( -1)^{|x|-1 } \sum_{i=1}^q ( -1)^{\alpha_i } [ y_1| \cdots | xy_i | \cdots | y_n],\ ] ] where @xmath275 . thus , since @xmath276 , we have that @xmath271\{[y_1 | \cdots | y_q]\ } ) & = ( -1)^{|x| - 1 } \sum_{i=1}^q ( -1)^{\beta_i}[y_1| \cdots | x | y_i | \cdots | y_n ] + \\&(-1)^{|x| - 1 } \sum_{i=1}^q ( -1)^{\gamma_i}[y_1| \cdots | y_i | x | \cdots | y_n ] , \end{aligned}\ ] ] where @xmath277 and @xmath278 thus , @xmath279 so , we have a telescoping sum with remaining terms as follows , @xmath280 + ( -1)^{\gamma_n + |x| - 1}[y_1| \cdots | y_n | x],\ ] ] and it is easy to verify that the signs are correct . we define @xmath281\{[y_1|\cdots|y_q]\}$ ] using identity [ mult ] . identity [ mult ] is consistent with associativity of the multiplication , i.e. , @xmath282 and so identity [ mult ] will necessarily be satisfied for the first brace . similarly , identity [ mult ] is consistent with identity [ diff ] , i.e. , @xmath283 and so by induction identity [ diff ] is also satisfied in general for the first brace . next , we need to define the higher braces . we start with the initial condition @xmath284\{\beta_1,\ldots,\beta_p\ } = 0 $ ] for @xmath285 . then , we extend using identity [ mult ] . now , the following expression , @xmath286\{\beta_1\}\beta_2 + ( -1)^{|[x]||\beta_1| + |\beta_1|}\beta_1 ( [ x]\{\beta_2\})\\ - [ x]\{\beta_1 \beta_2\ } + d([x])\{\beta_1 , \beta_2\}\end{aligned}\ ] ] is zero because of the definition @xmath287\{\beta_1 , \beta_2\ } = ( -1)^{|[x_2]||\beta_1| + |[x_1]| + |\beta_1| } [ x_1]\{\beta_1\}[x_2]\{\beta_2\}.\ ] ] thus , our initial condition is consistent with identity [ diff ] in the case @xmath288 . the corresponding formulas for @xmath289\{\beta_1,\ldots , \beta_p\})$ ] for @xmath290 are completely trivial . as above , by consistency , it follows that identities [ diff ] and [ mult ] hold in general for all braces . finally , we need to look at identity [ assoc ] . consider the case @xmath268\{[y_1|\cdots|y_l]\}\{\beta_1,\ldots,\beta_n\}\ ] ] where @xmath291 . since the required operations mostly act by zero , we only need to look at one sign . one can compute that @xmath292 @xmath268\{[y_1|\cdots|y_l ] \}\{\beta_1,\ldots,\beta_n\ } = ( -1)^{n+ n|[x]| } [ x]\{[y_1|\cdots|y_l]\{\beta_1 , \ldots , \beta_n\}\}\ ] ] by definition , we have @xmath284\{[y_1|\cdots|y_l]\ } = ( -1)^\alpha [ x^{(1)}y_1 |\cdots |x^{(l)}y_l]$ ] . thus , @xmath268\{[y_1|\cdots|y_l ] \}\{\beta_1,\ldots,\beta_n\ } = \ ] ] @xmath293[x^{(i)}y_i]\{\beta_1\}\cdots [ x^{(i+n)}y_{i+n}]\{\beta_n\}\}[\cdots|x^{(l)}a_l]\ ] ] on the other hand , @xmath294| } [ x]\{[y_1|\cdots|y_l]\{\beta_1 , \ldots , \beta_n\}\}=\ ] ] @xmath295|}\sum_{i=1}^{l - n } ( -1)^{\delta_i}[x]\{[y_1|\cdots][y_i]\{\beta_1\}\cdots [ y_{i+n}]\{\beta_n\}[\cdots|y_l]\}\ ] ] as a consequence of our definitions , we see that we get the same terms , and a trivial but tedious check shows that the signs work out . thus , we have a proof of identity [ assoc ] in this case . since identity [ assoc ] is consistent with identity [ mult ] , by induction , identity [ assoc ] holds in general . this concludes the proof that the braces defined above satisfy identities [ assoc ] , [ diff ] and [ mult ] . we have now established that if @xmath47 is a hopf algebra then @xmath253 is an @xmath4 algebra , and we clearly have a functor @xmath16 . we would now like to study the universal properties of @xmath296 and @xmath29 . suppose that @xmath47 is a hopf algebra , and let @xmath168 be a twisting morphism . to say that the associated coalgebra map @xmath167 is a map of hopf algebras boils down to the commutativity of the diagram : @xmath297\ar[d]&ba\otimes ba \ar[d]\\ c\ar[r ] & ba.}\ ] ] thus , we need to compare the two twisting morphisms @xmath298 coming from the two ways of following the diagram @xmath299 followed by the universal twisting morphism @xmath169 . consider @xmath300 and let @xmath301 if we let @xmath164 denote the original twisting morphism , going around the top leads to the expression @xmath302 where @xmath303 and going around the bottom leads to @xmath304 . this leads to the following characterization of hopf algebra maps @xmath167 . [ hopftwist]let @xmath47 be a hopf algebra and let @xmath21 be an @xmath4 algebra . a twisting morphism @xmath164 is called _ hopf _ if @xmath305 for all @xmath306 , where @xmath307 the canonical bijection between twisting morphisms @xmath168 and morphisms of coalgebras @xmath167 restricts to a bijection between hopf twisting morphisms @xmath168 and morphisms of hopf algebras @xmath167 . we now establish a technical theorem about hopf twisting morphisms that is useful for constructing such morphisms . this will play a key role in section [ tildecobar ] . [ techtwist]let @xmath47 be a hopf algebra and let @xmath21 be an @xmath4 algebra . let @xmath308 be a subcoalgebra with the property that if @xmath309 , then @xmath310 , and @xmath311 . let @xmath312 be a twisting morphism that satisfies the hopf twisting morphism formula ( see definition [ hopftwist ] ) on elements of the form @xmath313 for @xmath309 . let @xmath314 , and define @xmath315 where @xmath316 . then , @xmath317 the following composition induced by @xmath148 @xmath318 is a twisting morphism . let @xmath319 be the submodule generated by the images of @xmath320 and @xmath321 . then , @xmath322 is a subcoalgebra , and the product defines a map @xmath323 . now , we can use this map together with @xmath148 to induce a twisting morphism @xmath324 we see by assumption that on elements of @xmath322 , @xmath325 . thus , we have the following equations : @xmath326 @xmath327 and @xmath328 the conclusion follows from the fact that @xmath329 is a twisting morphism . [ hopftwistdef]let @xmath47 be a connected nonnegatively graded hopf algebra and let @xmath21 be an @xmath4 algebra . let @xmath306 . let @xmath164 be defined on all elements @xmath330 satisfying @xmath331 so that the twisting morphism formula @xmath332 holds , and if @xmath333 with @xmath334 , then @xmath335 if we set @xmath336 then the twisting morphism formula @xmath337 holds . let @xmath338 be the subcoalgebra given by elements in degrees less than @xmath339 , and apply theorem [ techtwist ] . as a direct corollary of this we have the following useful result . let @xmath340 be a hopf algebra whose algebra structure is free on the generators @xmath341 , where @xmath341 is non - negatively graded and @xmath342 for @xmath343 . let @xmath344 be the sub - hopf algebra given by the generators in degrees @xmath345 . assume that @xmath346 is a hopf twisting morphism , and there is an extension to @xmath347 that is a twisting morphism . then , the unique extension @xmath348 given by the formula of lemma [ hopftwistdef ] is a hopf twisting morphism . observe that the extension is unique by identity [ assoc ] , comparing terms , and sign calculations . the importance of this corollary is that it allows one to construct hopf twisting morphisms with source a free hopf algebra ( of the form @xmath340 as an algebra ) inductively on the generators @xmath341 . for example , the hopf algebra could be of the form @xmath349 , this example will be used in a future paper . in a different direction , we have the following variation . [ freehopf ] let @xmath341 be a free graded @xmath46-module with the zero differential , and define @xmath340 to be the hopf algebra with @xmath341 as its primitive elements . let @xmath21 be an @xmath4 algebra . then , any degree @xmath48 @xmath46-linear map @xmath350 extends uniquely to a hopf twisting morphism @xmath351 . by induction on length , theorem [ techtwist ] implies that the unique extension @xmath351 is still a twisting morphism . if @xmath352 then @xmath353 $ ] , and by definition above , we have @xmath354\{[{\overline}c_2^{(1)}],\ldots , [ { \overline}c_2^{(k)}]\ } & = ( -1)^{|c_1| -1}[c_1]\{[c_2]\}\\ & = [ c_1c_2]\end{aligned}\ ] ] and so the universal twisting morphism @xmath264 is a hopf twisting morphism and thus induces a map of hopf algebras @xmath258 . we now have the following generalization of the main result of @xcite to any ground ring @xmath46 . [ unit]the cobar construction induces a functor @xmath16 . furthermore , the natural weak equivalence @xmath355 is a map of hopf algebras . in @xcite , hess parent tonks study @xmath356 , where @xmath357 is a @xmath57-reduced simplicial set , as a coalgebra , and its relationship to @xmath358 , where @xmath359 is the kan loop group on @xmath357 . note that the product @xmath360 induces a hopf algebra structure on @xmath358 , thus , @xmath361 is an @xmath4 algebra . by unwinding the definitions used in @xcite and interpreting them in our language , the authors construct a natural hopf twisting morphism @xmath362 where @xmath357 is a @xmath57-reduced simplicial set , and @xmath359 is the kan loop group , which induces a weak equivalence @xmath363 . this directly implies that there is a weak equivalence of hopf algebras @xmath364 . thus , by interpreting theorem 4.4 of @xcite in our context , and applying theorem [ unit ] , we have the following . [ hessthm]if @xmath357 is a @xmath57-reduced simplicial set , then there is a natural equivalence of hopf algebras @xmath365 . thus , @xmath356 and @xmath358 are naturally equivalent as hopf algebras . back to generalities , it would be nice to complete the duality picture by showing that the natural map of algebras @xmath366 is a map of @xmath4 algebras if @xmath21 is an @xmath4 algebra . however , this is not the case in general as can be seen by considering the higher braces . for example , @xmath367\{[[y]],[[z]]\ } = 0 $ ] in @xmath23 but @xmath368 ) = a$ ] , and in general @xmath369 need not be zero in @xmath21 . essentially , the functor @xmath296 is not free enough as an @xmath4 algebra . we will address this problem in the following section . in this section , we freely use both the sequence notation for elements of @xmath4 introduced above and the brace notation that we have primarily used so far , for example @xmath370 . as is clear by now , @xmath16 is not the left adjoint to @xmath7 ; it is simply not free enough as an @xmath4 algebra to have the appropriate universal property . we now consider the true left adjoint @xmath45 . our definition of hopf twisting morphisms [ hopftwistdef ] makes the construction clear . first , given any hopf algebra @xmath47 , we can define a differential @xmath371 on @xmath372 ( the free @xmath4 algebra on @xmath373 defined in section [ conv ] ) by @xmath374 ) = ( -1)^{|{\overline}c^{(1)}|}[{\overline}c^{(1 ) } | { \overline}c^{(2)}]$ ] and extending by the leibniz rule . it is easy to see that this differential commutes with the natural differential @xmath320 , and thus we have a differential @xmath375 . define @xmath376 to be the quotient of @xmath377 by the ideal @xmath378 generated by @xmath379 - \sum ( -1)^\alpha [ c_1]\{[{\overline}c_2^{(1)}],\ldots , [ { \overline}c_2^{(k)}]\}\ ] ] for all @xmath306 and where @xmath303 . @xmath380 it suffices to show that for a generator @xmath381 - \sum ( -1)^\alpha [ c_1]\{[{\overline}c_2^{(1)}],\ldots , [ { \overline}c_2^{(k)}]\},\ ] ] we have @xmath382 . this is essentially an exercise in manipulating identity [ diff ] and computing signs ; we omit the details . thus , @xmath383 defines a differential on @xmath376 . therefore , a map @xmath384 of @xmath4 algebras is the same thing as a map @xmath385 that extends to a map @xmath386 such that every generator in @xmath378 is sent to zero . this description makes the following proposition clear . the functor @xmath45 is left adjoint to @xmath7 . the map @xmath387 is evidently a hopf twisting morphism , which is in particular a twisting morphism , and we have already seen that the map @xmath264 is a hopf twisting morphism . thus , we obtain maps @xmath388 . the first of these is a map of algebras , and the second is a map of @xmath4 algebras . furthermore , the composition is clearly the identity . the rest of this section is devoted to proving the following theorem and its corollaries . [ barthm]the @xmath4 algebra map @xmath389 is a strong associative algebra deformation retraction , with section @xmath390 . [ adjcor ] the unit map @xmath391 and the counit map @xmath392 are weak equivalences . if @xmath393 is a map of algebras that is a homotopy equivalence of complexes , then using the word length filtration , we see that @xmath394 is a weak equivalence . thus , we apply two - out - of - three to @xmath395 to obtain the first equivalence . the second follows similarly . [ equivthm]@xmath23 is weakly equivalent to @xmath21 as an @xmath4 algebra . we have the following zig - zag of @xmath4 algebra equivalences : @xmath396 assume that @xmath46 is a field . then , the adjunction @xmath397^{\widetilde \omega } & & { \mathcal{s}_{2}\mathit{alg } } \ar@<1ex>[ll]^b\ar@<1ex>[ll]}\ ] ] is a right semi - model approximation in the sense of @xcite . see definition @xmath398 of @xcite for the axioms . the category @xmath32 is a `` category with weak equivalences '' given by the homology isomorphisms . according to mandell @xcite(section 13 , see also fresse @xcite ( part iii section 12 ) ) , @xmath399 is a semi - model category . since @xmath46 is a field , @xmath400 and @xmath29 preserve weak equivalences , i.e. , they are `` homotopy meaningful '' . if @xmath384 is a weak equivalence , then its adjoint is given by the composite @xmath401 . the first map in the composition is a weak equivalence by corollary [ adjcor ] . the second map is a weak equivalence because @xmath29 preserves weak equivalences . we will now prove theorem [ barthm ] . first , we state a technical lemma that will help to construct the required homotopy . [ freesub]let @xmath340 be a tensor hopf algebra on the free graded @xmath46-module @xmath341 , where @xmath341 consists of primitive elements . then , the composition @xmath402 is injective . the natural degree @xmath48 map @xmath403 extends uniquely to a hopf twisting morphism @xmath404 by corollary [ freehopf ] . this defines a retraction @xmath405 of @xmath147 . let @xmath47 be a hopf algebra . we will use the notation : @xmath406 , @xmath407 , and @xmath408 . the hopf algebra embedding @xmath355 , together with the bar filtration on @xmath409 , induces a hopf algebra filtration @xmath410 on @xmath47 such that @xmath411 . this filtration induces an @xmath4 algebra filtration on @xmath412 and @xmath253 . we will now define a natural homotopy @xmath413 , such that @xmath414 by induction on the filtration . we consider the case that @xmath47 has the zero differential . if @xmath47 has a differential @xmath415 , then we can ignore the differential and obtain an @xmath416 that , once it has been defined , clearly satisfies @xmath417 . we begin by setting @xmath418 , \ldots , [ c_k ] ) ) = 0 $ ] with @xmath419 primitive , and @xmath420 . this defines a derivation homotopy on filtration level @xmath421 satisfying ( [ f1 ] ) and ( [ f2 ] ) . we next define @xmath422 for @xmath423 , \ldots , [ c_k])$ ] , where the @xmath424 are all primitive , by induction on the degree of a single sequence @xmath425 . by naturality , we can assume that @xmath340 and @xmath341 is freely generated by primitives @xmath426 . consider @xmath427 . observe that @xmath428 , and so by lemma [ freesub ] , there is a unique @xmath429 such that @xmath430 , \ldots , [ c_k]))$ ] . since @xmath431 is the sum of elements of the form @xmath432 , \ldots , [ c_k])$ ] , where @xmath433 , we have by induction that @xmath434 , therefore @xmath435 , and thus @xmath436 . our goal now is to find @xmath437 such that @xmath438 . define @xmath439 to be the first repeated integer to occur in the sequence @xmath440 from left to right , and let @xmath441 be the set of integers that occur before @xmath439 . we now define a general class of operators on @xmath442 . let @xmath443 , and let @xmath444 such that @xmath445 . then , we define @xmath446 to be given on a sequence by inserting a @xmath447 , after the first integer in the sequence that is not in the set @xmath448 . this is a variation of a homotopy given by mcclure smith ( which they credit to benson ) @xcite ( p. 689 ) to prove that @xmath34 is an @xmath5 operad . thus , we can see that the following equation is satisfied for sequences in which the elements of @xmath448 appear exactly once : @xmath449 where @xmath450 is zero if @xmath447 occurs more than once . if @xmath447 occurs exactly once and is the first integer in the sequence not in the set @xmath448 , it gives minus the sequence . otherwise , it gives minus the sequence obtained by deleting the original occurence of @xmath447 and putting it just before the first integer not in the set @xmath448 . then , given @xmath451 we define @xmath452 . since @xmath451 is a cycle , and the elements of @xmath441 do not repeat in @xmath451 , we have that @xmath453 . we will now verify two things about the construction above , first that @xmath454 and second , that @xmath455 is indeed in @xmath456 and not just in @xmath442 , as defined . both will require analyzing the @xmath4 operations on @xmath253 , and we will now proceed by checking special cases . if @xmath439 occurs at least three times in @xmath440 then @xmath457 , and @xmath439 occurs at least twice in the terms in @xmath458 . in this case , @xmath459 . now , we focus on the case when @xmath439 occurs exactly twice . the same will be true for all terms in @xmath431 , except for those two that come from deleting @xmath439 , these terms have opposite signs and cancel after applying @xmath416 and then @xmath460 . now we observe that @xmath461 or @xmath462 has only terms with @xmath439 as a repeat . thus , in this case @xmath459 , as well . if @xmath147 is an integer appearing in @xmath440 that does not occur between repeats , then the same is true of @xmath451 and thus @xmath455 . now , in @xmath440 , @xmath439 is not contained between repeats , thus @xmath463 . we now make some observations about @xmath416 . by induction on the degree of @xmath440 we can show that @xmath464 or it is a sum of sequences that have at least one integer that occurs at least three times . from this , it follows that @xmath465 on filtration level zero , since @xmath253 vanishes on such sequences applied to singletons . we now observe from the formulas for the action of @xmath4 on singletons in @xmath253 that , for the sequences appearing in @xmath462 , every integer occurs at most twice and if @xmath447 occurs twice then the subsequence between the two occurrences of @xmath447 is a singleton or of the form @xmath466 , for some @xmath98 . we can again prove by induction on degree that @xmath467 if @xmath440 is of this form , and thus @xmath468 on filtration level zero . we can now see by induction on the degree of @xmath440 that for @xmath469 , \ldots , [ c_k])$ ] , we have @xmath470 , and thus @xmath471 on filtration level zero . note that for every sequence @xmath472 in @xmath455 , @xmath473 , we see from above that @xmath474 , and @xmath475 , \ldots , [ c_k ] ) ) \\ & = h(x - p(x ) - h\partial x)\\ & = h(x)\\ & = z,\end{aligned}\ ] ] by induction and the fact that @xmath468 . thus , @xmath476 , and it follows that @xmath470 , and hence @xmath471 . since the map @xmath477 is an algebra map , and since we define @xmath416 from the left , we can show by induction on the degree of @xmath440 that @xmath416 is a derivation homotopy from @xmath57 to @xmath477 on filtration level zero , that is @xmath478 . we assume that @xmath440 can be subdivided into two sequences @xmath479 , @xmath480 such that @xmath481 , we assume that @xmath482 , \ldots , [ c_k])$ ] and @xmath483 , \ldots , [ c_k])$ ] , and thus @xmath484 , \ldots , [ c_k])$ ] . if @xmath479 contains no repeats , then @xmath467 and the formula is clear . if @xmath479 contains repeats then @xmath485 , @xmath486 , and @xmath487 , thus @xmath488 . now , by induction we have the following , @xmath489 @xmath490 the terms @xmath491 give @xmath492 , the term @xmath493 gives @xmath494 , and the rest of the terms will vanish after applying @xmath495 to the corresponding operation . finally , we proceed with the induction by filtration . assume we have defined the derivation homotopy @xmath416 satisfying ( [ f1 ] ) and ( [ f2 ] ) for all elements in filtration @xmath140 , and let @xmath423 , \ldots , [ c_k])$ ] be in filtration level @xmath236 . observe that @xmath431 is in filtration level @xmath140 , so @xmath496 is defined , and let @xmath497 . it follows from naturality that @xmath498 , \ldots , [ c_k])$ ] is in filtration level @xmath140 . thus , we define @xmath499 , \ldots , [ c_k ] ) ) + b_\alpha([c_1 ] , \ldots , [ c_k]).\ ] ] let @xmath500 , \ldots , [ c_k])$ ] . by induction , and the fact that @xmath501 we have @xmath502 now , a consistency check shows that @xmath503 . thus , we have @xmath504 finally , induction and similar arguments to those given above show that @xmath416 is a derivation homotopy satisfying ( [ f1 ] ) and ( [ f2 ] ) . let s trace through the first non - trivial example to see how the homotopy works ; we work with @xmath505 to avoid the signs . consider @xmath506 , [ c_2 ] , [ c_3])$ ] , where @xmath507 are all primitive , and we let @xmath508 . then , we have @xmath509 , but @xmath510 [ c_3 ] + [ c_2][c_1 c_3 ] = ( ( 1,2,1,3)+ ( 2,1,3,1))([c_1 ] , [ c_2 ] , [ c_3]),\ ] ] and thus @xmath511 , [ c_2 ] , [ c_3])$ ] , and so @xmath512 , @xmath513 and @xmath514 , and therefore @xmath515 . so , finally , @xmath516 , [ c_2 ] , [ c_3 ] ) = [ c_1]\{[c_2 ] , [ c_3]\}.\ ] ] in this section , we assume that @xmath517 . the purpose of this section is to prove that , up to homotopy , @xmath41 is commutative as an @xmath4 algebra . the purpose for this is first , to give a nice application of the machinery introduced above , and second , to study @xmath35 as an @xmath4 algebra , where @xmath36 is finite - type . in a forthcoming paper , we prove that under certain assumptions on @xmath36 that @xmath38 is equivalent to @xmath518 as a hopf algebra . it will then follow from theorem [ barthm ] , and this section , that under these assumptions @xmath35 is equivalent as an @xmath4 algebra to @xmath44 ( the chevalley - elilenberg cochain complex on the lie algebra @xmath519 ) , which is a commutative algebra . let @xmath520 be a connected , finite - type , nonnegatively graded differential graded lie algebra . then , @xmath40 is a hopf algebra , and thus @xmath41 is an @xmath4 algebra by our earlier result . we want to show that this algebra is equivalent as an @xmath4 algebra to a commutative algebra . consider @xmath521 , and define @xmath522 to be the largest cocommutative subcoalgebra that contains @xmath523 as its primitives ; this is the well - known chevalley eilenberg cochain complex . it is a classical result of differential homological algebra , see @xcite ( section 3 ) for example , that the inclusion @xmath524 is a weak equivalence of coalgebras . define @xmath525 , then @xmath42 is a commutative algebra , and the dual of @xmath524 is a weak equivalence of algebras @xmath526 . thus , we have a twisting morphism @xmath527 which is the dual of the twisting morphism @xmath528 . let @xmath530 be linear maps , corresponding to elements of @xmath531 . then , we obtain the product @xmath532 by dualizing the coproduct on @xmath533 , and thus we have that @xmath532 is defined on @xmath534 by @xmath535^{{\overline}\delta } & { \overline}{ul}\otimes { \overline}{ul } \ar[r]^{f\otimes g } & r\otimes r \cong r,}\ ] ] and @xmath532 maps to the element in @xmath42 under @xmath440 given by restricting the above to @xmath520 . since @xmath520 consists of primitive elements , @xmath532 maps to zero under the twisting morphism @xmath536 .

it is known that there exists a multiple - unicast network which has a rate linear network coding solution if and only if the characteristic of the finite field belongs to a given finite or co - finite set of primes . in this paper , we show that for any non - zero positive rational number , there exists a multiple - unicast network which has a rate fractional linear network coding solution if and only if the characteristic of the finite field belongs to a given finite or co - finite set of primes .

1610.08004

the concept of network coding came into light in the year 2000 by a seminal work of ahlswede _ et al._. in @xcite the authors showed that the min - cut bound in a multicast network can be achieved if the intermediate nodes are allowed to do operations on incoming messages before forwarding . this kind of operation of intermediate nodes has been termed as network coding . linear network coding refers to the scheme when all such operations are linear . in subsequent works it has been shown that linear network coding is sufficient to achieve the capacity of multicast networks @xcite . moreover , a capacity achieving network code can be designed efficiently @xcite . as far as a multicast network is concerned , a scalar linear network code over a sufficiently large finite field suffices to achieve the capacity . however , the requirement of field size can be reduced if vector linear network coding is used @xcite . in @xcite , it was shown that binary field is sufficient if the vector length is large enough . these results related to multicast networks were known by the year 2005 . after a decade later , there has been a few interesting works related to intricacies involved in linear network coding for multicast network . in one such result it has been shown that a multicast network being linearly solvable over a sufficiently large finite field does not necessarily mean that over all larger fields it is also solvable @xcite . it was also shown that existence of a vector linear solution for a certain vector dimension does not necessarily mean that there exists a vector linear solution for all larger vector dimensions @xcite . it is important to note that for a multicast network , the characteristic of the finite field does not play an important role in the sense that there does not exist a multicast network which has a scalar / vector linear solution only over a finite field of certain characteristic . the non - multicast networks have very different properties . for non - multicast networks it has been shown that there exists a network where better throughput ( capacity ) can be achieved if the nodes are allowed to do non - linear operations @xcite . there are non - multicast networks where scalar / vector linear solution not only depends on the size of the field but also on the characteristic of the field @xcite . in particular , it has been shown in @xcite that for any system of polynomial equations over integers , there exists a network which has a scalar linear network solution over a finite field if and only if the system of polynomial equation has a root in the same finite field . therefore , there exist networks which are solvable only over fields of certain characteristics . interesting examples are the fano and non - fano network presented in @xcite . the fano network has a vector linear solution for any vector dimension if and only if the characteristic of the finite field is even . over a finite field of odd characteristic it has linear coding capacity equal to @xmath2 . the non - fano network has a vector linear solution for any vector dimension if and only if the characteristic of the finite field is odd . if the characteristic is even then its linear coding capacity is equal to @xmath3 . in the references @xcite and @xcite the authors considered networks where all terminals demand the sum of the symbols generated at the sources . such networks have been given the name sum - networks . in @xcite it was shown that for any finite set of primes there exists a sum - network which has a vector linear solution for any vector dimension if and only if the characteristic of the finite field belongs to the given set . in @xcite it was shown that for any sum - network , there exists a multiple - unicast network which has a rate @xmath0 solution if and only if the sum - network has a rate @xmath0 solution . they also show that for any co - finite set of primes there exists a sum - network which has a vector linear solution for any vector dimension if and only if the characteristic of the finite field belong to the given set @xcite . thus , these three results when combined , shows that for any given finite / co - finite set of primes there exists a multiple - unicast network which has a vector linear solution if and only if the characteristic of the finite field belong to the given set . however , in the works of @xcite , the dependency on the characteristic of the field is shown only for either scalar linear network coding @xcite or vector linear network coding of any vector length @xcite . in this paper , we generalise the previous results to show that for any non - zero positive rational number @xmath1 and for any given finite / co - finite set of prime numbers , there exists a non - multicast network which has a rate @xmath1 fractional linear network code solution if and only of the characteristic of the finite belongs to the given finite / co - finite set of primes . the organization of the paper is as follows . in section [ sec2 ] we reproduce the standard definitions of fractional linear network coding , vector linear network and scalar linear network coding . in section [ sec3 ] we show that for any non - zero positive rational number @xmath1 , and for any finite / co - finite set of primes , there exists a network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belongs to the given set . in section [ sec4 ] we extend the results of section [ sec2 ] to multiple - unicast networks . the paper is concluded in section [ sec5 ] . a network is represented by a graph @xmath4 . the set @xmath5 is partitioned into three disjoint sets namely , the set of sources @xmath6 , the set of terminals @xmath7 , and the rest of the nodes are called the intermediate nodes and their collection is denoted by @xmath8 . without loss of generality the sources are assumed to have no incoming edge and the terminals are assumed to have no outgoing edge . each source generates an i.i.d random process uniformly distributed over an alphabet @xmath9 . the source process at any source is independent of all source processes generated at other sources . each terminal demands to compute the information generated at a subset of the sources . an edge @xmath10 originating from node @xmath11 and ending at node @xmath12 is denoted by @xmath13 ; where @xmath11 is denoted by @xmath14 , and @xmath12 is denoted by @xmath15 . for an node @xmath16 , the set of edges @xmath10 for which @xmath17 is denoted by @xmath18 . the information carried by an edge @xmath10 is denoted by @xmath19 . without loss of generality it is assumed that all the edges in the network are unit capacity edges . in a @xmath20 fractional linear network code the alphabet @xmath9 is taken as a finite field @xmath21 . each source @xmath22 generates a symbol @xmath23 from the finite field @xmath24 . for any edge @xmath10 , if @xmath25 , then @xmath26 where @xmath27 , @xmath28 and @xmath29 . if @xmath14 is any intermediate node @xmath30 , then @xmath31 where @xmath32 , and @xmath33 . for any terminal @xmath34 , if @xmath35 computes symbol @xmath36 , then @xmath37 where @xmath38 and @xmath39 . the matrices @xmath40 and @xmath41 shown above are called as the local coding matrices . if using a @xmath20 fractional linear network code all terminals can compute @xmath42 respective symbols in @xmath43 uses of the network , then the network is said to have a @xmath20 fractional linear network coding solution . the ratio @xmath1 is called the rate . a network is said to have a rate @xmath1 fractional linear network coding solution if it has a @xmath44 fractional linear network coding solution for any non - zero positive integer @xmath45 . a @xmath46 fractional linear network code is called as a @xmath42 dimensional vector linear network code and @xmath42 is called the vector dimension or as the message dimension . if a network has a @xmath46 fractional linear network coding solution then it is said that the network has a vector linear solution for @xmath42 message dimension . if a network has a @xmath47 vector linear network coding solution then the network is said to be scalar linearly solvable . first we show that for any positive non - zero rational number @xmath1 , and for any given finite set of primes , there exists a network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belongs to the given set . towards this end , we consider the network @xmath48 presented in fig . [ genfano1/n ] . the network shown here in fig . [ genfano1/n ] has both the generalized fano network shown in @xcite and the fano network shown in @xcite as a sub - network . as it can be seen , the network has @xmath49 sets of sources namely , @xmath50 , @xmath51 for @xmath52 and @xmath53 . in the figure , the individual source nodes are indicated by the source message it generates . a source @xmath54 generates the message @xmath55 . for @xmath52 and @xmath56 a source @xmath57 generates the message @xmath58 . and a source @xmath59 generates the message @xmath60 . there are @xmath61 sets of terminals namely , @xmath62 , @xmath63 and @xmath64 for @xmath52 ; and each of these sets contains @xmath43 terminals . each individual terminal is indicated by the source message it demands . below we list the set of edges which has a source node as its tail . 1 . @xmath65 for @xmath66 . 2 . @xmath67 for @xmath68 . 3 . @xmath69 for @xmath70 . @xmath71 for @xmath70 . @xmath72 for @xmath73 , @xmath74 , and @xmath56 @xmath75 for @xmath73 , @xmath74 , and @xmath56 . @xmath76 for @xmath77 , and @xmath56 . we now list the edges which originates at an intermediate node and ends at a intermediate node . 1 . @xmath78 for @xmath79 . 2 . @xmath80 for @xmath81 . 3 . @xmath82 , @xmath83 , and @xmath84 4 . @xmath85 for @xmath52 5 . @xmath86 for @xmath52 6 . @xmath87 and @xmath88 for @xmath52 7 . @xmath89 and @xmath90 for @xmath52 8 . @xmath91 for @xmath52 for any terminal @xmath92 there exists an edge @xmath93 and @xmath94 demands the message @xmath60 . for any terminal @xmath95 for @xmath96 , there exits an edge @xmath97 where the terminal @xmath98 demands the message @xmath58 . for any terminal @xmath99 there exits an edge @xmath100 and @xmath94 demands the message @xmath55 . for @xmath52 , a terminal @xmath101 for @xmath56 is connected to the node @xmath102 by the edge @xmath103 and @xmath98 demands the message @xmath104 . the local coding matrices are shown alongside the edges . [ lem3 ] the network in fig . [ genfano1/n ] has a rate @xmath105 fractional linear network coding solution if and only if the characteristic of the finite field divides @xmath106 . consider a @xmath107 fractional linear network coding solution of the network @xmath48 where @xmath45 is any non - zero positive integer . then the sizes of the local coding matrices are given in the following . for @xmath70 , the matrices @xmath108 and @xmath109 are of size @xmath110 and it left multiplies the information @xmath55 which is a @xmath45 length vector . matrices @xmath111 , @xmath112 , @xmath113 , @xmath114 and @xmath115 for @xmath116 and @xmath56 are of size @xmath110 and it left multiplies the @xmath45 length vector @xmath58 . for @xmath70 , the matrices @xmath117 and @xmath118 are of size @xmath110 and it left multiplies the information @xmath60 . the following matrices are of size @xmath119 : @xmath120 for @xmath121 , @xmath122 , @xmath123 and @xmath124 for @xmath52 . and the following are the matrices of size @xmath125 : @xmath126 and @xmath127 for @xmath52 and @xmath56 . also let @xmath128 be a @xmath129 identity matrix . then , rl y_(u_1,u_3 ) & = _ i=1^n d_1ia_i + _ i=1^q-1 _ j=1^n a_ijb_ij + y_(u_2,u_4 ) & = _ i=1^q-1 _ j=1^n b_ijb_ij + _ i=1^n d_2ic_i + y_(u_5,u_7 ) & = m_1y_(u_1,u_3 ) + m_2y_(u_2,u_4 ) = _ i=1^n m_1d_1ia_i + & + _ i=1^q-1 _ j=1^n ( m_1a_ij + m_2b_ij ) b_ij + _ i=1^n m_2d_2ic_i + y_(u_6,u_8 ) & = m_3y_(u_1,u_3 ) + _ i=1^n d_3ic_i + & = _ i=1^n m_3d_1ia_i + _ i=1^q-1 _ j=1^n m_3a_ijb_ij + _ i=1^n d_3ic_i + y_(u_9,u_10 ) & = m_4y_(u_5,u_7 ) + m_5y_(u_6,u_8 ) + & = _ i=1^n ( m_4m_1d_1i + m_5m_3d_1i)a_i + & + _ i=1^q-1 _ j=1^n \ { m_4(m_1a_ij + m_2b_ij ) + m_5m_3a_ij } b_ij + & + _ i=1^n ( m_4m_2d_2i + m_5d_3i)c_i + y_(u_11,u_12 ) & = _ i=1^n ( d_4i + q_1m_1d_1i)a_i + & + _ i=1^q-1 _ j=1^n ( m_1a_ij + m_2b_ij ) b_ij + _ i=1^n m_2d_2ic_i rl + y_e_i & = w_iy_(u_2,u_4 ) + _ j=1,ji^q-1 _ k=1^n u_jkib_jk = _ k=1^n w_ib_ikb_ik + & + _ j=1,ji^q-1 _ k=1^n ( w_ib_jk + u_jki)b_jk + _ k=1^n w_id_2kc_k + + y_(v_i , v_i^ ) & = k_iy_(u_9,u_10 ) + _ j=1,ji^q-1 _ k=1^n j_jkib_jk + & = _ k=1^n k_i(m_4m_1d_1k + m_5m_3d_1k)a_k + & + _ j=1^n k_i\ { m_4(m_1a_ij + m_2b_ij ) + m_5m_3a_ij } b_ij + & + _ k=1,ki^q-1 _ j=1^n \{j_kji + k_i(m_4(m_1a_kj + m_2b_kj ) + & + m_5m_3a_kj ) } b_kj + & + _ j=1^n k_i(m_4m_2d_2j + m_5d_3j)c_j + + y_(w_i , w_i^ ) & = v_iy_e_i + _ j=1^n e_ijb_ij = _ k=1^n ( v_iw_ib_ik + e_ik ) b_ik + & + _ j=1,ji^q-1 _ k=1^n \ { v_i(w_ib_jk + u_jki ) } b_jk + & + _ k=1^n v_iw_id_2kc_k + + & + _ i=1^q-1 _ j=1^n \{q_2m_3a_ij + r_iw_ib_ij + & + _ k=1,ki^q-1 r_k(w_kb_ij+u_ijk)}b_ij + & + _ i=1^n\ { q_2d_3i + ( _ k=1^q-1r_kw_k ) d_2i } c_i since a terminal @xmath92 computes @xmath60 , we have the following inequalities . for @xmath130 since the component of @xmath55 is zero at all @xmath131 , @xmath132 as the components of for @xmath58 is also zero at all @xmath133 , for @xmath52 and @xmath134 we have , @xmath135 now , since the terminal @xmath92 retrieves @xmath60 for @xmath130 and @xmath136 we have , l g_i(q_1m_2d_2i ) = i[etc3a ] + g_i(q_1m_2d_2j ) = 0[etc3b ] now consider the @xmath43 terminals in the set @xmath63 for @xmath52 . since the component of @xmath137 for @xmath138 at @xmath95 for @xmath56 is zero , we have , for @xmath52 and @xmath134 : @xmath139 since the terminal @xmath95 computes the information @xmath58 , we have , for @xmath140 , @xmath141 and @xmath142 : l r_ijk_i\ { m_4(m_1a_ij + m_2b_ij ) + m_5m_3a_ij } = i[etbb1 ] + r_ijk_i\ { m_4(m_1a_im + m_2b_im ) + m_5m_3a_im } = 0[etbb2 ] + r_ij\{k_i ( m_4(m_1a_kl + m_2b_kl ) + m_5m_3a_kl ) + j_kli } = 0 since the component of @xmath143 for @xmath138 is zero at @xmath95 , we have for @xmath52 and @xmath134 , @xmath144 let us consider the terminals in the set @xmath145 . since @xmath99 computes the message @xmath55 , for @xmath130 and @xmath136 , we have : l l_iq_2m_3d_1i = i [ etaa1 ] + l_iq_2m_3d_1j = 0 [ etaa2 ] at any @xmath146 for @xmath52 and @xmath147 the component of @xmath58 is zero . so we have , @xmath148 at a terminal @xmath149 , since the component of @xmath60 is zero , for @xmath130 : @xmath150 now consider the terminals in the set @xmath64 for @xmath151 . since at @xmath152 the component of @xmath58 for @xmath52 and @xmath147 is zero , we have , @xmath153 for @xmath154 the component of @xmath155 for @xmath56 is zero as well , @xmath156 since @xmath157 computes @xmath158 , for @xmath159 , we have , l z_ilv_iw_id_2l = i [ etcc1 ] + z_ilv_iw_id_2 m = 0 [ etcc2 ] the rest of the proof requires a lemma and a corollary which are presented next . let us consider @xmath160 and @xmath161 where @xmath162 and @xmath163 for @xmath70 are matrices of size @xmath125 and @xmath110 respectively . let @xmath164 denote an identity matrix of size @xmath165 . [ lem1 ] if @xmath166 but @xmath167 for @xmath168 , then @xmath169 ; and both @xmath170 and @xmath171 has an unique inverse . rl ab & = a_1 + a_2 + + a_n b_1 & b_2 & & b_n + & = a_1b_1 & a_1b_2 & & a_1b_n + a_2b_1 & a_2b_2 & & a_2b_n + & & & + a_nb_1 & a_nb_2 & & a_nb_n + & = i & 0 & & 0 + 0 & i & & 0 + & & & + 0 & 0 & & i = i_dn now since both @xmath170 and @xmath171 are matrices of size @xmath119 , for @xmath169 to hold , both of @xmath170 and @xmath171 has to be full rank matrices . this completes the proof . [ coro1 ] for @xmath130 , if @xmath167 , then @xmath172 . we now define the following matrices . note that all of these newly defined matrices are square matrices of size @xmath165 . l g = g_1 & g_2 & & g_n ^t [ g ] + q_1(m_1a_i + m_2b_i ) = q_1(m_1a_i1 + m_2b_i1 ) + q_1(m_1a_i2 + m_2b_i2 ) + + q_1(m_1a_in + m_2b_in ) ^t[q ] + q_1m_2d_2 = q_1m_2d_21 & q_1m_2d_22 & & q_1m_2d_2n [ q1 ] + r_ik_i = r_i1k_i & r_i2k_i & & r_ink_i ^t[r ] l m_4m_1d_1 + m_5m_3d_1 = m_4m_1d_11 + m_5m_3d_11 + m_4m_1d_12 + m_5m_3d_12 + + m_4m_1d_1n + m_5m_3d_1n ^t[m4 ] + m_4(m_1a_i + m_2b_i ) + m_5m_3a_i + = m_4(m_1a_i1 + m_2b_i1 ) + m_5m_3a_i1 + m_4(m_1a_i2 + m_2b_i2 ) + m_5m_3a_i2 + + m_4(m_1a_in + m_2b_in ) + m_5m_3a_in ^t[m4m2m5 ] + m_4m_2d_2 + m_5d_3 = m_4m_2d_21 + m_5d_31 + m_4m_2d_22 + m_5d_32 + + m_4m_2d_2n + m_5d_3n ^t[m4m5 ] + l = l_1 & l_2 & & l_n ^t[l ] + q_2m_3d_1 = q_2m_3d_11 & q_2m_3d_12 & & q_2m_3d_1n [ q2 ] l q_2m_3a_i + r_iw_ib_i + _ k=1,ki^q-1 r_k(w_kb_i+u_ik ) + = q_2m_3a_i1 + r_iw_ib_i1 + _ k=1,k i^q-1 r_k(w_kb_i1+u_i1k ) + q_2m_3a_i2 + r_iw_ib_i2 + _ k=1,k i^q-1 r_k(w_kb_i2+u_i2k ) + + q_2m_3a_in + r_iw_ib_in + _ k=1,k i^q-1 r_k(w_kb_in+u_ink ) ^t [ q2r ] + q_2d_3 + ( _ k=1^q-1r_kw_k ) d_2 + = q_2d_31 + ( _ k=1^q-1r_kw_k ) d_21 + q_2d_32 + ( _ k=1^q-1r_kw_k ) d_22 + + q_2d_3n + ( _ k=1^q-1r_kw_k ) d_2n ^t[q2d2 ] + z_iv_i = z_i1v_i & z_i2v_i & & z_inv_i ^t[z ] + w_ib_k + u_ki = v_i(w_ib_k1 + u_k1i ) + v_i(w_ib_k2 + u_k2i ) + + v_i(w_ib_kn + u_kni ) ^t[u ] + w_id_2 = w_id_21 & w_id_22 & & w_id_2n [ wd2 ] for the rest of this section we redefine @xmath128 as an identity matrix of size @xmath165 . using the corollary [ coro1 ] , from equations ( [ etcba ] ) , ( [ g ] ) and ( [ q ] ) , we have , for @xmath52 : @xmath173 using lemma [ lem1 ] , from equations ( [ etc3a ] ) , ( [ etc3b ] ) , ( [ g ] ) and ( [ q1 ] ) we get , @xmath174 using corollary [ coro1 ] and equations ( [ etba1 ] ) , ( [ r ] ) and ( [ m4 ] ) we have , for @xmath52 : @xmath175 using lemma [ lem1 ] and equations ( [ etbb1 ] ) , ( [ etbb2 ] ) , ( [ r ] ) and ( [ m4m2m5 ] ) we have , for @xmath52 : @xmath176 using corollary [ coro1 ] on equations ( [ etbc1 ] ) , ( [ r ] ) and ( [ m4m5 ] ) we have , for @xmath52 : @xmath177 using lemma [ lem1 ] on equations ( [ etaa1 ] ) , ( [ etaa2 ] ) , ( [ l ] ) and ( [ q2 ] ) for we get : @xmath178 using corollary [ coro1 ] and equations ( [ etab1 ] ) , ( [ l ] ) and ( [ q2r ] ) we get , for @xmath52 : @xmath179 using lemma [ lem1 ] , and equations ( [ etac1 ] ) , ( [ l ] ) and ( [ q2d2 ] ) we have : @xmath180 using corollary [ coro1 ] and equations ( [ etci1 ] ) , ( [ z ] ) and ( [ u ] ) we get , for @xmath181 , @xmath182 using lemma [ lem1 ] , and equations ( [ etcc1 ] ) , ( [ etcc2 ] ) , ( [ z ] ) and ( [ wd2 ] ) we have : @xmath183 now , from equation ( [ etc3c ] ) the matrices @xmath184 , @xmath185 , @xmath186 and @xmath187 are invertible . from equation ( [ etbb3 ] ) , for @xmath52 , @xmath188 and @xmath189 are invertible . @xmath190 , @xmath191 and @xmath192 is invertible from equation ( [ etaa3 ] ) . from equation ( [ etcc3 ] ) @xmath193 and @xmath194 are invertible . from ( [ etcb2 ] ) since @xmath184 and @xmath187 both are invertible , we have : @xmath195 since bot @xmath189 and @xmath188 are invertible , we have from equation ( [ etba2 ] ) : @xmath196 and from ( [ etbc2 ] ) we have : @xmath197 since @xmath190 is also invertible , from equation ( [ neweq1 ] ) we must have , @xmath198 substituting ( [ etcb3 ] ) in ( [ etbb3 ] ) we have @xmath199 since @xmath194 and @xmath193 are invertible , from equation ( [ etci2 ] ) we must have , for @xmath200 : @xmath201 substituting equation ( [ etci3 ] ) in equation ( [ etab2 ] ) , and noting that @xmath191 is invertible , we have for @xmath52 : @xmath202 @xmath203 and @xmath162 are invertible from ( [ etbb4 ] ) for @xmath52 , and since @xmath192 is invertible , @xmath204 is invertible , which leads that @xmath205 is invertible from equation ( [ etab3 ] ) , and hence @xmath163 is invertible . so from ( [ etab3 ] ) , for @xmath52 , @xmath206 since @xmath191 is invertible , we have from ( [ etac2 ] ) @xmath207 since @xmath185 is invertible , we have : @xmath208 substituting @xmath209 for @xmath210 , from ( [ etab4 ] ) we have , l q_2d_3d_2 ^ -1 + _ k=1^q-1-q_2m_3a_kb_k^-1 = 0 [ e1 ] since @xmath211 and @xmath163 are invertible for @xmath52 , from ( [ etcb3 ] ) @xmath212 is also invertible , and hence we have , for @xmath52 : @xmath213 substituting equation ( [ etcb4 ] ) in equation ( [ e1 ] ) we have : l q_2d_3d_2 ^ -1 + _ k=1^q-1-q_2m_3(-m_1 ^ -1m_2 ) = 0 since all constituent matrices are square , l q_2d_3d_2 ^ -1 + _ k=1^q-1q_2m_3m_1 ^ -1m_2 = 0 [ e2 ] since @xmath214 is invertible from equation ( [ etbb4 ] ) , and @xmath212 is invertible as shown above , from equation ( [ etba3 ] ) we have : @xmath215 substituting equation ( [ etba4 ] ) in equation ( [ e2 ] ) we have : l q_2d_3d_2 ^ -1 + _ k=1^q-1q_2(-m_5 ^ -1m_4)m_2 = 0 [ e3 ] now , from equation ( [ etbc3 ] ) we have : @xmath216 substituting this in ( [ e3 ] ) we have : l q_2d_3d_2 ^ -1 + _ k=1^q-1q_2d_3d_2 ^ -1 = 0 + ( q)q_2d_3d_2 ^ -1 = 0 since @xmath192,@xmath217 and @xmath185 are all invertible matrices , it must be that @xmath218 . now the fact that an element is equal to zero in a finite field if and only if the characteristic divides the element proves that the network has a rate @xmath105 fractional linear network coding solution only if the characteristic of the finite field divides @xmath106 . we now show that the network @xmath48 has a @xmath219 fractional linear network coding solution if @xmath218 . for this section , let @xmath220 denote an @xmath43-length vector whose @xmath221 component is @xmath55 and all other component is zero . let @xmath222 to denote an @xmath43-length vector whose @xmath221 component is @xmath60 and all other component is zero . also let @xmath223 denotes an @xmath43-length vector which has zero in all of its components but the @xmath224 one , which is equal to @xmath58 . note that @xmath55 , @xmath60 for @xmath70 and @xmath58 for @xmath96 are the source processes . now , it can be seen that by choosing the appropriate local coding matrices , the messages shown below can be transmitted by the corresponding edges . l y_(u_1,u_3 ) = _ i=1^n |a_i + _ i=1^p-1 _ j=1^n |b_ij + y_(u_2,u_4 ) = _ i=1^q-1 _ j=1^n |b_ij + _ i=1^n |c_i + y_(u_5,u_7 ) = y_(u_1,u_3 ) - y_(u_2,u_4 ) = _ i=1^n |a_i - _ i=1^n |c_i + y_(u_6,u_8 ) = y_(u_1,u_3 ) - _ i=1^n |c_i = _ i=1^n |a_i + _ i=1^q-1 _ j=1^n 1iq-1 : y_e_i = _ j=1^n |b_ij + _ i=1^n |c_i + y_(u_9,u_10 ) = y_(u_6,u_8 ) - y_(u_5,u_7 ) = _ i=1^q-1 _ j=1^n |b_ij + y_(u_13,u_14 ) = y_(u_6,u_8 ) - _ i=1^q-1 y_e_i = _ i=1^n |a_i - _ i=1^n |c_i - _ i=1^q-1 _ i=1^n |c_i + = _ i=1^n |a_i - _ i=1^q _ i=1^n |c_i = _ i=1^n |a_i - _ i=1^n q|c_i = _ i=1^n y_(u_11,u_12 ) = _ i=1^n |a_i - y_(u_5,u_7 ) = _ i=1^n |c_i + 1iq-1 : + y_(v_i , v_i^ ) = y_(u_9,u_10 ) - _ k=1,ki^q-1 _ j=1^n |b_ij = _ j=1^n |b_ij + 1iq-1 : y_(w_i , w_i^ ) = y_e_i - _ j=1^n |b_ij = _ i=1^n |c_i let @xmath225 be a unit row vector of length @xmath43 which has @xmath221 component equal to one and all other component equal to zero . then from the vector @xmath226 , @xmath55 for any @xmath70 can be determined by the dot product @xmath227 . similarly for any @xmath52 , @xmath228 . for @xmath70 , @xmath60 can be determined similarly from @xmath229 . [ thm1 ] for any non - zero positive rational number @xmath1 and for any finite set of prime numbers @xmath230 there exists a network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belong to the given set of primes . let us consider the union of @xmath42 copies of the network @xmath48 shown in fig . [ genfano1/n ] each for @xmath231 . denote the @xmath221 copy as @xmath232 . note that each source and each terminal has @xmath42 copies in the union . join all copies of any source or terminal into a single source or terminal respectively . name this new network as @xmath233 . we show below that @xmath233 has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belong to the set @xmath230 . before we proceed further , consider the following property of @xmath48 and @xmath233 . [ lem2 ] if @xmath233 has a @xmath44 fractional linear network coding solution for any non - zero positive integer @xmath45 , then @xmath48 has a @xmath234 fractional linear network coding solution . this is true since the information that can be sent using the network @xmath233 for @xmath235 times , can be sent using the network @xmath48 @xmath236 times . this is because @xmath233 has @xmath42 copies of @xmath48 . first consider the only if part . say @xmath233 has a rate @xmath1 fractional linear network coding solution even if the characteristic does not belong to the set @xmath230 . then from lemma [ lem2 ] , the network @xmath48 has a rate @xmath105 fractional linear network coding solution even if the characteristic does not belong to the given set of primes . however , as shown in lemma [ lem3 ] , the @xmath48 has a rate @xmath105 fractional linear network coding solution if and only if @xmath218 over the finite field . but , as @xmath231 , @xmath218 if and only if one of the prime number from the set @xmath230 is zero over the finite field . the latter is the case if and only if the characteristic of the finite field is one of the primes in the set . hence this is a contradiction to the case that @xmath48 has a rate @xmath105 fractional linear network coding solution even if the characteristic does not belong to the given set of primes . now consider the if part . since @xmath232 for @xmath237 has a @xmath219 fractional linear network coding solution , a @xmath20 fractional linear network coding solution can be constructed by keeping the same local coding matrices in all of the copies and sending the @xmath221 component of each source through @xmath232 . the outline of the contents in this section is similar to that of the last sub - section . consider the network @xmath238 shown in fig . [ nofano1/n ] . the sources are partitioned into @xmath49 sets : @xmath50 and @xmath51 for @xmath239 . each of these sets has @xmath43 sources . in fig . [ nofano1/n ] the sources are indicated by the message symbol it generates . a source @xmath240 generates the message @xmath55 . similarly a source @xmath57 generates the message @xmath58 for @xmath241 . the set of terminals are partitioned into @xmath242 disjoint sets namely , @xmath243 and @xmath63 for @xmath239 . the sets @xmath244 and @xmath245 has @xmath43 terminals in each . each individual terminal is indicated by the source message it demands . we have the following edges in the network . 1 . @xmath246 and @xmath247 2 . @xmath85 and @xmath248 for @xmath239 3 . @xmath249 for @xmath250 4 . @xmath251 for @xmath239 and @xmath252 5 . @xmath253 for @xmath254 6 . @xmath255 and @xmath256 7 . @xmath257 8 . @xmath258 for @xmath239 9 . @xmath259 for @xmath239 10 . @xmath260 for @xmath239 from each of the nodes @xmath261 , @xmath262 for @xmath239 and @xmath263 , @xmath43 outgoing edge emanates and the @xmath264 node of all such edges is a terminal . the set of @xmath43 terminals which have a path from node @xmath261 are denoted by @xmath244 . similarly , the set of @xmath43 terminals which have a path from node @xmath265 are denoted by @xmath245 . and the @xmath43 terminals in the set @xmath63 for @xmath239 are connected from the node @xmath262 by an edge . before we show that for any non - zero positive rational number @xmath1 and for a given co - finite set of primes there exist a network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belong to the given set , consider the following lemma . [ lem4 ] the network shown in fig . [ nofano1/n ] has a rate @xmath105 fractional linear network coding solution if and only if the characteristic of the finite field does not divides @xmath106 . consider a @xmath107 fractional linear network coding solution of the network . the local coding matrices are shown next to the edges . the matrices @xmath266 for @xmath70 and @xmath111 for @xmath267 are of size @xmath268 and left multiplies the massage @xmath55 . the matrices @xmath269 for @xmath241 , @xmath270 for @xmath271 , and @xmath272 for @xmath241 left multiplies @xmath58 and are of size @xmath268 . the matrices @xmath273 and @xmath274 and @xmath275 for @xmath239 are of sizes @xmath165 . and the matrices of size @xmath276 are @xmath277 and @xmath278 for @xmath239 and @xmath56 . also let @xmath128 be a @xmath129 identity matrix . the following is the list of messages carried by some of the edges of the network . ll y_e_a & = _ i=1^n q_ia_i + _ i=1^q _ j=1^n c_ijb_ij + + y_e_i & = _ j=1^n a_jia_j + _ j=1,ji^q _ k=1^n b_jkib_jk + y_e_b & = _ i=1^q _ j=1^n d_ijb_ij + y_e_a^ & = m_2y_e_a + m_3y_e_b = _ i=1^n m_2q_ia_i + & + _ i=1^q _ j=1^n ( m_2c_ij + m_3d_ij)b_ij + + y_e_i^ & = k_iy_e_a + r_iy_e_i = _ j=1^n ( k_iq_j + r_ia_ji)a_j + & + _ k=1^n k_ic_ikb_ik + _ j=1,ji^q _ k=1^n ( k_ic_jk + r_ib_jki)b_jk + y_e_b^ & = _ i=1^q u_iy_e_i + m_4y_e_b = _ i=1^q _ j=1^n u_ia_jia_j + & + _ j=1^q _ k=1^n ( ( _ i=1,ij^q u_ib_jki ) + m_4d_jk ) b_jk because of the demands of the terminals the following inequalities must be satisfied . since any terminal @xmath279 computes @xmath55 , we have , for @xmath280 : l e_im_2q_i = i[gq1 ] + e_im_2q_j = 0[gq2 ] at @xmath281 the component of @xmath58 is zero for @xmath282 we have : @xmath283 now consider the terminals in the set @xmath63 for @xmath239 . since at any terminal @xmath95 for @xmath56 the component of @xmath137 for @xmath138 is zero , we have , @xmath284 because @xmath95 computes @xmath58 for @xmath285 we have : l g_ij(k_ic_ij ) = i[gq5 ] + g_ij(k_ic_ik ) = 0[gq6 ] as the component of any @xmath286 at @xmath95 is zero if @xmath287 , we have for @xmath288 we have : @xmath289 we now consider the set @xmath245 . since the terminal @xmath290 computes @xmath55 we have , for @xmath280 l v_i(_k=1^q u_ka_ik ) = i[gq8 ] + v_i(_k=1^q u_ka_jk ) = 0[gq9 ] the component of @xmath291 is zero at @xmath290 for @xmath292 , and hence we have , @xmath293 we now define the following matrices . l e = e_1 & e_2 & & e_n ^t[e ] + m_2q = m_2q_1 & m_2q_2 & & m_2q_n [ m2q ] + m_2c_i + m_3d_i = m_2c_i1 + m_3d_i1 + m_2c_i2 + m_3d_i2 + + m_2c_in + m_3d_in ^t[m2ci ] + g_i = g_i1 & g_i2 & & g_in ^t[gi ] + k_iq + r_ia_i = k_iq_1 + r_ia_1i + k_iq_2 + r_ia_2i + + k_iq_n + r_ia_ni ^t[kiq ] + k_ic_i = k_ic_i1 & k_ic_i2 & & k_ic_in [ kici ] + k_ic_k + r_ib_ki = k_ic_k1 + r_ib_k1i + k_ic_k2 + r_ib_k2i + + k_ic_kn + r_ib_kni ^t[kick ] + v = v_1 & v_2 & & v_n ^t[v ] + _ k=1^q u_ka_k = _ k=1^q u_ka_1k + _ k=1^q u_ka_2k + + _ k=1^q u_ka_nk ^t[sumukak ] + ( _ r=1,rj^q u_rb_jr ) + m_4d_j = ( _ r=1,rj^q u_rb_j1r ) + m_4d_j1 + ( _ r=1,rj^q u_rb_j2r ) + m_4d_j2 + + ( _ r=1,rj^q u_rb_jnr ) + m_4d_jn ^t [ urbjr ] note that all of these newly defined matrices are square and are of size @xmath165 . for the rest of the paper let @xmath128 denote an identity matrix of size @xmath119 . applying lemma [ lem1 ] on equations ( [ gq1 ] ) and ( [ gq2 ] ) and using the newly defined matrices in equation ( [ e ] ) and ( [ m2q ] ) we get : @xmath294 from corollary [ coro1 ] and equations ( [ gq3 ] ) , ( [ e ] ) and ( [ m2ci ] ) we get , for @xmath239 : @xmath295 similarly using corollary [ coro1 ] and equations ( [ gq4 ] ) , ( [ gi ] ) and ( [ kiq ] ) we get , for @xmath239 : @xmath296 using lemma [ lem1 ] and equations ( [ gq5 ] ) , ( [ gq6 ] ) , ( [ gi ] ) and ( [ kici ] ) we get , for @xmath239 : @xmath297 from corollary [ coro1 ] and equations ( [ gq7 ] ) , ( [ gi ] ) and ( [ kick ] ) we have , for @xmath298 : @xmath299 employing lemma [ lem1 ] and equations ( [ gq8 ] ) , ( [ gq9 ] ) , ( [ v ] ) and ( [ sumukak ] ) we have : @xmath300 from corollary [ coro1 ] and equations ( [ gq10 ] ) , ( [ v ] ) and ( [ urbjr ] ) we have , for @xmath301 : @xmath302 the matrices @xmath303 and @xmath304 are invertible from equation ( [ em2q ] ) . matrix @xmath305 and @xmath306 for @xmath239 is invertible from equation ( [ gikici ] ) . and @xmath5 is invertible from equation ( [ vsum ] ) . since @xmath307 is invertible we have from equation ( [ em2ci ] ) : @xmath308 as both @xmath186 and @xmath306 are invertible matrices , their product is a full rank matrix , and hence from equation ( [ 1 ] ) , @xmath203 is an invertible matrix . this comes from the fact that a matrix of rank equal to a certain value can not multiply with any other matrices and result in a matrix of rank equal to a greater value . since @xmath309 is invertible for @xmath239 , we have from equation ( [ gikiq ] ) : @xmath310 since both @xmath188 and @xmath304 are invertible matrices , their product is a full rank matrix , and hence @xmath189 is an invertible matrix for @xmath239 . also from equation ( [ gikick ] ) we have , for @xmath311 : @xmath312 and since @xmath5 is invertible we have from equation ( [ vbigsum ] ) , for @xmath239 : @xmath313 substituting @xmath314 from equation ( [ 1 ] ) in equation ( [ 4 ] ) we get , for @xmath239 : l ( _ r=1,ri^q u_rb_ir ) - m_4m_3 ^ -1m_2c_i = 0 + we get : + - ( _ r=1,ri^q u_rr_r^-1k_rc_i ) - m_4m_3 ^ -1m_2c_i = 0 + we get : + ( _ r=1,ri^q u_ra_rq^-1c_i ) - m_4m_3 ^ -1m_2c_i = 0 + + ( _ r=1,ri^q u_ra_rem_2c_i ) - m_4m_3 ^ -1m_2c_i = 0 + ( ( _ r=1,ri^q u_ra_re ) - m_4m_3 ^ -1)m_2c_i = 0 + : + ( _ r=1,ri^q u_ra_re ) - m_4m_3 ^ -1 = 0 + : + ( v^-1 -u_ia_i)e - m_4m_3 ^ -1 = 0 + v^-1 -u_ia_i = m_4m_3 ^ -1e^-1 + u_ia_i = v^-1 - m_4m_3 ^ -1e^-1[5 ] now , substituting equation ( [ 5 ] ) in equation ( [ vsum ] ) we get l v(_i=1^q v^-1 - m_4m_3 ^ -1e^-1 ) = i + _ i=1^q v(v^-1 - m_4m_3 ^ -1e^-1 ) = i + _ i=1^q i - vm_4m_3 ^ -1e^-1 = i + ( q-1)i = qivm_4m_3 ^ -1e^-1[6 ] in equation ( [ 6 ] ) , if @xmath218 , then the equation becomes @xmath315 . hence @xmath316 is a necessary condition for the network @xmath238 to have a rate @xmath105 fractional linear network coding solution . then from the fact that an element in a finite field is equal to zero if and only if the characteristic of the finite field divides that element , it can be concluded that @xmath316 if and only if the characteristic of the finite field does not divides @xmath106 . we now show that @xmath238 has a @xmath219 fractional linear network coding solution if the @xmath106 has an inverse in the finite field . note that , as discussed above , @xmath106 has an inverse if and only if the characteristic of the finite field does not divides @xmath106 . let an @xmath43-length vector whose @xmath221 component is @xmath55 and all other components are zero be denoted by @xmath220 . also let @xmath43-length vector whose @xmath224 component is @xmath58 and all other components are zero be denoted by the notation @xmath223 . note that @xmath317 and @xmath58 for @xmath241 are the source processes . then it can be seen that for proper local coding matrices the following information can be transmitted by the corresponding edges . l y_e_a = _ j=1^n |a_j + _ i=1^q _ j=1^n |b_ij + 1iq : y_e_i = _ j=1^n |a_j + _ k=1,ki^q _ j=1^n |b_kj + y_e_b = _ i=1^q _ j=1^n |b_ij + y_e_a^ = y_e_a - y_e_b = _ j=1^n |a_j + 1iq : y_e_i^ = y_e_a - y_e_i = _ j=1^n |b_ij + y_e_b^ = q^-1\{_i=1^q y_e_i - ( q-1)y_e_b } = q^-1\{q(_j=1^n |a_j ) + + ( q-1)(_i=1^q _ j=1^n |b_ij ) - ( q-1)(_i=1^q _ j=1^n |b_ij ) } = _ j=1^n |a_j let @xmath318 be a unit row vector of length @xmath43 which has @xmath224 component equal to one and all other component equal to zero . then from the dot product of @xmath318 and @xmath319 , message @xmath317 can be retrieved . similarly from the dot product of @xmath318 and @xmath320 , @xmath58 can be determined . [ thm2 ] for any non - zero positive rational number @xmath1 and for any finite set of prime numbers @xmath230 there exists a network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field does not belong to the given set of primes . let @xmath106 be equal to @xmath321 in @xmath238 . let us construct @xmath322 by joining @xmath43 copies of @xmath238 at the corresponding sources and the terminals , in a similar way @xmath233 was constructed from @xmath48 . it can be also seen that lemma [ lem2 ] holds true when @xmath48 and @xmath233 are replaced by @xmath238 and @xmath322 respectively . so if @xmath322 has a @xmath20 fractional linear network coding solution then @xmath238 has a @xmath323 fractional linear network coding solution . now say @xmath322 has a rate @xmath1 fractional linear network coding solution even if the characteristic of the finite belongs to the set @xmath230 . since @xmath324 , note that @xmath218 over such a field . then , @xmath238 has a rate @xmath325 fractional linear network coding solution over a finite field in which @xmath218 . however , this is in contradiction with lemma [ lem4 ] . if however , the characteristic does not belong to the given set of primes , then , since there are @xmath43 copies of @xmath238 in @xmath322 , and each copy has a @xmath219 fractional linear network coding solution , a @xmath20 fractional linear network coding solution can easily be constructed for @xmath322 . are sources and source @xmath326 generates the messages @xmath327 for @xmath328 . the nodes @xmath329 are terminals and @xmath94 demands @xmath330 for @xmath328 . the nodes @xmath331 and @xmath332 were terminals in the original network ( network indicated by the dotted line ) and both of these nodes demanded the message @xmath333 . the rest of the naming convention has been kept the same as it was in @xcite.,scaledwidth=48.0% ] in this section we show that for any non - zero positive rational number @xmath1 and for any finite / co - finite set of primes , there exits a multiple - unicast network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belongs to the given set . to prove this result , we first show that for each of the networks @xmath48 and @xmath238 presented in section [ sec3 ] , there exists a multiple - unicast network which has a @xmath219 fractional linear network coding solution if and only if the the corresponding network @xmath48 or @xmath238 has a @xmath219 fractional linear network coding solution . in a multiple - unicast network , by definition , each source process is generated at only one source node and is demanded by only one terminal . additionally , each source node generates only one source process , and each terminal demands only one source process . in both the networks @xmath48 and @xmath238 there exists no source processes which is generated by more than one source node , or no source node generates more than one source process . moreover , there does not exist any terminal which demands more than one source process . however , there exists more than one terminal which demands the same source process . this is fixed in the following way . in @xcite it was shown that for any network there exists a solvably equivalent multiple - unicast network . to resolve the case of more than one terminals demanding the same source message , the authors considered two such terminals at a time and added a gadget to the two terminals . the same procedure is followed here , only the gadget has been modified . this modified gadget is shown in fig . it is assumed that the nodes @xmath331 and @xmath332 both demanded the same message @xmath333 in the original network ( network before attaching the gadget ) . after adding the gadget , the modified network has @xmath43 more source nodes @xmath334 , and @xmath335 new terminal nodes @xmath336 . nodes @xmath331 and @xmath332 becomes intermediate nodes in the modified construction . this process has to be repeated iteratively for every two terminals in the original network that demand the same source process . in the same way as shown in _ theorem ii.1 _ of @xcite , it can be shown that after the completion of this process , the resulting network has a @xmath219 fractional linear network coding solution if and only if the original network has a @xmath219 fractional linear network coding solution . hence , as shown above , corresponding to each of the networks @xmath48 and @xmath238 , there exist multiple - unicast networks @xmath337 and @xmath338 which have a @xmath219 fractional linear network coding solution if and only if @xmath48 and @xmath238 have a @xmath219 fractional linear network coding solution respectively . now by connecting @xmath42 copies of @xmath337 and @xmath338 in the same way as @xmath233 and @xmath322 was constructed from @xmath48 and @xmath238 respectively , the following theorem can be proved in a similar way to theorem [ thm1 ] and theorem [ thm2 ] . for any non - zero positive rational number @xmath1 and for any finite / co - finite set of prime numbers there exists a multiple - unicast network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belongs to the given set of primes . in this paper we have shown that for any non - zero positive rational number @xmath1 and any finite / co - finite set of prime numbers there exists a multiple unicast network which has a rate @xmath1 fractional linear network coding solution if and only if the characteristic of the finite field belongs to the given set . to prove the existence , we have explicitly presented networks having desired properties . the generalized fano and generalized non - fano networks presented in @xcite are special cases of the networks presented in this paper . s. jaggi , p. sanders , p. a. chou , m. effros , s. egner , k. jain , and l. m. g. m. tolhuizen , `` polynomial time algorithms for multicast network code construction , '' _ ieee transactions on information theory _ , 2005 .

the theory of zero - error communication is re - examined in the broader setting of using one classical channel to simulate another exactly in the presence of various classes of non - signalling correlations between sender and receiver i.e. shared randomness , shared entanglement and arbitrary non - signalling correlations . when the channel being simulated is noiseless , this is zero - error coding assisted by correlations . when the resource channel is noiseless , it is the reverse problem of simulating a noisy channel exactly by a noiseless one , assisted by correlations . in both cases , separations between the power of the different classes of assisting correlations are exhibited for finite block lengths . the most striking result here is that entanglement can assist in zero - error communication . in the large block length limit , shared randomness is shown to be just as powerful as arbitrary non - signalling correlations for exact simulation , but not for asymptotic zero - error coding . for assistance by arbitrary non - signalling correlations , linear programming formulas for the asymptotic capacity and simulation rates are derived , the former being equal ( for channels with non - zero unassisted capacity ) to the feedback - assisted zero - error capacity derived by shannon . finally , a kind of reversibility between non - signalling - assisted zero - error capacity and exact simulation is observed , mirroring the usual reverse shannon theorem .

1003.3195

much of classical and quantum information theory is concerned with the use of one resource ( a channel , an entangled state , etc . ) to simulate another . typically errors are allowed in the simulation protocol if they vanish asymptotically as the number of resources involved grows . one then asks for the asymptotic rates of resource exchange : shannon s channel coding theorem @xcite tells us the asymptotic rate at which we need to make use of a given discrete memoryless channel to simulate a perfect bit channel . the quantum reverse shannon theorem @xcite shows that a single number associated to quantum channels , the entanglement assisted classical capacity @xmath0 , determines the rate at which it can simulate another when entanglement is a free resource . since @xmath0 reduces to the shannon capacity for classical channels , the availability of entanglement does not affect the rate at which one classical channel can simulate another , in the setting where errors which vanish in the large block length limit are tolerated . since it is often unrealistic to assume that arbitrarily long block lengths can be used in encoding and decoding , an alternative , idealised , task of _ zero - error coding _ @xcite has been considered since the seminal 1956 paper of shannon @xcite and more recently in quantum information theory @xcite . for a suitable definition of decoding error probability @xmath1 , both asymptotic and zero - error coding theory make statements about the region of triples @xmath2 which can be achieved by codes which use @xmath3 channel uses to transmit @xmath4 bits ( or , equivalently , one of @xmath5 symbols ) . the full characterisation of this achievable region is normally far from tractable . whereas the freedom granted by demanding only that @xmath6 as @xmath7 admits simplification via random coding arguments ( for example ) in the asymptotic theory , the zero - error theory ( which studies the restriction of the region to the plane @xmath8 ) is attractive because the problem becomes essentially combinatorial nevertheless , it is a source of hard mathematical problems : in shannon s groundbreaking work on the subject @xcite he made a conjecture ( on the zero - error capacity of the pentagon channel ) which had to wait over twenty years before it was proven by lovsz @xcite . many related open problems remain @xcite . in this paper we consider both zero - error coding and the `` reverse '' problem of exact simulation of noisy channels when various types of correlations between sender and receiver are freely available . this leads to various relaxations of the combinatorial problems posed by the unassisted theory , some of which have complete and general solutions . this section introduces the central concepts and quantities dealt with in the rest of the paper ( please note that an index of notations is provided as an appendix ) . @xmath9 denotes the set of discrete , memoryless , classical channels ( i.e. conditional probability distributions ) with inputs in @xmath10 and outputs in @xmath11 , @xmath10 and @xmath11 being finite sets . @xmath12 means the set of bipartite conditional probability distributions , with inputs in the set @xmath13 and outputs in @xmath14 for alice , and inputs in @xmath15 and outputs in @xmath16 for bob . we will frequently consider bipartite distributions that are non - signalling , which we will refer to as _ correlations_. a _ class _ @xmath17 of correlations is a subset of all possible bipartite conditional probability distributions defined by some property such that the set is closed under local operations by either party , in additon to all distributions in @xmath18 being non - signalling . we denote the subset of @xmath12 which is in the class @xmath17 by @xmath19 . here we deal with the following classes of correlations : a bipartite channel is in @xmath20 if it can be implemented by local operations alone there are no correlations between the two parties at all . correlations belong to @xmath21 if they can be obtained using ( classical ) shared randomness ( and local operations ) ; to @xmath22 ( shared entanglement ) if they can be obtained from local operations on a shared quantum state ; and to @xmath23 if the correlation is non - signalling in both directions : that is , the marginal distribution of alice s output is independent of bob s input and vice versa . each class in this list has a strictly weaker defining property than the last , so we have @xmath24 . if alice and bob are connected by a classical channel @xmath25 and have access to any correlation in class @xmath17 ( shared randomness , entanglement etc . ) then they can _ exactly simulate _ @xmath26 if there is a _ local _ protocol whereby alice takes an input @xmath27 and , through local operations and a single use of @xmath28 and any use of @xmath17 , bob produces an output @xmath29 , such that the conditional probability of @xmath30 given @xmath31 is exactly @xmath32 . to say that @xmath3 uses of @xmath28 can exactly simulate @xmath33 uses of @xmath34 means that @xmath35 can exactly simulate @xmath36 . on pairs consisting of a bipartite correlation @xmath37 and a classical channel @xmath25 we define a bilinear map @xmath38 which corresponds to ` wiring ' alice s output of @xmath39 to the input of @xmath28 and the output of @xmath28 to bob s input to @xmath39 to produce a new classical channel @xmath40 $ ] , with @xmath41 because of the time ordering involved , this only makes operational sense if @xmath17 is non - signalling from bob to alice , and if it is not then @xmath34 may not be a valid conditional distribution . see figure [ fig : wiring ] for a diagram of the operational meaning . set valued arguments to @xmath38 are given the natural interpretation as yielding the image sets of classical channels . $ ] : a pre - shared resource ( randomness , entanglement or maybe something non - physical ) is portrayed by a dotted line . alice and bob interact with this , resulting in the non - signalling correlation @xmath42 : alice goes first , and obtains @xmath43 which she inputs into the channel @xmath25 . then , based on the channel output @xmath44 , bob interacts with the correlation resource , obtaining an outcome @xmath30 . for example , if the resource is an entangled system , then w.l.o.g . both parties interactions consist in choosing from a set of generalised measurements to perform on their local system . this all results in the channel @xmath40 \in \ccs{q}{r}$ ] . ] since classes of correlations are closed under local operations , a channel @xmath26 can be exactly simulated by a single use of @xmath25 and correlations in @xmath17 if and only if @xmath45.\ ] ] now , we can ask for the optimal use of one channel to simulate another one in the presence of some class of correlations @xmath17 . in this paper , we shall concentrate on the simulation of perfect ( i.e. identity ) channels by noisy ones ( `` zero - error capacity '' ) and the reverse ( `` exact simulation cost '' ) . [ ozec ] for a classical channel @xmath25 , and free correlations from the class @xmath17 , let @xmath46 denote the maximum alphabet size @xmath47 such that one symbol from the alphabet can be sent without error using @xmath17 and a single use of @xmath28 : @xmath48}{x}{y}{[c ] } , \mc{n } ] \ } , \end{gathered}\ ] ] where @xmath49 is the classical identity channel on @xmath47 symbols . since clearly @xmath50 , fekete s lemma guarantees existence of the _ @xmath17-assisted zero - error capacity _ of a channel @xmath28 defined by @xmath51 in this paper `` @xmath52 '' is base 2 so this is the capacity in bits . we use `` @xmath53 '' for the natural logarithm . for a classical channel @xmath25 , and free correlations in class @xmath17 , let @xmath54 denote the minimum alphabet size @xmath4 such that perfect transmission of one symbol of the alphabet allows exact simulation of one use of the channel : @xmath55}{[k]}{y } , \id_k ) ] \}. \end{gathered}\ ] ] similarly , we define @xmath56 as the asymptotic rate at which perfect classical bits must be transmitted to perfectly simulate @xmath28 , if correlations in class @xmath17 are free . the existence of the limit is once more guaranteed by fekete s lemma , because @xmath57 is clearly submultiplicative : @xmath58 the classical reverse shannon theorem @xcite assures us that in a setting of asymptotically vanishing simulation errors , all channels can reversibly simulate each other when shared randomness between sender and receiver is freely available : the rate at which @xmath59 can simulate @xmath60 being the ratio of their shannon capacities , @xmath61 . this remains true when entanglement and even more general non - signalling resources are shared by sender and receiver . the exact coding and simulation problem will be shown to have a much more complex structure . in the next section we review some of the classical theory of zero - error coding , discuss the correlation assisted zero - error quantities @xmath62 and @xmath63 , and then show several separations between them for different classes @xmath17 of assisting correlation . the most striking results here are a complete solution for the non - signalling assisted case and the construction of channels where entanglement assists for the one - shot scenario ( i.e. where @xmath64 ) . in section [ sec : sim ] we explore the quantities @xmath65 , which we show to be all different ( in general ) for @xmath66 ; and the simulation rates @xmath67 , which turn out to be all the same for @xmath68 , and indeed are given by a simple formula . we even find a kind of combinatorial reverse shannon theorem for zero - error communcation / noisy channel simulation in the presence of general non - signalling correlations : the simulation rate minimised over all channels with the same pattern of zeroes as the matrix @xmath69 is the same the non - signalling assisted zero - error capacity of @xmath28 . we conclude with some open questions . we start with the least powerful resources , @xmath20 and @xmath21 . the former simply describes arbitrary encoding and decoding maps . shared randomness does nt change anything since any value of the shared randomness will have to yield a zero - error coding if the randomised protocol does . for the same reason nothing is lost by requiring deterministic encoding and decoding maps , so the coding can be fully specified by giving a subset of input symbols to use as codewords . thus we are in shannon s original zero - error setting @xcite , and we shall write @xmath70 and @xmath71 . the fundamental observation is that , for zero - error coding over a channel @xmath25 , two symbols can both be used as codewords only if they are not _ confusable _ , that is , only if the corresponding output distributions have disjoint support . therefore , a zero - error code is just a set of pairwise non - confusable input symbols in @xmath10 , and @xmath72 is the largest size of such a set . in general , it is not hard to see that for any of our resources @xmath17 , only the pattern of zeroes in @xmath69 can affect @xmath62 and @xmath63 , so that the zero / one matrix @xmath73 encodes all the relevant information . this motivates the introduction of the following combinatorial representations of channels . the _ hypergraph _ @xmath74 of a channel @xmath25 has vertex set @xmath10 and hyperedges @xmath75 capturing the equivocation of each output symbol @xmath76 . note that different output symbols can give rise to the same hyperedge , so that the number of hyperedges may be less than the number of output symbols . looking back at definition @xmath77 , let @xmath78 denote the probability distribution on bob s output from the correlation conditional on alice having input @xmath79 , bob having input @xmath44 , and alice having obtained output @xmath43 . this is not well defined if @xmath79 never occurs for @xmath43 , and in this case we set @xmath80 ( so it is not in fact a distribution ) . when , bob obtains an output @xmath44 he knows that there is non - zero probability that alice made input @xmath43 iff it belongs to the hyperedge @xmath81 . the correlation @xmath39 yields a zero - error coding iff @xmath82 for all @xmath83 whenever @xmath84 for every hyperedge @xmath81 in @xmath85 . therefore , a correlation assisted zero - error capacities depends only on the channel hypergraph . to compute the unassisted zero - error capacity an even coarser representation of the channel will suffice : [ cg ] the _ confusability graph _ @xmath86 of a channel @xmath25 has vertices @xmath10 and an edge between input symbols @xmath43 and @xmath87 iff they are confusable , i.e. @xmath88 . with this notation , @xmath72 is simply @xmath89 : the _ independence number _ of @xmath86 . clearly @xmath86 can be obtained from @xmath90 by taking the vertex set of @xmath91 as the vertex set of @xmath92 and joining vertices with an edge iff there is a hyperedge of @xmath91 containing both . on the other hand , given a graph @xmath92 with vertex set @xmath10 , there are generally many hypergraphs on @xmath10 which are mapped to @xmath92 by this rule . hypergraphs on a given vertex set form a lattice when ordered by inclusion of their sets of hyperedges . the supremum of the set of hypergraphs with confusability graph @xmath92 is the clique hypergraph of @xmath92 , @xmath93 , whose hyperedges are all of the cliques in @xmath92 . from the point of view of zero - error coding , extra hyperedges can only be a bad thing , and this represents the worst case : for all hypergraphs @xmath91 with a given confusability graph @xmath92 , @xmath94 . for two graphs @xmath95 with vertex sets @xmath96 their strong product @xmath97 is the graph on @xmath98 with an edge @xmath99 iff @xmath100 or @xmath101 or @xmath102 . in terms of confusability graphs , @xmath103 . for two hypergraphs @xmath104 with vertex sets @xmath105 and edges @xmath106 , ( @xmath107 ) , we define the product @xmath108 on vertex set @xmath109 to have the hyperedges @xmath110 . the hypergraph of a product channel is the product of the individual hypergraphs , and the clique hypergraph of a strong graph product is the product of the individual clique hypergraphs . the _ shannon capacity _ of a graph is the asymptotic behaviour of the independence number of the strong product of @xmath3 copies @xmath111{\alpha(g^{\ox n})}.\ ] ] the zero - error capacity of @xmath28 is the same quantity but measured in bits per channel use @xmath112 the smallest example where the supermultiplicitvity of @xmath113 is strict is the pentagon graph @xmath114 , for which @xmath115 but @xmath116 . shannon conjectured that @xmath117 , which was only shown to true by lovsz @xcite . determining whether @xmath72 is greater than a given integer @xmath4 is np - complete ( indeed it is trivially equivalent to @xmath4-clique ) . whether @xmath118 is larger than some number is not even known to be decidable . shannon @xcite found an upper bound on the zero - error capacity by considering _ feedback _ assistance . in this scenario , as soon as bob receives an output @xmath44 from the channel , alice gets to know this @xmath44 with perfect reliability . while this is no advantage if only a single use of the channel is made , it is sometimes useful given multiple uses ( an observation shannon attributes to elias @xcite ) . shannon goes on to give a general formula for the asymptotic feedback assisted zero - error capacity @xmath119 . it is zero whenever @xmath120 is zero ( i.e. whenever the confusability graph of the channel is complete ) but otherwise is precisely the _ fractional packing number _ of the channel hypergraph . [ defi : fractional ] a _ fractional packing _ of a hypergraph @xmath91 with vertex set @xmath121 is an assignment of non - negative weights @xmath122 to all vertices @xmath43 such that @xmath123 a _ fractional covering _ of a hypergraph @xmath91 with vertex set @xmath121 is an assignment of non - negative weights @xmath124 to all hyperedges @xmath125 such that @xmath126 ( for weights in @xmath127 we recover the combinatorial notions of packing and covering . ) the _ fractional packing number _ @xmath128 is the maximum total weight allowed in fractional packing of @xmath91 and the _ fractional covering number _ @xmath129 is the minimum total weight required for a fractional covering of @xmath91 . these are clearly dual linear programs , which for a channel hypergraph @xmath90 have the formulation @xmath130 note that the possibility of redundant hyperedges in this representation ( as compared with the purer one in terms of sets ) has no effect on either quantity . the fractional packing problem is always feasible . on the other hand , the fractional covering problem is feasible if and only if the union of all hyperedges covers @xmath10 . where both are feasible @xmath131 by the strong duality theorem for linear programs . in particular , this holds for a channel hypergraph , since the fractional covering problem is always feasible ( as every input symbol always results in _ some _ output symbol occurring ) . from the definition of @xmath132 , we have @xmath133 for any hypergraph with confusability graph @xmath92 . but it turns out that @xmath132 is even an upper bound on @xmath134 : [ prop : duality1 ] @xmath132 is multiplicative with respect to the direct hypergraph product : @xmath135 . to show multiplicativity , the strong duality means that it suffices to show supermultiplicativity of @xmath132 , and submultiplicativity of @xmath136 , i.e. @xmath137 these are easy , because it is straightforward to confirm that the tensor product of two feasible vectors @xmath138 and @xmath139 ( dual feasible vectors @xmath140 and @xmath141 ) for @xmath142 and @xmath143 , respectively , is feasible ( dual feasible ) for @xmath144 . therefore , for any integer @xmath3 , @xmath145 , so @xmath146 . but this bound is often not tight . for example for the pentagon @xmath114 , it yields @xmath147 ; the above two - copy consideration shows on the other hand that @xmath148 . the celebrated result of lovasz @xcite says that the lower bound is tight , @xmath117 . he proved this by introducing another , tighter , but still multiplicative relaxation for @xmath149 , denoted @xmath150 . now that we have reviewed the state of the art regarding @xmath151 and @xmath152 , we go on to present our complete solution for @xmath153 and @xmath154 . [ nszelp ] for a classical channel @xmath155 with hypergraph @xmath90 @xmath156 where @xmath157 is the fractional packing number of @xmath90 . being a linear program , this can be efficiently computed from the channel . furthermore , since @xmath132 is multiplicative , the ns - assisted zero - error capacity of a channel is @xmath158 @xmath25 can exactly simulate a @xmath159-message identity channel with non - signalling correlations if and only if there exists @xmath39 in @xmath160}{x}{y}{[g]}$ ] such that @xmath161 without loss of generality , we can assume that a simplified form of non - signalling correlation is used : suppose some @xmath39 satisfies the above condition . then the symmetry of the identity channel under simultaneous permutation of the input and output alphabets means that we can always construct a new @xmath162 which is symmetrised by the following ` twirling ' procedure @xmath163 where @xmath164 is the symmetric group of order @xmath159 and @xmath165 is the image of @xmath79 under the permutation @xmath166 . this clearly simulates the same channel as @xmath39 , but it is highly symmetric in that @xmath167 with this simplification in mind , we maximize @xmath159 such that a valid non - signalling correlation @xmath162 allows the simulation of a @xmath159-message identity channel . we enumerate the constraints on @xmath162 in terms of @xmath168 and @xmath169 . \(1 ) @xmath162 is a valid conditional probability distribution iff @xmath170 and @xmath171 \(2 ) the non - signalling condition from bob to alice is given by : @xmath172 for some @xmath173 , whereas the condition that alice can not signal to bob is @xmath174 ( 3 ) the resulting channel is the @xmath159-message identity iff @xmath175 eliminating @xmath168 using condition ( 2 ) , the full set of constraints ( in terms of @xmath169 and @xmath176 ) can be simplified : @xmath177 @xmath178 is the largest integer smaller than the largest real number @xmath159 satisfying these constraints , which we now show is the @xmath179 of the theorem . defining @xmath180 , the largest feasible value of @xmath159 is @xmath181 by a simple application of the linear - fractional programming technique @xcite this optimisation can be recast as a linear program : making the substitutions @xmath182 , @xmath183 , @xmath184 yields @xmath185 this is equivalent to the linear program @xmath186 the would just be a reorganisation , except that we have also replaced the equality constraints on the second line with inequalities . this does nt change the value of the linear program : in any optimal solution with the inequalities , the sum over @xmath187 will be at least one . therefore , lowering the values of the @xmath188 , a solution to the lp where the equalities hold can be found which has the same objective value . finally , note that the @xmath188 are redundant in the above formulation . indeed , we may always set @xmath189 unless we are forced to take @xmath190 due to @xmath191 or , equivalently , due to @xmath192 . therefore , @xmath193 precisely the fractional packing number of @xmath90 . [ cor : cns - lower bound ] if a channel @xmath28 with @xmath3 inputs has at most @xmath33 non - zero entries @xmath69 for each @xmath44 ( i.e. the hyperedges of the equivocation graph are all of size @xmath194 ) . then , @xmath195 the _ proof _ is by checking that the assignment @xmath196 is feasible . we now show that @xmath197 can be arbitrarily larger than @xmath198 . in fact , there are channels for which the latter is @xmath199 while the former is positive ! let @xmath200}{m}$ ] denote the set of all size-@xmath33 subsets of @xmath201 $ ] . for all @xmath202 , define the channels @xmath203}{\binom{[n]}{m}}$ ] , each maps @xmath204 $ ] to a random subset of @xmath201 $ ] of cardinality @xmath33 containing @xmath43 , i.e. @xmath205 for all these channels , @xmath206 because any two inputs @xmath43 and @xmath87 are contained in a common set , hence they are confusable . ( in other words , the confusability graph is the complete graph @xmath207 . ) on the other hand , by corollary [ cor : cns - lower bound ] , all of these channels have @xmath208 , a strictly positive non - signalling - assisted capacity . the smallest parameters for which this effect can be seen on the single - shot level are @xmath209 and @xmath210 : @xmath211 is a channel with @xmath212 inputs and @xmath213 outputs , and theorem [ nszelp ] gives @xmath214 . how can this be ? define a non - signalling correlation @xmath215}{\binom{[4]}{2}}{\{0,1\}}$ ] as follows . alice s input is a bit @xmath79 , her output @xmath87 is a random element of @xmath216 $ ] . bob s input is a subset @xmath217}{2}$ ] . if @xmath218 then bob s output bit @xmath219 is @xmath79 and otherwise is @xmath220 . clearly , bob s output is independent of alice s input and vice versa so it is indeed non - signalling . suppose alice wires her output into the channel @xmath211 ( so @xmath221 ) and bob uses the output of @xmath211 as his input to @xmath39 ( so @xmath222 ) . the behaviour of the channel ensures that @xmath223 will always contain @xmath87 and therefore bob s output @xmath219 will always be equal to @xmath79 . a bit is transmitted from alice to bob with perfect reliability and that despite the fact that any two inputs of the channel can not be told apart with certainty by bob ! whenever @xmath224 , the non - signalling assisted zero - error capacity @xmath225 is precisely the same as the feedback assisted zero - error capacity . this is especially remarkable because the corresponding quantities for a finite number of channel uses are not necessarily the same , and the proofs of the capacity formulas are very different @xcite . also interesting is the fact that non - signalling proves strictly more powerful than feedback . in fact , when the capacities differ , the feedback assisted capacity must be zero . in @xcite we show that , like @xmath72 and unlike @xmath178 , the one - shot ( and hence also , asymptotic ) entanglement assisted zero - error capacity depends only on @xmath86 . an immediate corollary of this is that if @xmath226 then @xmath227 . proposition [ pw - g ] of the appendix , shows that these facts hold for assistance by any class of correlations with a certain operational property , which is possessed by @xmath22 but not @xmath23 . here , for the reader s convenience we repeat the proof of @xcite . [ sea ] for any channel @xmath228 with inputs @xmath10 and outputs @xmath11 , @xmath229 subject to the constraint that there exists a density matrix @xmath230 and positive semidefinite operators @xmath231 for all @xmath232 $ ] , @xmath233 , on some hilbert space such that , @xmath234 consequently , @xmath235 depends only on @xmath86 . we call the shared entangled state @xmath236 . without loss of generality , to send message @xmath79 , alice performs a measurement with povm elements @xmath237 , and with probability @xmath238 \;)$ ] , obtains outcome @xmath43 . conditional on the knowledge of @xmath79 and @xmath43 , the residual state of bob s system is @xmath239 . letting @xmath240 , for all messages @xmath79 @xmath241 reflecting the fact that without information from the classical channel , bob has no idea which message alice sent ( i.e. causality is respected ) . conversely , any set of positive operators @xmath231 which satisfy this condition for some @xmath230 can be realised by a suitable choice of @xmath236 and generalised measurements . alice puts the outcome @xmath43 into the channel @xmath28 . bob obtains the channel output @xmath44 , in addition to a quantum state left in his half of the entangled system . this bipartite state on bob s side is given by : @xmath242 the encoding works if and only if bob can distinguish perfectly between all the @xmath243 , i.e. for all distinct @xmath244 $ ] @xmath245 entanglement can still help , though . as shown in theorem [ kschan ] ( and previously in @xcite ) there are channels with @xmath246 . whether there are channels @xmath28 exhibiting an asymptotic separation @xmath247 remains an open question at this time . the efficiently computable formulae for @xmath153 and @xmath154 derived in the previous section provide upper bounds on entanglement assistance in both the one shot and asymptotic cases , but a tighter bound is known : duan _ et al . _ @xcite have defined a generalisation of the lovsz theta function @xcite for quantum channels . it is multiplicative for tensor products of channels and reduces to the classical lovsz theta function when the channel is classical . they show that this function is an upper bound on the entanglement assisted one - shot zero - error capacity ( for sending classical messages ) for any _ quantum _ channel . therefore , the classical theta function is an upper bound on @xmath248 for classical channels . a short and direct proof of this fact was derived independently by beigi @xcite . the bound is a strict improvement over the fractional packing bound and since @xmath249 , like @xmath250 , is multiplicative , it too can be immediately applied to the asymptotic rate : @xmath251 in terms of trying to decide whether separations exist between @xmath252 and @xmath152 , this is a rather frustrating result because @xmath253 is typically also the best bound we have on @xmath152 ! exceptions to this have been found , by haemers @xcite , but then the problem is to determine whether entanglement assisted protocols exist which beat the best known upper bound on @xmath152 for those special cases , and even then , only a positive answer would settle the general problem . another intriguing corollary of the result is that for the channels with @xmath254 made according to our construction from @xcite , the lovsz theta function coincides exactly with the _ lower _ bound on @xmath248 provided by the explicit protocol we give , so for these channels we know the precise value of @xmath255 and furthermore that it is achieved by repeating the optimal protocol for a single use of the channel . we now review the proof of the above statement as well as the construction from @xcite to which it applies . let @xmath92 be a graph with vertex set @xmath10 . an _ orthonormal representation _ @xmath256 of @xmath92 in @xmath257 is an assignment of unit vectors in @xmath257 to the vertices of @xmath92 such that if two vertices connected by an edge then their assigned vectors are orthogonal ( where orthogonality is with respect to the usual inner product @xmath258 ) : @xmath259 [ partg ] suppose that @xmath92 is a graph with an orthonormal representation in @xmath257 whose vertices can be partitioned into exactly @xmath31 cliques @xmath260 each of size @xmath261 . then there is a one - shot zero - error communication protocol assisted by a rank-@xmath261 maximally entangled state , which shows that @xmath262 . also , @xmath263 , and since @xcite proved that @xmath264 . therefore , @xmath265 . first , we describe the entanglement assisted protocol . alice and bob share @xmath266 , with @xmath267 the computational basis vectors for each party . the @xmath31 cliques of size @xmath261 which partition the vertices of the graph correspond to @xmath31 complete orthonormal bases for @xmath257 given by @xmath268 for @xmath269 to @xmath31 . to encode the message @xmath79 , alice measures her half of the shared state along the basis @xmath270 ( obtained by conjugating each state in @xmath271 ) . if the outcome corresponds to @xmath43 , bob s subsystem is left in the state @xmath272 . , then , the postmeasurement state is @xmath273 . ] alice inputs @xmath43 to the channel . bob s output @xmath44 from @xmath28 tells him a clique @xmath81 in @xmath92 that contains @xmath43 , which is not necessarily one of cliques in the partition . so bob s subsystem must be in one of the corresponding set of _ mutually orthogonal _ states @xmath274 . therefore , he can perform a projective measurement on his subsystem to determine exactly which state he has , from which he can deduce @xmath43 and , _ a fortiori _ , the symbol @xmath275 $ ] which alice chose , with certainty . second , to obtain @xmath150 , note that it can only increase if edges are removed , and it is multiplicative under strong graph product we have @xmath276 where @xmath207 and @xmath277 are the complete and empty graphs on @xmath3 , which have lovsz theta values of @xmath278 and @xmath3 , respectively . using the result from @xcite that @xmath264 , and putting both parts together , @xmath279 [ ksset ] we call a set @xmath280 of @xmath31 complete orthogonal bases @xmath281 for @xmath257 a _ ks basis set _ , if it is impossible to pick one vector from each basis so that no two are orthogonal . that such sets exist is a simple corollary of the kochen - specker theorem @xcite . an example of a ks basis set with 6 bases for @xmath282 taken from a proof of the kochen - specker theorem by peres @xcite is given in @xcite . [ kschan ] for any ks basis set @xmath280 in @xmath257 of @xmath31 bases , one can construct a classical channel @xmath283 ( with @xmath284 input symbols ) with @xmath285 and @xmath286 . construct the graph @xmath287 on @xmath288\times[d]$ ] with @xmath289 connected to @xmath290 iff @xmath291 and @xmath292 are orthogonal . clearly , @xmath287 partitions into @xmath31 cliques corresponding to the @xmath31 bases in @xmath293 , so @xmath294 , and if there was an independent set in @xmath295 of size @xmath31 , it would have to have one element in each of the @xmath31 cliques . but this would correspond to a selection of one vector in each basis in @xmath293 such that no two are orthogonal , in contradiction to the fact that @xmath293 is a ks basis set . letting @xmath296 be a channel with confusability graph @xmath297 , we have just shown that @xmath298 . on the other hand , since @xmath299 clearly defines an orthonormal representation of @xmath287 in @xmath257 , theorem [ partg ] tells us that @xmath300 ( and that this can be achieved using a rank-@xmath261 maximally entangled state ) . this section concerns the `` reverse '' problem to the zero - error channel coding problem : how much zero - error communication is _ required _ to exactly simulate a _ noisy _ channel . it will turn out that the one - shot communication cost can differ wildly between availability of no correlation , shared randomness , entanglement and non - signalling resources . however , in the many - copy limit they all turn out to give the same rate , as long as shared randomness is available . under a relaxed ( namely : combinatorial ) notion of channel simulation , we find complete reversibility between non - signalling assisted zero - error channel coding and channel simulation . what does it mean to simulate a channel @xmath25 ? with a @xmath4 symbol identity channel and no other correlations between sender and receiver the most general protocol is simply this : alice applies a local channel @xmath301}$ ] and sends to bob the result , on which he applies a channel @xmath302}{y}$ ] . the composition should be the desired channel @xmath303 . [ thm : no - resources - sim ] for a channel @xmath25 , @xmath304 equals the _ positive - rank _ @xcite of the transition probability matrix @xmath69 , i.e. the smallest number @xmath4 of probability distributions on @xmath11 such that their convex hull contains all of the output distributions @xmath305 . since positive - rank is lower bounded by linear rank , we get the following lower bounds : @xmath306 the latter because the rank is multiplicative . for instance , the channel @xmath307}{[n]}$ ] with @xmath308 if @xmath309 and @xmath310 if @xmath311 , will have @xmath312 , the same as the perfect channel , even though both its shannon and zero - error capacities are much lower . the set of channels one can perfectly simulate by sending one of @xmath4 symbols when arbitrary shared randomness is available is simply the convex hull of the set ( just described ) that can be achieved without shared randomness : [ ksr ] for a channel @xmath25 , @xmath313 is the minimum integer @xmath4 such that @xmath314 where we view @xmath315 naturally as a subset of @xmath9 . in fact , on the right hand side , we may replace the sets @xmath315 with their corresponding subsets of deterministic channels . as matrices , these are zero / one stochastic matrices , with rank @xmath316 and a channel @xmath28 has @xmath317 iff its matrix is a convex combination of these . any protocol to simulate @xmath28 exactly using shared randomness and @xmath4 messages amounts to writing @xmath28 as a convex ( probability ) combination of product channels , @xmath318 with @xmath319}$ ] , @xmath320}{y}$ ] , and @xmath321 . since the extreme points in the set of channels are the deterministic channels , we may push the randomness involved in forming a stochastic map into the shared randomness . but since @xmath322 has only @xmath4 outputs symbols and @xmath323 is deterministic , also the composition @xmath324 can have only @xmath4 possible output symbols forming a subset @xmath325 , @xmath326 . in other words , @xmath28 is a convex combination of deterministic channels in @xmath327 , for @xmath4-element subsets @xmath328 . just as in the case of zero - error communication , making non - signalling correlations freely available gives a very tractable structure to the problem of perfectly simulating noisy channels and the one - shot communication cost @xmath329 has a correspondingly simple form : it is the smallest integer greater than or equal to a certain simple norm of the conditional probability matrix . this norm is multiplicative under tensor products , so the corresponding asymptotic rate @xmath330 is just its logarithm . [ nsrate ] for a channel @xmath25 , @xmath331 and since @xmath332 , ( note that this function is a norm on stochastic matrices ) , the corresponding asymptotic rate is just @xmath333 if @xmath28 is in @xmath9 , it can be simulated with a @xmath4-input identity channel and non - signalling correlations if and only if there exists @xmath39 in @xmath334}{[k]}{y}$ ] such that @xmath335 again the twirling procedure in the proof of theorem [ nszelp ] simplifies things , but now the symmetry is in the identity channel used for the simulation . defining @xmath336 where @xmath337 is the symmetric group of order @xmath4 and @xmath165 is the image of @xmath79 under the permutation @xmath166 yields a correlation which simulates the same channel when used in place of @xmath39 ( summing over @xmath338 is the same as summing over @xmath339 ) , but where @xmath340 we now list the conditions on @xmath162 ( in terms of @xmath168 and @xmath169 ) : \(1 ) the correctness of the simulation is given by eq . : ( 2a ) the conditions for no signalling from alice to bob are @xmath342 } p'(z , y|x,\hat{z } ) = \sum_{z\in [ k ] } p'(z , y|x',\hat{z})\ ] ] for all @xmath343 , which reduce to @xmath344 so we write @xmath345 . clearly we require that @xmath346 and @xmath347 ( in fact , @xmath348 is just the marginal distribution of the output @xmath44 which is independent of both inputs , like in the pr box ) . ( 2b ) the conditions for no signalling from bob to alice , @xmath349 which reduce to @xmath350 , are already ensured by the condition that @xmath351 and @xmath352 for all @xmath43 ( these mean that @xmath353 for all @xmath43 ) . \(3 ) the only other constraint is that the entries of @xmath169 are positive . putting these constraints together , we see that a suitable @xmath162 ( and hence @xmath39 ) exists if and only if there is a probability vector @xmath176 such that the resulting @xmath169 matrix has positive entries , i.e. @xmath354 for all @xmath355 . such a @xmath176 is possible if and only if @xmath356 . it is not hard to verify directly that the bit rate needed to perfectly simulate @xmath28 with free non - signalling correlations is greater than the shannon capacity of @xmath28 . if the channel input is the random variable @xmath357 where @xmath358 and the resulting channel output is the random variable @xmath359 then @xmath360 the shannon capacity of @xmath28 is obtained by maximising the left - hand side over all input distributions @xmath361 . shared randomness is one type of non - signalling correlation so it is clear that @xmath363 . it turns out that there can be an arbitrarily large gap between these two costs . this is the case for the `` universal channels '' to be defined below . [ uchans ] recall that the set of all size-@xmath33 subsets of @xmath201 $ ] is denoted by @xmath364 \choose m}$ ] . the _ universal channel _ @xmath365 is the channel in @xmath366 \choose m}}{[n]}$ ] with , @xmath367 in words , the channel takes as input a set @xmath368 \choose m}$ ] and outputs a random element of that set . note that @xmath369 introduced earlier in this section is a special case of the universal channel with @xmath370 . the universal channels have a great deal of symmetry . the symmetric group @xmath371 acts on both the input and the output alphabet of @xmath365 : on the latter naturally as permutations of the symbols ( written @xmath372 ) , on the former as simultaneous permutations of all elements in the sets @xmath373 $ ] ( written @xmath374 ) . with these actions , @xmath365 is @xmath371-covariant : @xmath375 a beautiful consequence of the covariance is that it specifies @xmath365 almost uniquely : @xmath365 is the only channel satisfying eq . and in addition @xmath376 if @xmath377 . to simulate @xmath365 with zero error when assisted by arbitrary non - signalling correlations , theorem [ nsrate ] shows one needs a noiseless channel of @xmath378 many symbols , and this is sufficient . the minimal asymptotic rate of communication needed given free non - signalling correlations is @xmath379 . on the other hand , when only shared randomness is available , the communication cost can be much higher : [ ksr ] for any @xmath380 , @xmath381 we first show that @xmath382 is not sufficient by contradiction . recall theorem [ ksr ] and consider an element @xmath28 from @xmath315 , with @xmath328 , @xmath383 in the convex decomposition of @xmath365 that comes with a strictly positive weight @xmath361 . that means , for any input @xmath43 , @xmath384 in the sense of element - wise ordering of the probability vectors . choosing @xmath385\setminus z$ ] which has cardinality @xmath33 leads to the desired contradiction : restricted to @xmath293 , @xmath386 is the zero vector ( see the definition ) , whereas @xmath305 has all of its probability mass in @xmath293 . on the other hand , there is a protocol that uses only @xmath387 messages : the shared randomness is a uniformly distributed subset @xmath388 \choose n - m+1}$ ] . on input @xmath43 , alice selects a uniformly random element @xmath389 , which is non - empty by the pigeonhole principle . to send @xmath44 to bob , she needs only a number from @xmath278 through @xmath387 to specify where @xmath44 occurs in @xmath16 . clearly , this protocol , and hence the simulated channel , has the same @xmath371-covariance as @xmath365 . furthermore , the simulated channel assigns zero conditional probability to all @xmath390 for input @xmath43 . thus , the simulated channel must be @xmath365 . these universal channels provide simple and highly structured examples for separating @xmath362 , @xmath391 , and @xmath391 . we have already seen that for @xmath370 , @xmath392 but prop . [ ksr ] says that @xmath393 . in a different regime , for example when @xmath3 is even and @xmath394 , @xmath395 but @xmath396 . in both cases , the separation is of order @xmath3 which is maximal given the size of the input alphabet . the possibility of large separations between @xmath329 and @xmath313 , raises the question of where the power of the intermediate shared entanglement class fits in between the two . while we do not have a general understanding of this matter yet , we can at least give examples where entanglement beats shared randomness and cases where general non - signalling correlations can beat entanglement . let @xmath397 denote the ternary erasure channel with transmission probability @xmath361 : @xmath398 we can use the ideas of the appendix to show that non - signalling correlations can beat shared entanglement for one - shot simulation of @xmath399 . whereas @xmath400 , @xmath401 . therefore , using a single ( perfect ) bit of communication , strictly more channels can be simulated if generalised non - signalling correlations are available rather than entanglement . by using the twirling procedure of theorem [ nsrate ] ( which can be assumed w.l.o.g . whenever shared randomness is available ) , and considering the simplified non - signalling constraints which result , it is not hard to see that exact simulation of @xmath399 with a single bit is equivalent to the ability to realise a particular non - signalling correlation @xmath402 defined , in the notation of theorem [ nsrate ] , by @xmath403 , @xmath404 and @xmath405 . this @xmath402 must be in the class of available correlations . applying the observations of the appendix , and looking at the 8 conditional channels @xmath406 , one finds that they are all pair - wise distinguishable and therefore @xmath402 can not be in the class @xmath22 , by proposition [ pw - bound ] . strictly more classical channels can be simulated using shared entanglement than can with shared randomness . in particular , there are channels @xmath28 for which @xmath407 but @xmath408 . we construct these channels and demonstrate the separation in three steps . first , we show that @xmath399 can be simulated using one bit of communication and a certain non - signalling correlation called the `` pr - box '' . then , using the same protocol , but replacing the pr - box by a weaker correlation obtainable via shared entanglement , we write down the channel @xmath28 that is being simulated . finally , we explain why @xmath28 can not be simulated with shared randomness and one bit of communication . the pr - box ( introduced by popescu and rohrlich @xcite ) is a particular correlation @xmath409 given by : @xmath410 in other words , the outputs @xmath411 are random bits except for the constraint @xmath412 . we let @xmath79 to be the message sent via the classical channel . we first give an explicit method for simulating @xmath399 using a bit of communication and a single use of a pr box . bob always chooses his pr - box input @xmath413 to be the channel output @xmath79 and outputs @xmath414 for the simulation . alice has a message @xmath415 . if @xmath416 , she sets her pr - box input to be @xmath417 . the pr - box outputs a random bit @xmath418 . she sends @xmath419 to bob , who puts it in the pr - box and obtains @xmath420 . thus , bob outputs @xmath421 or @xmath422 randomly . if @xmath423 , alice sets @xmath424 and her pr - box outputs a random bit @xmath418 . she chooses @xmath419 . for either value of @xmath418 , @xmath425 . thus , bob outputs @xmath421 or @xmath426 randomly . if @xmath427 , alice sets @xmath417 and @xmath428 . thus , bob outputs @xmath421 or @xmath429 randomly . identifying the outputs @xmath421 , @xmath422 , @xmath426 and @xmath429 with the erasure symbol e , @xmath199 , @xmath278 , and @xmath430 respectively , the above simulates @xmath399 perfectly . in the second step , we generalize the pr - box to correlations @xmath431 given by : @xmath432 where @xmath433 . note that @xmath434 is the pr - box . the pr box can be approximated using a maximally entangled pair of qubits . an optimal approximation , in terms of the chsh violation , yields the correlation @xmath431 with @xmath435 . if this entanglement based approximation of the pr box is substituted into the protocol given above , the resulting channel @xmath436 is given by @xmath437 where @xmath438 and @xmath439 . finally , we check with a computer that @xmath436 is not a convex combination of rank - two , zero - one , stochastic matrices and so , according to theorem [ ksr ] , ca nt be exactly simulated using one bit of communication if only shared randomness is available . among the results of this section so far are channels proving separations between the communication costs of ns- , se- and sr - assisted channel simulation for a single channel use . in the case of ns vs. sr , the universal channels of definition [ uchans ] show that this gap can be arbitrarily large . despite this , we will prove in theorem [ srns ] that when simulating many uses of a channel , a protocol using shared randomness can achieve an _ asymptotic _ rate of communication as low as the optimal rate with non - signalling assistance derived in theorem [ nsrate ] . since the rate with entanglement assistance is sandwiched between these two rates , it follows that @xmath440 , for all channels @xmath28 . the proof is structured into two steps . first , we show that the asymptotic equality discussed above holds for all the universal channels . then , roughly speaking , any channel can be exactly simulated by a universal channel with the same value of @xmath441 . it is this ability of the channels @xmath365 that earns the name `` universal . '' we need a lemma for this proof : [ tset ] we call a set @xmath442^q$ ] _ @xmath33-touching _ if @xmath443 \choose m } \quad \catchset \cap ( x_1\times x_2\times\cdots\times x_q ) \neq \emptyset.\ ] ] there is an @xmath33-touching set of cardinality @xmath444 . if a set is populated by picking @xmath445 elements of @xmath201^q$ ] picked uniformly at random ( with replacement ) , the probability that it is not @xmath33-touching is bounded above by @xmath446 with the simple estimates @xmath447 and @xmath448 , @xmath449 so a set with the desired property and cardinality must exist . indeed , the probability that a set chosen in the way described above is nt @xmath33-touching is exponentially small in @xmath450 . [ uchanrate ] for any universal channel @xmath365 , @xmath451 by definition , @xmath452 , so it suffices to exhibit a protocol using only shared randomness that achieves this bound . to be precise , for @xmath31 copies of the channel , we prove the existence of such a protocol which uses the transmission of one of @xmath453 symbols . the rate is @xmath454 , which approaches @xmath379 as @xmath455 . the protocol works as follows : alice and bob agree on an @xmath33-touching set @xmath456 of size @xmath4 ( see lemma [ tset ] ) . they share randomness in the form of @xmath31 uniformly random permutations @xmath457 . on input @xmath458 alice picks a uniformly random element @xmath459 , where @xmath460 is the set @xmath456 with its elements permuted according to @xmath461 in coordinate @xmath462 . the intersection is guaranteed to exist because @xmath463 is also @xmath33-touching . to send @xmath44 , she only needs a number from @xmath278 through @xmath4 to specify the location within @xmath464 since the latter is known to bob . this protocol evidently simulates an @xmath465-covariant channel with the property that @xmath466 whenever @xmath467 . as discussed before , this means that the simulated channel must be @xmath468 . [ srns ] for any channel @xmath25 , @xmath469 first , suppose all the entries of @xmath470 are rational numbers , with common denominator @xmath471 , so that @xmath472 for integers @xmath473 . split up each output symbol @xmath44 into @xmath474 many , denoted @xmath475 , with @xmath476 . now define a new channel by letting @xmath477 be either @xmath199 or @xmath478 , in such a way that @xmath479 with the projection map / channel @xmath480 . clearly @xmath481 is a sub - channel ( i.e. a restriction on the input alphabet ) of the universal channel @xmath482 with @xmath483 . it can therefore be exactly simulated using shared randomness by the protocol of proposition [ uchanrate ] . this requires asymptotic communication rate @xmath484 which is precisely the lower bound set by @xmath330 , so the claim holds for rational @xmath28 . for the general case , pick a large integer @xmath471 and let , for all @xmath233 , @xmath485 , @xmath486 . now adjoin new elements @xmath87 ( @xmath233 ) to the output alphabet , i.e. define @xmath487 and a new channel @xmath488 with @xmath489 now , @xmath28 can be simulated by @xmath490 using post - processing by bob only : if @xmath485 is obtained , then it is left alone ; if @xmath491 is seen , then bob uses local randomness to output @xmath44 with probability @xmath492 so , extending @xmath169 to a proper channel by letting @xmath493 for @xmath494 iff @xmath495 , we have @xmath496 . second , the cost of simulating @xmath490 is @xmath497 letting @xmath498 , this rate approaches @xmath499 . to illustrate the idea , if @xmath500 denotes the partial order on channels `` @xmath501 is equal to @xmath502 , for some channels @xmath503 , @xmath504 '' , the proof uses @xmath505 to show that @xmath506 . the classical reverse shannon theorem @xcite also yields an exact simulation of the noisy channel using shared randomness . the difference to our result is explained by the different way to account for the communication : whereas @xcite shows that the _ expected _ rate of communication ( with respect to the shared randomness used in the protocol ) is the normal shannon capacity of the channel , here we consider the much more stringent _ worst case _ communication cost . looking over the formulas for @xmath154 and @xmath441 of a channel @xmath28 , we notice that the former only depends on the channel hypergraph , while the latter actually involves the transition probabilities . hence it is not surprising that the former is typically strictly smaller than the latter . however if we are content with the simulation of any channel that has the same hypergraph , we recover reversibility : let @xmath25 with channel hypergraph @xmath90 ( having hyperedges @xmath507 . then , @xmath508 where @xmath509 is the fractional covering number of the hypergraph of the channel . since the fractional covering number is equal to the fractional packing number @xmath128 , this minimum rate is also equal to @xmath225 . recall the formula for @xmath330 : it is the logarithm of the value of the following linear program ( all variables understood as non - negative ) : @xmath510 the additional minimisation over channels with prescribed hypergraph @xmath91 is also a linear program : @xmath511 but this is evidently equivalent to @xmath512 which is exactly the fractional covering number @xmath129 . for the other statements see proposition [ prop : duality1 ] . let us summarise the results ( both our own , and others ) discussed in this paper . for zero - error communication , we found both the one - shot and asymptotic non - signalling assisted capacities , upper bounding the chains of operationally obvious inequalities : @xmath513 @xmath514 these upper bounds on the entanglement assisted capacities from non - signalling are improved upon by the results of @xcite and @xcite which show that the lovsz theta bound applies even in the entanglement assisted case : @xmath515 while we proved that @xmath516 can be strict , we do nt yet know whether the same can be said of the asymptotic rates , and regard this as one of the main open problems . in the reverse problem of exactly simulating noisy channels , the non - signalling assisted case was again completely soluble , providing _ lower _ bounds on the chain @xmath517 for each inequality in this chain of one - shot costs , a channel showing that it can be strict was exhibited . some open questions remain regarding the potential sizes of these separations ( see section [ sec : sim ] ) . for the asymptotic rates of communication things were shown to be simpler : while large gaps can exist between the costs with free shared randomness and without ( @xmath518 ) , given free correlations from _ any _ class of non - signalling correlations which contains shared randomness the rates are equal : @xmath519 we would like to thank nicolas brunner , runyao duan , tsuyoshi ito , ashley montanaro , marcin pawowski , paul skrzypczyk and stephanie wehner for useful discussions . c. h. bennett , p. shor , j. smolin , a. v. thapliyal , `` entanglement - assisted capacity of a quantum channel and the reverse shannon theorem '' , ieee trans . theory 48 , 2637 - 2655 , ( 2002 ) arxiv : quant - ph/0106052 . w. haemers , `` on some problems of lovsz concerning the shannon capacity of a graph '' , ieee trans . theory * 25*(2):231 - 232 ( 1979 ) . w. haemers , `` an upper bound for the shannon capacity of a graph '' , coll . jnos bolyai * 25*:267 - 272 ( 1978 ) . if alice makes an input @xmath533 to her side of a bipartite correlation @xmath534 , and obtains the output @xmath43 then the conditional distribution on bob s side is a classical channel @xmath535 where @xmath536 is simply the conditional distribution @xmath537 given by bayes rules , but written differently to emphasise the fact that we regard @xmath533 and @xmath43 as fixed . similarly , there are @xmath538 such conditional channels @xmath539 on alice s side . we say that a class of bipartite correlations @xmath17 has property @xmath540 if the existence of a correlation @xmath541 , and @xmath542 , satisfying @xmath543 implies the existence of another correlation , @xmath544 , which is identical to @xmath39 when restricted to the input alphabets of @xmath39 , @xmath545 but which has an extra input @xmath546 on bob s side such that @xmath547 @xmath17 has property @xmath548 if it satisfies the same condition with the roles of the parties reversed . if @xmath17 has property @xmath540 and property @xmath548 then we simply say it has property @xmath549 . in other words , if a correlation @xmath39 belongs to a @xmath540 class @xmath17 , and the graph induced on @xmath550 by the pair - wise distinguishability relation between the conditional states @xmath551 associated with the vertices contains a clique @xmath14 , then there is another correlation @xmath162 in @xmath17 which behaves like @xmath39 except that bob has some extra output symbols ( possibly ) , and one new input symbol @xmath546 , which when input , yields pair - wise orthogonal output distributions on @xmath11 for all elements of @xmath14 so that they can be perfectly distinguished simultaneously . to illustrate this idea , we now show that the class @xmath23 of generalised non - signalling correlations is not @xmath549 : if alice and bob s shared correlation @xmath39 is the pr - box , then the conditional channels @xmath551 are given by @xmath552.\ ] ] these channels are all pair - wise distinguishable , but of course , the required input on bob s side depends on the pair . if ns were @xmath549 , then the existence of @xmath553 would imply the existence of another correlation in @xmath23 where a single input on bob s side would suffice to distinguish the @xmath212 residual states . but obviously this would allow bob to determine alice s input , so this is a contradiction : put another way , if a class is @xmath549 and contains the pr - box then it also contains signalling correlations . assuming w.l.o.g . that alice measures first : alice inputs @xmath533 ( corresponding to her measuring of some povm on her side ) and obtains outcome @xmath43 , leaving a residual state @xmath554 on bob s side . the conditional channel @xmath536 is given by @xmath555 so if @xmath556 then there must be some input @xmath413 on bob s side , corresponding to a povm with elements @xmath557 say , such that @xmath558 which implies that the residual states @xmath554 and @xmath559 are orthogonal ( i.e. have disjoint support ) . a clique of pair - wise distinguishable conditional channels on bob s side therefore corresponds to a clique of _ mutually _ orthogonal residual states on his side . therefore , there is a single measurement which perfectly distinguishes all members of the clique , which we can obviously use to construct a correlation @xmath162 in the class of correlations @xmath22 with the required properties . [ pw - g ] if @xmath17 is a pw class of correlations , then the one - shot ( and hence asymptotic ) @xmath17-assisted zero - error capacities @xmath46 and @xmath528 of a channel @xmath28 only depend on the confusability graph @xmath86 . write @xmath563 if there is a single input @xmath44 on bob s side such that @xmath564 : in words , @xmath563 means that if bob knows that the channel input was one of @xmath43 or @xmath87 then he can distinguish which @xmath79 alice chose by making input @xmath44 to his side of the correlation . when bob gets output symbol @xmath44 from the channel , let @xmath81 denote the set of possible inputs . he can decode @xmath79 perfectly iff @xmath565 . if we draw a graph on @xmath10 with edges labelled by outputs @xmath11 , with a @xmath44-edge between @xmath43 and @xmath87 iff @xmath566 , then ( ignoring edge labels and multiplicities ) this graph must contain @xmath86 . recalling the discussion after definition [ cg ] , we know that @xmath567 . by the pw property of @xmath17 it must be possible to find a new correlation in @xmath17 such that if bob knows that @xmath43 was in _ any _ clique in this graph , then he can still determine @xmath79 . so the @xmath17-assisted zero - error capacity of the clique hypergraph of @xmath92 is at least @xmath47 . [ pw - bound ] for a correlation @xmath534 , let @xmath569 ( @xmath570 ) be the graph on the @xmath571 ( @xmath538 ) conditional channels on bob s ( alice s ) side where edges denote pair - wise distinguishability . if @xmath39 belongs to a class which is both @xmath549 and non - signalling , then @xmath572 and @xmath573 where @xmath574 denotes the clique covering number . in particular , these bounds apply to the correlation class @xmath22 : the set of bipartite correlations which can be implemented using entanglement . let @xmath575 ( @xmath576 ) be a minimal clique covering of @xmath570 ( @xmath569 ) . suppose that @xmath39 is in @xmath17 which is @xmath549 and non - signalling . by repeated use of the definition of the @xmath549 property , @xmath17 contains a @xmath577 such that for @xmath578 @xmath579 this means that if alice inputs @xmath580 to @xmath162 and obtains output @xmath43 , and then tell s bob a clique in @xmath576 which contains @xmath581 , bob can determine @xmath581 exactly . in particular , he discovers alice s choice of input in @xmath13 without error , by alice transmitting one of @xmath582 messages . since non - signalling correlations ca nt increase the zero - error capacity of identity channels ( a simple consequence of theorem [ nszelp ] ) , if @xmath583 then @xmath162 can not be non - signalling ( and similarly if @xmath584 ) .

we present spectropolarimetry of sn 2009ip throughout the evolution of its 2012 explosion . during the 2012a phase , when the spectrum exhibits broad p - cygni lines , we measure a-band polarization of% at a position angle of , indicating substantial asphericity for the 2012a outflow . near the subsequent peak of the 2012b phase , when the spectrum shows signs of intense interaction with circumstellar material ( csm ) , we measure% and , indicating a separate component of polarization during 2012b , which exhibits a higher degree of asphericity than 2012a and an orthogonal axis of symmetry on the sky . around 30 days past peak , coincident with a substantial bump in the declining light curve , we measure% and another significant shift in . at this point , broad photospheric lines have again become prominent and exhibit significant variations in relative to the continuum , particularly he i / na i d. by 60 days past peak the continuum polarization has dropped below 0.2% , probably declining toward a low value of interstellar polarization . the results are consistent with a scenario in which a prolate ( possibly bipolar ) explosion launched during the 2012a phase impacts an oblate ( toroidal ) distribution of csm in 2012b . previous calculations that assumed spherical symmetry for the csm have substantially underestimated the required explosion energy , since only a small fraction of the sn ejecta appears to have participated in strong csm interaction . an ejecta kinetic energy of at least ergs is difficult to avoid , supporting the interpretation that the 2012 outburst of sn 2009ip was the result of a core - collapse explosion . [ firstpage ] circumstellar matter stars : evolution stars : winds , outflows supernovae : general supernovae : individual ( sn 2009ip )

1403.4240

interacting supernovae ( sne ) are stellar explosions that collide with dense circumstellar material ( csm ) produced by the progenitor star . these events are raising critical new questions about the final evolutionary phases of massive stars and the mass - loss episodes that ensue before core collapse ( smith & arnett 2014 ) . type - iin and ibn sne , in particular , are interacting sne that are characterized spectroscopically by the presence of relatively narrow emission lines of h and he in their spectra ( schlegel 1990 ; filippenko 1997 ; pastorello et al . 2008 ) , which arise from dense csm that becomes illuminated by the shock between the fast moving sn ejecta and slower moving csm ( chevalier & fransson 1994 ) . as such , observations of interacting sne probe the stellar progenitor s pre - sn mass - loss history , providing valuable information on its final evolutionary episodes . various lines of evidence show that interacting sne require eruptive pre - sn mass loss that is reminiscent of luminous blue variable ( lbv ) stars , like @xmath9 car , although observations indicate a wide range of mass - loss properties ( e.g. , see the review by smith 2014 ) . the eruptions are often detectable as extragalactic transients , commonly referred to as ` sn impostors ' ( van dyk 2000 ; smith et al . 2011a ; kochanek et al . 2012 ) , which can rival the luminosity of a sn and spectroscopically mimic sne iin . although the observational distinction between interacting sne and sn impostors is not always clear , a direct link connecting these phenomena has been established for several objects . the progenitors of sn 2006jc and sn 2011ht were detected undergoing luminous outbursts 12 yr prior to their core - collapse explosions ( foley et al . 2007 ; pastorello et al . 2008 ; fraser et al . sn 2010mc ( ofek et al . 2013a ) was also detected 12 months before dramatically brightening , although for this case it appears that the precursor event may actually have been the initial phases of the sn caught unusually early ( smith , mauerhan , & prieto 2014 ) . perhaps the most interesting transient observed to have undergone multiple phases of eruptive mass loss is sn 2009ip . originally classified as a sn ( maza et al . 2009 ) , this object was actually discovered during an lbv outburst that is , as a sn impostor ( smith et al . 2010 ; foley et al . 2011 ) . several years of continued activity were followed photometrically and spectroscopically , leading up to sn 2009ip s most extreme outburst in 2012 ( mauerhan et al . 2013 ; prieto et al . 2013 ; levesque et al . 2014 ; smith et al . 2013b ; smith , mauerhan , & prieto 2014 ; pastorello et al . 2013 ; fraser et al . 2013b ; margutti et al . 2014 ; graham et al . the 2012 event was comprised of two main components : the 2012a " phase , marked by an initial peak at @xmath10 mag that lasted for just over 1 month ; and the 2012b " phase , which began with a fast 10-day rise to a second peak of @xmath11 mag , followed by a bumpy decline down to a slowly declining floor in the light curve near @xmath12 mag . based on spectral similarities with known core - collapse sne , mauerhan et al . ( 2013a ) suggested that the relatively faint 2012a phase of sn 2009ip marked the initial stages of a core - collapse sn , while the subsequent 2012b brightening was the result of interaction between this sn and dense csm ejected during the earlier lbv outbursts . levesque et al . ( 2014 ) also shared the view that the 2012b brightening was the result of interaction between the 2012a outflow and existing csm . however , several authors have suggested potential alternatives to a core collapse explosion that sn 2009ip s 2012 evolution was possibly the result of one or more nonterminal outbursts ( pastorello et al . 2013 ; fraser et al . 2013b ; margutti et al . 2014 ) , perhaps caused by the pulsational pair - instability mechanism . the motivation for nonterminal scenarios has been based largely on the fact that the total radiated energy of the light curve ( @xmath13 erg ) can be explained by an explosion energy of @xmath14 ergs ( if spherical symmetry is assumed for the csm ) , and also because the late - time data were interpreted as looking different from what is expected for radioactive decay phases . more recently , smith , mauerhan & prieto ( 2014 ; hereafter smp14 ) showed that the 2012 light curve and spectral evolution are consistent with published models for core - collapse sne , which can be initially faint ( like sn 1987a ) owing to a relatively compact progenitor radius ( i.e. , a blue supergiant , as expected for an lbv , instead of a red supergiant ) , while the late - time characteristics could be explained if the mass of synthesized @xmath15ni was half that of sn 1987a . smp14 further argued that the radiated energy did not provide an argument against core collapse , since the csm is probably aspherical , leading to an inefficient conversion of kinetic energy into radiation . in line with the spectral modeling results and interpretation of levesque et al . ( 2014 ) , which are consistent with a toroidal distribution of csm , smp14 also proposed a disk - like distribution of csm , but further argued that significant asphericity is required by the fact that broad photospheric lines are still seen at late times , even after strong csm interaction ends . this requires that a large fraction of the total solid angle of ejecta was able to expand unimpeded by csm . smp14 also pointed out that the @xmath16100-day persistence of broad photospheric lines requires a large ejecta mass of at least a few m@xmath17 , and is incompatible with a 0.1 m@xmath17 shell , which would become optically thin much more quickly . a mass of a few m@xmath17 moving at high speeds ( @xmath168000 km s@xmath18 ) directly implies @xmath19 ergs of kinetic energy . in the case of sne iin , the system geometry is critical if the total radiated energy from csm interaction is to be used to infer the kinetic energy of the sn explosion . fortunately , spectropolarimetry can yield important clues about the geometry of sne , allowing us to stringently test the hypothesis of aspherical csm . here , we report multi - epoch spectropolarimetry of the 2012 evolution of sn 2009ip , from the end of the initial 2012a phase , through the peak of 2012b , and later into its decline . the results unambiguously demonstrate that the source of bright continuum emission during the 2012b phase ( i.e. , the csm interaction zone ) was aspherical , consistent with a toroidal or disk - like distribution of csm proposed by smp14 and levesque et al . furthermore , the results indicate that the initial 2012a phase has a separate polarization component having an axis of symmetry that is distinct from 2012b , which implies that the 2012a event did not create the csm responsible for the 2012b brightening . diminishing polarization at late times , in addition to similarities in wavelength - dependent polarization across lines for sn 2009ip and other sne , provides additional evidence for a persistent underlying sn photosphere . altogether , the available evidence supports the hypothesis of a sn explosion for sn 2009ip in 2012 , and argues strongly against nonterminal explosion models for this object ( at least those proposed thus far ) . spectropolarimetric observations were obtained at the following facilities utilised by the university of arizona : the 6.5 m multiple mirror telescope ( mmt ) on mt . hopkins during 2012 sep . 21 and 24 ( ut dates are used throughout this paper ) ; the kuiper 61-inch telescope on mt . lemmon during nov . 11 and 14 ; and the bok 2.3 m telescope on kitt peak during dec . 5 , 6 , and 7 . observations at these facilities all made use of the same ccd imaging / spectropolarimeter ( spol ; schmidt et al . the instrument contains a rotatable semiachromatic half - waveplate used to modulate incident polarization and a wollaston prism in the collimated beam to separate the two orthogonally polarized spectra onto a thinned , anti - reflection - coated @xmath20 pixel site ccd . spol was configured with a 600 line mm@xmath18 grating in first order . the slit selection was based on the seeing conditions and ranged from @xmath21 to @xmath22 . this setup resulted in spectral resolutions of @xmath23@xmath24 with useful wavelength coverage in the range @xmath2540007600 . a standard hoya l38 blocking filter was used to ensure that the first - order spectrum was not contaminated by second - order light for @xmath26 . the efficiency of the waveplate as a function of wavelength was measured and corrected at all epochs by inserting a fully polarizing nicol prism into the beam above the slit . the orientation of the slit on the sky was always set to a position angle of @xmath27 , aligned along north - south . a series of four separate exposures that sample 16 orientations of the waveplate yields two independent , background - subtracted measures of each of the normalised linear stokes parameters , @xmath28 and @xmath29 . on some nights several such polarization sequences of sn 2009ip were obtained and combined , with the weighting of the individual measurements based on photon statistics . the instrumental polarization of spol and the kuiper , bok and mmt telescopes is extremely small ( @xmath30% ) , as measurements of unpolarized standard stars over the past two decades have consistently shown . still , to confirm , we observed one or both of the unpolarized standard stars bd+28@xmath314211 and g191b2b ( schmidt et al . 1992b ) during each epoch , confirming that the instrumental polarization has to be less than 0.1% and that there is no low - level residual structure observed in the @xmath28 and @xmath29 spectra . the linear polarization position angle on the sky ( @xmath32 ) was determined by observing one or more of the interstellar polarization standards hiltner 960 , vi cyg # 12 , or bd+59@xmath31389 ( schmidt et al . 1992b ) during all epochs . the adopted correction from the instrumental to the standard equatorial frame for @xmath32 for all epochs was determined from the average position angle offset of the polarized standards . differences between the measured and expected polarization position angles were @xmath33 for all of the standard stars . the sep . 21 mmt observations were impacted by intermittent cloud cover , which degraded the signal - to - noise ratio of these observations . nonetheless , secure detections of polarization from sn 2009ip were obtained , owing to the dual - beam design of spol , which allows for variations in atmospheric transparency during integrations . measurements at all other epochs from kuiper and bok were obtained under clear conditions . from arizona latitudes , sn 2009ip was observed through airmass values of 2.052.32 . the shane 3 m reflector at lick observatory equipped with the kast spectrograph ( miller & stone 1993 ) was used on oct . 5 and 14 , and on nov . 6 and 14 . like spol , kast is a dual - beam spectropolarimeter that utilises a rotatable half - waveplate and wollaston prism . only the red channel of kast was used for spectropolarimetry . a gg455 order - blocking filter blocked all second - order light at wavelengths shortward of 9000 . observations were made with the 300 line mm@xmath18 grating and the 3 slit , yielding a spectral resolution of @xmath1620 . the orientation of the slit on the sky was always set to a position angle of @xmath27 , aligned along north - south . exposures of 900 s were obtained at each of four waveplate positions ( @xmath34 , @xmath35 , @xmath36 , and @xmath37 ) . for the oct . 14 , nov . 6 , and nov . 14 observations , three sequences were performed yielding a total on - source exposure time of 3 hr . the observations on oct . 5 were hampered by cloudy weather , and were halted after only one waveplate sequence ( 1 hr total exposure ) . from the latitude of lick observatory , sn 2009ip rose to a minimum airmass of 2.47 at transit . observations typically began at airmass values near 2.5 and ended near 3.2 . flat - field and wavelength calibration spectra were obtained immediately after each sequence , without moving the telescope . for polarimetric calibrations , standard stars were selected from the sample of schmidt et al . ( 1992a , b ) . the unpolarized standard stars bd+32@xmath313739 and hd 57702 were observed to verify the low instrumental polarization of kast . we constrained the average fractional stokes @xmath28 and @xmath29 values to @xmath38% . by observing bd+32@xmath313739 through a 100% polarizing filter , we determined that the polarimetric response is so close to 100% that no response correction was necessary , and we obtained the instrumental polarization position - angle curve , which we used to correct the data . we observed the high - polarization stars bd@xmath3959@xmath31389 and hd 204827 to obtain the zero - point of the polarization position angle on the sky ( @xmath6 ) and to determine the accuracy of polarimetric measurements , which were consistent with published values within 0.05% . since the oct . 5 observations were hampered by intermittent cloud cover , no polarization standard stars could be observed . thus , for this epoch we assumed negligible instrumental polarization , relying on the fact that the instrumental polarization was below 0.05% for the oct . 14 , nov . 6 , and nov . 14 observations , and we used the position - angle correction curve from oct . 14 , which we have verified to be very stable over time . we also adopted an offset angle of 5.9@xmath31 to further calibrate the oct . 5 observations ; this is the same offset value measured in the oct . 14 , nov . 6 , and nov . 14 observations , to within 0.5@xmath31 . all data obtained from the arizona and lick observatories were extracted and calibrated using generic iraf routines and our own idl functions . spectropolarimetric analysis was also performed in iraf and idl following the methods described by miller et al . ( 1988 ) and implemented by leonard et al . ( 2001 ) . we also obtained spectropolarimatric data on sn 2009ip from the european southern observatory ( eso ) public archive . we analysed data from the very large telescope ( vlt ) and focal reducer and low - dispersion spectrograph ( fors2 ) instrument ( appenzeller et al . 1998 ) obtained on 2012 nov . 7 , dec . 6 , and dec . 10 , which were part of the target - of - opportunity programme i d 290.d-5006 . we used the 300v grism data for nov . 7 and dec . 6 ; these data have spectral resolution of @xmath1613 and wavelength coverage of @xmath1644509500 . we also analyzed higher - resolution ( @xmath163 ) 1200r grism data for dates nov . 7 and dec . 10 , which covers a wavelength range of 58007300 . the gg435 order - blocking filter was used for all observations . flat - field , bias frames , and wavelength - calibration spectra were obtained for both grism settings . a master distortion frame and a waveplate chromatism data file were also obtained from the archive . sn 2009ip was observed at airmass values around 1.01.1 for the 300v and 1200r grism data , with respective exposure times of 330 and 470 s per waveplate position for the nov . observations , and approximately 900 s for the dec . observations . the position angle on the sky was always kept at 0@xmath31 by the instrument rotator . standard stars from fossati et al . ( 2007 ) were observed for calibrations . the high - polarization standard ngc 2024 - 1 was observed on nov . 6 and dec . 7 to calibrate the zero - point of the polarization position angle of the polarimeter on the sky , and the low - polarization standards wd 2039@xmath40202 and wd 2149@xmath39021 were observed on nov . 1 and dec . 7 , respectively , to gauge any instrumental polarization . the instrumental position angle spectrum was obtained directly from the eso public archive . on nov . 1 the observation of wd 2039@xmath40202 yielded linear stokes parameters of @xmath41% , after binning the @xmath0 band , while on dec . 6 , wd 2149@xmath39021 also yielded @xmath41% . such a small amount of instrumental polarization will not have a substantial impact on the science observations , and we thus make no attempt to remove it . reduction and spectropolarimetric analysis of the vlt / fors2 data was performed using the fors pipeline kit ( version 4.11.13 ) supplied by eso . specifically , the _ fors_pmos_calib _ and _ fors_pmos_science _ reduction recipes were used , and executed using the _ gasgano _ graphical user interface software . after all reduction and calibrations were performed , the data were corrected for the redshift of the host galaxy ngc 7259 ( @xmath42 ) . linear polarization is expressed as the quadratic sum of the @xmath28 and @xmath29 stokes parameters , @xmath43 , and the position angle on the sky is given by @xmath44 , while carefully taking into account the quadrants in the @xmath28@xmath29 plane where the inverse tangent angle is located . since @xmath7 is a positive - definite quantity , it is significantly overestimated in situations where the signal - to - noise is low . it is typical to express the de - biased " ( or bias - corrected " ) form of @xmath7 as @xmath45 , where @xmath46 and @xmath47 are the uncertainties in the @xmath28 and @xmath29 stokes parameters . if @xmath48 , then we set a 1@xmath49 upper limit on @xmath7 of @xmath50 . in cases where @xmath51 , @xmath6 is essentially undetermined . all polarized spectra presented herein are displayed in this manner . note , however , that at low signal - to - noise ratio @xmath52 is also not a reliable function , as it has a peculiar probability distribution ( miller et al . thus , for extracting statistically reliable values of polarization within a particular waveband , we have binned the calibrated @xmath28 and @xmath29 stokes spectra separately over the wavelength range of interest before calculating @xmath7 and @xmath6 . all quoted and tabulated values in this paper were determined in this manner , although spectra are displayed as @xmath52 , so they may exhibit slight offsets from our quoted values . 0.04 for both components , implying low interstellar absorption and , hence , low galactic interstellar polarization ( see 3.1 ) . na d is seen in weak emission at the redshift of sn 2009ip.,width=316 ] interstellar polarization ( isp ) is commonly a very problematic element of polarimetric analysis , as an accurate probe of the isp from both the milky way and the host galaxy can be very difficult to obtain . fortunately , several factors indicate that the isp in the direction of sn 2009ip must be very small , and should not significantly impact our polarimetry results and the physical interpretation that follows . first , the high - resolution vlt / fors2 flux spectrum on nov . 7 , shown in figure [ fig : na ] , reveals the resolved components of galactic interstellar na i d absorption lines at 5890.1 and 5896.3 . the equivalent width ( ew ) of each component is a very small @xmath53 m , which implies a galactic @xmath54 mag ( munari & zwitter 1997 ) . according to serkowski et al . ( 1975 ) , the value of @xmath55 can be used to derive an upper limit on galactic isp . the so - called serkowski relation is given by isp @xmath56 , which implies that isp @xmath30% in the direction of sn 2009ip . second , earlier photometric estimates of the total reddening toward the host galaxy ngc 7259 and sn 2009ip indicate a low value of @xmath57 mag ( schlegel et al . 1998 ; smith et al . 2010 ; foley et al . 2011 ) , which implies that the host galaxy also does not produce substantial isp along the line of sight . this is also consistent with the lack of na d absorption at the redshift of ngc 7259 ; in fact , na d appears in weak emission , perhaps from a local h ii region or from gas excited by sn radiation , with no indication of additional absorption components . the low host extinction is not particularly surprising , given sn 2009ip s large radial distance from the center of this nearly face - on spiral galaxy . if we were to naively apply the serkowski relation to the total measured reddening , we would obtain isp @xmath58% , still a very low value . however , as shown by leonard et al . ( 2002 ) , the serkowski relation does not necessarily apply to all galaxies outside the milky way , as different dust properties can result in variable efficiencies for isp . nonetheless , the data strongly suggest a low value of total isp , which justifies our decision to make no attempt to remove it from the data , as its effect should be minor and not significantly impact our results . we revisit the subject of isp at the end of the following section . , and position angle @xmath6 for sn 2009ip during its 2012 outburst , at epochs between sep . 21 and dec . the three kuiper / spol epochs ( _ red _ ) have been averaged for clarity . the sep . 21 and 24 mmt / spol data ( _ light blue _ ) have been averaged and binned to 300 ; all other epochs are binned to 50 . note the @xmath59 offset in @xmath6 between the 2012a phase and the peak of 2012b.,width=316 ] -band ( 50505950 ) polarization for sn 2009ip during the 2012a and 2012b phases . for reference , an arbitrarily scaled version of the @xmath0-band light curve is plotted in the background . _ lower panel _ : evolution in the @xmath28@xmath29 plane . interstellar polarization is constrained to @xmath600.2% , illustrated by the black dot near the origin . the dashed lines represent the position angles @xmath61 and @xmath62 measured at the first epoch on sep . 21 and peak polarization on oct . 14 , respectively . our latest continuum measurement ( 52005500 ) for dec . 6 is marked with a red cross near the origin , consistent with our isp limit . the approximate point - reflection symmetry between 2012a and the peak of 2012b suggests two separate and roughly orthogonal components of polarization on the sky . note the temporary shift in @xmath6 associated with the bump in the light curve at @xmath1630 days past peak.,title="fig : " ] -band ( 50505950 ) polarization for sn 2009ip during the 2012a and 2012b phases . for reference , an arbitrarily scaled version of the @xmath0-band light curve is plotted in the background . _ lower panel _ : evolution in the @xmath28@xmath29 plane . interstellar polarization is constrained to @xmath600.2% , illustrated by the black dot near the origin . the dashed lines represent the position angles @xmath61 and @xmath62 measured at the first epoch on sep . 21 and peak polarization on oct . 14 , respectively . our latest continuum measurement ( 52005500 ) for dec . 6 is marked with a red cross near the origin , consistent with our isp limit . the approximate point - reflection symmetry between 2012a and the peak of 2012b suggests two separate and roughly orthogonal components of polarization on the sky . note the temporary shift in @xmath6 associated with the bump in the light curve at @xmath1630 days past peak.,title="fig : " ] figure [ fig : pol_seq ] shows the total flux spectra , @xmath7 , and @xmath6 of sn 2009ip for epochs between sep . 21 and dec . 6 . to gauge the total @xmath0-band polarization @xmath7 and @xmath6 , we first binned @xmath28 and @xmath29 separately in the wavelength range 50505950 . the extracted values are listed in table 1 . figure [ fig : pol_lc ] shows the resulting polarization light curve and the temporal evolution in the @xmath28@xmath29 plane . in addition to @xmath0-band , we also sampled continuum " regions of 52005500 and 61006200 , because they appear mostly devoid of strong line features ; these ranges are close to the respective green " and red " wavelength regions used by hoffman et al . we caution , however , that there is no location within the spectrum that is completely free of line emission / absorption during all epochs , so the regions we have selected should be regarded as pseudocontinua that avoid the strong lines . [ cols="<,>,^,^,^,^,>",options="header " , ] [ tab : p48 ] @xmath63day is with respect to the day of peak in the @xmath0 band ( jd 2456207.72 ) . + @xmath64data for oct . 5 interrupted by poor weather ; required calibration using oct . 14 standard - star observations ; to be used with caution . + the sep . 21 and 24 epochs took place near the end of sn 2009ip s 2012a phase , before the sharp rise in brightness that marks the onset of the 2012b phase . the total flux spectrum during 2012a exhibits the characteristics of sne iin , displaying intermediate - width ( @xmath161000 km s@xmath18 ) emission lines superimposed on a spectrum with broad p - cygni lines ( @xmath168000 km s@xmath18 ) that is reminiscent of more common sne ii - p ( mauerhan et al . 2013 ) . during this early epoch , sn 2009ip is polarized at @xmath160.9% with position angle @xmath2 , and it occupies quadrant iv in the @xmath28@xmath29 plane . by oct . 5 , sn 2009ip has entered the 2012b phase and the flux spectrum changes substantially ; the broad emission components have become strongly diluted by the strengthening continuum , which is quite blue , while the intermediate - width emission features have developed broad lorentzian wings , a commonly observed characteristic of sne iin and an indication of high optical depth ( chugai 2001 ; dessart et al . 2009 ; smith et al . 2008 ) , probably resulting from csm interaction . at this time , @xmath7 has changed only marginally to @xmath160.8% but the position angle has shifted dramatically to @xmath65 , moving into quadrant ii of the @xmath28@xmath29 plane . interestingly , this dramatic change in position angle indicates separate polarization components for the 2012a and 2012b phases , with distinct axes of symmetry . the lack of a correspondingly significant change in fractional @xmath7 associated with the large rise in brightness could possibly be explained by the partial cancellation of the polarization vectors from the 2012a and 2012b components , as the latter rises in luminosity . by oct . 11 , the source has risen to peak flux . the spectrum has retained the overall appearance of oct . 5 , while the continuum becomes less blue and broad - line p - cygni absorption becomes apparent for he i / na d. our lick / kast and kuiper / spol measurements on oct . 14 show that @xmath7 has risen to a maximum of @xmath161.7% and @xmath6 has reached @xmath66 . at this time , strong deficits in polarization are seen for the intermediate - width emission components of h@xmath67 and , to a lesser extent , h@xmath68 . this is likely to be the result of dilution from the strong , intrinsically unpolarized emission lines in the outer csm . meanwhile , a slight increase in @xmath7 is seen at 56005800 , suggesting a potential association with broad he i @xmath695876 or na d p - cygni lines . figure [ fig : qu_full ] ( top panels ) shows the stokes spectra and the data distribution in the @xmath28@xmath29 plane for the oct . 14 lick epoch , binned to 100 . for reference , we also show scaled logarithmic flux spectra in the same figure , which has been normalised with a low - order spline ( avoiding the strongest emission features for the spline nodes ) . on oct . 14 , near peak brightness , the entire optical spectrum of sn 2009ip occupies a compact portion of the @xmath28@xmath29 plane , in quadrant ii . the points form a slightly elongated cluster that shows a continuous trend in wavelength , with the bluest points exhibiting the lowest position angle in the @xmath28@xmath29 plane . the cluster as a whole exhibits a small but non - negligible spread in position angle , which corresponds to a range of @xmath70 on the sky . the dilution in polarization associated with intermediate - width h@xmath67 emission is apparent as a protrusion from the cluster of data points which roughly points toward the origin . note that our binning causes the line to be unresolved , and this affects the strength and position angle of the protrusion ; the unbinned data show the protrusion extending very close to the origin . in total polarization @xmath7 , we see broad , shallow variations that appear to be associated with broad absorption components of h@xmath67 and he i / na d blueward of 5876 . it is interesting to note that the scaled flux spectrum included in figure [ fig : qu_full ] shows that the broad p - cygni emission features from fast ejecta , first seen during 2012a , are still present throughout the peak of 2012b ; this has important physical consequences for the energetics of sn 2009ip ( see 4 ) . by nov . 67 , roughly 30 days past peak , the continuum has faded substantially and broad p - cygni lines have once again become prominent in the flux spectrum . he i / na d absorption has deepened , and the intermediate - width component of h@xmath67 has developed an absorption component on top of the broad underlying feature , perhaps part of its own p - cygni profile . interestingly , this epoch coincides with the appearance of a temporary albeit substantial bump in the declining light curve . 7 vlt / fors2 spectropolarimetry data are shown in detail in the lower panels of figure [ fig : qu_full ] . at this time , the @xmath0-band polarization has dropped to @xmath160.7% and the position angle has shifted substantially to @xmath1650@xmath31 . the enhancement of polarization near he i / na i absorption feature has become more pronounced and sharp ; this feature is also apparent in lick data from nov . 6 , shown in figure [ fig : pol_seq ] . meanwhile , the sharp dip in polarization produced by the dilution from unpolarized intermediate - width balmer lines is no longer apparent . we examine the polarization enhancements of two specific line features in figure [ fig : qu_line1 ] , which show the stokes parameters plotted as a function of velocity , with respect to the he i @xmath695876 and h@xmath67 rest wavelengths . on nov . 7 , the enhanced polarization of he i / na d is substantially blueshifted to a velocity of 13,000 km s@xmath18 ( if it is associated with he ) and exhibits a velocity width of @xmath161500 km s@xmath18 . the peak is also blueshifted by @xmath161500 km s@xmath18 with respect to the flux minimum of the absorption line . the enhanced he i / na d feature forms a linear - shaped loop that roughly traces the average polarization axis of the system for that date . there also appears to be an enhancement in fractional polarization for the absorption component blueward of the intermediate - width h@xmath67 emission feature ( at @xmath71 km s@xmath18 ) . the origin of this absorption feature is unclear . it could be an absorption component of a p - cygni profile for the intermediate - width h@xmath67 component , or it could be the result of blueshifted absorption for he i @xmath696678 . if it is from h@xmath67 p - cygni , then this enhanced polarization feature does not appear blueshifted as in the case of the broad he i / na d absorption . in the @xmath28@xmath29 plane , the track across this potential h@xmath67/he i feature appears to be roughly consistent with the overall axis of symmetry . there also appears to be some indication of polarization enhancement for blend of absorption components seen at the higher velocities of @xmath72 km s@xmath18 and larger , which are probably from h@xmath67 . overall , the broad emission component of h@xmath67 appears to make a loop in the @xmath28@xmath29 plane . averaging all of our measurements from kuiper / spol on nov . 12 and 14 , we detect @xmath160.7% polarization in the @xmath0 band , similar to the value of the preceding vlt / fors2 epoch on nov . 7 . the lick / kast data solely from nov . 14 show a value of polarization similar to the nov.1214 combined average of the kuiper / spol data . the polarized spectra are not shown in figure [ fig : pol_seq ] , to avoid cluttering the figure , although their broadband measurements are presented in figure [ fig : pol_lc ] . interestingly , although the overall degree of polarization has not changed substantially between nov . 6 and 14 , the position angle has shifted again back toward the axis of symmetry of the 2012b peak . it thus appears that the bump in the light curve on nov . 67 is associated with another component of polarization , perhaps related the 2012a luminosity source . a few weeks later , on dec . 57 , we detect weak polarization in an average of three separate measurements with bok / spol , obtaining @xmath73 and only an upper limit for the continuum of @xmath74% . the last vlt / fors2 measurement takes place on dec . 6 , for which we obtain @xmath75% and @xmath76% . for the december epochs the position angle has changed significantly with the diminishing polarization , moving into quadrant iii with @xmath77 . the flux spectrum during this late epoch shows that the continuum has become redder , while the broad p - cygni features are still present . he i / na d absorption has become a double - dipped profile . figure [ fig : qu_line2 ] shows that by dec . 6 the enhanced polarization of this feature has shifted redward , and is probably associated with the bluer component of the double - dipped profile . a complicated blend of multiple absorption and perhaps emission features appears blueward of h@xmath67 at this time , and also exhibits evidence for enhanced polarization , roughly along the average axis of symmetry . examining more closely the temporal evolution of sn 2009ip in the @xmath28@xmath29 plane , shown in figure [ fig : pol_lc ] ( lower panel ) , the range of possible isp values constrained by the value of @xmath55 ( see 3.1 ) are illustrated by the radius of the black spot centered on the origin . it seems plausible that the path traced by @xmath28 and @xmath29 during dec . is the result of a migration back to the point of isp , somewhere near the origin . indeed , our latest measurement of the continuum polarization yielded @xmath74% from bok / spol for dec . 57 , and @xmath78% from vlt / fors2 on dec . 6 ( @xmath79% , @xmath80 ) , which is consistent with the isp limits we have estimated . an alternative , and less likely possibility , is that @xmath28 and @xmath29 are destined to drift back to the location in quadrant iv occupied by our earliest mmt / spol measurements in september ( i.e. , that the 2012a phase is intrinsically unpolarized , and instead reflects the location of isp ) . but such a large value of isp ( @xmath811% ) seems implausible given the constraints on isp indicated by the low values of @xmath55 and ew(na d ) ; the structure in @xmath7 for the september epochs also does not follow a serkowski law , which is inconsistent with the 2012a polarization being the result of isp . furthermore , it would be a rather remarkable coincidence for the position angle of isp to be orthogonal to the polarization axis of sn 2009ip at peak . therefore , the evidence is strong that our earliest polarization detections from mmt / spol in september reveal true polarization intrinsic to the source during the 2012a phase , and that the isp is within @xmath58% of the origin , and likely @xmath82% , as indicated by our latest continuum measurement . as one can see , any value of isp within this allowable range will affect our results only slightly , and will not change the fact that the 2012a and 2012b phases appear to have two separate components of polarization that will remain roughly orthogonal between 2012a and the peak of 2012b , no matter where the isp lies within the range illustrated by figure [ fig : pol_lc ] ( lower panel ) . the most natural explanation for linear polarization of a sn is thomson scattering by free electrons , either from the sn photosphere or within the zone of intense csm interaction , which includes the forward shock , the reverse shock , and the cold dense shell ( cds ) in between . for most sne iin , including sn 2009ip , evidence of electron scattering is also indicated by the broad lorentzian wings typically seen at the bases of the intermediate - width emission - line profiles ( chugai et al . 2001 ; smith et al . 2008 ; dessart et al . 2009 ) . but in the case of scattered photons emerging from a spherically symmetric distribution of material , the sum of all electric vectors will cancel out and yield zero net polarization . thus , the detection of net polarization from any sn directly implies aspherical structure , either intrinsic or imposed from partial obscuration of the polarized surface by intervening material ( see wang & wheeler 2008 for a review on sn polarization ) . core - collapse sne of type ii - p have been observed to be weak sources of polarization while in their plateau phases , which implies that the outer layer of ejecta is roughly spherical in some cases . however , near the end of their plateaus , sne ii - p can exhibit brief but significant increases in polarization ( leonard et al . 2000 , 2006 ; chornock et al . 2010 ) , because as the ejecta cool and recombine the photosphere recedes from the spherical outer layers of h - rich material into more aspherical he - rich and metal - rich layers deeper inside the ejecta . thus , to first order , sne ii - p can be characterized by a single component of photospheric polarization that scales inversely with the decreasing luminosity . sne iin , on the other hand , appear to exhibit a high degree of polarization while in their brightest phases , at least for the few objects that have been studied in sufficient detail ( e.g. , leonard et al . 2001 ; hoffman et al . this implies that csm interaction luminosity must be a major source of the net polarization from sne iin . in this case , since both the sn photosphere and the csm interaction zone are potential sources of polarized photons , changes in the relative luminosities of these components can lead to variations in net polarization , even if the fractional polarization and position angles of each component remain constant with time . below , we propose this to be the case for sn 2009ip . sn 2009ip s photospheric component is made evident by its spectropolarimetric similarities with the type iib sn 1993j , shown in figure [ fig:93j ] . near peak brightness , both sne exhibit a comparable degree of polarization across the optical band , with sn 2009ip being stronger in the continuum , and they both display remarkably similar wavelength - dependent variations across their polarization spectra , particularly evident at wavelengths of 52007000 . in the case of sn 1993j , these variations are associated with the broad p - cygni lines of h@xmath67 and he i@xmath39na d , and the same is probably true for sn 2009ip . the variations are produced by the depolarizing effect of line scattering ( trammell et al . 1993 ; hflich 1995 ; hflich et al . 1996 ; tran et al . this is not to be confused with the intrinsically unpolarized recombination emission from the outer optically thin csm that we see in sn 2009ip , which dilutes the polarized continuum and results in relatively narrow absorption " features of polarization for the h@xmath67 and h@xmath68 lines . rather , true depolarization is important for the broad emission lines that develop in the ionised ejecta above the electron - scattering photosphere of the sn . the persistence of these lines in sn 2009ip for @xmath16100 days implies a massive amount of high - velocity material ( a few m@xmath17 ) , consistent with a sn outflow ( see smp14 ) . csm is also a potential contributor to the polarization properties of sn 1993j ( hflich et al . 1996 ; tran et al . 1997 ) , via electron scattering within a sn / csm interaction zone and/or the scattering of photons by dust . after all , the continuum polarization of sn 1993j being entirely due to an aspherical ejecta photosphere is somewhat difficult to reconcile with the evolution of a highly spherical radio remnant revealed by vlbi ( bietenholz , bartel , & rupen 2003 ) , although this structure might only trace the morphology of the outermost h - rich layers , not the more asymmetric he - rich layers potentially responsible for the polarized continuum . still , the source did exhibit evidence for pre - existing csm at early times , in the form of narrow emission - line features exhibiting small velocity widths of @xmath16170 km s@xmath18 , indicative of an ambient stellar wind ( benetti et al . 1994 ) . however , these features were less pronounced than those of sn 2009ip and lasted for a shorter amount of time , indicating a much less dense csm . sn 1993j s x - ray properties were also indicative of csm interaction ( leising et al 1994 ) , while its radio evolution indicated a changing pre - sn mass - loss rate in the range @xmath83@xmath84 yr@xmath18 ( van dyk et al . 1994 ) , which , although roughly 100 times weaker than the pre - sn mass - loss rate inferred for sn 2009ip ( ofek et al . 2013b ) , could generate enough csm to influence the object s polarization properties . therefore , like sn 2009ip , sn 1993j could have two components of polarization photospheric and csm interaction . in the case of sn 2009ip , however , it appears rather obvious that csm interaction is the dominant source of polarization near photometric peak . the opposite could be true for sn 1993j : its polarization could be dominated by an aspherical explosion and only moderately influenced by the csm component . in this regard , a noteworthy difference between the two sne is that near peak brightness sn 1993j exhibits more variation than sn 2009ip in @xmath6 as a function of wavelength , suggesting that sn 2009ip has a more highly ordered axis of symmetry in 2012b resulting from its luminous and highly aspherical csm interaction zone . by 5060 days after peak , however , both sne exhibit remarkable similarity in @xmath7 and @xmath6 as a function of wavelength . it appears that as csm interaction fades after the peak of 2012b and the broad p - cygni lines return by nov . 6 , the polarization characteristics of sn 2009ip begin to look more like a sn photosphere again . it is thus interesting that the source makes a pronounced shift in position angle on nov . 6 , associated with the obvious bump in the light curve ( see figure [ fig : pol_lc ] ) , which could represent the temporary brightening of the photospheric component of polarization that dominated the 2012a phase . taken at face value , sn 2009ip s degree of peak polarization for the 2012a and 2012b phases ( from @xmath1% to @xmath85% ) indicates respective axial ratios of @xmath600.85 and @xmath600.7 for the aspherical source ( @xmath8615% and @xmath8630% asphericity ) , if we assume that polarization translates directly to the apparent geometry via comparison with the oblate electron - scattering atmosphere models of hlfich ( 1995 ) . however , the true physical axial ratios could be substantially more aspherical than these limits , for two key reasons : ( 1 ) inclination angle , and ( 2 ) partial cancellation between the 2012a and 2012b polarization vector components . if both the 2012a peak and the late decline of the 2012b phase ( from the nov . 67 bump onward ) are dominated by the sn photosphere component , as appears to be the case , then partial cancellation could be particularly important for the 2012b peak . indeed , the @xmath59 shift in position angle between the earliest 2012a observation and the peak of the 2012b phase implies two separate components of polarization that are roughly orthogonal to one another . since orthogonal polarization vectors cancel , the degree of intrinsic polarization for the 2012b component must be higher than it appears ( i.e. , even more aspherical ) , because the orthogonality of the separate 2012a vector component , if still present during 2012b , will partially cancel the 2012b component . we can estimate the true intrinsic polarization of the 2012b peak component as follows . if we rotate the @xmath28@xmath29 values from figure [ fig : pol_lc ] such that they are aligned with the axis of symmetry at maximum polarization ( @xmath87 ) , which roughly traces the average axis of symmetry for the entirety of 2012b , then the majority of polarization during the 2012b phase is shifted onto the positive @xmath28 axis , while the orthogonal 2012a polarization gets shifted onto the negative @xmath28 axis . figure [ fig : rsp ] shows the evolution of the resulting rotated stokes parameters . the @xmath88 curve represents variations along the main axis of symmetry , while @xmath89 traces deviations from the average axis of symmetry . the result implies that if the 2012a component maintains its orthogonal contribution to the total electric vector during 2012b , then the separation between the 2012a and 2012b components will approximately define the true degree of intrinsic polarization for the 2012b peak , which , therefore , could be as high as @xmath162.6% . in this case , comparison with the models of hlfich ( 1991 ) would imply a more highly aspherical axial ratio of @xmath900.6 for the 2012b component ( @xmath8640%50% asphericity ) . considering the additional effect of a @xmath91 inclination angle , hypothetically , it is possible that the intrinsic geometry of the 2012b component is a highly flattened one . below we discuss a plausible geometric scenario for sn 2009ip s 2012 evolution , facilitated by figures [ fig : cartoon1 ] and [ fig : cartoon2 ] . ) . the plot is meant to illustrate the full possible degree of intrinsic polarization for the 2012b component ( @xmath162.6% ) , after considering the effect of the partial cancellation , roughly orthogonal 2012a vector component ( see 4.1.2).,width=316 ] for the 2012b component of polarization the situation appears relatively clear . since csm interaction dominates the luminosity during this phase , and since the associated polarization source exhibits a rather steady and well - defined axis of symmetry , the net polarization must reflect the physical structure of the dense csm , which in this case could represent a toroidal geometry ( a disk or ring ) . toroidal geometry for the interacting csm has already been suggested for sn 2009ip by levesque et al . ( 2014 ) , based on spectral modeling and analysis of the balmer emission lines . those authors found that the h@xmath67 emission during the peak of 2012b requires a total radiating area of @xmath1620,000 au@xmath92 , while the balmer decrement requires a high csm density of @xmath93 @xmath94 , for case - b recombination ( note that such a high density would naturally explain the absence of forbidden emission lines in sn 2009ip s spectrum ) . the combination of high density and large area implies a flattened geometry ( i.e. , a disk / ring ) , possibly as thin as 10 au . alternatively , a lower density csm of @xmath95 @xmath94 in a limb - brightened spherical shell configuration was also proposed by levesque et al . ( 2014 ) as an additional possibility , although less favourably because it requires that the optical depth of h@xmath68 be lower than for h@xmath67 , which would be unusual . independently , smp14 also favoured disk - like geometry for the csm based on energetic arguments and on the persistence of broad p - cygni lines even after the phase of intense csm interaction had declined , which implies that a large fraction of the ejecta did not decelerate and , thus , must not have participated in strong csm interaction ( toroidal geometry would allow a large fraction of the ejected mass to bypass strong csm interaction in directions orthogonal to the csm plane ) . our spectropolarimetric results are consistent with toroidal geometry for the csm around sn 2009ip , and contradict models that invoke spherical distributions of csm . in this case , the axial ratio of the polarized emission source at the peak of 2012b ( constrained to @xmath96 , and probably @xmath600.6 ; see 4.1.2 ) is consistent with toroidal geometry tilted at some intermediate inclination angle with respect to our line of sight . the spectropolarimetric evolution is _ inconsistent _ with some physical scenarios previously proposed as possibilities for sn 2009ip s 2012 evolution . since it is the csm geometry that determines the polarization properties at the peak of the interaction phase , then the earlier eruption that produced such csm should reflect the same geometry . thus , the orthogonal position angle shift we observe between 2012a and the peak of 2012b is not compatible with the interpretation that these respective phases represent the launching and collision of successive shells of csm ( e.g. , from pair - instability eruptions ) , a possibility suggested by other authors ( pastorello et al . 2013 ; margutti et al . 2014 ) , since such a scenario should result in similar axes of symmetry for the two phases , not orthogonal geometries . rather , the orthogonality between the 2012a and 2012b events implies a physically distinct origin for each of the associated outflows . in addition , the large shift in polarization and position angle between the 2012b peak and the subsequent bump in the declining light curve on nov . 67 also suggests that the bump is the result of a polarized luminosity component that is separate from that of the 2012b peak for example , the sn photospheric component brightening again as the csm interaction component fades away . this is inconsistent with the interpretation that the bump in the light curve is the result of a fluctuating central source that variably illuminates the csm , such as activity from a central surviving star , as proposed by martin et al . ( 2013 ) . toroidal csm geometry would not be unique to sn 2009ip ; it has been proposed before for other sne iin , specifically in the cases of sn 1988z ( chugai & danziger 1994 , their model b ) , sn 1998s ( leonard et al . 2000 ) , and sn 1997eg ( hoffman et al . this type of geometry is also unsurprising in the context of sn 2009ip s stellar progenitor , since rings and tori are common phenomena around lbvs , b[e ] supergiants , and other evolved massive stars ( smith , bally , & walawender 2007 ; smith et al . not at all clear , however , is the connection of such geometry to the erratic behaviour of sn 2009ip over the last decade . it is possible that dense equatorial csm is the result of binary - influenced mass loss , as has been suggested to explain the toroidal csm of b[e ] supergiants . a connection between such objects and sn 2009ip was speculated upon by clark et al . ( 2013 ) in the case of lha 115-s 18 , a b[e ] supergiant exhibiting erratic long - term variability that is reminiscent of lbvs . it is thus interesting to note that the radial dimensions of b[e ] disks ( @xmath97@xmath98 au ; zickgraf et al . 1986 ) are comparable to the size estimates for sn 2009ip s toroidal csm ( smith et al . 2013 ; smp14 ) . represents the angle subtended by the csm from the point of view of the explosion.,width=316 ] and [ fig : qu_line2]).,width=316 ] the origin of the polarization for the 2012a phase is less clear . since the spectrum of sn 2009ip on sep . 2124 resembles that of more common core - collapse sne , the simplest explanation might be electron scattering within the photosphere of the early sn , which has yet to impact the dense csm lying farther out . in this case , the axial ratio of @xmath99 and the orthogonal geometric axis with respect to the peak of the csm interaction phase would suggest a prolate / bipolar shape for the 2012a outflow . interestingly , bipolar geometry has been suggested previously for other cases of highly polarized core - collapse sne ( e.g. , see wang et al . 2001 ) . alternatively , net polarization from the 2012a phase could also arise if the near side of the soon - to - be - shocked toroidal csm is partially blocking the waist of the sn photosphere ( figure [ fig : cartoon1 ] , inset ) , which could result in a polarization axis that is orthogonal to 2012b if the polarized poles of the photosphere remain relatively unobscured . finally , scattering of the sn photons by dust in the outer csm could provide another source of polarized photons . after all , infrared data obtained during the 2012a phase show strong evidence for circumstellar dust at a radius of @xmath16120220 au ( smith et al . we note that the strength of sn 2009ip s blue continuum detected at the same time indicates that the central source is probably not completely enshrouded by the dusty component , which is consistent with the toroidal geometry inferred for the csm . however , if dust scattering from the dense csm were an important source of polarized flux during 2012a , we would expect the geometric axis during this phase to reflect that of 2012b rather than being orthogonal to it . thus , electron scattering within the sn photosphere appears to be the most plausible explanation as the dominant source of polarized photons during 2012a , with net polarization either being the result of bipolar geometry for the photosphere or partial blocking of its waist by the toroidal csm . if strong csm interaction is confined to a toroidal geometry , then large amounts of high - velocity sn ejecta could expand relatively unimpeded . this is already indicated by the persistence of the broad p - cygni lines ( smp14 ) , but could also explain some of the higher - order effects we see in the spectropolarimetric data . for example , the sharp blueshifted enhancement in polarization associated with the he i / na d feature seen on nov . 67 ( figure [ fig : qu_line1 ] ) is an indication of high - velocity material . this feature has a blueshifted velocity of 13,000 km s@xmath18 , while the position angle across the feature approximately follows along the main axis of symmetry of this epoch in the @xmath28@xmath29 plane . the fact that the peak of the enhanced polarization feature is not coincident with the flux minimum of the absorption line implies that the enhancement can not simply be the result of absorption of unpolarized or less polarized flux ( which would increase the fractional polarization at that wavelength ) . rather , the net blueshift probably results from an intervening layer of fast he - rich ( or na - rich ) material that lies near the outer edge of the ejecta and is _ partially _ absorbing the inner polarized continuum . from the point of view of the observer , such partial blocking can increase the net polarization over that range of velocities by interfering with the cancellation of orthogonal polarization vectors distributed across the polarized source ( e.g. , also see kasen et al . the high - velocity he / na d - rich material was probably launched during the sn in 2012a phase , while the csm responsible for the 2012b phase was likely to have been created by the lbv progenitor eruptions during the years prior ( e.g. , see graham et al . 2014 , their figure 7 ) . a toroidal or disk - like distribution of csm around sn 2009ip has important consequences for the kinetic energy of the explosion that is inferred from the total radiated energy measured during the 2012b csm - interaction phase . taking the 2012a explosion time of @xmath100 days , suggested by smp14 , and the fastest ejecta velocity of 13,000 km s@xmath18 indicated by the blue edge of the h@xmath67 p - cygni line during 2012a , implies a radial distance of @xmath16350 au from the point of explosion to the inner edge of the dense csm ; this is consistent with the limits on the minimum radius of the dust component inferred from infrared measurements during 2012a ( @xmath86120 au ; smith et al . 2013b ) , and with the 300 au radial estimates by levesque et al . the latter authors showed that the csm need only be @xmath1610 au thick to explain the h@xmath67 luminosity in the toroidal scenario . for a 10 au toroidal scale height , the 300350 au distance from the explosion would imply a very small subtended angle of @xmath101 for the csm from the point of view of the explosion . such a configuration inclined at an angle of 3545@xmath31 with respect to our line of sight ( from edge - on ) would appear as an ellipse on the sky having an axial ratio that is consistent with the geometric constraints placed by the value of peak polarization during 2012b ( see 4.1.2 ) . for such a small subtended angle of csm , the fraction of the total explosion solid angle ( @xmath102/4@xmath103 ) intercepted by csm will be only 23% ( @xmath6010% conservatively ) . this minor amount of interacting area will result in an inefficient conversion of the total kinetic energy of the sn into radiation . thus , the @xmath104 ergs of kinetic energy inferred from the radiated energy of the explosion by fraser et al . ( 2013 ) and margutti et al . ( 2014 ) , both of whom assume a spherical csm configuration in their calculations , must be a significant underestimate , by 1 dex or more . a kinetic energy of @xmath19 ergs , as suggested by smp14 , seems more likely , given the demand for aspherical csm by spectropolarimetry . one must also keep in mind that the shape of the explosion is an influential factor as well . a bipolar explosion , which we have shown is a plausible geometry for the 2012a outflow , will intersect even less area of the toroidal csm . if the material is clumpy , then the effective intersecting area could be even lower than we have estimated , implying an even lower efficiency conversion of kinetic energy into radiation . we have presented multi - epoch spectropolarimetry of the 2012 outburst of sn 2009ip , covering the 2012a and 2012b phases . since the available data imply a very low amount of interstellar absorption and isp @xmath58% , we have not attempted to remove isp in our analysis . this could potentially lead to some small systematic errors in our results , although we expect any such changes will be minor and will not significantly influence our interpretation . the degree of polarization at the peak of 2012b implies an approximate axial ratio of @xmath600.7 ( probably @xmath600.6 ) for the polarized csm , which is consistent with the toroidal distribution of material suggested previously ( smp14 ; levesque et al . 2014 ) . in the future , modeling of the polarization properties of shocked toroidal csm could reveal how well the comparison to the oblate electron - scattering atmosphere models of hfilch ( 1991 ) can characterise the axial ratio of such geometry . details aside , the spectropolarimetric results demand substantial asphericity for the source of the 2012b luminosity and are thus inconsistent with models for sn 2009ip s 2012 explosion that invoke a spherical distribution of csm ( fraser et al . 2013 ; margutti et al . 2014 ) . for the 2012a phase , the detection of significant polarization at a position angle that is nearly orthogonal to 2012b is consistent with a bipolar geometry for the sn outflow or partial obscuration of a potentially more spherical sn photosphere by the toroidal csm . in any case , the orthogonality is inconsistent with the launching and collision of successive shells of csm ( pastorello et al . 2012 ) generated @xmath10540 days apart ( margutti et al . 2014 ) , since such a scenario should produce similar axes of symmetry for the 2012a and 2012b phases , not orthogonal geometries . furthermore , the geometric parameters constrained by spectropolarimetry imply that only a small fraction of the sn ejecta ( @xmath6010% ) participated in strong interaction with the toroidal csm . therefore , earlier estimates of a @xmath104 ergs explosion that are based on the total radiated energy , which assume spherical symmetry for the csm ( fraser et al . 2013 ; margutti et al . 2014 ) , are likely to be substantially underestimated by an order of magnitude . the strong similarities between sn 2009ip and sn 1993j , from the time near their peak through their decline , provide another strong line of evidence for the existence of an underlying sn photosphere from high - velocity ejecta ( 50008000 km s@xmath18 ) that persists throughout sn 2009ip s 2012 evolution ( @xmath16100 days ) . such a long - lasting optically thick component indicates a mass of at least a few m@xmath17 for the fast outer component of the ejecta , and the associated speeds imply a kinetic energy of @xmath19 ergs for the ejecta , as suggested by smp14 . after peak polarization , higher - order structure in the @xmath28@xmath29 plane from he - i / na d indicates the presence of fast metal - rich ejecta that have overrun the dense csm , consistent with a high - velocity flow that bypassed intense csm interaction . meanwhile , the weakening of polarization with declining continuum flux at late times suggests that the diminishing intensity of electron scattering is responsible for the decline in polarization . the fact that the flux and polarization drop more rapidly for sn 2009ip relative to other sne iin , such as sn 1997eg ( hoffman et al . 2008 ) or sn 2010jl ( grant williams , 2014 , private communication ) , also suggests that the csm around sn 2009ip has a compact configuration , compared to these other explosions . the lines of evidence presented here thus favour a scenario first proposed by mauerhan et al . ( 2013 ) , in which the luminosity of the 2012a phase of sn 2009ip was the result of a @xmath19 ergs explosion ( i.e. , a core - collapse sn ) that subsequently plowed into an aspherical ( probably toroidal ) distribution of csm during 2012b , creating the jump in luminosity and the strong polarization of the source . nonterminal scenarios are difficult to support in light of the current body of evidence , and we would have to invent a process capable of generating ejecta having @xmath19 ergs of kinetic energy without destroying the star . still , the best evidence of progenitor death is a vanishing of the star at late times . however , csm interaction can persist for decades or more , and generate enough luminosity to make progenitor disappearance a difficult observable to confirm . sn 1961v is a particularly controversial example ( filippenko et al . 1995 ; chu et al . 2004 ; smith et al . 2011b ; kochanek et al . 2011 ; van dyk et al . moreover , it is very plausible that the progenitor of sn 2009ip s 2012 explosion left a massive companion at the explosion site , not only because massive stars are rarely solitary ( sana et al . 2012 ) , but because toroidal or disk - like geometry for the csm is suggestive of binary influence . as discussed by smith & arnett ( 2014 ) , repeated brief pre - sn eruptions could result from binary interaction if the primary suddenly increases its radius during late nuclear burning stages , thereby triggering binary interaction and mass ejection . finally , we note that strong net polarization consistently observed from sne iin , as a class , implies that csm interaction is a dominanting influence on their polarization properties , and that an aspherical distribution of csm is a commonplace feature of the stellar progenitors of these explosions . we thank the referee for their review of this manuscript . we thank the staffs at lick , the mmt , and steward observatories for their excellent assistance . we thank s. bradley cenko for assisting with the lick observations . some observations reported here were obtained at the mmt observatory , a joint facility of the university of arizona and the smithsonian institution . this research was also based , in part , on observations made with eso telescopes at the la silla paranal observatory under programme i d 290d-5006 . we thank hien tran for supplying polarimetric data on sn 1993j for our comparison . thanks dan kasen at uc berkeley for insightful discussion , and we also thank leah n. huk at u. denver for helpful commentary . this research was supported by nsf grants ast-1210599 ( ua ) , ast-1211916 ( uc berkeley ) , ast-1210311 ( sdsu ) , and ast-1210372 ( u. denver ) . the supernova research of a.v.f.s group at u.c . berkeley is also supported by gary & cynthia bengier , the richard & rhoda goldman fund , the christopher r. redlich fund , and the tabasgo foundation . fossati , l. , bagnulo , s. , mason , e. , & landi deglinnocenti , e. 2007 , the future of photometric , spectrophotometric and polarimetric standardization , 364 , 503 fraser , m. , magee , m. , kotak , r. , et al . 2013a , apjl , 779 , l8 smith n. , chornock r. , li w. , et al . 2008 , apj , 686 , 467 smith n. , miller a. , li w. , et al . 2010 , aj , 139 , 1451 smith n. , et al . 2011a , mnras , 415 , 773 smith n. , mauerhan , j.c . , kasliwal , m. , & burgasser , a. 2013 , mnras , 434 , 2721

certain aspects of the behaviour of the gravitational field near null and spatial infinity for the developments of asymptotically euclidean , conformally flat initial data sets are analysed . ideas and results from two different approaches are combined : on the one hand the null infinity formalism related to the asymptotic characteristic initial value problem and on the other the regular cauchy initial value problem at spatial infinity which uses friedrich s representation of spatial infinity as a cylinder . the decay of the weyl tensor for the developments of the class of initial data under consideration is analysed under some existence and regularity assumptions for the asymptotic expansions obtained using the cylinder at spatial infinity . conditions on the initial data to obtain developments satisfying the peeling behaviour are identified . further , the decay of the asymptotic shear on null infinity is also examined as one approaches spatial infinity . this decay is related to the possibility of selecting the poincar group out of the bms group in a canonical fashion . it is found that for the class of initial data under consideration , if the development peels , then the asymptotic shear goes to zero at spatial infinity . expansions of the bondi mass are also examined . finally , the newman - penrose constants of the spacetime are written in terms of initial data quantities and it is shown that the constants defined at future null infinity are equal to those at past null infinity . pacs : 04.20.ha , 04.20.ex , 04.20.gz

gr-qc0701166

penrose s definition of _ asymptotic simplicity _ is an attempt to provide a characterisation of isolated systems in general relativity @xcite . it offers a framework on which diverse notions of physical interest can be defined and handled in a precise and rigorous manner . there is , however , a notorious lack of nontrivial examples where the associated formalism see e.g. @xcite has been actually been used to extract some physics . the obvious reason for this is the scarcity of radiative exact solutions to the einstein field equations other than the boost - rotation symmetric ones @xcite . a more conspicuous consideration in this regard is the fact that radiative spacetimes have to be constructed starting from an initial value problem where , say , some cauchy initial data are provided . up to fairly recently , there has not been a way of relating the properties of the initial data to the radiative properties of the development . for example , how does an initial data set have to be so that the resulting spacetime _ peels _ ? that is , the components of the weyl tensor in an adapted gauge having a distinctive decay in terms of powers of an affine parameter of the generators of outgoing light cones @xcite . another natural question on these lines would be how does the adm mass and the bondi mass relate ? similarly , is there any relation between the angular momentum defined at spatial infinity and that defined at null infinity ? which classes of initial data allow to select the poincar group out of the group of asymptotic symmetries the bms group in a canonical fashion ? arguably , there is a shortage of general results about the evolution of initial data sets for the einstein field equations near spatial infinity . this particular has made the idea of relating the structure of initial data with radiative properties a daunting endeavour . however , work by friedrich @xcite on the _ regular initial value problem at spatial infinity _ for time symmetric initial data sets has provided a tool to address questions similar to the ones raised in the first paragraph through a systematic approach . friedrich has introduced a representation of spatial infinity as a cylinder _the cylinder at spatial infinity _ , @xmath0 in stark contrast to the usual representation as a point . the cylinder at spatial infinity can be regarded as a limit set of incoming and outgoing characteristics of the einstein field equations , and as such it happens to be a _ total characteristic_. as such , it allows to transport information from the cauchy surface to null infinity without being contaminated by any sort of boundary conditions . the construction which led to the cylinder at spatial infinity allows an unfolding of the evolution process , which in turn can be analysed in all detail and to the desired order . the latter can be thought of as enabling the construction of a certain kind of asymptotic expansions which are completely determined by the cauchy data near infinity . friedrich s seminal work has been extended in several directions , automatising , in some sense , the calculation of the asymptotic expansions of the relevant field quantities @xcite and to include more general classes of initial data sets @xcite . the analysis described in these references has exhibited the existence of a hierarchy of obstructions to the smoothness of null infinity which goes beyond that arising from a mere consideration of the linear aspects of the field equations as described in e.g. @xcite . the existence of the hierarchy of obstructions has nurtured a _ rigidity conjecture _ on the smoothness of null infinity which , broadly speaking , states that the only cauchy initial data sets which have a development with a smooth null infinity . ] are those which are stationary close to infinity . similarly , a given degree of differentiability at null infinity , say @xmath1 , would only be possible if the initial data is stationary to a given order which depends on @xmath2 . the kind of initial data required can be constructed by means of the gluing techniques of @xcite . intuitively , the absence of gravitational radiation near spatial infinity which the rigidity conjecture asserts , suggests that the incoming radiation has to die off in the infinite future , and at the same time the system can not have emitted gravitational waves for all times in the infinite past . now , the calculation of the asymptotic expansions which arise from the construction of the cylinder at infinity is performed in a gauge which is adapted to a cauchy initial value problem remarkably , the gauge also allows to read off the structure and location of infinity directly from the initial data . on the other hand , the discussion of the gravitational field near null infinity is usually done using a gauge which is based on geometric structures of null hypersurfaces . having two different gauges , hampers , to a certain extent , extracting the physics of the system . it also difficults the attempts to assess the relevance of the presence of obstructions to the smoothness to null infinity in the asymptotic expansions . one of the aspects of the analysis of the asymptotic behaviour of the gravitational field via the definition of asymptotic simplicity and related constructions is that it allows to rephrase the decay of the fields in terms of differentiability at the conformal boundary . thus , it is natural to expect that the presence of obstructions to smoothness in the asymptotic expansions may result in a modified peeling behaviour this idea has already been explored in @xcite . a first study of the transformation between the gauge used in friedrich s analysis of spatial infinity and the gauge generally used to discuss null infinity has been given in @xcite . in the present article their analysis will be extended to the developments of a class of initial data sets which are not time symmetric . the consideration of initial data sets with a non - vanishing second fundamental form will allow us to examine some time asymmetric aspects of the structure of the conformal boundary of asymptotically flat spacetimes which up to now have remained _ terra incognita . _ on the other hand , the discussion shall be restricted to conformally flat initial data sets . this is done for conciseness and for the ease of calculations and presentation . previous analysis with conformally flat initial data sets see e.g. @xcite have shown that most of the crucial phenomena and structure observed in this setting admit a counterpart when discussing non - conformally flat data sets like in @xcite . furthermore , the detailed understanding of conformally flat initial data sets is of relevance in view of their use in the numerical simulation of black hole spacetimes see e.g. @xcite for a recent discussion on this . a cautionary note is , however , due : most of the conformally flat initial sets considered in the numerical simulations are boosted . this is done on physical grounds . here , initial data sets with linear momentum i.e . boosted will not be considered . the reason being that , as it will be discussed in the sequel , they are more technically involved and less smooth . the effects of this lower regularity are of interest in themselves and will be discussed elsewhere . the article is structured as follows : section [ section : basic_setup ] discusses some ideas of the basis set up and fixes some notation ; section [ section : f - gauge ] describes some aspects of the `` cylinder at spatial infinity''-formalism ; section [ section : constraints ] considers relevant aspects of the class of initial data sets to be used ; section [ section : expansions ] muses over the asymptotic expansions that can be calculated using the cylinder at spatial infinity ; section [ section : np - gauge ] goes briefly about the construction of a frame adapted to null infinity the np gauge while section [ section : relating_gauges ] discusses how it is related to the gauge used in the expansions of section [ section : expansions ] . this article revisits the analysis carried out in @xcite of `` bondi type '' systems what here is called the newman - penrose gauge and combines it with the asymptotic expansions for the development of conformally flat initial data sets which have been calculated in @xcite . in section [ section : peeling ] analysis of the implications of these expansions and the related obstructions to the smoothness of null infinity is done under some existence and regularity assumptions . in section [ section : sigma ] the machinery is used to calculate expansions of the spin coefficient @xmath3 on null infinity in the np gauge . the behaviour of this coefficient is related to various physical aspects of the theory of isolated systems . the implications of these expansions are analysed . it is shown that if the spacetime peels , then the behaviour of @xmath3 is such that it is always possible to single out the poincar group out of the asymptotic symmetry group in a canonical way . expansions of the bondi mass are also considered . in section [ section : np - constants ] the formalism is used to write down the newman - penrose constants of the developments of the class of initial data under consideration in terms of initial data quantities . these expressions allow to show that the constants at future null infinity are equal to those at past null infinity . there is much more to the aforementioned analysis than what is presented in this article . in particular , most of the calculations involved in the analysis have been carried out using _ ex profeso _ scripts in the computer algebra system maple v. a first subset of the scripts constructs the initial data ; another subset determines the coefficients of the asymptotic expansions in terms of initial data quantities ; finally a third subset calculates the transformation between gauges and allows to calculate the asymptotic expansions of np objects . in order to keep the presentation at a reasonable length , the explicit expressions of intermediate steps will be omitted , and more emphasis will be put on to the qualitative aspects of the results . some familiarity with the use of spinors in particular the space - spinor formalism of @xcite will be assumed . details of the diverse other formalisms used the cylinder at spatial infinity , the formalism of null infinity , the construction of gauges , and how to relate them is kept to the minimum necessary to motivate the analysis and to point out the many subtleties arising . in any case , the reader is referred to the various references cited for more complete details . finally , are some concluding remarks and a brief appendix containing some useful spinorial expressions . let @xmath4 be a vacuum spacetime arising as the development of asymptotically euclidean cauchy initial data @xmath5 . let @xmath6 be the conformally invariant weyl tensor of the metric @xmath7 . tilded quantities will refer to quantities in the _ physical spacetime _ , whereas untilded ones will denote , generically , quantities in an _ unphysical _ i.e . conformally rescaled spacetime . indices @xmath8 ( second half of the greek alphabet ) are spacetime indices taking the values @xmath9 ; the indices @xmath10 are spatial ones with range @xmath11 . the latin indices @xmath12 will be used in spatial expressions which are valid for a particular coordinate system ( usually a cartesian normal one ) . the indices @xmath13 are spatial frame indices , while @xmath14 will be spinorial indices taking the values @xmath15 . the initial data sets are taken to be maximal that is , @xmath16 , and the metric to be @xmath17 negative definite the latter in order to ease the later use of spinors . for simplicity and definiteness , it shall be assumed that the initial hypersurface contains only one asymptotic end that is , a region which is diffeomorphic to @xmath18 minus a ball . the analysis shall be concentrated in the region of spacetime near null and spatial infinity . in order to discuss the behaviour of the development of the initial data sets in the asymptotic region use will be made of the so - called conformal picture . let @xmath19 denote the point at infinity corresponding to the asymptotic end under consideration . the point @xmath19 is obtained by conformally compactifying the hypersurface @xmath20 with a conformal factor @xmath21 which can be obtained by solving the einstein constraint equations see section ( [ section : constraints ] ) . let @xmath20 be the resulting compact 3-dimensional manifold . the manifold is identified in a standard way with @xmath22 . most of the discussion will be carried out in a sufficient small neighbourhood @xmath23 of radius @xmath24 centred on @xmath19 and on the resulting associated domains of influence and dependence @xmath25 . as it is customary , the discussion of the initial data will be carried out in terms of the tensors @xmath26 on the conformally rescaled manifold @xmath20 . in this work attention will be restricted to conformally flat initial data sets and their developments . this assumption will prove fruitful and will allow to infer features of the development of more general types of initial data sets . under these assumptions coordinates @xmath27 can be chosen such that : @xmath28 unless otherwise stated , @xmath29 denote some normal coordinates with origin at @xmath19 based on a h - orthonormal frame @xmath30 . in @xcite a certain representation of the region of spacetime @xmath25 which can be thought of as that close to null infinity , @xmath31 , and spatial infinity , @xmath32 has been introduced . the standard representation of this region of spacetime depicts @xmath32 as a point . in contrast , the representation of @xcite depicts spatial infinity as a cylinder the cylinder at spatial infinity . the technical and practical grounds for introducing this description have been discussed at length in the seminal reference . the original construction in @xcite was carried out for the class of time symmetric metrics with analytic conformal metric @xmath33 . however , as seen in @xcite , the construction can be adapted to settings without a vanishing second fundamental form . friedrich s construction makes use of a blow up of the infinity of the initial hypersurface to @xmath34 . the blow up of @xmath19 to @xmath34 requires the introduction of a particular bundle of spin - frames over @xmath23 . let @xmath35 be the spinorial version of a timelike vector @xmath36 . assume the normalisation @xmath37 . let @xmath38 denote the fibre bundle of spinorial dyads @xmath39 over @xmath23 with projection @xmath40 , such that @xmath41 . the fibres of @xmath38 are of type @xmath42 . with this notation , define the set @xmath43 where @xmath44 is an affine parameter of horizontal curves on @xmath38 which project to geodesics on @xmath23 starting at @xmath19 . the matrices @xmath45 can be regarded as `` angular coordinates '' . crucial for the applications is that , @xmath46 , the fibre over @xmath19 , is such that @xmath47 . thus , working on @xmath48 provides a representation of @xmath23 where @xmath19 has been `` blown up '' to a sphere . note , however , that while @xmath23 is 3-dimensional , @xmath49 is 4-dimensional . this extra dimension can be interpreted as giving rise to the spin - weight of the diverse spinorial objects when lifted from @xmath23 to @xmath49 . this issue will arise again in section [ section : relating_gauges ] . the conformal structure of the spacetime @xmath50 is exploited by the introduction of conformal gaussian coordinates based on a certain type of conformal invariants : _ conformal geodesics _ see e.g. @xcite . accordingly , the spinor field @xmath35 will be thought of as being tangent to timelike conformal geodesics which are orthogonal to @xmath23 . for a gauge based on conformal gaussian coordinates , there exists a canonical conformal factor , @xmath51 i.e . a choice of representative of the conformal class of the metric @xmath52 . it is given by @xmath53 where @xmath54 is the conformal factor of the initial hypersurface usually obtained from solving the constraint equations , @xmath55 , with @xmath56 smooth , @xmath57 , and @xmath58 an affine parameter of the conformal geodesics . the function @xmath59 expresses the remaining conformal freedom in the construction . for the ease of future calculations the choice @xmath60 will be made . as it will be seen , with this choice , the _ locus _ of null infinity corresponds to @xmath61 . other choices are possible , and indeed , the calculations in @xcite use the apparently simpler @xmath62 which renders a more complicated expression for the locus of null infinity . in order to complete the construction of a representation of @xmath25 the region of spacetime close to null and spatial infinity consider the set @xmath63 of scaled spinor dyads @xmath64 . define the bundle manifold @xmath65 see figure 1 . the conformal factor @xmath51 given by ( [ theta ] ) can be used to define the following relevant sets : @xmath66 which will be referred to as , respectively , the _ cylinder at spatial infinity _ , the _ critical sets _ where null infinity touches spatial infinity and _ future and past null infinity_. .,title="fig : " ] ( -20,100)@xmath67 ( -25,15)@xmath68 consistent with the structure of the bundle space @xmath69 it is possible to introduce a calculus based on operators @xmath70 , @xmath71 and @xmath72 related to a basis of the lie algebra @xmath73 and the derivatives @xmath74 , @xmath75 . for this , use will be made of a frame @xmath76 and the associated spin connection coefficients @xmath77 of @xmath78 . conforming with the use of space - spinors one writes @xmath79 with @xmath80 it is also convenient to introduce an unprimed version of @xmath77 , @xmath81 which will be decomposed as @xmath82 where @xmath83 agrees on @xmath23 with the second fundamental form of the initial hypersurface . the discussion of the curvature will make use of the spinor @xmath84 , related to the ricci tensor of the weyl connection which can be constructed from @xmath78 and a spinor @xmath85 with spatial part given by @xmath86 @xmath85 is related to the acceleration of @xmath35 . for later use , define @xmath87 the remaining part of the curvature is encoded in the rescaled weyl spinor @xmath88 the spinorial counterpart of the rescaled weyl tensor , @xmath89 . [ proposition : f - gauge ] there is a frame @xmath76 on @xmath69 for which @xmath90 the description of @xmath91 in terms of the manifold @xmath69 , with the conformal factor @xmath51 given by ( [ theta ] ) , and the frame @xmath76 of proposition [ proposition : f - gauge ] , will be called the _ friedrich gauge _ f - gauge for short . in many parts of the sequel it will be necessary to _ lift _ objects defined in the base manifold @xmath91 to the manifold @xmath69 . scalar and space spinorial objects can be lifted in a natural way , with their `` angular dependence '' expressed in terms of the matrices @xmath45 . functions from @xmath92 to @xmath93 can be expanded in terms of certain functions @xmath94 which can be thought of as the lifts of spin - weighted spherical harmonics see @xcite for a more extended discussion : @xmath95 the operators @xmath70 , @xmath71 and @xmath72 can be defined by their action on @xmath94 . the @xmath94 s form a complete set of orthonormal functions with respect to the haar measure of @xmath92 this fact will be used several times . a discussion of conformally flat initial data which is well suited for the purposes here presented has been given in @xcite . its translation into the language of space - spinors has been carried out in @xcite . here , the main results required are presented . the space - spinorial versions of the momentum and hamiltonian constraints are given by @xmath96 the spinorial translation of the approach adopted in @xcite see e.g. @xcite suggests a solution of the form see the appendix for a definition of the irreducible spinors involved : @xmath97 where @xmath98 with and @xmath99 are referred to as , respectively , the _ electric _ and _ magnetic _ parts of @xmath100 . ] @xmath101 and @xmath102 is the lift of a @xmath103 function on @xmath104 . note that an overbar denotes complex conjugation . finally , @xmath105 in the above expressions , @xmath106 , denotes a constant related to the expansions conformal killing vector . the term @xmath107 expresses the angular momentum of the initial data , while @xmath108 is associated with the so - called conformal boosts . note that boosted initial data sets i.e . with linear momentum are excluded from the class under consideration . if the initial data is such that it has a non - vanishing linear momentum , then it can be shown that the asymptotic expansion of @xmath109 around infinity will contain logarithmic terms already at order @xmath110 . as firstly noted in @xcite , even completely regular initial data sets e.g . time symmetric , conformally flat may give rise to obstructions to the smoothness of null infinity . it is expected that initial data with logarithms will render further obstructions . this kind of obstructions arising from particular non - smooth features of the data have not yet been studied . thus , attention is restricted to initial data sets which are as regular as possible . even if the initial data sets are non - boosted , there is still the possibility of having solutions to the hamiltonian constraint whose asymptotic expansions contain logarithms . the conditions under which the solutions to the euclidean momentum constraint give rise to _ logarithm - free _ solutions of the hamiltonian constraint have been spelled in @xcite . from their results it follows that if @xmath111 where @xmath112 and @xmath113 are the lifts of @xmath103 functions on @xmath23 , then the asymptotic expansion of the conformal factor @xmath109 contains no logarithms . recalling that the lift of an analytic function @xmath114 to @xmath49 admits an expansion of the form @xmath115 where @xmath116 if @xmath117 is even and @xmath118 if @xmath117 is odd , the following assumption is made : [ assumption : initialdata ] the class of initial data @xmath119 under consideration is conformally flat and with a solution to the momentum constraint given by ( [ psi_abcd ] ) where the lift of the complex function @xmath102 to @xmath49 is of the form @xmath120 for the class of initial data comprised by assumption [ assumption : initialdata ] , the lift of the conformal factor @xmath109 to @xmath69 admits the parametrisation @xmath121 where @xmath122 with @xmath123 the adm mass , and @xmath124 satisfying reality conditions so that @xmath125 . attention will be now given to a different type of regularity condition on the initial data . in @xcite it has been shown that , generically , a specific class of obstructions to the smoothness of null infinity will be present unless certain coefficients in the expansion of the spinor @xmath88 on the initial hypersurface , @xmath20 , satisfy a particular relation . in what follows , objects expressible in terms of initial data which conrol the regularity of the development at null infinity will be referred to as _ obstructions_. the obstructions described in @xcite are related to the structure of the principal part of the evolution equations ( [ evolution_bianchi ] ) , and are shared by zero - rest - mass fields see @xcite . for the case of time symmetric data sets these conditions on the weyl spinor can be rewritten as a condition on the cotton - bach tensor . namely : @xmath126 where @xmath127 is the spinorial version of the cotton - bach tensor . the way the above formula has to be modified in order to accommodate initial data sets which are not time symmetric is not known . however , the conditions on the coefficients of the expansion of the weyl tensor given in @xcite can be calculated explicitly up to any desired order for conformally flat initial data sets . one finds that cfr . with @xcite : @xmath128 with @xmath129 , @xmath130 . an explicit calculation shows that _ at least for @xmath131 _ the conditions ( [ lambda : conditions ] ) are equivalent to @xmath132 where @xmath133 @xmath134 , with @xmath135 being the part of the general solution to the euclidean momentum constraint which is obtained arising from the real part of @xmath102 , that is , @xmath136 . is given by @xmath137 ] it is noted that for the class of data under consideration , @xmath138 , hence this particular condition poses no restriction on @xmath102 . the equivalence of the conditions ( [ lambda : conditions ] ) and ( [ regularity_c ] ) to all orders @xmath117 will be discussed elsewhere . the evolution of the spinorial quantities @xmath139 , @xmath140 , @xmath141 and @xmath88 describing the geometry of the manifold @xmath69 will be given by means of the _ conformal propagation equations _ the procedure described in section [ section : f - gauge ] , by which the point @xmath32 is replaced by the cylinder at spatial infinity @xmath0 , renders regular data and equations for the unphysical spacetime . moreover , it leads to an unfolding of the evolution mechanism which allows to analyse the process at arbitrary order and in all details . this kind of analysis is made possible by the _ total characteristic _ character of the cylinder at spatial infinity with regard to the conformal propagation equations @xmath0 can be understood as a limit set of incoming and outgoing light cones . in order to briefly discuss the equations implied by the extended conformal equations it is convenient to write : @xmath142 in this manner , one can write the propagation equations concisely in the form @xmath143 where @xmath144 , @xmath145 denote , respectively , linear and quadratic functions with constant coefficients , @xmath146 denotes a linear function with coefficients depending on the coordinates via @xmath51 and related objects , while @xmath147 denotes the @xmath148 unit matrix , @xmath149 are @xmath148 matrices depending on the coordinates and @xmath150 is a linear matrix - valued function of the connection @xmath140 with constant coefficients acting on @xmath151 . the reader is referred to @xcite for a full list of the equations . the evolution equations ( [ evolution_upsilon ] ) and ( [ evolution_bianchi ] ) are to be supplemented with initial data which can be readily calculated once an initial data set @xmath119 satisfying the vacuum constraint equations is known see @xcite for the details . the unfolding of the evolution process referred to in the previous paragraphs can be thought as the calculation of a certain type of asymptotic expansions for the diverse field quantities grouped in the vectors @xmath152 and @xmath151 . consistent with the class of initial data under consideration which is expandable in terms of powers of @xmath44 one can put forward the following expansion ansatz : @xmath153 the upper limits of the above summations have been intentionally left open to emphasise some convergence issues which will be discussed later . transport equations for the above equations can be calculated by differentiating the propagation equations ( [ evolution_upsilon])-([evolution_bianchi ] ) and then evaluating at @xmath0 that is , @xmath154 . these equations are schematically of the form @xmath155 these equations will be referred to as the _ transport equations of order _ a crucial feature of the transport equations is that their principal part is universal . the solution for @xmath156 is also universal that is independent of the initial data and agrees with the solution for minkowski spacetime . for @xmath157 they are linear differential equations for the unknowns of order @xmath117 . furthermore , note that the subsystem ( [ v_transport_eqns ] ) consists only of differential equations . the transport equations ( [ v_transport_eqns])-([bianchi_propagation_transport_eqns ] ) are decoupled in the following sense : knowledge of @xmath158 , @xmath158 , @xmath159 , together with the initial data @xmath160 allows us to solve the subsystem ( [ v_transport_eqns ] ) to obtain the quantities @xmath161 . with @xmath162 , @xmath163 , @xmath164 and @xmath165 and the initial data @xmath166 at hand one could , in principle , solve the equations ( [ bianchi_propagation_transport_eqns ] ) to get @xmath167 . in practise , in order to carry out the above described algorithm one makes use of the fact that all the field quantities appearing in the evolution equations ( [ evolution_upsilon])-([evolution_bianchi ] ) have a definite spin - weight and expansion type . thus , the coefficients @xmath168 , @xmath163 , @xmath169 can be rewritten in terms of the functions @xmath170 and @xmath58-dependent coefficients . for example , for the coefficients appearing in the components of the rescaled weyl tensor one writes : @xmath171 the products of functions @xmath170 which naturally arise in ( [ v_transport_eqns])-([bianchi_propagation_transport_eqns ] ) are linearised by means of the use of the clebsch - gordan coefficients of @xmath92 note that the products appearing in the principal part of the equations are somehow trivial as they involve the function @xmath172 . the above `` decomposition in terms of spin - weighted spherical harmonics '' allows for an implementation in a computer algebra system whereby the diverse transport equations in the hierarchy and their solutions are explicitly calculated to a given order @xmath117 . the order @xmath117 is restricted by the computer resources available generally up to @xmath173 for a spacetime without axial symmetry . the assumption of axial symmetry allows to carry the calculations up to a couple of orders more . an important question in the programme is the sense if any in which the expansions ( [ v_ansatz ] ) and ( [ phi_ansatz ] ) approximate the solutions of the equations ( [ evolution_upsilon])-([evolution_bianchi ] ) . in particular , one would like to be able to estimate the residues @xmath174 the estimation of the above residues and proving that the asymptotic expansions actually do approximate the solution to the conformal field equations is one of the remaining outstanding problems in the analysis of the structure of spatial infinity . it is expected that some suitable extension of the estimates discussed in @xcite in the case of the spin-2 massless field ( linearised gravity ) will permit to remove this hurdle . some of the technical complications of the problem lie in the degeneracy of the principal part of the equations ( [ evolution_bianchi ] ) and ( [ bianchi_propagation_transport_eqns ] ) at the critical sets @xmath175 , which precludes the use of standard methods of the theory of partial differential equation . further , as discussed in @xcite , this degeneracy translates in a non - smooth behaviour of the solutions of the transport equations which presumably will also extend to the actual solutions of the field equations . it can be seen that for the data of assumption [ assumption : initialdata ] and if the further regularity conditions ( [ lambda : conditions ] ) hold , then up to @xmath176 the coefficients in ( [ v_ansatz ] ) and ( [ phi_ansatz ] ) have all polynomial dependence in @xmath58 . starting with @xmath177 , logarithmic terms which are singular at @xmath178 appear . this adds further complexity to the analysis of the asymptotic expansions . hence , most of the calculations of expansions are stopped once the first logarithmic divergences have appeared . in order to proceed with the analysis of this article , the following assumption will be made : [ assumption : existence ] for the class of initial data under consideration , the propagation equations ( [ evolution_upsilon])-([evolution_bianchi ] ) have a unique solution in @xmath69 . further , the expansions ( [ v_ansatz ] ) and ( [ phi_ansatz ] ) where their coefficients are given by the solutions of equations ( [ v_transport_eqns ] ) and ( [ bianchi_propagation_transport_eqns ] ) approximate these solutions in the sense that the residues ( [ residue_v ] ) and ( [ residue_phi ] ) can be suitably estimated and are finite in the region of interest . the regularity of the solutions is at least that of the @xmath58-dependence of the expansions . this section discusses a certain class of gauge conditions on the spacetime @xmath179 near each component of null infinity i.e @xmath67 and @xmath68 known as the _ newman - penrose gauge _ np gauge . this class of gauge conditions is convenient for the analysis of radiative properties of the spacetime , like its peeling behaviour and its mass loss . the np gauge consists of certain requirements on the conformal gauge of the spacetime , together with adapted coordinates and frame . the np gauge conditions have to be implemented independently for @xmath67 and @xmath68 . here , for conciseness the discussion is concentrated on @xmath67 . the analogous construction for @xmath68 can be obtained , in principle , in the same way by interchanging the role of the relevant vectors . the construction of the np gauge is closely related to the so - called _ asymptotic characteristic initial value problem _ in which initial data is prescribed on an outgoing null cone and a portion of @xmath67 and allows to recover a portion of the spacetime to the past of these null hypersurfaces see @xcite . alternatively , one could prescribe data on an incoming null cone and on a portion of @xmath68 . here , however it will be assumed that the spacetime has been already constructed , and a suitable conformal factor , @xmath51 , has been found , allowing to locate the conformal boundary of the spacetime . let @xmath180 denote the metric of the conformally rescaled ( unphysical ) spacetime . our discussion will consider frame fields @xmath181 such that @xmath182 . the correspondence to the standard np notation is given by @xmath183 , @xmath184 , @xmath185 and @xmath186 . the spin connection coefficients associated to the frame @xmath181 are given by @xmath187 the correspondence with the standard np notation can is shown in the table 1 . [ table : spin_coefficients ] .the correspondence between the spin coefficients @xmath188 and the standard np notation . [ cols="^,^,^,^",options="header " , ] consider a subset @xmath189 containing a portion of @xmath67 . and let @xmath190 be a _ fiduciary cut _ of null infinity : @xmath191 , @xmath192 . it shall be assumed that @xmath193 , @xmath194 that is , it contains an open set of cuts . note that because of the assumption that the spacetime @xmath195 arises as the development of some asymptotically euclidean initial data set , it follows that the cuts of @xmath67 have the topology of @xmath34 . the following result shows that there is a certain gauge adapted to the description of @xmath67 . [ proposition : np - gauge ] assume that @xmath196 is suitably small . then , there exists on @xmath196 a conformal factor @xmath197 , a frame @xmath198 and coordinates @xmath199 such that @xmath200 and @xmath201 where @xmath202 are the connection coefficients with respect to the frame @xmath198 of the levi - civita connection @xmath203 of the metric @xmath204 , where @xmath205 . furthermore , @xmath206 and @xmath207 where @xmath208 and @xmath209 denote , respectively , the ricci scalar and ricci tensor of the metric @xmath210 . the coordinate @xmath211 on @xmath67 is an affine parameter of the generators of null infinity . the vector @xmath212 on @xmath67 is tangent to the generators of null infinity and satisfies @xmath213 . on @xmath196 , @xmath211 satisfies the eikonal equation @xmath214 , and the vector @xmath215 is tangent to the generators of the hypersurfaces @xmath216 and @xmath217 an affine parameter . the coordinates @xmath218 and @xmath219 are chosen such that the metric induced on the cuts of @xmath67 is the standard @xmath34 metric . * remarks . * 1 . proofs of the above statements can be found in the discussions of @xcite . the conformal factor , frame and coordinates given by the above result will be referred collectively to as the _ newman - penrose ( np ) gauge for _ @xmath67 . a corresponding np gauge consisting of a conformal factor @xmath220 , a frame @xmath221 and coordinates @xmath222 can be obtained by performing the interchange @xmath223 , @xmath224 in the spinorial indices , so that e.g. @xmath225 on @xmath226 for some suitably small @xmath227 . here @xmath228 denote the spin coefficients with respect to @xmath229 of the levi - civita connection @xmath230 of the metric @xmath231 , with @xmath232 . the only non - vanishing coefficients on @xmath233 are @xmath234 , @xmath235 and @xmath236 . in particular @xmath237 encodes information of physical interest . having introduced two different gauges for the analysis of the class of spacetimes under consideration , it is now necessary to discuss the way they are related to each other . the np - gauge is hinged on null infinity and is adapted to a characteristic initial value problem , while the f - gauge is based on a cauchy initial value problem . in the f - gauge the conformal factor @xmath51 , and hence the location of null infinity is known _ a priori _ by inspection of the initial data on @xmath20 as discussed in section ( [ section : f - gauge ] ) . further , because of the asymptotic expansions of section [ section : expansions ] which allow to examine the evolution of the field up to a given order @xmath117 in terms of cauchy initial data , the setting on which the f - gauge is implemented will be regarded as the starting point . as in the previous section , in order to fix ideas , the discussion will be mainly concerned with the np - gauge associated to @xmath67 . a first important observation is that the construction of the f - gauge and the associated regular cauchy initial value problem near spatial infinity see @xcite lives in the 5-dimensional bundle manifold @xmath69 , and not in the spacetime manifold , @xmath195 . the correspondence between objects on @xmath69 and @xmath195 is carried out by the projection @xmath40 . clearly , the correspondence between objects in the two manifolds is not one to one . this ambiguity is related to the spin - weight of the diverse spinorial objects under discussion put in other way , when working on @xmath69 one works with all possible choices of frame vectors @xmath238 , @xmath239 . assume that the region @xmath196 considered in section [ section : np - gauge ] is such that @xmath240 . in order to relate the frames @xmath76 living on @xmath69 and @xmath241 living on @xmath196 it is convenient to fix a section of @xmath69 to be thought of as a choice of a tetrad @xmath242 on @xmath196 . let @xmath243 be an open subset of @xmath34 . let @xmath244 be a local section of the hopf fibration @xmath245 . the map @xmath246 induces a smooth section @xmath247 . the image of @xmath246 will be denoted by @xmath248 . the vector fields tangent to @xmath249 which have a projection identical to that of @xmath250 are of the form @xmath251 with some functions @xmath252 on @xmath249 which satisfy @xmath253 . due to the commutation relations obeyed by @xmath250 and @xmath72 , the functions @xmath252 do not vanish on open subsets of @xmath249 . thus , the fields @xmath242 on @xmath248 which satisfy @xmath254 are given by @xmath255 the spin connection coefficients associated to the frame @xmath242 defined on @xmath248 are given by @xmath256 in what follows , objects defined on @xmath196 will be lifted to the section @xmath257 . in order to avoid complicating the notation unnecessarily , lifted objects will be designed with the same symbols as their original version on @xmath196 . from the context it should be clear where the diverse objects live . let @xmath258 the spin connection coefficients associated to the frame @xmath259 . from the discussion in @xcite it follows : there is a lorentz transformation , @xmath260 such that @xmath261 where @xmath262 and @xmath197 are the frame and conformal factor of proposition [ proposition : np - gauge ] . * remarks . * 1 . the matrix elements @xmath263 on @xmath67 can be determined entirely by means of a construction intrinsic to null infinity . furthermore , for the class of spacetimes under consideration + @xmath264 + where @xmath265 is the @xmath266 term in the expansion of @xmath267 see equation ( [ w_definition ] ) . latter discussion of the newman - penrose constants will require the knowledge of the quantities @xmath268 . one has that @xmath269 remarkably , the gauge conditions ( [ np_conditions : scri ] ) enable a calculation of @xmath268 completely intrinsic to @xmath67 . 2 . the determination of the values of @xmath197 and @xmath270 on @xmath67 which is equivalent to the knowledge of the quantities @xmath271 can also performed in a way which is intrinsic to @xmath67 . 3 . consistently with the discussion of the previous section there is a function , @xmath272 , on @xmath67 such that @xmath273 4 . the retarded time @xmath211 on @xmath67 has the expansion @xmath274 where @xmath275 is an arbitrary function of the angular coordinates associated to the supertranslational freedom . this expansion has been calculated up to @xmath276 inclusive . the retarded time @xmath277 has a similar expansion on @xmath68 : @xmath278 in the sequel and in particular for the evaluation of the np constants it will be natural to push the fiduciary cuts of null infinity to the critical sets where null infinity touches spatial infinity . thus , the following assumption is made . [ assumption : limits ] the fiduciary cuts @xmath190 and @xmath279 of , respectively , @xmath67 and @xmath68 can be made coincide with the critical sets @xmath280 and @xmath281 . in view of the existence of obstructions to the smoothness of null infinity of the developments of the class of initial data sets of assumption [ assumption : initialdata ] discussed in sections ( [ subsection : regularity_condition ] ) and ( [ section : expansions ] ) it is natural to ask how do they affect the asymptotic behaviour near null infinity . the translation between gauges discussed in section [ section : relating_gauges ] provides an adequate way of performing this discussion . the _ peeling behaviour _ is a specific type of decay of the components of the weyl tensor with respect to a frame satisfying the np - gauge . the classical discussion of the peeling behaviour is generally carried out in the _ physical spacetime _ a manifold without boundary . let @xmath7 the _ physical metric _ obtained from @xmath210 via @xmath282 where @xmath283 . coordinates in @xmath284 can be introduced by considering @xmath285 . recall that @xmath286 i.e . @xmath287 gives the locus of @xmath67 . a frame , @xmath288 , and a spinor dyad , @xmath289 on the physical spacetime can be introduced via @xmath290 in the coordinates @xmath285 one has that @xmath291 . the original peeling behaviour described in the classical literature assumes sufficient regularity of the solutions of the einstein field equations at , say , future null infinity so that in the physical spacetime the components of the weyl tensor admit an expansion of the form see e.g. @xcite : @xmath292 where @xmath293 , @xmath294 , etc . the decay conditions for the components @xmath295 are not independent . assuming a particular decay for @xmath296 , it is possible to deduce the decay of the other components . note that _ smoothness at null infinity _ is a much stronger requirement than that of _ peeling behaviour _ for which only a finite degree of differentiability is needed . in order to proceed some observations of generic nature are made . it is recalled that the rescaled weyl spinor , @xmath88 , is related to the ( conformally invariant ) weyl spinor @xmath297 by @xmath298 so that if @xmath299 , then @xmath300 . further , the components @xmath297 are related to those in the physical manifold via : @xmath301 moreover , it is noted that @xmath302 close to @xmath67 . using the above relations , and assuming that assumption [ assumption : existence ] holds , it is possible to read off more or less what the peeling behaviour of the developments of the initial data sets of assumption [ assumption : initialdata ] should be . it is nevertheless necessary to analyse whether there occur any cancellations which may improve the decay of the weyl tensor . the best behaved case that will be considered here is that of an initial data set where the regularity condition ( [ regularity_c ] ) holds up to @xmath176 inclusive . in this case , the main result of @xcite shows that all the functions @xmath305 appearing in the expansions ( [ phi_expansion ] ) up to @xmath177 are polynomials in @xmath58 save for the functions @xmath306 , @xmath307 , with @xmath308 which contain logarithmic terms . more precisely , one has : @xmath309 where @xmath310 are certain coefficients ( obstructions ) completely specifiable in terms of the initial data , @xmath311 denotes a certain polynomial of degree @xmath312 in @xmath58 such that @xmath313 , and @xmath314 is an arbitrary polynomial . now , it is noted that from formula ( [ theta ] ) for the conformal factor @xmath51 it follows that close to , say @xmath67 , @xmath315 further , @xmath316 is analytic close to @xmath317 . hence , a direct use of the relations ( [ case1 ] ) , ( [ psi_phi ] ) and ( [ rescaling_psi ] ) , shows that for this class of developments : @xmath318 for @xmath319 . the first non - smooth terms in the expansions of @xmath295 arise in @xmath296 at order @xmath320 , where a term of the form @xmath321 @xmath308 , is to be found . it is mentioned by passing that in this case the obstructions @xmath310 correspond to the polyhomogeneous constants of motion discussed in @xcite . as discussed in @xcite , the obstructions @xmath310 are special in the sense that they are the first truly non - linear obstructions that is , they arise as a product of the specific non - linear features of the conformal einstein equations . in contrast , the regularity conditions ( [ regularity_c ] ) arise from features which are present already at the linear level . nevertheless , it is of interest to analyse how these kind of obstructions affect the peeling behaviour of the components of the weyl tensor . as discussed in @xcite , if @xmath323 , for a certain @xmath117 , then the functions @xmath324 in the expansion ( [ phi_expansion ] ) contain logarithmic terms . more precisely , @xmath325 with the obstruction @xmath326 such that @xmath327 if and only if @xmath323 . as discussed in section [ subsection : regularity_condition ] , for @xmath328 one can explicitly show that it is equivalent to the condition ( [ regularity_c ] ) . it is plausible that this kind of logarithms will propagate due to the hyperbolic nature of the propagation equations and may produce further non - smoothness at higher order . this process is , however , not yet understood . according to the above discussion , if for @xmath176 the regularity condition ( [ regularity_c ] ) does not hold , then the resulting peeling spacetime will contain in addition to the logarithmic terms of ( [ singular_terms]) the following singular terms : @xmath329 note that in contrast to @xmath310 , which is an object of quadrupolar nature , @xmath330 is of @xmath331-polar nature . the above argumentation can be easily extended to the case where the obstructions @xmath326 with @xmath332 are present . in these cases , the component @xmath296 of the physical weyl tensor contains singular terms of the form @xmath333 it can be checked explicitly that when performing the lorentz transformation implied by the matrix @xmath334 of section [ section : relating_gauges ] , there is no cancellation taking place which would improve the regularity of the components of the weyl tensor . thus , in the case @xmath335 the singular terms arise sufficiently low in the expansions to preclude the peeling behaviour . one obtains the following . under assumption [ assumption : existence ] , the developments of initial data sets of assumption [ assumption : initialdata ] with @xmath336 peel in the sense that @xmath337 on the other hand , if @xmath338 then @xmath339 and accordingly the development does not peel . * remarks . * 1 . note that the requirements on conformally flat initial data to have peeling are very mild . in particular initial data sets like the brill - lindquist @xcite , misner @xcite and bowen - york @xcite will satisfy them . thus , modulo assumption [ assumption : existence ] , the developments of these initial data sets will peel . however , the weyl tensor of the unphysical spacetime will not be smooth due to the presence of the terms in ( [ singular_terms ] ) the obstruction @xmath340 has been explicitly been calculated for this case in @xcite and shown to be non - zero unless the data is schwarzschildean . these properties have been used in @xcite to show the non - existence of conformally flat slices in the kerr spacetime . 2 . on the other hand , as shown in @xcite , the einstein field equations are compatible with characteristic initial data with a decay of up to @xmath341 . the question is , which kind of cauchy data would produce such type of behaviour ? from the discussion in this work , it is to be expected that conformally flat initial data with linear momentum and/or for which the complex function @xmath102 has a singular behaviour worse than that of ( [ lambda : ansatz ] ) will give an example of initial data which would render this kind of spacetimes . 3 . the definition of the newman - penrose constants see section [ section : np - constants] requires certain degree of smoothness of the weyl tensor . from the discussion presented in this section , the required condition on the initial data is @xmath342 so that @xmath343 for spacetimes whose initial data do not satisfy the above it is not hard to see that the obstructions @xmath344 and @xmath345 are related to the conserved constants and logarithmic np constants of @xcite . as another application , the expansions of the newman - penrose coefficient @xmath346 on @xmath67 the _ asymptotic shear_ are discussed . the relevance of this stems from three in principle unrelated directions . firstly , as discussed in @xcite , if @xmath347 the electric part of @xmath348 satisfies @xmath349 that is , the limit as @xmath350 , then there is a canonical way of selecting the poincar group out of the asymptotic symmetry group the bondi - metzner - sachs ( bms ) group . secondly , @xmath351 has been called in older accounts the _ news function _ , and enters in the expression of the mass loss formula . here and in the sequel , the overdot denotes differentiation with respect to @xmath211 . thirdly , as discussed in @xcite , @xmath352 is related to the memory of non - linear tail waves first discussed by christodoulou @xcite . direct computer algebra calculations with the asymptotic expansions of section [ section : expansions ] and the transformation between the f - gauge and the np - gauge render the following result . [ theorem : sigma ] for the class of initial data of assumption [ assumption : initialdata ] and under the existence assumptions [ assumption : existence ] and [ assumption : limits ] one has that @xmath353 as @xmath350 . * remarks . * 1 . on the light of this result it is natural to ask what kind of initial data is required to obtain a development such that the decay condition @xmath354 occurs . preliminary calculations suggest that boosted conformally flat initial data would give rise to developments for which the decay ( [ bad ] ) occurs . another alternative would be to move away from the choice ( [ lambda : ansatz ] ) of the complex function @xmath102 giving rise to the higher multipoles of the second fundamental form . in any case , it seems that in order to obtain the asymptotic decay ( [ bad ] ) one has to consider solutions to the momentum constraint with logarithms in their asymptotic expansions . these issues will be explored elsewhere . 2 . another natural question to ask is : what is to be expected if one moves away from the restriction to conformally flat data ? it is to be presumed that as long as one restricts to a class of `` reasonably smooth '' conformal metrics , @xmath355 , the conclusion of theorem [ theorem : sigma ] will still hold . as an example of what one would call `` reasonably smooth '' consider metrics with the same smoothness as the conformal metric of stationary solutions that is , @xmath356 , where @xmath357 denotes the standard hlder space , see e.g. @xcite . 3 . the above two observations and the results of section [ section : peeling ] give credence to the conjecture that _ any _ peeling spacetime will have @xmath358 as @xmath359 . if the initial data set satisfies the regularity condition ( [ regularity_c ] ) to some orders , then it is possible to provide further information about the expansions of @xmath348 on null infinity . in particular , for a spacetime such that the first obstructions to null infinity appear at order @xmath176 in the components of the weyl tensor , one has that @xmath360 from the expansion ( [ sigma : expansion ] ) it is possible to derive an expansion for the bondi mass of the development , @xmath361 , close to the critical set @xmath280 . in the np - gauge one has that @xmath362 the above expression can be lifted in a natural way to the manifold @xmath69 , by noting that surface element @xmath363 has to be replaced with @xmath364 where @xmath365 is a parameter along the fibres . the functions @xmath366 form an orthonormal basis with respect to the haar measure on @xmath92 which is given by @xmath367 , thus only terms with @xmath172 contribute to the integration . one finds that : for the class of initial data of assumption [ assumption : initialdata ] and under the existence assumptions [ assumption : existence ] and [ assumption : limits ] one has that @xmath368 that is , the bondi mass agrees with the adm mass on the critical set @xmath280 . * remarks . * 1 . versions of the above result with different regularity assumptions can be found in the literature see e.g. @xcite . the present form has been anticipated in @xcite . using the expansions ( [ sigma : expansion ] ) which are valid for developments for which the first obstructions to the smoothness of null infinity appear at order @xmath176 one can obtain a generalisation of the expansion for the bondi mass given in @xcite : + @xmath369 + where + @xmath370 in @xcite , frauendiener has identified the non - linear memory effect of tail waves in terms of the difference , @xmath371 , between the value of @xmath372 at a cut of @xmath67 given by @xmath373 and its value at some fiduciary cut @xmath190 . in the current context it is natural to let the critical set @xmath280 to be this cut . in this light theorem [ theorem : sigma ] states that there is no `` residual '' distortion contained in the initial data , and that , in principle there is no restriction to the possible values of @xmath371 which can be attained along @xmath67 . on the other hand , if @xmath374 as @xmath375 , then there are values of the memory effect which are not accessible for @xmath376 , as it satisfies the equation @xmath377 the newman - penrose constants are a set of nontrivial complex quantities of quadrupolar nature defined on cuts of @xmath67 ( five of them ) and @xmath68 ( another five ) which are absolutely conserved see @xcite . here , absolutely conserved means that their value is independent of the choice of cut , and that they are supertranslation invariant . further , the two sets of quantities transform adequately under lorentz transformations i.e . they form a representation of the lorentz group . the argumentation leading to the np constants makes certain assumptions on the regularity of the spacetime at null infinity . let @xmath293 . assume that on @xmath378 the component @xmath296 of the weyl tensor has the decay @xmath379 then the quantities ( the np constants for @xmath67 ) @xmath380 with @xmath381 are supertranslation invariant and absolutely conserved that is , @xmath382 . * remarks . * 1 . note that the spacetime is required to have a bit more regularity at null infinity than what is needed for the peeling behaviour . 2 . an analogous result holds for @xmath68 if on @xmath383 the component @xmath384 has the decay @xmath385 where here @xmath386 . then the np constants for @xmath68 are given by @xmath387 3 . note that in principle there is no reason to expect a relation between @xmath388 and @xmath389 , unless the spacetime is time reflexion symmetric like in the case analysed in @xcite . a clear cut physical interpretation of the np constants has remained , so far , elusive . from their evaluation in particular examples and from comparison with linear theories , it follows that the content of the constants is nontrivial @xcite . there has been some work attempting to relate the constants with the radiation content of the spacetime at late times and the value of the weyl tensor at timelike infinity , @xmath390 see e.g. @xcite . remarkably , it has been recently shown by wu & shang @xcite that the np constants vanish for algebraically special spacetimes . this was a long - standing conjecture see @xcite . from here it follows that the constants vanish for the schwarzschild spacetime ( which was known ) and for kerr i.e . for the ( presumably ! ) asymptotic states of spacetimes containing black holes . this result , together with conjectures given in @xcite support the idea that the np constants measure in some way the gravitational radiation content of the spacetime at earlier and later times . in @xcite it has been shown that it is possible to reexpress the np constants in terms of quantities defined on the unphysical manifold , @xmath195 . for this , note that @xmath391 , so that one obtains @xmath392 in order to perform the obvious lift to the bundle space @xmath248 recall the correspondence @xmath393 if @xmath394 denotes the haar measure on @xmath92 , then @xmath388 lifts to @xmath395 where @xmath365 denotes a parameter on the fibres of the fibre bundle @xmath69 . the second integration does not alter the result as the integrand is independent of @xmath365 . for the np constants on @xmath68 one obtains in a similar way that @xmath396 expanding the integrands one finds that @xmath397 the strategy is now to calculate @xmath398 in order to obtain an expression for the np constants in terms of initial data quantities . a direct inspection using the results of the asymptotic expansions discussed in section [ section : expansions ] reveals that there are potentially conflictive terms of order @xmath399 . remarkably enough these cancel out . a computer algebra calculation renders the following result . [ theorem : equality ] for the class of initial data given by ( [ assumption : initialdata ] ) and under assumption [ assumption : existence ] , the newman - penrose constants at @xmath67 are equal to those at @xmath68 . more precisely , one has that @xmath400 where @xmath401 it is noted that the integrand of the formula ( [ monster_1 ] ) can be written concisely as : @xmath402 from where it can more easily be seen that the constants have the structure : @xmath403 in the view of the above formulae , it is natural to ponder what would happen if one were to consider a more general class of initial data sets . it is to be expected that the theorem [ theorem : equality ] will still hold if one considers non - conformally flat initial data , for the time asymmetry in the solutions of the einstein constraint equations is fed by the second fundamental form , not by the conformal metric . if one were to consider boosted initial data , it is quite plausible that null infinity may not be regular enough recall the existence of logarithms in the initial data discussed in section [ section : constraints] to ensure the constancy of @xmath404 and @xmath405 unless , perhaps if the initial data set is not maximal , like in the case of boosted slices of schwarzschild . the present article has been concerned with an analysis of the physical implications of certain asymptotic expansions for asymptotically flat spacetimes which allow to relate the behaviour near null and spatial infinity to properties of the initial data . these asymptotic expansions are obtained in a gauge which exploits to the maximum the conformal structure of the spacetime . hence , in order to gain intuition on the physics of these expansions one has to transform to a more `` physical gauge '' . crucial feature of the expansions is that they bring to the foreground the fact that the development of generic initial data will render a non - smooth conformal boundary of the spacetime . this non - smoothness manifests itself in the appearance of logarithmic terms in the solutions to the transport equations which generate the terms in the asymptotic expansions . it has to be emphasised that these logarithmic terms are conformal invariants of the spacetime and not the product of some bad choice of gauge . in other words one could completely forget about the f - gauge and the cylinder spacetime , but one would still have to confront in one way or another their implications as the discussion about the peeling behaviour in section [ section : peeling ] evidences . on a more practical side , it is not clear what the implications for numerical simulations of the non - smoothness of null infinity are . as seen here , as long as no linear momentum is included in the reference asymptotic end , the smoothness of most conformally flat initial data sets of interest is enough peeling so that most of the @xmath406 formalism remains untouched . an interesting and in principle not evident result from our calculations is the fact that the np constants in the cases for which they are well defined offer a set of absolute conserved quantities for the whole spacetime , and not too different for each component of null infinity . it has been argued that this feature will be preserved in the case of more general initial data sets . further , a simple computation also reveals that the logarithmic np constants should also share this feature . finally , it has been seen that under certain existence assumptions having the peeling behaviour automatically implies the right kind of behaviour of the asymptotic shear at null infinity to obtain a canonical representative of the poincar group out of the bms group . this result exemplifies the power and insights that can be gained by methods presented in this work . at this point it would be natural to proceed with a discussion of angular momentum . however , as the result of the non - existence of conformally flat slices in the kerr and other ( non - static ) stationary spacetimes exposes , there is a certain non - cooperation between conformal flatness and the presence of angular momentum at null infinity see e.g. @xcite . so it seems that a more natural setting for this analysis is that of non - conformally flat metrics . this last issue and other questions raised in this article will be further dwell upon elsewhere . i would like to thank h. friedrich , s. dain and r. beig for valuable discussions on diverse aspects of this research . i thank also cm losert - vk for a careful reading of the manuscript . i am grateful to the isaac newton institute , cambridge for its hospitality while some parts of this projects were in progress . this research is funded by an epsrc advanced research fellowship . in the text , use of the following spinors has been made :

we compare results from a relativistic and a nonrelativistic set of 2d axisymmetric jet simulations . for a set of five relativistic simulations that either increase the lorentz factor or decrease the adiabatic index we compute nonrelativistic simulations with equal useful power or thrust . we examine these simulations for morphological and dynamical differences , focusing on the velocity field , the width of the cocoon , the age of the jets , and the internal structure of the jet itself . the primary result of these comparisons is that the velocity field of nonrelativistic jet simulations can_not _ be scaled up to give the spatial distribution of lorentz factors seen in relativistic simulations . since the local lorentz factor plays a major role in determining the total intensity for parsec scale extragalactic jets , this suggests that a nonrelativistic simulation can not yield the proper intensity distribution for a relativistic jet . another general result is that each relativistic jet and its nonrelativistic equivalents have similar ages ( in dynamical time units , r/ , where r is the initial radius of a cylindrical jet and is the sound speed in the ambient medium ) . also , jets with a larger lorentz factor have a smaller cocoon size . in addition to these comparisons , we have completed four new relativistic simulations to investigate the effect of varying thermal pressure on relativistic jets . the simulations confirm that faster ( larger lorentz factor ) and colder jets are more stable , with smaller amplitude and longer wavelength internal variations . however , an exception to this occurs for the hottest jets , which appear the most stable . the apparent stability of these jets does not follow from linear normal mode analysis , which suggests that there are available growing kelvin - helmholtz modes . however , these modes are not excited , due to a lack of perturbations able to couple to them . as an example of how these simulations can be applied to the interpretation of observations , we use our results to estimate some parameters of . although none of these estimates alone can determine if the jets in are relativistic or nonrelativistic , estimates for the age and the jet - to - ambient density ratio confirm values for these parameters estimated by other means . accepted for publication by the astrophysical journal to appear in the may 10 , 1999 issue

astro-ph9901046

jet flows in extragalactic radio sources and those associated with the galactic superluminals could have relativistic velocities for a significant fraction of their length ( for example , for extragalactic sources see the arguments in cawthorne 1991 , and for galactic superluminals see the introduction in hardee et al . 1998 , hereafter hrhd ) . recently , a number of groups have begun to simulate such relativistic flows ( @xcite ; @xcite ; komissarov & falle 1997 ; @xcite ; and @xcite ) . there have even been preliminary investigations of relativistic magnetohydrodynamics ( @xcite ; van putten 1996 ) , including one attempt at combining mhd with general relativity in 3 + 1 dimensions ( koide , shibata & kudoh 1998 ) . while these papers and a review ( @xcite ) have discussed some of the qualitative differences between relativistic and nonrelativistic jets , a more thorough comparison is essential in order to understand whether nonrelativistic simulations can be applied to the interpretation of observed relativistic flows , and whether a comparison of nonrelativistic and relativistic simulations can provide a diagnostic of flow speed . many researchers have performed simulations of nonrelativistic jets over the last 1015 years , with reviews of the early work in @xcite and @xcite . investigations frequently focused on the question of jet disruption by fluid instabilities , and in this light @xcite discuss the susceptibility of the jet to the ordinary or fundamental and reflecting modes of the kelvin - helmholtz ( hereafter kh ) pinch axisymmetric instability . much recent literature uses the terms surface wave and body wave for the ordinary and reflecting mode , respectively , and we adopt that here . for each of the pinch , helical , triangular , rectangular and higher order normal kh _ modes _ there is one surface and many body _ waves_. axisymmetric simulations , like the ones discussed here , only allow waves of the pinch mode , since higher order modes require non - axisymmetry . note that the higher order modes have smaller growth lengths ( or faster growth rates ) than the pinch mode , so axisymmetric jets remain essentially stable for longer distances than in simulations where higher order modes are allowed ( e.g. , a 2d slab jet simulation : hardee & norman 1988 ; bodo et al . 1994 ; bodo et al . 1995 ; or any 3d jet simulation : e.g. , clarke 1996 ) . another reason for expecting that the flows studied here will exhibit little instability is that previous analytic and numerical work has shown that jets with relativistic flow speeds are more stable than jets with slower flow speeds ( hrhd ; @xcite ) , primarily because the former jets behave as if they are more dense . in this paper we will compare 2d axisymmetric nonrelativistic jet simulations performed with zeus-3d and simulations performed with a relativistic godunov ( shock - capturing ) scheme discussed in dh94 . for each relativistic simulation we have computed nonrelativistic equivalents as suggested by komissarov ( 1996 ) , based on equating the _ useful _ power ( i.e. , the enthalpy flux minus the mass - energy flux ) or the thrust of the jet at the inlet . we show that one of these sets of nonrelativistic equivalents should have stability properties ( i.e. , spatial growth rates , etc . ) that are similar to those of many of the relativistic simulations . these similar stability properties will lead to similar structures within the jet : for example , small - scale features along the jet axis that may grow to become shocks . also , we investigate the effect of jet temperature on the structure of the relativistic simulations . as opposed to the morphology of nonrelativistic jets , which primarily depends on jet - to - ambient density and mach number ( assuming that jet and ambient thermal pressures are equal ; norman , winkler , & smarr 1984 ) , the morphology of relativistic jets depends additionally on temperature as an independent parameter . one natural consequence of a hot " relativistic jet , where the thermal velocities of particles are at least mildly relativistic , is that the classical or newtonian mach number ( @xmath2/@xmath3 , where @xmath4 is the jet velocity relative to the laboratory frame and @xmath3 is the sound speed within the jet ; hereafter we refer to this as the mach number " ) is small enough that the flow is transonic ( with a mach number between one and three ) . however , linear stability analysis suggests that the spatial growth length of the kh instability is roughly proportional to the product of the lorentz factor and mach number , and not the mach number alone ( hrhd ) . thus , a hot relativistic flow with low mach number should be less prone to instability , although still not completely stable , than the mach number alone would indicate . in fact , mart et al . ( 1997 ) see this relative stability in their hot jets , which have fewer internal features . however , their simulations spanned a wide range of specific internal energies with hot jets quite hot and cold jets quite cold . we have simulated jets with a similar range , 34 orders of magnitude , of specific internal energies , but with more resolution in temperature space and with specific internal energy density in some cases closer to unity . we also wish to reiterate the main point of hrhd : that relativistic jets are not unconditionally stable many simulations that have been computed contain perturbations that only weakly excite any of the available growing modes . in 2 , we describe the simulations with particular emphasis on the nonrelativistic equivalents , in 3 , we discuss the various results while focusing on large scale structure and the growth of the kh instability , and in 4 , we use some of the analysis presented here to make estimates of some fundamental quantities ( e.g. , the age , jet - to - external ratios for density and enthalpy , and the mach number ) for the jets in , a well - studied radio source . in addition , we include two appendices ; in appendix a we derive the relativistic form of the jet head advance speed and in appendix b we develop a simple model to estimate the cocoon radius in either relativistic or nonrelativistic jets . for the nonrelativistic simulations , we used a cylindrical coordinate system and the same boundary conditions as did dh94 . specifically , the boundary conditions allow outflow everywhere with two exceptions : at the inlet with inflow boundary conditions and along the symmetry axis , which has a reflecting boundary . dh94 describe a relativistic code that uses adaptive mesh refinement ( amr ) , a general purpose scheme written by quirk ( 1991 ) , to gain spatial and temporal resolution . the simulations reported in dh94 employed one level of refinement : the coarse grid had 6 zones / jet radius and the refined grid had 24 zones / jet radius . we have run the nonrelativistic simulations using zeus-3d with a resolution of 12 zones / jet radius and the van leer linear advection scheme , which is a combination that resolves features of the same size as the relativistic code , as determined by examining schlieren ( density gradient ) plots . the grid extends to 41@xmath5 jet radii along the jet axis and either 10 r or 16@xmath5 r radially , depending on whether the relevant relativistic simulation had a lorentz factor below or above 5 , respectively . the nonrelativistic simulations were run on grids of either 500 @xmath6 120 or 500 @xmath6 200 zones . the relativistic simulations are as described in dh94 . in order to fill a gap in the dh94 simulations at small lorentz factors , we include an additional relativistic simulation that has a lorentz factor of 2.5 and was labeled run e in hrhd . run e fits between runs a and b in the original dh94 sequence . we point out that hrhd adopted incorrect values for the lorentz factor when analyzing runs e ( 2.55 instead of 2.5 ) , b ( 5.5 instead of 5.0 ) and c ( 14.35 instead of 10.0 ) and that we use the correct values here . since the correct lorentz factors reduce by @xmath7 30% the maximum allowable wavelengths and the wavelengths associated with the fastest growth in the most important modes in runs b and c , this correction does not alter significantly either the specific results or the overall conclusions in hrhd . for the relativistic simulations this paper generally uses the notation of dh94 ; @xmath8 is the proper mass density , @xmath9 is the speed of light , @xmath10 is @xmath11 , @xmath12 is the lorentz factor ( @xmath13 ) , @xmath14 is the adiabatic index , @xmath15 is the internal ( or thermal ) pressure , @xmath16 is the proper mass - energy density and is @xmath17 , and the enthalpy is @xmath18 ) . one change from dh94 is that the mass density in the lab frame for relativistic simulations is designated @xmath19 and not r , which we use for the initial jet radius . in this paper , we designate the jet - to - ambient density ratio as @xmath20 , and the jet - to - ambient enthalpy ratio as @xmath21 . we will discuss @xmath20 primarily in the nonrelativistic simulations , but where we specifically use @xmath20 in regard to the relativistic simulations , this refers the ratio of proper densities and is 0.1 for all of our relativistic simulations . we use the variable @xmath22 exclusively for the mass density in the nonrelativistic simulations . additionally , when comparing a similar quantity between relativistic and nonrelativistic simulations , we occasionally refer to each with @xmath23 or @xmath24 subscripts . for the nonrelativistic simulations , we need specify only two parameters , assuming that the jet and ambient thermal pressures are equal , as in the simulations of dh94 . the morphology of 2d axisymmetric nonrelativistic jets depends on only @xmath20 and m , the mach number with respect to the sound speed within the jet ( norman , winkler , & smarr 1984 ) . in this paper , mach number always refers to this internal " value and we will refer to the mach number with respect to the ambient or external medium as either m@xmath25 ( @xmath26 ) or m@xmath27 ( @xmath28 ) , respectively . here , the ambient medium refers to the fluid into which the jet is propagating and the external medium refers to the fluid in contact with the length of the jet , which can be either the shocked ambient medium or the cocoon and is important for the development of the kh instability . in jets with flow speeds that are relativistic , there is a quantity analogous to the mach number , the relativistic or proper mach number , @xmath29 @xmath30 , as defined in the discussion of time - independent relativistic flows by @xcite ; the subscript @xmath31 refers to the sound speed ( e.g. , @xmath32 ) . since the nonrelativistic simulations are scale - free and @xmath20 and m are the main independent variables , the ambient medium density and sound speed are set initially to unity in the zeus code . we will discuss two sets of nonrelativistic equivalents , both of which are suggested by komissarov ( 1996 ) . one can determine @xmath20 and m for a nonrelativistic simulation by assuming that the velocities and pressures are equal in a relativistic simulation and its nonrelativistic equivalent and equating _ one _ of the following fluxes : useful power , thrust , or mass . we feel that the mass flux equivalent is less interesting , because the mass flux is physically less important than the other two fluxes and quantities that are associated with the mass flux , such as the mass of a radio lobe , are rarely estimated from observational data . thus , we have chosen not to compute the mass flux equivalent simulations . equating the momentum flux of jets in the two types of simulations leads to a more interesting situation . since momentum flux and ram pressure are equivalent within either the relativistic or nonrelativistic formulation , there is a similar expected jet - head - to - jet velocity ratio in both a relativistic simulation and its nonrelativistic momentum flux equivalent . another interesting case involves the useful power ( @xmath33 in the relativistic simulations ) , which is the energy flux that feeds the observed luminosity of radio lobes . for the useful power and thrust equating cases , we list below the equivalent nonrelativistic mass density and mach number in terms of the relativistic fluid variables for each region of the flow . recall that @xmath20 = @xmath34 and thus we need to calculate the equivalent nonrelativistic mass density in both the jet and ambient medium . the relationships for @xmath22 and m in the useful power equivalent are @xcite : @xmath35 @xmath36.\ ] ] the relationships that follow from equating thrust are : @xmath37 @xmath38 note that equation ( 4 ) reduces to m@xmath39 = @xmath12m@xmath40 . of the relativistic simulations , the pressure in the ambient medium is most significant in runs c and d , increasing the equivalent @xmath22 and decreasing @xmath20 by a factor of three ( for useful power ) or two ( for thrust ) . we will refer to each of the nonrelativistic equivalents by the relativistic simulation to which they are similar and two letters describing the nature of the equivalence , e.g. , b@xmath41 and b@xmath42 for the useful power and thrust equivalent , respectively , for the relativistic simulation run b. since we computed only one nonrelativistic equivalent of run a , we will refer to this as run a@xmath41 . the values of @xmath20 , internal mach number ( m ) and ambient mach number ( m@xmath25 ) are shown in table 1 for both sets of nonrelativistic simulations . note that run d in dh94 used a softer adiabatic index , @xmath43 , for the jet and ambient medium , and that we have reproduced this in the nonrelativistic equivalents of run d. since our relativistic simulations have the same @xmath20 , for these simulations we list the enthalpy ratio @xmath21 , which is one measure of relativistic thermal motions , in table 1 . the progression of @xmath21 is related to the increase in @xmath15 as @xmath12 is increased in the set of relativistic jets . since these simulations model light jets pressure - matched with the ambient medium , increasing @xmath15 leads to @xmath21 asymptotically approaching unity from below . the nonrelativistic equivalent jets are frequently denser than the external medium , which is a consequence of relativistic jets behaving as if they have a higher density than their proper density . specifically , @xmath20 would have to be @xmath44 in the most relativistic simulations ( runs c and d ) for nonrelativistic equivalents to have @xmath45 1 . examining the values of m@xmath25 reveals that the most relativistic simulations have power- and thrust - equating nonrelativistic equivalents that are transonic ( 1 @xmath7 m@xmath46 3 ) with respect to the ambient medium . one goal of this analysis is to compare the analytically predicted growth of the kh instability modes with the internal structure found in the two sets of simulations . we can derive mach numbers and jet - to - external ( i.e. , cocoon , _ not _ ambient ) density ratios for a nonrelativistic jet with stability properties similar to those of a relativistic jet by considering the approximation for the wave speed of kh - induced disturbances at resonance ( or maximum growth ) , @xmath47 , for a relativistic cylindrical jet ( from hrhd , equation 6c ) : @xmath48^{1/2 } } { [ { \rm m}_{\rm x}^2-\beta^2]^{1/2}+\gamma [ { \rm m}^2-\beta^2]^{1/2 } } v .\ ] ] the product @xmath12 m in relativistic jets appears approximately in the same role as the mach number in nonrelativistic jets , and is the same form used for m in the thrust - equating case . in order to estimate the jet - to - external density ratio , @xmath49 , we take the nonrelativistic limit for @xmath50/@xmath4 : m/(m@xmath27 + m ) . assuming that the jet and external pressures are equal , this term reduces to @xmath51 . using values of @xmath10 , m , and m@xmath27 from the relativistic simulations , we solve for desired values of @xmath49 in the nonrelativistic simulations by equating @xmath52 in eq . ( 5 ) to its nonrelativistic limit . note that typically m@xmath53 m@xmath25 , but that we do list m and m@xmath27 for runs e , b , and c below . for the nonrelativistic equivalents of all but relativistic run a , @xmath49 corresponds to an overdense jet ( see table 1 ) , dense enough that there is no significant difference between @xmath49 and @xmath20 . additionally , these values are roughly similar to the set of density ratios used in the thrust - equating simulations , which thus have _ both _ independent parameters appropriate for matching kh stability properties . even though runs e@xmath42 and b@xmath42 have density ratios quite different from the desired @xmath49 , the density ratios in runs c@xmath42 and d@xmath42 are a closer match . also , note that the @xmath20 in run b@xmath42 is similar to the desired @xmath49 for run e , with only a small difference in mach number between e@xmath42 and b@xmath42 . this suggests that the jet in run b@xmath42 should have similar stability properties as those of the jet in relativistic run e. we show schlieren plots , in which darker regions indicate a larger density gradient , for all of the simulations as the bow shock nears the right edge of the grid ( @xmath54 40 r ) in figure 1 . typical jet features include a bow shock , a terminal shock or mach disk of the jet flow , a cocoon of material lighter than the jet that has passed through this shock , and internal structure in the form of biconical shocks along the jet . in the smaller lorentz factor relativistic simulations , there is some curvature of the terminal shock , which is the very thin line close to the leading edge of the jet and almost perpendicular to the jet axis in the schlieren images . there is some difference in this curvature between the simulations , particularly between runs e and b , and in general the shape and curvature of the terminal shock varies with time . in some of the simulations and in particular run c , there is a secondary shock roughly 1020r behind and parallel to the bow shock , which is caused by shocked ambient material expanding supersonically backwards . this secondary shock is weaker , or nonexistent in run d , a consequence of slower expansion accompanying the softer adiabatic index . some of the schlieren plots show a nonphysical reflection of the bow shock at the outer radial boundary , generated by the edge of the numerical grid . the strength of this feature is somewhat exaggerated in a schlieren plot and the reflected bow shock does not interact significantly with any of the more important jet structures . despite the appearance that the reflection coincides with an enhancement in cocoon width in run b , we have determined that these coincident structures are unrelated . in this case , the expansion of the cocoon occurs before the reflection arrives . from the schlieren images in figure 1 , we see that the morphology of the jets changes with lorentz factor in a similar manner for all three sets of simulations . however , the two codes do not give identical results for the relativistic simulation with a nonrelativistic jet flow speed and for its nonrelativistic equivalent , i.e. runs a and a@xmath55 . while the cocoon width is similar , the detailed structure of the cocoon with the relativistic code is more regular and appears somewhat underesolved . this is caused by the relatively poor capturing of contact discontinuities by the relativistic harten - lax - van leer - einfeldt ( rhlle , see einfeldt 1988 and dh94 ) solver used in the relativistic simulations . these differences within the cocoon lead to the less regularly spaced biconical shocks within the jet in the nonrelativistic equivalent ( run a@xmath41 ) . given these differences between the numerical schemes for the same simulation , we will focus primarily on large - scale differences such as in the cocoon , jet head , and bow shock between a relativistic simulation and its nonrelativistic equivalents . with the aim of examining any differences in the appearance between relativistic simulations and their nonrelativistic equivalents and since in the nonrelativistic simulations we can rescale the velocity field , we do so by equating inlet velocities for a relativistic simulation and its nonrelativistic equivalent . from the rescaled @xmath10 in each zone in the nonrelativistic simulations , we compute a scaled lorentz factor . doppler boosting is to first order the primary contributor to the observed intensity from the flow . because doppler boosting dominates variations in the inferred intrinsic emissivity , for a given angle of view the lorentz factor should provide a useful ( if coarse ) estimator of the appearance of the jets at radio wavelengths . we display contours of two values of lorentz factor for the relativistic simulations and for the same two values of scaled lorentz factor for the useful power set of nonrelativistic simulations in figure 2 . the adopted contours in these plots are at a low lorentz factor , which is near unity but different for each simulation ( see fig . 2 for the values ) , and at the inflow lorentz factor in the appropriate relativistic simulation . assuming that the fluid flow is along the jet and that an observer is viewing the source close to the critical cone ( with a viewing angle @xmath56 1/@xmath12 ) , the radial width of these contours suggests an effective radius of observed intensity for the flow . since we will not compare simulated intensity _ within _ any set of simulations , assuming different viewing angles for each relativistic simulation does not invalidate the following analysis . we show only the useful power nonrelativistic equivalents in figure 2 because the thrust - equated set has a similar appearance . since nonrelativistic hydrodynamics imposes no upper bound on velocity , rescaled velocities above @xmath9 are unavoidable . rescaled @xmath10s above unity occur in most of the nonrelativistic equivalents of runs c and d ( the only exception of the four cases is run d@xmath41 ) , although the maximum rescaled @xmath10 in any simulation is 1.02 in run c@xmath57 . therefore , these equivalents are able to accommodate velocity rescaling reasonably well . the maxima of the rescaled @xmath10 is smaller for equivalents of run d than for those of run c. correspondingly , the maximum speed in relativistic run d is smaller than the maximum speed in relativistic run c ; one effect of a softer adiabatic index is to reduce the acceleration of the jet . in general , the relativistic simulations have a much larger width as measured by these lorentz factor contours than do the nonrelativistic simulations . in runs a and e , there are only small regions of high @xmath12 along the jet axis . this is also true for run a@xmath41 , but not for run e@xmath41 . this similarity between run a and a@xmath41 suggests that the different numerical methods admit similar accelerations in the simulations that are most similar ( note that a @xmath12 of 1.05 represents an acceleration of less than 2% above the initial jet velocity in run a ) , even though the appearance differs considerably in detail . while some aspects of the contours in runs a and a@xmath41 are similar , the most similar contours of any relativistic - nonrelativistic pair shown in figure 2 are the contours in runs d and d@xmath41 , which are narrow over most of the jet length and only display a small cocoon - like bulge near the terminal shock . also , noticeable in the high-@xmath12 contour is the pressure wave close to the inlet , which was discussed in hrhd ( the contour starts at @xmath58 r near the inlet and moves inward farther along the jet axis ) . this pressure wave is easiest to see in runs c , d , and d@xmath55 , and is perhaps seen in runs b and e@xmath41 as well . given our assumption that there is a correspondence between the lorentz factor and the perceived radio intensity from simulated jets , we see from figure 2 that the shape of emission regions are quite different between the pairs of simulations . specifically , high lorentz factor or bright " regions are smaller near the jet head and the internal structure is more varied in the nonrelativistic equivalents than in their relativistic counterparts . these differences between relativistic and nonrelativistic equivalents may also be seen in lighter jets ( komissarov & falle 1996 ) , but there the cocoon in the relativistic simulation contains significant backflow , suggesting that the cocoon could be doppler enfeebled and much dimmer than the jet flow if the jet is directed toward the observer . the doppler enfeeblement of radiation from a cocoon with a significant backflow could dim the cocoon by a large enough amount that light relativistic jets appear very narrow when the jet is directed nearly toward the observer . this narrow appearance of the cocoon and jet from a relativistic flow could reduce the difference between the apparent lobe widths of relativistic jets and their nonrelativistic equivalents . in both relativistic and nonrelativistic sets shown in figure 2 , the simulations with the most contours crossing the jet axis are for jets that have an underlying flow with a moderate @xmath12 , between roughly 2 and 5 . although this comparison only shows contours of @xmath12 and is not a true representation of the total intensity from a radio source , this is highly suggestive that jet flows of moderate lorentz factor may lead to the most knotty " appearance , which confirms a result of @xcite . we conclude that , despite the modest success of predicting the maximum flow velocity when rescaling a nonrelativistic simulation to the relativistic inflow speed , the rescaled velocities underscore the necessity of performing fully relativistic hydrodynamic simulations : the entire velocity field computed by a nonrelativistic simulation can not be simply rescaled to yield the velocity field of a relativistic jet . by comparing the width of the cocoon seen in simulations with analytic estimates based on simple assumptions , we determine the viability of such an analytic model for the interpretation of data , and gain insight into the origin of different flow speed - dependent morphologies . we derive equations for the ratio of cocoon to jet radius ( r@xmath59/r ) for each of the relativistic and nonrelativistic cases in appendix b. from these equations , we estimate a cocoon radius based on the relativistic form ( equation [ 6 ] ) for each of the relativistic simulations . while this relation is most applicable to run d , because the derivation assumes @xmath14 = 4/3 , we expect that the different adiabatic index in runs a , e , b , and c will have little effect on the cocoon radius and that it is valid to generalize this simple model to include other adiabatic indices . this follows from the results of duncan , hughes , & opperman ( 1996 ) , who recomputed run c using a modified version of the relativistic hydrodynamic code that allowed for a variable adiabatic index . the recomputed simulation demonstrated that even though the simulation displayed significant spatial and temporal variations in @xmath14 , there were minimal differences in the flow morphology . we display contours of r@xmath59/r in figure 3 from the analytic model in @xmath21@xmath12 space for the relativistic case from eq . 6 ( derived in appendix b.1 ) and in @xmath20m space for the nonrelativistic case from eq . 7 ( derived in appendix b.3 ) : @xmath60 @xmath61 for both cases , we show the positions of the appropriate simulations in parameter space . we have not placed the nonrelativistic equivalents for run d in fig . 3b , since the contours are shown for @xmath14 = 5/3 . however , there is only a small difference between the expected cocoon radius between the two adiabatic indices , with the difference being less than 20% for any of our simulations . we display the theoretical values of r@xmath59/r for each simulation as vertical lines in plots of radial profiles of density and pressure ( figure 4 ) , which have been measured at the position of local minimum density that is indicated by an arrow in each of the schlieren images in figure 1 . in the panels that show data from a relativistic simulation , both the lab frame density and proper density radial profiles are shown . in all three simulation sets , the cocoon radius expected from the simple model typically decreases as the lorentz factor is increased , with the exception of both nonrelativistic equivalents of run c , which have an expected r@xmath59 slightly larger than that of each run b equivalent . frequently the theoretical values in figure 4 are coincident with a local minimum in density or pressure , usually well within the cocoon , which we interpret as the cocoon center . the pressure usually increases outward from this radial position , consistent with a nearly fully expanded cocoon . while all three data profiles typically are similar within each relativistic simulation , there is one exception . the expected r@xmath59 in run c occurs at a smaller radius than the position of the minimum proper density and a larger radius than the position of the minimum lab frame density , but it is well - centered within a wide minimum in pressure . thus , the analytic estimate of r@xmath59 is a better indicator of the radial position of minimum pressure . also , this estimate is closer to local extrema in the relativistic cases than in the nonrelativistic simulations . specifically , in the nonrelativistic equivalents of runs c and d , the expected r@xmath59 is within the shocked ambient medium or farther from the jet axis than any cocoon - like feature in the radial profile . some of the underlying assumptions for the analytic expressions in appendix b are demonstrably poor . specifically , two of these assumptions are unlikely to be satisfied : 1 ) pressure balance between the jet , cocoon , and the ambient medium and 2 ) a slow advance speed of the terminal shock in the lab frame . the radial profiles in figure 4 show that the variation of pressure is typically an order of magnitude between the jet and the cocoon , demonstrating the weakness of the first assumption . the larger effective density of relativistic jets suggests a jet head advance speed nearly that of the jet flow speed ( based on the relation for @xmath62 in eq . [ a10 ] ) , which would violate the assumption of a slowly moving jet head . this is confirmed by comparing the average velocities of the terminal shock , @xmath63 in table 2 , with @xmath64 0.9@xmath9 for runs e , b , c , and d ( @xmath10 = 0.3 in run a ) . the approximation of a slowly moving head is poorly satisfied for all simulations but runs a or a@xmath41 . nevertheless , we see that the predicted cocoon radii are close to those simulated , which suggests that the overall simple model is a valid framework in which to interpret the results . the analytic expressions for r@xmath59/r successfully mimic the typical reduction in cocoon size that accompanies the increase in lorentz factor in the relativistic runs and for the most part their nonrelativistic equivalents . primarily because our sets of simulations do not have a large range in estimated r@xmath59/r , this estimate is not a good discriminate between a relativistic and nonrelativistic flow in our simulations . nevertheless , we see that either a small @xmath20 with moderate mach number or a small @xmath21 with moderate lorentz factor is required to generate the typically large cocoon of a fr ii source . balancing the sum of ram plus thermal pressure at the jet head between the jet and the ambient medium leads to a predicted jet head advance speed , which we will designate @xmath65 and is equal to @xmath66 . in appendix a , we show a slightly different method for deriving this than given by mart et al . ( 1997 ) , clarify the approximations used , and introduce notation used in appendix b. we define the efficiency as the mean velocity divided by this predicted jet head velocity , i.e. @xmath63/@xmath65 . we list the advance speeds and the efficiency for each simulation in table 2 , where we have again used inlet velocities to rescale all velocities in the nonrelativistic simulations . from the data in table 2 , it can be seen that the efficiencies in the nonrelativistic equivalents are within 25% of the efficiencies in the relativistic simulations . since the only inaccuracy in the position of the terminal shock is the numerical and artificial viscosities that smear the shock over a few zones , the differences in efficiencies between a relativistic and nonrelativistic pair is larger than the error in the measured position . there is a smaller difference between the efficiencies of these relativistic simulations and those of each run s nonrelativistic equivalent than between the efficiencies of relativistic simulations in mart et al . ( 1997 ) and those of nonrelativistic simulations of cylindrical jets in norman , winkler , & smarr ( 1983 ) . the main reason for this smaller difference is that these two sets of simulations have dissimilar initial conditions . specifically , efficiencies as large as 1.25 in mart et al . ( 1997 ) were associated with highly supersonic , cold jets with an adiabatic index of 4/3 , which is unlikely to be equivalent in any sense ( e.g. , equal in thrust , power , or adiabatic index ) to the nonrelativistic simulations in norman , winkler , & smarr ( 1983 ) or those presented here . the hot relativistic jets in mart et al . ( 1997 ) have efficiencies near unity , similar to those of our hot relativistic runs c and d. our simulations generally have efficiencies larger than the smallest of those in norman , winkler , & smarr ( 1983 ) , who point out that for nonrelativistic jets of equal @xmath20 the efficiency is at a minimum for mach numbers of roughly 3 . the nonrelativistic equivalent simulations have mach numbers between 6.0 and 25.4 , so the efficiencies in our set are understandably larger than the smallest in norman , winkler , & smarr ( 1983 ) , which had an efficiency of 0.49 . of our nonrelativistic simulations , run a@xmath41 has the smallest mach number and efficiency of our nonrelativistic simulations , and it is also one of the norman , winkler , & smarr ( 1983 ) set ; for the efficiency that we list as 0.61 , they find 0.59 . thus , we see that our relativistic simulations have efficiencies smaller than those in mart et al . ( 1997 ) , and our nonrelativistic simulations have efficiencies larger than in norman , winkler , & smarr ( 1983 ) . despite the implication of mart et al . ( 1997 ) that only hypersonic ( m @xmath67 30 ) nonrelativistic jets have efficiencies similar to those of relativistic jets , we are able to match the efficiencies of relativistic jets with our set of power- or thrust - equivalent supersonic , dense jets . as with the simulations of mart et al . ( 1997 ) , we find that jets with higher lorentz factors and their nonrelativistic equivalents have greater efficiencies , which is related to the ability of faster , effectively denser jets to have a narrow jet head ( see figure 1 ) . since the derivation for @xmath65 assumes that the areas on both sides of the terminal shock are equal , the degree of collimation of the jet flow is a primary cause for differences between estimated and computed jet head speeds in any of the simulations . in fact , an explanation for efficiencies greater than unity in the simulations of mart et al . ( 1997 ) is that the softer adiabatic index in a _ cold _ cylindrical relativistic jet can focus the jet flow and accelerate the terminal shock . in comparing the efficiencies of run c and d , we see that the smaller adiabatic index in a hot relativistic jet does not significantly affect the efficiency . the useful power and thrust equating cases for runs c and d display bow shocks that are weaker in the schlieren images than for the other simulations because the jet is transonic with respect to the ambient medium ( see m@xmath27 in table 1 ) . although the strength of the bow shock is different between the relativistic and nonrelativistic simulations , bow shocks are not usually identifiable in extragalactic radio sources . indeed , the faintness of bow shocks has been suggested by radiative transfer calculations using these relativistic simulations ( mioduszewski , hughes , & duncan 1997 ) . we conclude that this difference can not be used to distinguish relativistic jet flows in actual sources . however , the bow shock can affect the evolution of the cocoon via the shocked ambient medium . for example , a larger @xmath29 reduces the mach angle between the bow shock and the jet axis and leads to greater confinement of the cocoon via an enhanced thermal pressure in the shocked ambient medium . we list the ages of the jets ( in units of a dynamical time , t@xmath68 r/@xmath1 ) in table 2 . in general , the ages for nonrelativistic equivalents are similar to those for the corresponding relativistic simulation . of particular note is the closeness in ages of the thrust equating cases with the relativistic simulations , which is related to the equivalence of momentum flux and ram pressure , as discussed in 2 . note that a comparison of ages in dynamical times is more relevant between a relativistic simulation and its nonrelativistic equivalents than within any set of simulations , since the concept of a fiducial dynamical time " is most useful for cases where r and @xmath1 can be scaled to the same value in each compared simulation . this can not be true for this relativistic set of simulations because the relativistic equations require that velocities are not scale - free ( and the temperature in the ambient medium is different for each of these relativistic simulations ) , which is not a requirement of the nonrelativistic simulations . larger temperatures and sound speeds in the relativistic simulations lead to smaller dynamical times , and explain the larger ages in the hotter and faster relativistic simulations . motivated by the result of hrhd that a larger lorentz factor is associated with a smaller variation of pressure along the jet axis , we analyze these pressure variations in the nonrelativistic simulations . hrhd pointed out that the variations in runs a , e , and b are typically larger than those of runs c and d. since the thrust - equating nonrelativistic simulations and the relativistic simulations should have similar stability properties ( e.g. , similar wave speeds and growth lengths , see discussion in 2 ) , we display the pressure variations along an axial slice in both sets of simulations in figure 5 . these slices are taken from a row of zones near r = r/8 to avoid uncharacteristically large deviations that occur along the symmetry axis in a cylindrical jet . a comparison of the axial pressure slices reveals that there are fewer and smaller deviations along the sequence for the nonrelativistic simulations as well . in addition , in the thrust - equating cases we reproduce the fairly abrupt break between many large - amplitude variations in runs a , e , and b and the few small - amplitude variations in runs c and d. recall that the analysis in 2 suggests that the stability properties of run b@xmath42 is most similar to that of run e of the relativistic set . since we have grouped the simulations into two groups between runs b and c , this distinction is not very useful . the trend of fewer and smaller variations in pressure for faster jets is also evident in axial slices of the rescaled axial velocity and to some extent in the rescaled radial velocity , neither of which is shown . although we do not display the useful power - equating set of simulations , the set has similar behavior and demonstrates the general trend toward smaller pressure variations as the lorentz factor is increased in the relativistic set . thus , the amplitudes of pressure variations within the length of the jet vary similarly in both relativistic simulations and their nonrelativistic equivalents , and the behavior of both sets may readily be understood through linear stability analysis . however , the match between the stability properties of relativistic and nonrelativistic equivalent simulations is less than perfect . perhaps most importantly , for the nonrelativistic equivalents of the largest lorentz factor simulations ( runs c and d ) , variations of the rescaled velocity are larger than the velocity variations in the relativistic simulations because the nonrelativistic simulations have no upper limit on their speed . this difference re - emphasizes the need for fully relativistic simulations . a detailed stability analysis of runs a , e , b , c , and d has appeared elsewhere ( hrhd ) with the conclusion that relativistic jets are _ not _ unconditionally stable " ( mart et al . 1997 ) but that the simulations which show jets that appear stable have not excited any instabilities . for example , hrhd interpret the apparent relative stability of run c as due to only a weak coupling of the first body wave of the pinch mode with a pressure wave that originates at the inlet , while the lack of cocoon vorticities prevents any perturbations to the jet surface from jet material that has passed through the terminal shock . note that this is a common mechanism for exciting the kh instability in supersonic jets . ferrari , trussoni , & zaninetti ( 1978 ) state that small wavelength perturbations are ineffective for @xmath69 ( d/@xmath70r)@xmath71 while large wavelength perturbations are stable , where d is the length of the jet ( e.g. , the distance between the parent galaxy and a hot spot ) and @xmath70 is a scaling parameter between 0.01 and 0.1 . this confirms a general result of the stability analysis in hrhd , who found a maximum wavelength for body waves of the pinch mode in a relativistic jet . the condition on @xmath12 suggests that instabilities can grow for slow " relativistic jets or within faster jets of sufficient length . for example , this criterion suggests that in our simulations , which terminate when the bow shock reaches @xmath72 40r , the jets can show evidence of instability if @xmath73 20 ( if @xmath70 is at its maximum value ) , which is a condition that all of our relativistic jets satisfy . however , as described above , little instability was seen due to the absence of a strong coupling of a kh mode with available perturbations . in comparison , the fastest simulations in mart et al . ( 1997 ) have @xmath12 = 22.4 and they were computed out to only 50r , so the jet speed is sufficiently large that very little internal structure formed . evidently , the condition on @xmath12 alone is insufficient to assess the stability properties of flows , and careful consideration must also be given to how available perturbations couple with the available modes . in the set of relativistic simulations , both the bulk flow ( the lorentz factor ) and the random motions ( the internal pressure ) were increased in tandem : the jets in run a and run e were cold , the jet in run b was warmer , and the jets in runs c and d were hot . one indicator of this is the mach number ( _ not _ the relativistic mach number ) in each of these simulations . since the sound speed within the jets varied from 0.05@xmath9 to nearly 0.8@xmath9 , the mach numbers varied between 6 in run a and 1.28 in run c. alternatively , we can calculate the ratio of mass - energy density to the pressure - dependent portion of enthalpy , @xmath74 , which is a ratio bicknell ( 1994 ) has designated @xmath75 . for the @xmath14 = 5/3 cases , @xmath75 = 1/2.5@xmath15 ( with the jet proper density and the speed of light set to unity ) , which varies by an order of magnitude between each of runs e , b , and c , with values of 15.6 , 1.4 , and 0.098 , respectively . for run d ( with @xmath76 ) , @xmath75 is the more familiar 1/4@xmath15 and equal to 0.091 . this shows that run b is the closest to equipartition between mass and thermal energies . run a is a very cold simulation with @xmath77 250 . our simulations represent an evenly distributed sample over two orders of magnitude in @xmath75 , and include a simulation near equipartition between mass and thermal energy . by comparison , the simulations in mart et al . ( 1997 ) have @xmath75 of either @xmath78(10 ) or @xmath78(0.01 ) , but nothing in between . thus , our simulations investigate a new region of parameter space for differences between hot and cold jets . in order to disentangle the effects of increasing lorentz factor from those of increasing pressure , we have completed another set of relativistic simulations using the same code and numerical techniques as for the first set . these new simulations fill in places in lorentz factor - internal pressure ( @xmath12@xmath15 ) space beyond runs e , b , and c ( see table 3 ) . all of the new runs have @xmath79 . we name the new simulations with the designation of an original simulation with the same lorentz factor and a subscript indicating a different temperature ( e.g. , b@xmath80 has @xmath12 = 5.0 as did run b , but a pressure and @xmath75 similar to those of the cooler run e ) . we also list the relativistic mach number , @xmath29 , and the mach number , m , for each case in table 3 , which shows the decrease in m that accompanies an increase in thermal pressure . schlieren plots of the new set of simulations and runs e , b , and c are displayed in figure 6 , which has columns of equal @xmath15 increasing to the right and rows of equal @xmath12 increasing from top to bottom . as with the schlieren plots in figure 1 , all plots in figure 6 are at the time when the bow shock nears the edge of the grid . we list the age of each simulation in units of a dynamical time in table 3 . these new simulations have been analyzed for their stability properties , as was done in hrhd . this analysis can explain much of the internal structure in the schlieren images of figure 6 . the normal mode analysis suggests that faster , cooler relativistic jets are more stable , so run c@xmath80 should have the least internal structure of the simulations in figure 6 . any simulation in a panel to the right of ( hotter than ) or above ( slower than ) run c@xmath80 should be more unstable and have more structure . hotter simulations should be more unstable because the increased sound speed shortens a jet crossing time for a perturbation . faster flows are more stable because they naturally increase the spatial growth length . with the exception of the hottest simulations , we do see the expected instability trends , with the slow , hot run e@xmath81 appearing the most internally unstable . therefore , we confirm the most basic results of the stability analysis . the stability analysis of hrhd also suggests an explanation for the relative lack of structure in the hot " simulations . run b@xmath82 appears to have the least structure and as we have already discussed run c has only a few small internal variations in pressure . not only do the conditions in the two hot simulations admit growing modes , as shown by the normal mode analysis , but both have external mach numbers , m@xmath27 , near 1.3 ( see table 3 ) . the stability analysis suggests that this is near a critical value , where the solution to the dispersion relation changes dramatically and quickly . we display pinch mode solutions from the linear analysis for the three simulations with @xmath12 = 5 , i.e. runs b@xmath80 , b , and b@xmath83 , in figure 7 . there is a noticeable difference between the shape of the growth rate of the surface wave of the pinch mode for run b with m@xmath27 = 1.56 and for run b@xmath83 with m@xmath84 = 1.31 , and this difference exists even between run c with m@xmath84 = 1.33 and run b@xmath83 . as the external mach number decreases toward the critical value of 1.3 , the growth rate of the surface wave increases at all frequencies so that it dominates the growth rates of all the body waves , whose solutions remain similar . in addition , the dominant surface wave for m@xmath85 1.3 does not pass through a resonance , so there is no frequency associated with a peak growth rate ( see figure 7 ) . this contrasts with the body waves , which still have a peak growth rate while the jet is supersonic ; body waves are suppressed in a subsonic flow . as with the hot jets in hrhd , the stable appearance of jets in these simulations is most likely due to a weak coupling between the instability at all frequencies and the jet . additionally , some simple physical arguments for the amplitude of displacement surfaces ( e.g. , as in hrhd ) indicate that short wavelength perturbations will saturate at small amplitudes . these small wavelengths would need adequate grid resolution to grow within a numerical simulation . we will now apply our results to available observations in the manner of williams ( 1991 ) . for the following comparison with observations , we will concern ourselves with data for a single source . a large fraction of the observations of this source have recently been compiled in a review article by carilli & barthel ( 1996 ) . at two orders of magnitude above the fr i fr ii break and a redshift of 0.0562 ( @xcite ) , is a very powerful and very close double - lobed radio source . based on results from a 3d mhd nonrelativistic jet simulation with zeus-3d , @xcite suggested that a dense jet with @xmath12 m , which can be thought of as @xmath29 in a cold jet , @xmath56 10 and a jet - to - cocoon density ratio of @xmath86 6 is needed in order to reproduce the observed helical and filamentary structure . a dense jet is appropriate for such a simulation since the ambient medium represents the cocoon , which is lighter than the jet . while komissarov & falle ( 1998 ) have modeled cygnus a with some success by using the self - similar properties of an expanding jet , we apply the results from our cylindrical jets to cyg a by first estimating the age of the source . the distance of the hotspot or farthest edge of the lobe from the core , d , is 50 h@xmath87 kpc , where h@xmath88 is the hubble constant normalized by @xmath89 . we adopt a jet radius @xmath72 0.5 kpc , based on the results of perley , dreher , & cowan ( 1984 ) who observed that the jet has a constant fwhm of 07 for the inner d ; at the redshift quoted above this implies a radius of 575 h@xmath87 pc . therefore the jet has propagated @xmath72 100 r , and it would take roughly 2560 dynamical times to reach that distance ( based on 1025 t@xmath90 to reach @xmath72 40 r in simulations with @xmath91 8 , see table 2 ) . high resolution rosat results indicate the inner 50 kpc of cluster has a temperature of 3 @xmath6 10@xmath92 k ( reynolds & fabian 1996 ) , so the sound speed in the external medium is roughly @xmath93 and 2560 dynamical times corresponds to roughly 1545 myrs . this age range straddles the observationally determined value of 30 myr ( carilli et al . 1991 ) , which is from a self - consistent model based on the spectral aging of the lobes . since the age estimates for the nonrelativistic and relativistic are close , we can not use this estimate to determine whether the jet in propagates at a relativistic speed . from the analytic expressions for the radius of the cocoon ( see eqs . [ 6 ] and [ 7 ] ) , we can also estimate @xmath21 in the relativistic case or @xmath20 in the nonrelativistic case . the typical radius of the cocoons in is 10 kpc , so r@xmath59/r is roughly 20 . from eq . ( 6 ) , assuming that the lorentz factor is in the range of 510 , and accounting for the fact that the analytic expression frequently predicted the cocoon _ center _ , we estimate an enthalpy ratio of roughly 10@xmath94 , with smaller lorentz factors requiring smaller values of @xmath21 ( as seen in figure 3a ) . from eq . ( 7 ) and assuming m is roughly 510 , @xmath20 should be 3 @xmath95 . this suggestion of a very light jet is consistent with an estimate by clarke ( 1996 ) , who from 3d nonrelativistic simulations proposed that @xmath20 is 3 @xmath6 10@xmath94 and m is 15 . either of the estimates for @xmath21 and @xmath20 suggest the jet in is very light , and possibly cold . also , it should be noted that this value of @xmath20 is smaller than that of any jet simulation yet attempted . in this paper , we have examined morphological and dynamical differences between simulations of relativistic and nonrelativistic jets , including the age , efficiency , velocity field , and internal structure of the jets as well as the size of the cocoon . an important result of these comparisons is that the velocity field of nonrelativistic jet simulations can_not _ be scaled up to give the spatial distribution of lorentz factors seen in relativistic simulations . specifically , such a scaling substantially underpredicts the size of the region of significant lorentz factor . since the local lorentz factor is the primary factor determining the brightness of the source , this suggests that a nonrelativistic simulation can not yield the proper intensity distribution from a relativistic jet . from the rescaled velocity plots , we see evidence for a result more definitively seen in @xcite : jet flows of moderate lorentz factor ( @xmath96 25 ) generate the most knotty appearance . other general results are that relativistic simulations and their nonrelativistic equivalents have similar ages ( in dynamical time units , @xmath0 r/@xmath1 ) , efficiencies , and that jets with a larger lorentz factor have a smaller cocoon size . the ages and efficiencies in the nonrelativistic simulations with equal thrust are a particularly good match to the relativistic simulations . also , for both relativistic simulations and their nonrelativistic equivalents with @xmath14 = 5/3 , the widths of the cocoon radii predicted from simple models frequently match a density minimum within the cocoon , and therefore are adequate estimators of cocoon centers . however , due to the limited range of estimated cocoon widths in all of our relativistic and nonrelativistic sets of simulations , these simple models do not discriminate well between relativistic and nonrelativistic flow speeds . slices of pressure near the symmetry axis vary similarly in both relativistic and nonrelativistic simulation sets . even in the nonrelativistic simulations , these data slices show fewer and smaller deviations as the lorentz factor in the equivalent relativistic simulation is increased . in addition to this comparison , we also have completed four new relativistic simulations to investigate the effect of varying thermal pressure on the structure of simulated relativistic jets . these simulations confirm that faster ( larger lorentz factor ) , colder jets are more stable , with smaller amplitude and longer wavelength internal variations . however , an exception to this occurs for hot jets , where we see evidence beyond that presented in hrhd that hot , relativistic jets appear stable because of a weak coupling between perturbations and available growing kh instability modes . the stability analysis also suggests a different behavior as the external mach number decreases through @xmath72 1.3 . jets with such a small m@xmath27 should have large growth rates at high frequencies and small wavelengths . these jets appear stable also because of the weak coupling of between the instability and the perturbations within the jet . a higher grid resolution may allow the smaller wavelength perturbations to grow . as an example of how these simulations can be applied to observations , we use our results to estimate the age , jet - to - ambient density , and jet - to - ambient enthalpy of . although none of these estimates can determine if the jet in is relativistic or nonrelativistic , they confirm independent estimates of these parameters . our results suggest that the jet is light with @xmath97 , and possibly cold with @xmath98 if the jet has a relativistic flow speed . future work will focus on extending the techniques used in the relativistic jet simulations ( i.e. , amr and rhlle ) to allow for high resolution three dimensional relativistic simulations and the analysis thereof . since the growth rates of nonaxisymmetric modes in nonrelativistic jets are larger than the growth rates of the symmetric modes , nonaxisymmetric relativistic jets should be more unstable than the relatively stable simulations presented here . thus , investigating the stability of 3d relativistic jets will reveal much about the nature of astrophysical jets . the momentum ` discharge ' of a jet is @xmath99a,\ ] ] where @xmath16 is total internal energy density , @xmath15 is pressure , @xmath100 is the four - velocity , and @xmath101 is the area of the flow . an estimate of the advance speed of the head of a jet may be made by balancing the discharges associated with the jet and ambient media in the frame of the advancing mach disk - contact surface - bow shock structure . since we assume the area is common , we have @xmath102 where we have used @xmath103 , and subscript @xmath104 refers to the jet , while subscript @xmath105 refers to the ambient medium . subscript @xmath9 refers to the jet speed measured in the advancing head frame , so that @xmath106 and @xmath107 if the jet and ambient pressures are equal , or the jet flow is hypersonic , the above simplifies to @xmath108 making the following definitions : @xmath109 @xmath110 and @xmath111 we see that @xmath112 substituting for @xmath113 and @xmath114 from equations ( a3 ) and ( a4 ) , after some algebra we find @xmath115 which is the speed of ambient flow towards the head , and thus the speed of the head through the ambient medium . note that the quantity @xmath116 is the predicted head velocity , which is designated @xmath65 in the text . this is the relativistic generalization of the familiar newtonian form @xmath117 the particle ` discharge ' of a jet is @xmath118 where @xmath8 is the baryon density , and other quantities are as defined above . jet material thermalized at a mach disk expands laterally to from a cocoon , but the particle discharge is conserved , so @xmath119 where the subscript @xmath9 refers to the cocoon . as @xmath120 and @xmath121 ( the jet displaces cocoon material ) , we have @xmath122 if we adopt the picture that shocked jet material streams beyond the mach disk to the vicinity of the contact surface , and expands laterally to form the cocoon , then , until further longitudinal expansion has given the material a back flow velocity in the frame of the head , this material is moving at the speed of the head as characterized by @xmath116 and @xmath123 above . thus @xmath124 we may write @xmath125 if the jet is in pressure balance with the cocoon , so that @xmath126 . using the previously - derived relation for @xmath127 , the term @xmath128 may be recast , and after some algebra we get @xmath129 the terms @xmath130 and @xmath131 are determined by the rankine - hugoniot conditions at the mach disk . the simplest case ( applicable to run d in dh94 ) is for @xmath76 ; then these ratios are functions only of @xmath132 . let primes denote values measured in the frame of the shock discontinuity , and subscript @xmath133 signify the upstream state . given the extreme equation of state , from knigl ( 1980 ) we have @xmath134 because @xmath135 and @xmath136 for the adopted adiabatic index . from mass conservation @xmath137 so that @xmath138 for the adopted adiabatic index , @xmath139 ( for _ any _ flow speed ) , and after some algebra the above equation may be recast as @xmath140 note that both pressure and density depend on primed quantities those measured in the head frame . for situations in which the head moves slowly compared with the jet flow speed , it will be a reasonable approximation to use the jet parameters directly . using arguments similar to those above we have @xmath141 so that @xmath142 it follows as before that @xmath61 from the nonrelativistic rankine - hugoniot conditions we may write @xmath143\times \left[\frac{2\gamma{\rm m}^2-\left(\gamma-1\right ) } { \left(\gamma+1\right)}\right]^{1/\gamma}.\ ] ] note again that mach number is determined by the upstream flow speed in a frame moving with the shock , but that for a fast jet and slowly moving bow , the jet speed will yield a good approximation to this . cccccccccccc a & @xmath145 & 0.10 & 6 & 1.05 & 0.11 & 6.1 & 18.1 & 0.11 & 6.0 & 18.1 e & @xmath145 & 0.11 & 8 & 2.5 & 1.03 & 10.2 & 9.9 & 0.70 & 8.3 & 9.9 & 8.8b & @xmath145 & 0.16 & 8 & 5.0 & 6.70 & 12.6 & 4.9 & 3.96 & 9.4 & 4.7 & 41.8c & @xmath145 & 0.56 & 8 & 10.0 & 73.1 & 18.0 & 2.1 & 55.5 & 12.8 & 1.7 & 83.0d & @xmath146 & 0.57 & 15 & 10.0 & 74.4 & 25.4 & 2.9 & 57.1 & 18.0 & 2.4 & 93.1 lccccc a & 0.017 & 11.0 & 0.058 & 0.075 & 0.77 a@xmath41 & & 14.0 & 0.044 & 0.072 & 0.61 e & 0.092 & 8.1 & 0.314 & 0.417 & 0.75 e@xmath41 & & 9.8 & 0.433 & 0.464 & 0.93 e@xmath42 & & 8.5 & 0.373 & 0.419 & 0.89 b & 0.208 & 14.7 & 0.554 & 0.652 & 0.85 b@xmath41 & & 12.3 & 0.646 & 0.707 & 0.91 b@xmath42 & & 14.5 & 0.575 & 0.654 & 0.88 c & 0.581 & 26.9 & 0.828 & 0.877 & 0.94 c@xmath41 & & 21.5 & 0.868 & 0.891 & 0.97 c@xmath42 & & 26.5 & 0.840 & 0.876 & 0.96 d & 0.418 & 19.5 & 0.845 & 0.879 & 0.96 d@xmath41 & & 15.5 & 0.876 & 0.891 & 0.98 d@xmath42 & & 19.4 & 0.852 & 0.878 & 0.97 lccrrrrccc e & 0.051 & 0.275 & 7.8 & 2.5 & 8.0 & 3.33 & 11.5 & 2.86 b@xmath80 & 0.051 & 0.275 & 7.8 & 5.0 & 17.1 & 3.56 & 5.7 & 1.96 c@xmath80 & 0.051 & 0.275 & 7.8 & 10.0 & 34.8 & 3.62 & 4.6 & 1.88 e@xmath81 & 0.277 & 0.522 & 1.4 & 2.5 & 3.7 & 1.76 & 24.1 & 1.67 b & 0.277 & 0.522 & 1.4 & 5.0 & 8.0 & 1.88 & 14.7 & 1.56 c@xmath81 & 0.277 & 0.522 & 1.4 & 10.0 & 16.3 & 1.91 & 10.5 & 1.63 b@xmath83 & 4.096 & 0.779 & 0.1 & 5.0 & 3.9 & 1.26 & 30.8 & 1.31 c & 4.095 & 0.779 & 0.1 & 10.0 & 8.0 & 1.28 & 26.9 & 1.33

we have imaged in co(2 - 1 ) the molecular gas in ngc 1275 ( perseus a ) , the cd galaxy at the center of the perseus cluster , at a spatial resolution of kpc over a central region of radius kpc . per a is known to contain of molecular gas , which has been proposed to be captured from mergers with or ram - pressure stripping of gas - rich galaxies , or accreted from a x - ray cooling flow . the molecular gas detected in our image has a total mass of , and for the first time can be seen to be concentrated in three radial filaments with lengths ranging from at least 1.12.4 kpc all lying in the east - west directions spanning the center of the galaxy to radii of kpc . the eastern and outer western filaments exhibit larger blueshifted velocities with decreasing radii , whereas the inner western filament spans the systemic velocity of the galaxy . the molecular gas shows no signature of orbital motion , and is therefore unlikely to have been captured from gas - rich galaxies . instead , we are able to reproduce the observed kinematics of the two outer filaments as free - fall in the gravitational potential of per a , as would be expected if they originate from a x - ray cooling flow . indeed , all three filaments lie between two prominent x - ray cavities carved out by radio jets from per a , and closely resembles the spatial distribution of the coolest x - ray gas in the cluster core . the inferred mass - deposition rate into the two outermost filaments alone is roughly . this cooling flow can provide a nearly continuous supply of molecular gas to fuel the active nucleus in per a.

0712.2979

rich groups and clusters of galaxies are immersed in hot ( @xmath0@xmath4@xmath5 k ) x - ray - emitting gas that constitutes a large fraction of their baryonic mass @xcite . radiative cooling of this gas , by virtue of its x - ray emission , should result in an inflow of relatively cool gas to the cluster center ; i.e. , a x - ray cooling flow @xcite . in the absence of any reheating , this flow is inferred to have mass - deposition rates ranging from @xmath0@xmath6 to @xmath0@xmath7 depending on the given cluster , and over a hubble time ( @xmath0@xmath8 ) should deposit @xmath0@xmath9 to @xmath0@xmath10 of relatively cool gas in the central cd ( giant elliptical ) galaxy ( e.g. , reviews by * ? ? ? * ; * ? ? ? numerous searches have been made for the predicted mass sink from x - ray cooling flows . these searches cover a broad range of temperatures , from ionized gas at temperatures of @xmath0@xmath11 ( e.g. , through the ovi line in the ultraviolet ; * ? ? ? * and references therein ) down to @xmath0@xmath12 ( e.g. , h@xmath13 , @xmath14}$ ] and @xmath15}$ ] lines in the optical ; * ? ? ? * and references therein ) , to neutral gas at temperatures @xmath16 in atomic hydrogen ( 21-cm line ; * ? ? ? * and references therein ) , hot molecular hydrogen ( near - ir vibrational lines ; * ? ? ? * and references therein ) , as well as cool molecular hydrogen ( traced using rotational lines of carbon monoxide at mm wavelengths ; * ? ? ? * ; * ? ? ? * and references therein ) . none of these searches have found gas at quantities anywhere near the abovementioned predicted levels ( prior to the era of the xmm - newton satellite as explained below ) in putative cooling - flow clusters ( e.g. , review by * ? ? ? * ; * ? ? ? * ) . over the past few years , observations of hot gas in galaxy clusters with the xmm - newton and chandra x - ray observatories have radically altered our concept of x - ray cooling flows . spectra taken with xmm - newton at high energy resolution and sensitivity indeed show a decrease in gas temperature towards the cluster center , but fail to reveal any detectable gas at temperatures below about one - third the bulk ambient temperature @xcite . these results provide no evidence for a cooling flow , and imply that any gas cooling to temperatures @xmath17@xmath18 has a mass - deposition rate at least ten times lower than that previously inferred . furthermore , images taken with chandra at high angular resolutions reveal that the x - ray gas at the centers of putative cooling - flow clusters is usually disturbed ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? x - ray cavities are commonly seen , spatially coincident with radio jets from the central galaxy where detectable ( * ? ? ? * and references therein ) , indicating that energetic particles injected from an active galactic nucleus ( agn ) are responsible for the observed disturbances . the work needed to inflate these x - ray cavities implies that the energy output of agns is more than sufficient to balance radiative losses from the x - ray gas , and can therefore greatly diminish if not quench the cooling flow @xcite . against this backdrop , in the last few years searches targeting putative strong cooling - flow clusters have become relatively successful at detecting massive amounts of cool molecular hydrogen gas as traced in carbon monoxide ( co ) at the centers of these clusters @xcite . about twenty such examples are now known , whereas earlier searches ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) detected co gas at the center of only one cluster , the perseus cluster @xcite . the earlier searches targeted relatively nearby clusters which , in the light of the downward revisions in mass - deposition rates , are now known to mostly have very weak if any cooling flows . on the other hand , the more recent surveys @xcite target more distant clusters with higher mass - deposition rates as inferred in the traditional picture of x - ray cooling flows . follow - up imaging where performed reveals that the molecular gas is usually located within a region of @xmath19 kpc coincident with the central galaxy @xcite . the inferred mass of molecular hydrogen gas spans the range @xmath0@xmath20@xmath9 , still much lower than that predicted in the traditional picture of x - ray cooling flows , but closer now to the revised picture given the current upper limits in mass - deposition rates . the amount of cool molecular gas far outweighs all other gas components detected at temperatures below @xmath0@xmath21 combined . these results leave open the possibility that x - ray cooling flows may deposit what is still a massive amount of cool gas compared with that normally present in galaxies ( especially elliptical or cd galaxies ) . a cooling flow , however , is not the only possible source of molecular or other relatively cool gas in cd galaxies . ram - pressure stripping of the interstellar medium of gas - rich galaxies by the hot intracluster medium , or cannibalisms of gas - rich galaxies by the cd galaxy , provide plausible if not more attractive alternatives ( e.g. , reviews by * ? ? ? * ; * ? ? ? * ) . unlike cooling flows , ram - pressure stripping is firmly established to occur in cluster galaxies ( e.g. , review by * ? ? ? * ) , and likewise mergers and cannibalisms in field galaxies ( e.g. , review by * ? ? ? indeed , cd galaxies presumably attained their present - day mass through repeated cannibalisms of cluster galaxies , or repeated merging of subclusters and galaxy groups before or during cluster virialization ( e.g. , see discussion in * ? ? ? * ) . in support of these arguments , a number of elliptical galaxies in the field , members of groups not recognized to have x - ray cooling flows , or those residing in but not at the centers of clusters have been detected in co with inferred molecular hydrogen gas masses of up to several @xmath0@xmath22 ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . at least one of these galaxies , ngc 759 , resides in but not at the center of a putative cooling - flow cluster , abell 262 . ngc 759 has an inferred mass in molecular hydrogen gas of @xmath23 , at least two orders of magnitude higher than the upper limit placed on the mass of molecular hydrogen gas in the central cd galaxy , ngc 708 @xcite . these results clearly demonstrate that a cooling flow is not the only mechanism capable of depositing large quantities of molecular gas in cluster elliptical galaxies . at the present time , the available observations do not allow us to confidently discriminate between the different possible origins for the molecular ( or other relatively cool ) gas in cd galaxies . yet , naively , such a distinction should be possible simply from the observed spatial distribution and kinematics of the molecular gas . in the absence of any disturbances , the molecular gas deposited by a x - ray cooling flow should simply flow radially inwards along the gravitational potential of the cd galaxy . on the other hand , gas captured from a gas - rich galaxy is expected to orbit ( and hence exhibit a global rotation ) about the cd galaxy , reflecting the orbit ( or passage ) of the gas - rich galaxy around ( or through ) the cluster center or cd galaxy . here , we present a high angular - resolution co observation of the cd galaxy perseus a ( per a ; ngc 1275 ) that reveals for the first time an ordered spatial - kinematic pattern in its molecular gas . located at a comparatively close distance of @xmath074 mpc ( redshift @xmath24 , and assuming @xmath25 and @xmath26 ) , per a lies at the center of the perseus cluster ( abell 426 ) , the x - ray - brightest cluster in the sky . the radiative cooling time in the inner few tens of kpc of the perseus cluster is less than @xmath0@xmath27 , leading to a mass - deposition rate of a few @xmath28 in the absence of any heat input @xcite , the highest of any cluster within its distance . for over a decade , perseus a was the only cd galaxy in a putative cooling - flow cluster known to contain molecular gas , which together with its other peculiarities and relative proximity have made it the subject of intensive studies . this galaxy has been mapped with single - dish telescopes in the @xmath29 @xcite , @xmath30 @xcite , and @xmath31 @xcite transitions of co. the most recent map in co(2 - 1 ) by @xcite , which has the highest angular resolution and covers the widest area , shows that the molecular gas is elongated approximately east - west across the center of the galaxy . in agreement with previous single - dish maps , the molecular gas shows no apparent ordered spatial - kinematic pattern , with predominantly blueshifted velocities both east and west of the nucleus , and velocities near the systemic velocity immediately around the nucleus . the inferred mass of molecular hydrogen gas in per a is @xmath0@xmath1 ( fig . 4 of * ? ? ? * ) , far larger than that yet detected in field or non - cd cluster elliptical galaxies in the local universe ( e.g. , review by * ? ? ? * ) . the only previous inteferometric map of the molecular gas in per a was made in co(1 - 0 ) with the nobeyama millimeter array ( nma ) by @xcite . this map differs in a number of important respects to single - dish maps ( see discussion in * ? ? ? * ) , and bears little resemblence to that shown here in co(2 - 1 ) . we suspect that the co(1 - 0 ) interferometric map may be compromised by inaccurate bandpass calibration , as explained in the next section . the only other published attempt to image the molecular gas in per a with an interferometer was made by @xcite in co(2 - 1 ) with the iram plateau de bure interferometer ( pdbi ) , where virtually all the emission was resolved out . readers interested in the technical details of our observations and data reduction should now proceed to @xmath32 . those interested only in the results can skip ahead to @xmath33 . in @xmath34 we compare the observed properties of the molecular gas with predictions for capture from gas - rich galaxies , and in @xmath35 with predictions for accretion from a x - ray cooling flow . in @xmath36 , we discuss the implications of our results for the cooling flow in the perseus cluster . readers interested only in a concise summary of our results and interpretation can proceed directly to @xmath37 . mapping of line emission from per a at radio wavelengths is made difficult by the much stronger continuum emission of the galaxy . extended jets and lobes dominate the continuum at centimeter wavelengths @xcite , whereas a compact nuclear source dominates the continuum at millimeter wavelengths @xcite . in such a situation , accurate measurements of the spectral response ( passband ) of each antenna and its variation with time are required . any measurement errors introduce a frequency and time dependence in the point spread function of the telescope , which when convolved with the continuum source leave spatially - extended and frequency - dependent artifacts in the map that can be mistaken for or alter the line emission . we suspect that such artifacts may be responsible for the differences seen between the interferometric map of @xcite in co(1 - 0 ) ( rest wavelength at 2.6 mm ) and the single - dish maps mentioned in @xmath38 , as well as the map made here . the continuum emission of per a is weaker at shorter wavelengths , and hence the required measurement accuracy and stability of antenna passbands become correspondingly less stringent at higher transitions of co. we have therefore chosen to map per a in co(2 - 1 ) at 1.3 mm , using the submillimeter array ( sma ) . the observations spanned @xmath05.5 hrs with an on - source time of @xmath04 hrs on oct 22 , and spanned @xmath010 hrs with an on - source time of @xmath07 hrs on oct 24 . the weather was relatively poor on oct 22 , and so we report here only the observation on oct 24 . the sma was in its compact configuration , and with 6 antennas operating provided projected baselines over the observation ranging from about 868 m ( 652 k@xmath39 ) . we observed the co(2 - 1 ) line in the upper sideband of the double - sideband receivers . the correlator was configured to provide twenty - four independent spectral windows ( `` chunks '' ) in each sideband . each window was divided into 128 channels with a channel width of 0.8125 mhz ( i.e. , @xmath0@xmath40 ) , giving a total frequency coverage in each sideband ( with a small overlap between windows at their edges ) of 1.97 ghz ( i.e. , @xmath02590 km / s ) . to accurately measure and track variations in the antenna passbands with time , we observed mars at hourly intervals over the duration of our observation . we found that the antenna passbands changed only slowly with time , and at a level smaller than that required to keep artifacts below the level of random noise fluctuations . we have therefore averaged all our measurements of mars to derive a precise passband solution for each antenna . callisto served as the absolute flux calibrator . we used the strong ( @xmath03.4 jy ) and compact ( unresolved ) nuclear continuum source in per a to derive the complex gains ( amplitude and phase ) of each antenna as a function of time , solving for the phase at 3 min intervals and amplitude at 20 min intervals . the full - width half - maximum ( fwhm ) of the sma primary beam at a frequency of 226.6 ghz in co(2 - 1 ) is 55 , corresponding ( with @xmath41 in the assumed cosmology ) to a circle of radius @xmath010 kpc . the co(2 - 1 ) map of @xcite shows that the molecular gas extends to radii of @xmath014 kpc both east and west of the nucleus , and hence our map spans only the inner central extent of the molecular gas . our observation was centered on the nucleus of per a , located at @xmath42 and @xmath43 . we attained a synthesized beam ( with natural weighting of the visibilities ) of @xmath44 ( @xmath45 ) , with major axis at a position angle of @xmath46 . to make maps in co ( 2 - 1 ) , we first derived the continuum emission from the line - free channels and subtracted this emission from the measured visibilities . we then averaged the data in velocity channels of @xmath47 , and deconvolved the point spread function of the telescope from the image in each channel . to make the corresponding maps in total intensity ( zeroth moment ) and intensity - weighted mean velocity ( first moment ) , we first convolved the channel maps with a triangular function ( i.e. , hanning smoothing ) having a full - width at zero intensity of @xmath48 , and then discarded pixels with intensities less than @xmath49 ( where @xmath50 is the root - mean - square noise fluctuation in one channel ) . we then summed the remaining data in the channel maps ( with no smoothing ) to produce the total intensity map , and computed the mean velocity of each pixel weighted by its intensity in each channel to produce the intensity - weighted mean velocity map . the co(2 - 1 ) channel maps are shown in figure 1 , and the corresponding moment maps in figure 2 . in figure 1 , the line intensity in each channel is plotted as contours on a grayscale map of the integrated line intensity . in figure 2 , the integrated line intensity is plotted as contours on a color - coded map of the intensity - weighted mean velocity . the location of the central continuum source , corresponding to the agn , is indicated by a cross in both figures . the velocities are measured with respect to the systemic heliocentric velocity of per a derived in the optical of @xmath51 @xcite , where @xmath52 is the speed of light and @xmath53 the redshift . for clarity , no primary beam correction has been applied to the channel maps shown in figure 1 . on the other hand , to faithfully reproduce the overall intensity distribution of the molecular gas , a primary beam correction has been made to the integrated line intensity map shown in both figure 1 and figure 2 . as in single - dish maps @xcite , the molecular gas in our interferometer map is concentrated primarily in the east - west directions . also as in single dish maps , the emission away from the nucleus on both the east and west , as well as north and south , is entirely blueshifted , with emission closest to the nucleus redshifted . for the first time , our interferometer map reveals that the molecular gas in the central region of radius @xmath010 kpc is concentrated primarily in three radial filaments , labeled in figure 2 as the inner , western , and eastern filaments in order of increasing projected radii from the nucleus . ( as explained below , the inner filament may not actually comprise a single integral filament but rather two separate structures . ) although the inner filament appears to connect with the western filament , they span very different velocities and have entirely different radial velocity gradients . these filaments appear as distinct spatial - kinematic features in the channel maps ( fig . 1 ) , and are clearly separate structures . similarly , the feature labeled e1 in figure 2 is discontinuous in velocity from the eastern filament , and is regarded here as a separate structure ( blob ) . another blob labeled e2 can be seen just to the north of the eastern filament . an apparently isolated blob labeled w1 lies to the west of the western filament , located along a line from center that nearly passes through the inner and western filaments . the only features located away from the east - west directions are the isolated blobs n1 and s1 located to the north and south of center respectively . the spectra of all the features labeled in figure 2 are shown in figure 3 . our co(2 - 1 ) interferometer map bears little resemblance to the co(1 - 0 ) interferometer map of @xcite , which as mentioned in @xmath38 also shows significant differences compared with single - dish maps . first , none of the eastern features seen in figure 2 are detected in the co(1 - 0 ) interferometer map , despite the presence of gas at this location in single - dish maps . second , the inner and western filaments do not appear as distinct features in the co(1 - 0 ) map , but coincide with a complex and spatially extended distribution of gas in the co(1 - 0 ) map . third , gas in the co(1 - 0 ) map at the location of the western filament exhibits both blueshifted and ( primarily ) redshifted velocities ; in our map , the western filament exhibits only blueshifted velocities , consistent with that seen at this location in single - dish maps . finally , the co(1 - 0 ) map shows gas towards the south - east of the nucleus that is not seen in our co(2 - 1 ) map . we conclude that the co(1 - 0 ) interferometer map of @xcite is unreliable , and can not be meaningfully compared with our co(2 - 1 ) map . the radial positions and extents , as well as velocities , of all the features labeled in figure 2 are listed in table 1 . to estimate the lengths of the filaments , we measure from the channel maps the difference in radial positions of the emission centroids at opposite ends in velocity spanned by that filament . because the emission is spatially resolved in many channels , this method provides a conservative estimate of the true length of the filaments , but is necessary in particular for the eastern and western filaments as they do not have a smoothly - varying intensity profile along their lengths . as measured from the emission centroid in the channel maps , the inner filament can be traced from a ( projected ) radius of @xmath54 ( @xmath55 pc ) where it has a ( projected ) velocity of @xmath48 to a radius of @xmath56 ( @xmath57 kpc ) westwards where it has a velocity of @xmath58 , thus spanning a ( projected ) length of @xmath030 ( @xmath01.1 kpc ) . a two - dimensional gaussian fit to the inner filament in the integrated intensity map of figure 2 gives a length ( at full - width half maximum ) of @xmath59 ( @xmath60 kpc ) . this filament straddles the systemic velocity of the galaxy , and exhibits a radial velocity gradient with smaller blueshifted velocities or larger redshifted velocities with decreasing radii ( fig . 2 ) . the western filament can be traced from a radius of @xmath61 ( @xmath62 kpc ) where it has a velocity of @xmath63 to a radius of @xmath64 ( @xmath65 kpc ) westwards where it has a velocity of @xmath66 , thus spanning a length of @xmath067 ( @xmath02.4 kpc ) . the eastern filament can be traced from a radius of @xmath67 ( @xmath68 kpc ) where it has a velocity of @xmath69 to a radius of @xmath70 ( @xmath71 kpc ) eastwards where it has a velocity of @xmath72 , thus spanning a length of 61 ( 2.2 kpc ) . unlike the inner filament , the two outer filaments ( and all the blobs ) exhibit entirely blueshifted velocities , consistent with the velocities measured in single - dish maps at their locations . these filaments can be seen to exhibit a radial velocity gradient with larger blueshifted velocities at smaller radii , opposite to that exhibited by the inner filament . in figure 4 , we plot the co(2 - 1 ) intensity of each filament along their radial extents as a function of velocity ( i.e. , position - velocity ( pv- ) diagrams ) . for clarity , no primary beam correction has been applied to these plots . both the eastern and western filaments exhibit a simple linear change in velocity with radius from the center of the galaxy . the feature at a position of 26 and velocity of @xmath66 in the pv - diagram of the eastern filament ( bottom panel of fig . 4 ) corresponds to the feature e1 ( fig . 2 ) , which as mentioned earlier probably consitutes a separate structure . the pv - diagram of the inner filament is more complicated : the velocity changes linearly with radius at both blueshifted and redshifted velocities , but with different velocity gradients on either sides of the systemic velocity . it is possible that the redshift and blueshifted sides of the inner filament correspond to separate structures that are poorly resolved spatially in our map . indeed , a close examination of figure 2 suggests that the redshifted and blueshifted portions of this filament lies at somewhat different position angles , producing a visible twist in the major axis of this filament between the two portions . the integrated line intensities , corrected for the primary beam response of the sma , and corresponding molecular hydrogen ( h@xmath73 ) gas masses of all the features labeled in figure 2 are listed in table 1 . to facilitate a direct comparison with the molecular gas masses derived in @xcite , we adopt like them a brightness temperature in co(2 - 1 ) of 0.7 times that in co(1 - 0 ) , and the standard co(1 - 0 ) to h@xmath73 conversion ratio ( applicable to gas with solar metallicity ) of @xmath74 @xcite . altogether , the features have a total gas mass of @xmath75 , about half that detected in observations with single - dish telescopes over the same radius of @xmath0@xmath76 ( fig . 4 of * ? ? ? the three filaments alone have a total gas mass of @xmath77 , about @xmath78 of that contained in all the features . our interferometric observation is preferentially sensitive to relatively bright and compact features , and hence traces regions where the molecular gas is most strongly concentrated ( i.e. , highest filling factor ) . to determine if there is emission external to the features detected in figure 2 that lies below our brightness sensitivity , we have integrated the channel maps over the entire sma primary beam from a velocity of @xmath79 to @xmath80 . the integrated line intensity determined in this manner is @xmath81 , corresponding to a molecular gas mass of @xmath82 . this is higher than the total molecular gas mass of the features detected in figure 2 , and comparable with the abovementioned molecular gas mass detected with single - dish telescopes over the same region . per a exhibits a complex h@xmath83 nebula that extends beyond the optically visible confines of the galaxy to nearly 70 kpc @xcite . the origin and excitation of this nebula is very poorly understood . in figure 5 , we plot the integrated co(2 - 1 ) intensity on a color map of the inner region of this h@xmath83 nebula . all the molecular filaments and blobs lying in the east - west directions , but not the blobs lying north and south of center , appear to be associated with the h@xmath83 nebula . the inner filament closely coincides with a bright western extension in h@xmath83 from a compact nuclear component . the inner radius of the western filament closely coincides with an apparent extension of the bright central h@xmath83 feature further west . the eastern filament coincides with local brightenings in h@xmath83 , as do the nearby blobs e1 and e2 . @xcite have measured the velocity of the h@xmath83-emitting gas within a radius of @xmath010 of the center of per a , approching to within 26 of its nucleus . along the inner filament , the h@xmath83 emission has a velocity of about @xmath84 at a radius of 26 and about @xmath85 at a radius of 35 . this radial range coincides with the outer portion of the inner filament where the co(2 - 1 ) emission is primarily blueshifted : the h@xmath83 velocity compares well with the range of co(2 - 1 ) velocities seen here , including the trend of increasingly blueshifted velocities towards larger radii . along the eastern filament , the h@xmath83 emission has a velocity of about @xmath86 at a radius of 57 , which coincides quite closely to the inner portion of the eastern filament . here , the co(2 - 1 ) velocity reaches about @xmath63 , again comparable with that seen in h@xmath83 . @xcite have measured the velocity of the h@xmath83-emitting gas out to a radius of nearly 10 on the eastern side and nearly 20 on the western side , but with poorer spatial sampling . at and close to the nucleus , the h@xmath83 emission is redshifted , like in in our co(2 - 1 ) map . further west at the location of the western filament , the gas is blueshifted , again as is seen in our co(2 - 1 ) map . we therefore conclude that the co(2 - 1 ) features are not just spatially coincident with bright h@xmath83 features but , where the velocity of the latter has been measured , also have comparable velocities . in figure 6 , we overlay ( black ) contours of the integrated co(2 - 1 ) intensity on a x - ray image of the center of the perseus cluster taken from @xcite . we also plot in ( red ) contours the radio emission of per a at 90 cm , produced by relativistic jets from the central agn , taken from @xcite ( see also * ? ? ? the x - ray image shows cavities to the north and south of the nucleus that coincide with the radio jets . these cavities are therefore believed to be giant bubbles blown by the radio jets into the surrounding x - ray gas . the bulk of the molecular gas specifically , all the features located east and west of center lies ( in projection ) between the x - ray cavities . the isolated blob n1 lies close ( in projection ) to a local x - ray brightening projected against the northern x - ray cavity , and the blob s1 near the inner rim of the southern x - ray cavity . in figure 7 , we overlay contours of the integrated co(2 - 1 ) intensity on a map of the lowest temperature ( 0.5 kev , or @xmath0@xmath87 ) x - ray gas detected by chandra towards the center of the perseus cluster , taken from @xcite . the bulk of this relatively cool x - ray gas lies in an east - west band ( curving north ) across the center of the galaxy , located between the two x - ray cavities . as remarked by @xcite , the bulk of this relatively cool x - ray gas closely resembles the spatial distribution of the h@xmath83 gas ( see also fig . 12 of * ? ? ? * ) , although the mass of the x - ray gas at this temperature alone is an order of magnitude larger than that of the h@xmath83 gas . the bulk of the molecular gas detected in our image also has a spatial distribution that closely matches that of this relatively cool x - ray gas . the origin of the molecular gas in per a has long been a subject of debate and controversy . arguments have been made for mergers with or cannibalisms of gas - rich galaxies , ram - pressure stripping of gas - rich galaxies passing close to or through per a , and a x - ray cooling flow . here , we examine whether the observed properties of the molecular gas are consistent with capture from gas - rich galaxies . suggestions that the molecular gas in per a is captured from either ram - pressure stripping or mergers / cannibalisms of gas - rich galaxies are bolstered by evidence that per a may currently be interacting with a gas - rich galaxy . the latter has been detected through its emission in h@xmath83 @xcite , as well as absorption in both atomic hydrogen ( hi ) gas @xcite and lyman @xmath83 @xcite against the continuum emission of per a at a velocity around @xmath88 . despite having a velocity that is @xmath0@xmath89 higher than that of per a , and hence commonly referred to as the high - velocity system ( hvs ) , the detection of line absorption in both hi and lyman @xmath83 , as well as continuum absorption in x - rays @xcite , implies that the hvs lies in front of per a. evidence that this galaxy is not just a chance foregound galaxy but in the close vicinity of per a comes from the detection of h@xmath83 emission at velocities intermediate between those of per a and the hvs @xcite . this intermediate - velocity gas may comprise gas from the hvs entrained by the hot intracluster x - ray gas . furthermore , the ionized gas in the hvs appears to exhibit a complex velocity field ( e.g. , * ? ? ? * ; * ? ? ? * ) , which together with the active star formation in this galaxy @xcite suggests that the hvs has been disturbed by the intracluster medium or interactions with per a. it is highly unlikely , however , that the molecular gas in per a was captured from the hvs . first , the velocity of the molecular gas is @xmath0@xmath89 lower than the velocity of the hvs , and is predominantly blueshifted with respect to the systemic velocity of per a rather than redshifted like the hvs . second , the hvs has not been detected in hi emission nor co , placing a 3@xmath50 upper mass limit of @xmath90 in atomic hydrogen gas @xcite and @xmath91 in molecular hydrogen gas @xcite . the latter is about a factor of four lower than the mass of molecular hydrogen gas in per a. consistent with these arguments , @xcite place a lower limit of @xmath060 kpc on the separation between the hvs and perseus a , and argue that any collision between the two galaxies still lies in the future . if captured , the molecular gas in per a more likely originated from previous encounters with other gas - rich galaxies . gas captured in this manner should exhibit a transverse ( i.e. , perpendicular to radial ) as well as any radial spatial - kinematic components following the path of the interacting galaxy through or decaying orbit of the cannibalized galaxy in the cd galaxy . in both cases , one might expect the captured gas to orbit and hence exhibit a global rotation about the cd galaxy . where imaged and well resolved , the co gas in elliptical galaxies has until now been invariably found to be distributed in rotating disks ( e.g. , see * ? ? ? * and references therein ) , suggesting that their gas was indeed captured from gas - rich galaxies . where only co spectra ( at good signal - to - noise ratio ) are available , many exhibit double - horned line profiles indicative of rotating disks ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in per a , however , we detect no transverse spatial - kinematic components nor any ordered global motion indicative of rotation . instead , the most prominent features that we detected in per a the three filaments described in @xmath33 exhibit only radial spatial - kinematic components . if each filament is attributed to capture from an individual gas - rich galaxy passing through per a , this would require three separate events with each gas - rich galaxy taking a projected path that crosses the center of per a. although this could be the future fate of the hvs , there is no independent evidence that three such highly unlikely events have occurred in the relatively recent past . the excellent spatial correspondence between the molecular gas and the relatively cool x - ray gas ( fig . 7 ) , but not the warmer x - ray gas ( fig . 12 of * ? ? ? * ) , in per a provides circumstantial evidence that the molecular gas originates from a cooling flow . in the simplest possible situation imaginable , gas accreted from such a cooling flow should simply flow radially inwards along the gravitational potential of the cd galaxy . the gas should therefore exhibit only radial spatial - kinematic components , as is indeed observed for the three filaments . such a signature alone , however , is not a unique indicator of radial inflows . in principle , gas undergoing radial outflows also can exhibit the same spatial - kinematic pattern . we now examine in greater detail the properties of these filaments to see if we can distinguish between inflows or outflows . the eastern and western filaments , as well as the other blobs of molecular gas , exhibit only blueshifted velocities . this implies inflow if the filaments and blobs are located on the far ( back ) side of per a , or outflows if located on the near ( front ) side . to deduce which is the case , we have compared the spatial distributions of these filaments and blobs to optical images showing dust ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) seen in silhouette against starlight from per a. if these filaments and blobs are located on the near side of the galaxy , they ought to show a good correspondence with the silhouette dust . on the contrary , if these filaments and blobs are located on the far side of the galaxy and at a large inclination to the plane of the sky , they may not show up strongly if at all in silhouette dust . examples of silhouette dust seen in elliptical galaxies with the hst - wfpc2 can be found in @xcite , @xcite , @xcite , and @xcite . inferred dust masses as small as 10@xmath92 are detectable , which for a ratio in mass of molecular hydrogen gas to dust of about 100 ( as in our galaxy ) corresponds to molecular gas masses as small as just @xmath93@xmath94 . in figure 8 , we overlay the integrated co(2 - 1 ) intensity on the hst - wfpc2 image of per a taken from @xcite . an especially prominent and complicated network of dust can be seen in silhouette towards the north of the nucleus , extending both east and even further west . much if not all of this dust likely belongs to the foreground infalling galaxy ( e.g. , * ? ? ? * ; * ? ? ? first , the overall spatial distribution of this dust network closely resembles that of the h@xmath83 gas in the hvs , but not that in per a itself ( e.g. , compare the silhouette dust in fig . 7 with the individual h@xmath83 images of the hvs and per a in fig . 3 of * ? ? ? in addition , the morphology of this dust closely matches that of gas detected in x - ray absorption against per a , attributed to the hvs @xcite . whether belonging to the hvs or per a , or portions in both , neither the eastern nor western filaments , nor the blobs e1 , e2 , w1 , and n1 coincide with any of the silhouette dust ( s1 is located beyond the region spanned by the hst - wfpc2 image ) . specifically , the complicated network of dust does not extend to the location of the eastern filament and blobs , where the light distribution appears to be smooth . dust in silhouette is visible to the north , but not at the location , of the western filament and w1 . the light distribution at and immediately around the western filament and w1 appears to be quite smooth , although its proximity to the prominent silhouette dust does make any weaker silhouette features difficult to discern . the lack of correspondence with silhouette dust suggests that both the eastern and western filaments , as well as the other blobs of molecular gas , are located on the far side of the galaxy and at a relatively large inclination to the plane of the sky . their blueshifted velocities therefore imply inflow . we disregard the possibility that these filaments and blobs lack dust because they formed from metal - poor gas ; on the contrary , the iron abundance of the x - ray gas is at least @xmath00.5 solar @xcite within @xmath0100 kpc . to permit a more detailed comparison between the inner molecular filament and the silhouette dust , in figure 9 we overlay the integrated co(2 - 1 ) intensity on just the inner portion of hst - wfpc2 image of per a taken from @xcite . the latter is a transmission map , representing the residual intensities after subtracting smooth model profiles for both the foreground and background galaxy ( for more details , see * ? ? ? brighter regions therefore indicate intensities above the smooth background , and darker regions intensities below the smooth background ( i.e. , dust extinction ) . confirming what was seen in figure 8 , neither the western molecular filament nor blob w1 obviously coincide with the silhouette dust . there is a patch of dust at the innermost radii of the western filament where the co(2 - 1 ) emission is strongest , but this dust patch does not trace the length of the filament and hence is likely coincidental . there is patchy filamentary dust at and around the location of the inner molecular filament , although once again this dust does not obviously trace out the filament . if the dust seen here also belongs to the hvs , this would make any silhouette dust close to the nucleus of per a difficult to discern . thus , at the present time , we can not be certain whether the inner filament is associated with any silhouette dust , and hence whether this filament is located on the near or far side of per a. the relative locations of the molecular features and radio jets from per a were shown earlier in figure 6 . if the molecular filaments we observe correspond to gas entrained by the radio jets , they should lie in projection against , or at the edges of , these jets . in addition , the filaments should have their long axes aligned approximately north - south , along the axis of the radio jets . instead , the molecular filaments have their long axes aligned approximately east - west , closely orthogonal to the axis of the radio jets . furthermore , all the filaments lie neatly between the x - ray cavities , in regions where there is presumably no energetic particles to excavate the x - ray gas . thus , the molecular filaments are unlikely to comprise gas entrained by the radio jets or other energetic particles . the same argument applies to the blobs e1 , e2 , and w1 . on the other hand , the blobs n1 and s1 do lie in projection against the radio jets . s1 is spatially resolved , and appears to be elongated in an approximately west - west direction , closely parallel with the western filaments . if entrained by the radio jet , one would expect this blob to be elongated approximately north - south , along the axis of the radio jet . both the eastern and western filaments exhibit a linear increase in blueshifted velocities with decreasing radii . here , we examine whether such a motion can be produced by gas in free - fall along the gravitational potential of per a. to model the gravitational potential of per a , we use an analytical model proposed by @xcite that has been found to reproduce the projected surface brightness profile of elliptical galaxies parameterised only by their mass and effective radius . in this scheme , the mass density as a function of radius is given by + @xmath95 + where @xmath96 is the total mass of the galaxy , @xmath97 the radius , and @xmath98 a scale length . the scale length @xmath98 is related to the half - mass radius , @xmath99 , which contains half of the mass of the galaxy by + @xmath100 + the gravitational potential is then given by + @xmath101 + where @xmath102 is the gravitational constant . the velocity , @xmath103 , of an object in free - fall can be computed from the relation + @xmath104 + where @xmath105 is the initial radius at which the object is released and @xmath106 its initial velocity . taking values from @xcite but scaled to the cosmology used here , per a has an optical v - band luminosity of @xmath107 and an estimated mass - to - light ratio in this band of @xmath108 . the inferred mass of the galaxy is therefore @xmath109 ; given the uncertainty in @xmath110 ( which , moreover , may change with radius ) , there is probably considerable uncertainty in the estimated galaxy mass . @xcite quote an effective radius , @xmath111 , for per a of 12.4 kpc . the effective radius is related to the scale length @xmath98 by + @xmath112 + and so @xmath113 . the x - ray gas is very nearly in hydrostatic equilibrium , and so a given parcel of cooling gas is initialy at rest with respect to the cluster . in figure 10 , we plot from eq . ( 4 ) how an object , which is initially at rest , changes in velocity with radius when dropped from two different initial radii , at 8.5 kpc ( upper panel ) and 12.2 kpc ( lower panel ) . ( the choice of initial radii is explained in the next paragraph . ) for comparative purposes , but also because of the considerable uncertainty in the estimated galaxy mass , we show the results for two different masses for per a of @xmath114 ( solid curve ) and the value estimated above of @xmath109 ( dashed curve ) . as can be seen , beyond a distance of @xmath01 kpc from which the object is dropped , the velocity increases essentially linearly with decreasing radius , as is observed for both the eastern and western filaments , irrespective of the assumed galaxy mass . the challenge here is to reproduce not just the observed velocity gradient , but also the actual velocities spanned by the two filaments . with a suitable choice for the inclination of the filaments as the only free parameter , we find that this simple model can reproduce the measured range of velocities of both the eastern and western filaments provided we adopt the lower value for the galaxy mass of @xmath114 . in figure 10 , we overplot the pv - diagrams of the western and eastern filaments shown in figure 4 but deprojected in both velocity and radius for inclinations that give the best model fit . as discussed in @xmath115 , these two filaments must be located at a relatively large inclination to the plane of the sky ( and on the far side of per a ) for dust in silhouette not to be seen associated with these filaments . we find relatively good fits to the predicted curves for an inclination of about @xmath116 for the western filament and @xmath117 for the eastern filament . assuming larger inclinations would result in a larger velocity gradient consistent with a higher galaxy mass ( dashed curves in fig . 9 ) , but lower deprojected velocities placing these filaments on the part of the curve where the velocity increases non - linearly with decreasing radius . assuming smaller inclinations , on the other hand , would result in a smaller velocity gradient inconsistent with the model ( unless one is prepared to accept an even lower galaxy mass ) . the choice of initial radii mentioned in the previous paragraph is therefore determined by our model once a galaxy mass is adopted and the best fit inclinations for the filaments determined . for an inclination of @xmath116 , the western filament ( upper panel ) spans deprojected radii of 4.27.4 kpc . in this model , the outermost tip of the filament lies just 1.1 kpc away from the location where it first began its free fall in the gravitational potential of per a. the deprojected velocity of this filament at its outermost radius is about @xmath118 compared with the computed model value of about @xmath119 , and at its innermost radius about @xmath120 compared with the computed model value of about @xmath121 . for an inclination of @xmath117 , the eastern filament ( lower panel ) spans deprojected radii of 8.111.3 kpc , with its outermost tip lying just 0.9 kpc away from where it began its free fall ( i.e. , at a radius of 12.2 kpc ) . the deprojected velocity of this filament at its outermost radius is about @xmath122 compared with the computed model value of about @xmath123 , and at its innermost radius about @xmath124 compared with the computed model value of about @xmath125 . we regard the agreement between this simple model and our observations to be satisfactory given that we have ignored a number of effects , primarily the outward pressure the x - ray gas exerts on the infalling molecular gas . the latter should cause an infalling object to increase more slowly in velocity with decreasing radius ( i.e. , resulting in a shallower velocity gradient ) than predicted by our simple model , thus requiring a larger galaxy mass to produce the observed velocity gradient than assumed in our best - fit model . this would bring our results into better agreement with model predictions for the inferred larger galaxy mass for per a of @xmath109 . the scenario we have in mind then is the following . the co(2 - 1 ) transition is excited at a critical density in molecular hydrogen gas of @xmath0@xmath126 . from an analysis of the line ratios in co(2 - 1 ) and co(1 - 0 ) based on single - dish observations , @xcite also infer a density for the molecular hydrogen gas of @xmath0@xmath126 , and a temperature of 2030 k. by comparison , the density of the x - ray gas at radii of 510 kpc from the center of per a lies in the range 0.05@xmath127 @xcite . the western filament corresponds to molecular gas that has condensed from x - ray gas located at a distance of @xmath08.5 kpc from the center of per a. the cooling gas decouples from the x - ray gas , and then free falls in the gravitational potential of per a. by the time it has travelled 1.1 kpc from its initial location , taking about 18.2 myr , the cooling infalling gas has already been ( largely ) converted to molecular gas and has attained a density sufficiently high and temperature sufficiently low to be detectable in co(2 - 1 ) . because the filament is relatively dense , it is unlikely to be ( significantly ) disrupted during its passage through the more diffuse x - ray gas . furthermore , its linear appearence suggests that the filament is not significantly perturbed by turbulence or any other disturbances . the same picture applies for the eastern filament , except that in this case the molecular gas condensed from x - ray gas initially located at a distance of @xmath012.2 kpc from the center of per a. the time it takes for the cooling infalling gas to travel 0.9 kpc from its initial location to the outermost tip of the eastern filament is about 21.1 myr . the measured properties of the inner filament , on the other hand , are difficult to explain with a simple model . consider the situation where this filament also corresponds to radially - infalling molecular gas , as we have shown is likely the case for the two outer filaments . the blueshifted portion of the filament would then have to lie on the far side of the galaxy , and the redshifted portion on the near side . as mentioned in @xmath33 , these two portions of the filament may actually be separate features that are poorly resolved spatially in our map . unlike the two outer filaments , however , the blueshifted portion of the inner filament exhibit smaller blueshifted velocities at smaller radii ( see top panel of fig . 4 ) . it may be that , close to the center of the galaxy , the infalling molecular gas is being retarded by the high pressure of the x - ray gas , such that the blueshifted portion of the filament experiences an inward decrease in velocity . the redshifted portion of the filament , on the other hand , exhibits larger redshifted velocities at smaller radii , implying an inward increase in velocity . alternatively , and perhaps more plausibly , this filament may correspond to infalling molecular gas that is settling into the gravitational potential well of per a. specifically , the gas may be settling into orbits to form a dynamically unrelaxed disk around the nucleus . @xcite tentatively infer an approximately constant stellar orbital velocity of @xmath0@xmath128 over radii of @xmath0525 in per a. it is reasonable to expect that as one approaches the center of per a , the x - ray gas increasingly partakes in the overall galaxy rotation . in such a case the molecular gas deposited by the x - ray cooling flow should also possess angular momentum , and can not directly flow into the center but must eventually form a rotationally - supported structure ( i.e. , disk ) around the nucleus . in support of this picture , hot ( @xmath01000 k ) molecular hydrogen gas has been detected in the near - ir vibrational line of h@xmath73 @xmath129 s(1 ) around the nucleus of per a @xcite . this gas appears to be distributed in a disk - like structure oriented east - west with a radius of @xmath050 pc , and therefore shares the same morphological orientation as the inner filament . the eastern side of the h@xmath73 @xmath129 s(1 ) disk is redshifted in velocity by @xmath0@xmath48 and the western side blueshifted by @xmath0@xmath48 , therefore spanning the same range of velocities and also having the same sense in velocity gradient as the inner filament . the inner filament may therefore trace the outer , cooler , dynamically unrelaxed portion of the inner , warmer , and more dynamically relaxed molecular hydrogen disk , and is responsible for despositing gas into this circumnuclear molecular disk . given that the eastern and western filaments exhibit a linear dependence in velocity with radius , we can compute their dynamical ages from the relationship + @xmath130 + where @xmath131 is the deprojected velocity at the maximum deprojected radial extent @xmath132 , and @xmath133 is the deprojected velocity at the minimum deprojected radial extent @xmath134 . for the values inferred in @xmath135 , we derive a dynamical age of @xmath016 myr for the western filament and @xmath029 myr for the eastern filament . the analytical model described in sect 5.3.2 ( where the predicted velocity changes close to but not perfectly linearly with radius ) gives a dynamical age of 21.7 myr for the western filament and 16.5 myr for the eastern filament . changing the inferred inclinations of the filaments by @xmath136 only changes their dynamical age by a factor of @xmath137 ( the dynamical age is proportional to tan(@xmath138 ) , where @xmath138 is the inclination of the filaments to the plane of the sky ) . given that the western filament has a molecular gas mass of @xmath0@xmath139 and the eastern filament @xmath0@xmath140 as listed in table 1 , the total mass - deposition rate into these two filaments is therefore @xmath0@xmath3 . this value should be considered as a very rough estimate that is probably accurate to only a factor of a few . it would be instructive to compare the inferred mass - deposition rate into these two molecular filaments with the upper limit in mass - deposition rate inferred for the x - ray gas at the same radius . at the present time , however , no such information for the perseus cluster has been published . instead , below we compare the inferred mass - deposition rate with that inferred for other gas components below x - ray temperatures . @xcite have detected o vi lines tracing ionized gas at temperatures of @xmath0@xmath11 in a 30 square region centered on per a. they attribute these lines to gas from a x - ray cooling flow , and infer a mass - deposition rate of @xmath141 within a radius of @xmath06 kpc . the inferred mass - deposition rate is roughly comparable with that inferred for the band of x - ray gas at @xmath0@xmath87 located between the x - ray cavities , provided this gas originates from a cooling flow @xcite . if responsible in its entirety for producing ( at the present mass - deposition rate ) the inner molecular filament , which has a mass of @xmath142 ( table 1 ) , the cooling flow in the inner @xmath06 kpc region of the cluster must have lasted at least @xmath056 myr . the mass - deposition rate inferred by @xcite is , coincidentally , roughly comparable with the mass - deposition rate we infer for the two outer molecular filaments . these filaments , however , lie beyond the region spanned by the observed o vi lines . in addition , there is molecular gas lying even further out than the region we have mapped ( and which also may originate from a cooling flow ) , suggesting that the x - ray cooling flow in the perseus cluster may have a significantly higher mass - deposition rate than that inferred from the o vi lines alone . the close spatial - kinematic correspondence between the cool molecular gas detected here and ionized gas at @xmath0@xmath143 traced in h@xmath83 suggests that at least a portion of the latter could also originate from a x - ray cooling flow . the mass in h@xmath83 gas over its entire visible extent is only @xmath0@xmath144 @xcite , two orders of magnitude below the mass in molecular gas of the filaments reported here , and three orders of magnitude below the total mass of molecular gas in per a. the bulk of the gas from the x - ray cooling flow must therefore spend relatively little time at temperatures of @xmath0@xmath143 . one of the outstanding challenges is to understand the time evolution of the cooling flow as it transforms from x - ray - emitting plama into molecular gas . at a radius of @xmath010 kpc where we infer the two outer molecular filaments to first cool from the x - ray gas , the azimuthally - averaged cooling time of the x - ray gas is @xmath0500 myr @xcite . although the cooling flow must spend relatively little time at temperatures of @xmath0@xmath12 , it may well spend much longer times at other temperatures before condensing into molecular hydrogen gas . specifically , the time it spends at temperatures above those traced in co(2 - 1 ) may be considerably longer than the inferred free - fall time of just @xmath020 myr to reach the outer tips of the molecular filaments . if the cooling gas remains coupled to the x - ray gas by magnetic fields , and strong magnetic fields have been inferred in the central region of per a @xcite , then this gas may not start to free - fall until atomic hydrogen has ( almost fully ) recombined at temperature below @xmath04000 k. molecular gas accreted from a x - ray cooling flow is an excellent candidate for fueling the agn observed in many cd galaxies lying in rich clusters . in the case of per a , the circumnuclear hot molecular hydrogen disk detected by @xcite may be responsible for fueling the agn ; this disk in turn may have been accreted from or is replenished by the inner cool molecular filament observed here ( 5.3.2 ) . the powerful radio jets produced by the agn reheats the surrounding x - ray gas , thereby diminishing if not quenching the cooling flow ( e.g. , * ? ? ? * ; * ? ? ? if the cooling flow is severely disrupted , the agn may eventually run out of fuel and turn off . this allows the cooling flow to eventually resume and deposit a fresh supply of fuel to the agn . in the absence of any disturbances ( i.e. , reheating of the x - ray gas ) , one would expect the gas deposited by a x - ray cooling flow to stream inwards along all radial directions . instead , all three molecular filaments detected in per a are confined to the east - west directions , reflecting the global distribution of the molecular gas in this galaxy @xcite . this is just the configuration expected if the cooling flow is quenched ( or reduced below the level of detectability ) over a large solid angle in the north - south direction by the radio jets from per a , but is still able to operate in directions orthogonal to these jets . indeed , as pointed out in @xmath145 , the bulk of the molecular gas detected coincides with a band of relatively cool x - ray gas lying between the x - ray cavities inflated by the radio jets . our observations implicate this relatively cool x - ray gas as part of the cooling flow . note that , when deprojected in three dimensions , the molecular gas is not infalling along all directions orthogonal to the radio jets , but is instead confined to certain radial directions mostly toward the far side of the galaxy . this suggests that , even orthogonal to the radio jets , large portions of the x - ray gas may be experiencing reheating , reducing their cooling flow if any to below the level of detectability . indeed , the x - ray image of the perseus cluster shows ripples in virtually all directions projected in the plane of the sky , but which is strongest over a broad range of angles centered along the north - south directions @xcite . each ripple is believed to be an acoustic wave driven by episodic expansions in the x - ray bubbles inflated by the radio jets from per a. once a given bubble reaches a sufficiently high pressure to become buoyant it rises through the x - ray gas , whereby the radio jet creates new x - ray bubbles at the center . the observed ripples ( acoustic waves ) are believed to gently heat the surrounding gas . our results indicate that a ( detectable ) cooling flow is still able to operate in a few localized regions where presumably the x - ray gas is not being strongly reheated . even at these locations , the cooling flow is likely intermittent or modulated in time by disturbances propagating through the x - ray gas . @xcite infer an average duty cycle for the bubbling activity in putative cooling - flow clusters of at least @xmath0@xmath146 , and perhaps as high as @xmath0@xmath147 . if the cooling flow is severely disrupted if not completely quenched when the agn is on , such a high duty cycle may not permit the x - ray gas to both cool and reach the central supermassive black hole in the cd galaxy during the relatively short time when the agn is off . instead , our results imply that a residual cooling flow occurs even when the agn is on , and that this cooling flow is able to provide a nearly continuously supply of fuel most likely in the form of molecular gas to the agn . to better understand the origin of the molecular gas in per a , we have imaged this galaxy in co(2 - 1 ) at an angular resolution of 3 ( spatial resolution of 1 kpc ) with the submillimeter array ( sma ) . single - dish maps with angular resolutions as high as 12 ( 4 kpc ) have previously shown that the molecular gas is concentrated in the east - west direction , and extends @xmath040 ( @xmath014 kpc ) in both directions from center . this gas shows no discernible ordered spatial - kinematic pattern , exhibiting blueshifted velocities both east and west of center , and redshifted velocities at or close to the center . our sma map ( figs . 1 and 2 ) , which covers a region of radius @xmath028 ( @xmath010 kpc ) centered on per a , reveals that : * the molecular gas extends to the edge of our field , and is concentrated primarily in three radial filaments with projected lengths ranging from at least 1.12.4 kpc spanning the galaxy center to 7.7 kpc . these filaments , along with several blobs , have a total mass in molecular gas of @xmath75 , with the three filaments alone containing @xmath0@xmath78 of the detected gas mass . * the bulk of the molecular gas detected ( i.e. , all three filaments and all except two blobs of molecular gas ) is aligned roughly east - west and lie between two x - ray cavities to the north and south of center . this gas is aligned approximately orthogonal to the axis of the radio jets from the agn in per a. * the two outer ( western and eastern ) filaments , and the other blobs of molecular gas , exhibit only blueshifted velocities . the velocities of both these filaments change linearly with radius to larger blueshifted velocities at smaller radii . neither these filaments nor the blobs are coincident with detectable silhouette dust , suggesting that they lie at a relatively large inclination to the plane of the sky on the far ( back ) side of per a. * the inner filament exhibits both blueshifted and redshifted velocities that straddle the systemic velocity of the galaxy . unlike the two outer filaments , the velocity of the inner filament does not change linearly with radius ; furthermore , this filament exhibits smaller blueshifted or larger redshifted velocities at smaller radii . it is not clear whether this filament is coincident with silhouette dust associated with per a. * all the molecular gas detected in the east - west direction is spatially coincident with h@xmath83 gas , often lying against or close to local h@xmath83 brightenings . where measured , specifically at the locations of the inner and western filaments , the velocities of the h@xmath83 gas are comparable with that of the molecular gas . the molecular gas also coincides spatially with a band of relatively cool x - ray gas lying between the two x - ray cavities . * the filaments are unlikely to have been captured from one or more gas - rich galaxies . gas captured in this manner should orbit and hence show a global rotation about the cd galaxy , as has hitherto been the case in all other elliptical galaxies that have been well mapped in co. * lying on the far ( back ) side of per a , the two outer filaments are therefore flowing radially inwards towards the center of the galaxy . the only known process that can naturally deposit gas in this manner is a x - ray cooling flow . * the velocity pattern of the two outer filaments can be easily reproduced as free fall in the gravitational potential of per a , with their outer tips lying @xmath01 kpc downstream from where the cooling gas first decouples from the x - ray gas . both filaments have a dynamical age of @xmath020 myrs , implying a total mass - deposition rate into these filaments of roughly @xmath0@xmath3 ( likely accurate to only a factor of a few ) . * the velocity pattern of the inner filament can not be explained with a simple model . we speculate that this filament likely traces gas settling into the gravitational potential well of per a , and perhaps corresponds ( in part ) to the outer regions of a putative hot molecular hydrogen gas disk with a radius of @xmath050 pc around the agn . despite the strong and widespread heating of the surrounding gas inferred from x - ray observations , our results provide the most direct evidence yet for a ( residual ) x - ray cooling flow in the perseus cluster . this cooling flow may provide a nearly continuous supply of fuel for the agn in per a , which in turn regulates the cooling flow . we thank a. c. fabian and j. s. sanders for kindly providing a fits file of the x - ray image , and c. j. conselice for a fits file of the h@xmath83 image , used in this paper . we thank the second of the two anonymous referees for suggesting changes that improved our presentation . y .- p . ao acknowledges the asiaa for supporting his stay as a visiting scholar , when much of this work was done . j. lim thanks jan vritlek and bill forman for useful discussions during a visit to the harvard cfa , and acknowledges a grant from the national science council of taiwan in support of this work . the submillimeter array ( sma ) is a collaborative project between the smithsonian astrophysical observatory and the academia sinica institute of astronomy & astrophysics of taiwan . lim , j. , leon , s. , combes , f. , & dinh - v - trung 2004 , in penetrating bars through masks of cosmic dust : the hubble tuning fork strikes a new note , ed . d. l. block , i. puerari , k. c. freeman , r. groess , & e. k. block ( dordrecht : kluwer ) , 783 mcnamara , b. r. , wise , m. w. , nulsen , p. e. j. , david , l. p. , carilli , c. l. , sarazin , c. l. , odea , c. p. , houck , j. , donahue , m. , baum , s. , voit , m. , oconnell , r. w. , & koekemoer , a. 2001 , , 562 , l149 cccccccc inner filament & @xmath148 & @xmath149 & @xmath0@xmath150 & @xmath151 & @xmath152 & @xmath153 & @xmath154 + western filament & @xmath155 & @xmath156 & @xmath0@xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + w1 & @xmath162 & & @xmath0@xmath163 & @xmath164 & & @xmath165 & @xmath166 + eastern filament & @xmath167 & @xmath168 & @xmath0@xmath169 & @xmath170 & @xmath171 & @xmath172 & @xmath173 + e1 & @xmath174 & & @xmath0@xmath169 & @xmath175 & & @xmath176 & @xmath177 + e2 & @xmath178 & & @xmath0@xmath179 & @xmath159 & & @xmath180 & @xmath181 + n1 & @xmath182 & & @xmath0@xmath183 & @xmath184 & & @xmath185 & @xmath186 + s1 & @xmath187 & & @xmath0@xmath188 & @xmath189 & & @xmath190 & @xmath191 + @xcite . contour levels are plotted at -3 , 3 , 4 , 5 , 6 , 7 , @xmath192 ( root - mean - square noise level ; @xmath193 ) . no primary beam correction has been applied to these maps . the gray background corresponds to the integrated co(2 - 1 ) intensity after primary beam correction as shown in figure 2 . the position of the nuclear continuum source in per a is indicated by a cross . the synthesized beam is shown as an ellipse at the lower left corner of the top left panel , and has a size of @xmath44 and position angle @xmath46 . the filaments indicated in figure 2 are not uniformly bright in intensity along their lengths , and hence may not appear continuous in these channel maps.,width=585 ] , and color - coded map of the intensity - weighted co(2 - 1 ) mean velocity measured with respect to the systemic heliocentric velocity of @xmath194 @xcite as indicated by the right vertical bar . the full - width half - maximum of the sma primary beam is 55 ( 20 kpc ) , as indicated by the dotted circle . the integrated co(2 - 1 ) intensity has been corrected for the primary beam response . the map is centered on the active nucleus of per a ( @xmath195 , @xmath196 ) , which was detected in the continuum at the location indicated by a cross . the synthesized beam is shown as a filled ellipse at the lower left corner , and has size of @xmath44 with major axis at a position angle of @xmath46 . the inner , western , and eastern filaments , along with the features e1 , e2 , w1 , n1 , and s1 , are labeled as described in the text . , width=585 ] , ( b ) western filament along a position angle of @xmath197 , and ( c ) eastern filament along a position angle of @xmath198 as a function of velocity . contour levels are plotted at @xmath199 , 2 , 3 , 4 , 5 , 6 , and @xmath200 ( @xmath193 ) . positions correspond to offsets from the center of per a with negative values indicating west , and velocities are measured with respect to the systemic heliocentric velocity of @xmath194 @xcite . in panel ( b ) , the feature at positions @xmath201 and velocities @xmath202 corresponds to the inner filament . in panel ( c ) , the feature at a position of 26 and velocity of @xmath66 that is discontinous in velocity from the eastern filament corresponds to e1 ( fig . 2 ) , which is probably a separate structure as described in the text . ] @xmath116 so that it spans 4.27.4 kpc as indicated by the double - headed arrow . panel ( b ) shows the computed velocity profile of an object dropped from a radius of 12.2 kpc , and superposed the pv - diagram ( contours ; from fig . 4 ) of the eastern filament deprojected by an inclination of @xmath0@xmath117 so that it spans 8.111.3 kpc as indicated by the double - headed arrow . as can be seen , the computed velocity profile assuming a lower galaxy mass of @xmath114 ( solid curve ) better matches the observed pv - diagrams than the computed velocity profile for a galaxy mass of @xmath109 inferred from the luminosity of the galaxy and its estimated mass - to - light ratio as described in the text . ]

mesoscopic non - equilibrium thermodynamics is used to formulate a model describing non - homogeneous and non - debye dielectric relaxation . the model is presented in terms of a fokker - planck equation for the probability distribution of non - interacting polar molecules in contact with a heat bath and in the presence of an external time - dependent electric field . memory effects are introduced in the fokker - planck description through integral relations containing memory kernels , which in turn are used to establish a connection with fractional fokker - planck descriptions . the model is developed in terms of the evolution equations for the first two moments of the distribution function . these equations are solved by following a perturbative method from which the expressions for the complex susceptibilities are obtained as a functions of the frequency and the wave number . different memory kernels are considered and used to compare with experiments of dielectric relaxation in glassy systems . for the case of cole - cole relaxation , we infer the distribution of relaxation times and its relation with an effective distribution of dipolar moments that can be attributed to different segmental motions of the polymer chains in a melt .

1005.5713

relaxation processes in complex systems such as glass - forming liquids , polymeric melts , liquid crystals and even lipid membranes is a very active area of research @xcite . they have been studied using different techniques ranging from nmr and brillouin light scattering to dielectric spectroscopy @xcite . the variety of relaxation phenomena occurring in these systems are non - debye and therefore can not be characterized by a single relaxation time @xcite . these multiscale time relaxations suggest an accumulative effect of different elemental relaxation processes that can be associated with collective motions in the system @xcite . the average effects of these cooperative mechanisms observed in experiments can be accurately described by means of well known empirical relations such as the williams - watts function in time space , or through its conjugated frequency domain ( havrilak - nagami ) function . these functions , and many others that can be found in the literature @xcite , reflect the specific features of the micro- or mesoscopic relaxation dynamics of each system . the microscopic description of these features may involve the formulation of specific molecule - molecule interaction models that lead to particular results . in view of this , it seems important to have a general theoretical model able to cope with different relaxation processes that may describe , for example , the dielectric relaxation in diverse materials including solids , glass - forming liquids and polymer melts @xcite . these last systems may exhibit , for example , anomalous kerr effect which can be described in terms of models based on fractional fokker - planck equations ( ffpe ) @xcite . in this work we adopt a mesoscopic point of view for describing the relaxation of a system of non - interacting polar molecules inmersed in a time - dependent external electric field , because at this level of description it is still possible to formulate the mentioned general framework able to cope with different relaxation dynamics . in particular , we use the mesoscopic non - equilibrium thermodynamics ( mnet ) formalism in order to derive a non - markovian fokker - planck like equation ( nmfpe ) for the evolution of the probability distribution describing the orientation and position of the molecules . the mnet method uses the entropy produced by the system when it is driven out from equilibrium and is particularly useful for describing the stochastic evolution of systems at the mesoscopic scale @xcite . the advantage of this formalism lies in the fact that the nmfpe derived has a kinetic origin , and therefore there is no explicit assumption on the specific time correlation properties of the underlying gaussian stochastic process , contrary to what happens with equations derived from langevin formalisms @xcite . within this framework , linear relations between generalized forces and fluxes occurring in phase space are postulated , which allow us to introduce memory effects accounting for the non - instantaneous response of the system to the external perturbations . the effects of the non - markovian dynamics on a measurable property of the system like its complex dielectric susceptibility can then be analyzed in a systematic way by considering different memory kernels . in this work we will concentrate mainly on memory kernels decaying as power laws . we will show that these particular functions turn out to be important since they lead to the mentioned ffpe . in the section ii we will derive the nmfpe for a system of non - interacting polar molecules by using mnet . this equation describes both rotational diffusion and the diffusion of polarization , and we will show that it can be recast into a ffpe when proper memory kernels are considered . in section iii we will construct the equations for the first two moments of distribution . subsequently , in section iv we will present a perturbation procedure which allow for a solution of the obtained dynamic equations , when the system is subjected to an external harmonic electric field superposed to a large static electric field . consequently , we will obtain a closed expression for the non - homogeneous dielectric susceptibility of the system which incorporates non - markovian effects , and the effects of the three first moments of the distribution function . these effects will be analyzed systematically in section v by considering diverse limiting cases and memory functions . memory kernels decaying as power laws in time will be shown to generate ffpe and response functions of the cole - cole type @xcite . a comparison with recent experimental results is performed to validate the models derived . in section vi we will propose a model based on a theory of distribution of relaxation times which can be used to give an insight of the physical processes producing power law kernels and , accordingly , ffpe . finally , we will present some final remarks an discuss the limitations of our model in section vii . we shall consider a system of @xmath0 non - interacting polar molecules with dielectric moment , @xmath1 , of magnitude @xmath2 , which are in contact with a thermal bath at temperature @xmath3 . for convenience , in the following we will use the unit vector , @xmath4 , which represents the orientation of the molecular dielectric moment , and describe the system in terms of a single - molecule probability density . let @xmath5 denote the probability for finding one molecule at a position ranging from @xmath6 to @xmath7 , and oriented within the range @xmath8 and @xmath9 , at time @xmath10 . the normalization condition , @xmath11where @xmath12 is the numerical density of molecules , imposes the following restriction on the time evolution of the probability density function @xmath13 , @xcite @xmath14here , @xmath15 and @xmath16 are the conjugate velocities to @xmath17 and @xmath18 , respectively ; @xmath19 is the @xmath20-th spatial derivative ; and @xmath21 is the @xmath20-th component of the so called rotational operator@xmath22 in this expression @xmath23 represents the levi - civitta antisymmetric symbol and summation over repeated indices will be implicit in the next . consider now , that an external time - dependent and uniform electric field , @xmath24 , acts on the the system . then , the potential energy of one molecule in the presence of this field , @xmath25 , is an explicit function of the orientation vector and also depends on time , @xmath10 , since the electric field , @xmath26 , is time - dependent . in first approximation , the potential energy of the system is given by @xmath27 which holds for molecules in which the anisotropy of the electric polarizability is negligible @xcite . the explicit form of eq . ( [ mnet001 ] ) in terms of @xmath28 can be deduced by analyzing the entropy produced by the system during its time evolution . according to the so called gibbs entropy postulate @xcite , entropy is defined in terms of the probability distribution @xmath13 by @xmath29 + s^{\text{leq } } , \label{mnet003}\]]where @xmath30 is the boltzmann constant and @xmath31 is the local - equilibrium entropy . according to eq . ( [ u(t ) ] ) , the corresponding probability density in local - equilibrium is given by canonical distribution function @xmath32 , \label{mnet002}\]]where @xmath33 is the corresponding partition function . in order to find the entropy production , @xmath34 , we take the time derivative of eq . ( [ mnet002 ] ) from where , by using the normalization condition ( [ mnet000 ] ) and eq . ( [ mnet001 ] ) , we find @xmath35 + k_{b}\iint d\vec{r}d\vec{n}\ f\ \frac{\partial } { \partial t } \ln f^{\text{leq } } , \label{mnet004}\ ] ] where the non - equilibrium chemical potential , @xmath36 , has been defined by @xmath37 . \label{mnet005}\ ] ] now , in accordance with the assumption of local equilibrium , we will consider external fields that evolve in a time - scale larger than the time - scale of fluctuations . therefore , changes of the local equilibrium distribution are not significant over the short time - scale of the fluctuations , and it can be approximated by @xmath38 . consequently , the second term on the right hand side of eq . ( [ mnet004 ] ) vanishes . this approximation is also supported by the fact that later on we will consider electric fields consisting of the superposition of a static , @xmath39 , and a time varying , @xmath40 , contributions such that the magnitude of the time - dependent superposed field @xmath41 is small when compared with the static field @xmath39 , that is , we will assume @xmath42 with @xmath43 . under this assumption , an integration by parts of eq . ( [ mnet004 ] ) yields the following expression for @xmath44 , @xmath45where we have assumed an infinite system with vanishing probability density at the boundaries in @xmath6-space , and periodic conditions in @xmath8-space . the entropy production given by eq . ( [ mnet005a ] ) involves the sum of products of the generalized currents , @xmath15 and @xmath16 , and forces , @xmath46 and @xmath47 . thus , following a linear response scheme @xcite , we may assume that the currents are proportional to forces and a linear relation for @xmath15 and @xmath16 , in terms of @xmath48 and @xmath47 , can be written . for simplicity , we will not consider cross terms in these relations , i.e. we will assume that the generalized force @xmath48 is the unique responsible for producing the flux @xmath15 , and correspondingly for @xmath47 and @xmath16 . this assumption is in accordance with the corresponding onsager - casimir reciprocal relations given the vectorial character of the vectors involved in the description , see ref . @xcite . in the general case and according to the phenomenology observed , we expect that the response of the system is not instantaneous , and then memory effects should be considered . following refs . @xcite , these effects can be taken into account by introducing memory functions in the following linear relationships @xmath49 @xmath50 the memory functions @xmath51 and @xmath52 , are such that @xmath53 , for @xmath54 . they describe the delay in the response of the system to forces which vary in time . this delay is always present and is particularly important when forces vary fast in time @xcite . by replacing eqs . ( [ mnet006 ] ) and ( [ mnet007 ] ) into eq . ( [ mnet001 ] ) , we obtain @xmath55 \label{mnet008}\]]where for notation convenience , we have introduced the modified memory functions @xmath56 , and @xmath57 . in order to obtain eq . ( [ mnet008 ] ) , we have also used the relations @xmath58which follow directly from eq . ( [ mnet005 ] ) . ( [ mnet008 ] ) is the desired non - markovian fp equation for the probability distribution function @xmath13 . it provides a complete probabilistic description of the system at the mesoscopic level and constitutes a generalization of previous equations obtained in the literature for systems with debye relaxation dynamics @xcite . the memory kernels @xmath59 and @xmath60 , characterize the translational and rotational diffusion processes in the system , respectively . for anisotropic molecules such as rigid polymers in solution , the translational diffusion tensor depends in general on the orientational degrees of freedom @xmath18 @xcite . here , for simplicity , we will restrict ourselves to consider isotropic translational and rotational diffusion processes , i.e. @xmath61 , and @xmath62 . consequently , our results will be expected to describe better the dynamics of slightly anisotropic polar molecules . within this approximation eq . ( [ mnet008 ] ) can be recast in the final form @xmath63 . \label{mnet009}\ ] ] in addition , notice that we have restricted ourselves to consider memory kernels which only depend on time , and accordingly , we have neglected the possible non - local effects from interactions at different points in space . in the following sections , eq . ( [ mnet009 ] ) will be used to obtain a dynamical model for the time evolution of the average polarization @xmath64 and its associated order parameter , the quadrupole moment @xmath65 . diverse interesting limiting cases can be obtained from eq . ( [ mnet009 ] ) . firstly , we will consider the so called homogeneous approximation mendez001 , in which a simplified description in terms of the reduced orientational probability distribution , @xmath66is given . averaging eq . ( [ mnet009 ] ) over space coordinates , the corresponding non - markovian fp equation for @xmath67 is found to be @xmath68 . \label{frac001}\ ] ] clearly , moments of the reduced distribution @xmath67 will be uniform time - dependent quantities . therefore , in the homogeneous approximation spatial diffusion of these moments is ignored and the properties obtained from eq . ( [ frac001 ] ) must be considered as volume averages @xcite . it is interesting to notice that eq . ( [ frac001 ] ) leads to a ffpe when the memory function @xmath69 is conveniently chosen to be a power law . specifically , let @xmath70where @xmath71 represents the gamma function , @xmath72 is a dimensionless parameter within the range @xmath73 , and @xmath74 is a constant with units of inverse time that can be identified with the rotational diffusion coefficient of the molecules , as we will show below . it follows directly from eq . ( [ frac001 ] ) that when the memory function is given by eq . ( [ frac002 ] ) , the orientational distribution obeys the dynamic equation @xmath75 . \label{frac002a}\ ] ] on the right hand side of the last expression we can identify the generalized fractional derivative operator @xcite @xmath76 for @xmath77 . thus eq . ( [ frac002a ] ) can be recast in the form @xmath78 , \label{frac006}\ ] ] which is the referred ffpe for the orientational distribution . it can be noticed that eq . ( [ frac006 ] ) reduces to the usual fokker - planck equation for @xmath79 ( normal rotational diffusion ) , @xmath80 , \ ] ] where @xmath74 can be identified as the rotational diffusion coefficient @xcite . for @xmath73 , eq ( [ frac006 ] ) describes the non - markovian behavior associated with rotational subdiffusive processes . an interesting point concerning the applicability of the memory function ( [ frac002 ] ) is worth stressing . the presence of @xmath81 on the right hand side of eq . ( [ frac002 ] ) yields a negative memory function for @xmath73 . this implies that the memory function given by eq . ( [ frac002 ] ) superposes all the effects prior time @xmath10 with an intrinsic negative sign . this fact is rather masked by the fractional derivative operator appearing in eq . ( frac006 ) , and as far as we know it has not been noticed in literature where dynamic equations of the type ( [ frac006 ] ) are postulated for the probability distribution @xmath67 . the presence of the function @xmath81 is , however , necessary in order to obtain the ffpe , eq . ( [ frac002a ] ) , from a power law decay of the memory function . finally , we will consider the case in which both rotational and translational diffusion processes are present and can be described by power law decaying memory functions . for the translational kernel we introduce the function @xmath82 where @xmath83 , is a constant parameter , and @xmath84 is a generalized translational diffusion coefficient with units of length square times @xmath85@xcite . by substituting eqs . ( [ frac002 ] ) and ( [ frac007 ] ) into eq . ( [ mnet009 ] ) , it can be easily shown that the fpe describing the evolution of the probability density @xmath13 takes the form of a fractional equation , @xmath86 , \label{frac008}\]]which includes anomalous diffusion through the fractional derivative @xmath87 . in the limit @xmath88 , the markovian approximation is recovered . at the macroscopic level , the description of the system must be given in terms of the moments of the distribution function @xmath13 . for this purpose , eq . ( [ mnet008 ] ) can be used to construct a hierarchy of equations for the multipolar moments associated with the charge distribution . let us consider the dipole , @xmath89 , quadrupole , @xmath65 , and octupole , @xmath90 fields , which are defined by the following averages @xmath91@xmath92@xmath93 . \label{multi003}\ ] ] in the last definition , @xmath94 represents the symmetric part of the tensor @xmath95 , which has the explicit form @xmath96 multiplying eq . ( [ mnet008 ] ) by @xmath97 and averaging over the orientational degrees of freedom we obtain @xmath98 the first term at the right hand side of the last equation accounts for the diffusion of the polarization whereas the other terms are consequences of the orientational dynamics . analogously , multiplying eq . ( [ mnet008 ] ) by @xmath99 and averaging over @xmath100 , we obtain the dynamic equation for the quadrupolar moment,@xmath101 \!\right ] _ { ij}-\frac{6}{5\mathcal{n}k_{b}t}\int\limits_{-\infty } ^{t}ds\,\phi \left ( t - s\right ) q_{ijk}\left ( s\right ) e_{k}\left ( s\right ) . \notag\end{aligned}\]]where @xmath102 \!\right ] $ ] is the symmetric traceless part of the dyad @xmath103 evaluated at time @xmath10 , i.e.@xmath104 \!\right ] _ { ij}=e_{i}\left ( t\right ) p_{j}\left ( t\right ) + e_{j}\left ( t\right ) p_{i}\left ( t\right ) -\frac{2}{3}e_{k}\left ( t\right ) p_{k}\left ( t\right ) \delta _ { ij}.\ ] ] it can be noticed from eq . ( [ multi003 ] ) that the contraction @xmath105 yields a symmetric traceless tensor of rank two . consequently , eq . ( [ multi005 ] ) is in agreement with the symmetry properties of the order parameter tensor @xmath65 . for notation s simplicity , in eqs . ( [ multi004 ] ) and ( [ multi005 ] ) we have not written the dependence of the fields @xmath89 , @xmath65 and @xmath90 on @xmath6 . notice that eqs . ( [ multi004 ] ) and ( [ multi005 ] ) constitute the first two equations of an infinite recurrence hierarchy which includes the time evolution equation for the octupole moment , @xmath90 , and higher order moments . the problem of approximating this infinite hierarchy by a finite number of equations is a common feature of diverse systems in non - equilibrium statistical mechanics @xcite . in the following section we will describe in detail our closure approximation for the hierarchy of recurrence equations for @xmath89 , @xmath65 , @xmath90 . in eqs . ( [ multi004 ] ) and ( [ multi005 ] ) , the moments @xmath89 and @xmath65 turn out to be coupled only with moments of next superior order because translational diffusion has been assumed to be isotropic . otherwise the evolution equation for @xmath89 will not only involve @xmath65 but @xmath90 as well , while the dynamic equation for @xmath65 will include contributions of @xmath90 and the fourth order moment . these contributions have been shown to be important in recent literature for describing the dynamics of anisotropic molecules like those composing liquid crystals materials @xcite . ( [ multi004 ] ) and ( [ multi005 ] ) can be reduced to the homogeneous case in which both @xmath89 and @xmath65 can be approximated by their position - independent volume averages @xmath106 and @xmath107 then , it can be proved that @xmath108 and @xmath109 obey the dynamical equations @xmath110 and @xmath111 \ ! \right]_{ij } \notag \\ & & -\frac{6}{5\mathcal{n}k_{b}t } \int\limits_{-\infty } ^{t}ds\ , \phi \left(t - s\right ) q_{ijk}\left ( s\right ) e_{k}\left ( s\right ) , \label{multi010}\end{aligned}\ ] ] respectively . eqs . ( [ multi009 ] ) and ( [ multi010 ] ) can also be obtained from eq . ( [ frac001 ] ) for the reduced orientational probability distribution after multiplying and averaging over @xmath100 , as it was done in the case of eqs . ( [ multi004 ] ) and ( [ multi005 ] ) . in the following , we will focus on calculating the response function to a small oscillating field @xmath112 , which is superposed to a large static field @xmath39 , such that @xmath113 . accordingly , the total field acting on the system is @xmath114 . in this section we shall show that a perturbation theory in terms of the small parameter , @xmath115 , can be used in order to find the frequency and field - dependent dielectric susceptibility of the system . with this purpose , let @xmath116 , @xmath117 , etc . , denote the electric multipolar moments associated with the static field . it is clear that these quantities relax towards equilibrium by following equations which are completely analogous to eqs . ( [ multi004 ] ) and ( [ multi005 ] ) , with @xmath116 , @xmath117 , @xmath118 and @xmath119 , replacing the fields @xmath89 , @xmath65 , @xmath90 and @xmath120 , respectively . when both @xmath39 and @xmath121 act on the system , the response to the oscillating field can be identified from the dynamics of the perturbed fields @xmath122 , @xmath123 , etc . @xcite . by substracting the equations corresponding to the unperturbed fields from eqs . ( [ multi004 ] ) and ( [ multi005 ] ) , the following expressions are found for the time evolution of the perturbed fields @xmath124 , \notag\end{aligned}\ ] ] @xmath125 \!\right ] _ { ij } \notag \\ & & -\frac{6}{5\mathcal{n}k_{b}t}\int\limits_{-\infty } ^{t}ds\,\phi \left ( t - s\right ) \left [ q_{ijk}\left ( s\right ) e_{k}\left ( s\right ) -q_{ijk}^{(0)}\left ( s\right ) e_{k}^{(0)}\right ] . \notag\end{aligned}\ ] ] for small perturbative fields , @xmath126 , deviations with respect to @xmath127 , @xmath128 , @xmath129 , etc . , are expected to be also small . in this case , we can approximate @xmath130 @xmath131 \ ! \right]_{ij } \simeq \left [ \ ! \left [ \vec{p}^{(0)}\left ( t\right ) \vec{e}^{(1)}\left ( t\right ) \right ] \ ! } + \left [ \ ! \left [ \vec{p}^{(1)}\left ( t\right ) \vec{e}^{(0 ) } \right ] \ ! \right]_{ij } , \ ] ] @xmath132 moreover , for a large static field we can assume that the non - perturbed multipoles remain close to their equilibrium time - independent values , i.e. @xmath133 , @xmath134 , @xmath135 , where , in order to simplify the succeeding discussion , we have omitted to write the dependence of the equilibrium moments on @xmath39 . in the following it should be regarded that these quantities are of order zero in the perturbative field . finally , in order to close the hierarchy of recurrence equations , we will assume that moments of order @xmath90 and higher are not perturbed at all by the time dependent field , and consequently @xmath136 , for all @xmath10 . all these approximations yield the following closed system of equations for the time evolution of the perturbed dipole , @xmath137 , and quadrupole , @xmath138 , moments @xmath139 , \notag\end{aligned}\ ] ] @xmath140 \ ! \right]_{ij } + \left [ \ ! \left [ \vec{p}^{(1)}\left ( s\right ) \vec{e}^{(0 ) } \right ] \ ! \right]_{ij } \right\ } \notag \\ & & -\frac{6}{5\mathcal{n}k_{b}t } \int\limits_{-\infty } ^{t}ds\ , \phi \left(t - s\right ) q^{\text{eq}}_{ijk } e^{(1)}_{k}\left ( s\right ) . \notag\end{aligned}\ ] ] it can be observed that the solution of eqs . ( [ perturbation005 ] ) and ( [ perturbation006 ] ) , gives indeed the perturbed polarization as a linear function of the perturbative field . from this solution the dielectric susceptibility of the system to the small oscillating field can be identified . for this purpose , we consider the fourier transform of eqs . ( [ perturbation005 ] ) and ( [ perturbation006 ] ) , which can be written in the form @xmath141 \right\ } , \label{perturbation007}\ ] ] @xmath142 \!\right ] _ { ij } + \left [ \!\left [ \vec{p}^{(1)}\left ( \vec{k},\omega \right ) \vec{e}^{(0)}\right ] \!\right ] _ { ij } \right\ } \notag \\ & & -\frac{6}{5\mathcal{n}k_{b}t}\phi \left ( \omega \right ) . \label{perturbation008}\end{aligned}\ ] ] here we have introduced the notation @xmath143 , for the static dielectric susceptibility , and @xmath144 and @xmath145 for the propagators in fourier space defined by @xmath146@xmath147 in eqs . ( [ perturbation007 ] ) and ( [ perturbation008 ] ) we have also introduced the notation @xmath148 , for the space - time fourier transform of the perturbation field . the frequency - dependent dielectric susceptibility can be identified by solving eq . ( [ perturbation008 ] ) for @xmath149 , inserting the result into eq . ( [ perturbation007 ] ) , and solving for @xmath150 . after some algebraic manipulations this procedure finally yields @xmath151where the susceptibility , @xmath152 , can be written as the product @xmath153 , with the following definition for the matrices @xmath154 and @xmath155 , @xmath156 @xmath157 \right\ } . \label{perturbation012}\end{aligned}\ ] ] in the following section we will explicitely calculate the complex dielectric susceptibility , @xmath158 , for a specific geometry of the external fields and explore the effects of the memory functions , @xmath159 and @xmath160 , on this quantity . in oder to illustrate the effects of the different mechanisms considered in previous sections on the response of the system to the applied fields , in this section we consider a specific geometry in which the static electric field , @xmath39 , is directed along the @xmath161-axis of a cartesian coordinate system . furthermore , we will calculate the equilibrium tensors @xmath162 , @xmath163 and @xmath164 , by using the stationary solution of the reduced orientational equation , eq . ( [ frac001 ] ) , @xmath165 for this probability density and the geometry described above , we obtain the following expressions for the equilibrium averages of the moments entering in eqs . ( [ perturbation011 ] ) and ( [ perturbation012 ] ) : @xmath166@xmath167 @xmath168 \left ( \begin{array}{ccc } -1/2 & 0 & 0 \\ 0 & -1/2 & 0 \\ 0 & 0 & 1 \end{array } \right ) . \label{illustration003}\ ] ] in eqs . ( [ illustration001])-([illustration003 ] ) , the dimensionless parameter @xmath169 quantifies the electric energy of a molecule in the static field relative to the thermal energy , and @xmath170 represents the langevin function . as expected , from eqs . ( [ perturbation011 ] ) , ( [ perturbation012 ] ) and ( [ illustration001])-([illustration003 ] ) , we obtain that the matrices @xmath154 , @xmath155 and @xmath171 reflect the uniaxial symmetry of the problem and are given by @xmath172 after calculating the expressions for the components @xmath173 , @xmath174 , @xmath175 and @xmath176 ( see the appendix ) , the corresponding formulas for the components of the dielectric susceptibility perpendicular and parallel to the static field are found to be @xmath177 \right\ } , \label{illustration010}\]]@xmath178 \right\ } , \label{illustration011}\ ] ] where we have introduced the normalized propagators @xmath179 . the dependence of eqs . ( [ illustration010])-([illustration011 ] ) on the dimensionless parameter @xmath180 implies that the response of the system is in general a nonlinear function of the magnitude of the static component @xmath181 . this dependence also influences the characteristic relaxation time as known from experiments @xcite . our formalism reduces to well known phenomenological relations used to describe dielectric relaxation in diverse materials including solids , glass - forming liquids and polymer melts @xcite , when the appropriate limiting cases are considered . in particular , eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) can be reduced to the debye and cole - cole @xcite dielectric response functions if the infinite hierachy of dynamic equations for the moments of the distribution , starting with eqs . ( [ perturbation003 ] ) and ( [ perturbation004 ] ) , is truncated under the assumption that the order parameter tensor , @xmath65 , remains very close to its equilibrium value , i.e. @xmath182 , and @xmath183 . if in addition we neglect the diffusion of polarization by assuming @xmath184 , for all @xmath10 , then eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) , reduce to the simpler form @xmath185where the amplitudes of the parallel and perpendicular components , @xmath186 and @xmath187 , read @xcite @xmath188when the response of the system can be considered instantaneous , we have @xmath189 , i.e. @xmath190 , and therefore eq . ( [ limiting001 ] ) takes the form of two independent debye expressions @xmath191where the relaxation time , @xmath192 , is related with the rotational diffusion coefficient through @xmath193 . on the other hand , for the power law memory kernel given by eq . ( [ frac002 ] ) , we have @xmath194 , and eq . ( [ limiting001 ] ) reduces to two expressions of the cole - cole type @xmath195where the relaxation time appearing in this expression , @xmath196 , is given by @xmath197 . in the following subsections we will analyze in more detail the dependence on the frequency , the normalized dipole electrostatic energy , the rotational diffusion and the diffusion of polarization of the real and imaginary parts of the complex susceptibility , @xmath198 and @xmath199 . for simplicity , in the following the analysis will be carried out in different successive reduced levels of description in order to better quantify and elucidate the effects due to couplings , diffusion of polarization and the non - markovian character of the system . with this aim we have subdivided this section into subsections . we will consider first the effects of the coupling of the average polarization with the moments of second and third order on the dielectric susceptibility . from eqs . ( [ perturbation008a ] ) and ( [ perturbation008b ] ) , and ( [ illustration010])-([limiting002 ] ) , it follows that coupling effects are manifested by the terms proportional to @xmath200 , which is the propagator associated with the time and space evolution of @xmath65 . ignoring for convenience the effects associated to diffusion of polarization by making @xmath201 , and the non - markovian effects by assuming an instantaneous rotational diffusion kernel , i.e. @xmath202 , the normalized complex susceptibilities perpendicular and parallel to @xmath39 , eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) , are given by @xmath203 @xmath204 where the normalized frequency , @xmath205 , has been defined by @xmath206 when these expressions for @xmath207 and @xmath208 are used in eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) , the result will contain exclusively the effects of the dynamics of the second and third order moments on @xmath209 and @xmath210 . this result has to be compared with eq . ( [ limiting003 ] ) , in which these effects have been neglected . in figs . [ figure001 ] and [ figure002 ] we present this comparison for the real and imaginary parts of the normalized susceptibility @xmath211 , and four values of the normalized electrostatic energy , @xmath180 , ranging from small , @xmath212 , to large values , @xmath213 , as compared with the average thermal energy per molecule . it can be observed that when the external static field is small , the coupling with high order moments produce no significant effect on the the complex susceptibility . however , for moderate , @xmath214 , and large external static fields , @xmath215 , the polar moments of the distribution increase and the changes induced by the coupling between these moments in both @xmath216 and @xmath217 , are appreciable . figs . [ figure001 ] and [ figure002 ] show that the coupling of higher order moments reduces the susceptibility of the system with respect to the debye - like case , in which this coupling is neglected . it can be also noticed that the frequency at which @xmath218 takes its maximum value is shifted towards the region of large frequencies . however , the frequency dependence of @xmath219 is not modified in this limit since this quantity still decreases as @xmath220 . . all curves correspond to markovian processes in the homogeneous approximation . solid lines contain the effect of the coupling of high order moments of the distribution function as obtained from eq . ( [ illustration011 ] ) with @xmath207 and @xmath208 given by eqs . ( [ markovhomo001 ] ) and ( [ markovhomo002 ] ) . dashed lines are obtained from the imaginary part of eqs . ( [ limiting001 ] ) and ( [ limiting002 ] ) , in which the coupling with high order moments is not taken into account ( debye case).,width=302 ] for the real part of the parallel component of the dielectric susceptibility . solid lines represent the effect of the coupling of high order moments whereas the dashed lines correspond to the debye case . ] the diffusion of polarization may also influence dependence on the wave vector and frequency of the dielectric susceptibility as we will show in this section . in order to focus on these effects , in this case we will consider the markovian dynamics with memory functions @xmath202 , for the rotational diffusion , and @xmath221 , for the diffusion of polarization , that has associated the diffusion coefficient @xmath222 . we will also ignore the effects due to the coupling of the average dipole moment with moments of second and third order by making @xmath223 in eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) . within these approximations , only the normalized propagator @xmath207 is required for describing the behavior of @xmath224 , and it can be written in the form @xmath225 where the parameter @xmath226 , has been introduced to quantify the relative contribution of the diffusion of polarization with respect to rotational diffusion of the molecules . the normalized frequency @xmath205 is given as before by eq . ( [ markovhomo003 ] ) . when eq . ( [ markovnonhomo001 ] ) together with @xmath227 , are used in eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) , we obtain expressions which exclusively quantify the effects of the diffusion of polarization on the components of the dielectric susceptibility . since in this approximation the diffusion of polarization only modifies the real part of @xmath207 , it is clear that its general effect on the imaginary part of @xmath209 and @xmath210 will consist in decreasing the value and shifting the location of the maxima of these functions in an amount proportional to the parameter @xmath228 . however , it can be noticed that the general dependence of the dielectric susceptibility with respect to the frequency is not modified by a memory kernel of the form @xmath229 , because both @xmath230 and @xmath218 increase as @xmath231 , for @xmath232 , and decrease as @xmath220 , for @xmath233 . similar conclusions are valid for the real part of @xmath209 and @xmath210 . in this section , we will first consider the effects of the non - instantaneous response of the system to the imposed fields and afterwards both , the effects coming from coupling with higher order moments and those due to the diffusion of polarization . consequently , we set @xmath201 , and @xmath234 , in eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) . this yields eqs . ( [ limiting001 ] ) and ( [ limiting002 ] ) for the dielectric susceptibility components , in which the memory kernel , @xmath159 , has to be given in order to specify the behavior of the dielectric relaxation . for the sake of clarity , in the present paper , we will only analyze the effects of two types of memory kernels , namely , the power law memory function given by eq . ( [ frac002 ] ) , and a modified kernel that combines both exponential and power law decay . we will write the latter in the form @xmath235 with @xmath73 . let us discuss first the effects of the memory function given by eq . ( [ frac002 ] ) . previously , we have shown that this function yields the ffpe , eq . ( [ frac006 ] ) , and the cole - cole law for the dielectric susceptibility , eq . ( [ limiting004 ] ) . accordingly , the effects of this kind of non - markovian dynamics can be analyzed by simply comparing our previous results , namely , eq . ( [ limiting004 ] ) with respect to the debye expression , eq . ( [ limiting002 ] ) , in which such effects do not appear . in fig . [ figure003 ] we perform this comparison for the real and imaginary parts of @xmath236 . for convenience , we have illustrated our results in terms of the normalized susceptibility @xmath237 , and the normalized frequency @xmath205 , eq . ( [ markovhomo003 ] ) . ) . the dimensionless exponent @xmath238 , takes diverse values including @xmath239 , which corresponds to normal diffusion , and @xmath240 , which describe rotational subdiffusive processes . @xmath241 and @xmath242 are calculated from eqs . ( [ limiting004 ] ) and ( [ limiting003 ] ) , with @xmath237 . the value of the normalized field is taken as @xmath243 . ] curves in fig . [ figure003 ] exhibit the well known features of the cole - cole susceptibility . the dielectric peak in the cole - cole case is symmetric and behaves in the form @xmath244 , @xmath245 , in the limits of small and large frequencies , respectively . for @xmath246 , the width of the dielectric loss increases the smaller @xmath247 is , while the departure with respect to the debye case is eliminated for @xmath248 . the frequency dependence of the real part of the susceptibility is also modified by the power law memory function , eq . ( [ frac002 ] ) , and instead of the characteristic debye s decay , @xmath249 , a decay of the form @xmath250 , is observed . the modified memory function given by eq . ( [ homo000 ] ) yields an asymmetric shape for the dielectric peak . this result follows from the fourier transform of eq . ( [ homo000 ] ) : @xmath251 , and after replacing this expression into eq . ( [ limiting001 ] ) . the resulting formula for the normal and parallel components of the susceptibility can be written in the form @xmath252 where we used @xmath253 and the amplitudes @xmath254 are given by eq . ( [ limiting002 ] ) . in order to analyze the low and high frequency behavior of eq . ( [ homo003 ] ) , we will use the formula @xmath255 in the limit @xmath256 , the imaginary part of @xmath257 increases linearly with @xmath231 , i.e. @xmath258 ; while for @xmath259 , it decreases as @xmath260 . this asymmetry in the dielectric loss is not exhibited by the debye nor by the cole - cole functions , but resembles the dependence that is observed in systems described by the phenomenological cole - davidson expression @xcite . in a similar way , it can be shown that the real part of @xmath257 decreases as @xmath260 for @xmath259 , in contrast with the usual debye dependence . [ figure004 ] exhibits these effects on the real and imaginary parts of @xmath236 , in terms of the normalized susceptibility @xmath237 , and the normalized frequency @xmath205 , eq . ( [ markovhomo003 ] ) . ) . for @xmath239 , the usual decay in the dielectric loss @xmath261 , is observed . for @xmath262 , @xmath263 becomes asymmetric , increasing as @xmath205 and decreasing as @xmath264 . @xmath241 and @xmath242 are calculated from eqs . ( [ homo003 ] ) and ( [ limiting003 ] ) , with @xmath237 . the value of the normalized field is fixed at @xmath243 . ] the general case in which both , rotational diffusion and the diffusion of polarization processes are present and have a non - markovian character can be analyzed as a function of the parameter @xmath265 . in general terms , the qualitative correction arising from the coupling with higher orders moments only modulates the amplitudes of the obtained functions and their characteristic relaxation times ( their location along the @xmath231-axis ) , as shown in figs . [ figure001 ] and [ figure002 ] . thus , if for simplicity we consider small values of @xmath265 , such that the coupling of high order moments is neglected , then the expression for the components of @xmath266 takes the form @xmath267 with @xmath268 given by eq . ( [ limiting002 ] ) . when the memory kernels @xmath269 and @xmath270 are assumed to decay as power laws in time , eqs . ( [ frac002 ] ) and ( [ frac007 ] ) , then eq . ( [ homo005 ] ) reduces to @xmath271 which , as eq . ( [ homo003 ] ) , is also asymmetric . unlike in the previous case , now the low and high frequency behaviors of the real and imaginary parts of the susceptibilities will be determined by the exponents @xmath247 and @xmath272 . this result implies that the diffusion of polarization introduces asymmetries in the response functions of the system that depend on the viscoelastic properties of the medium . the representation of eq . ( [ homo006 ] ) contains a combination of the effects shown in figs . [ figure003 ] and [ figure004 ] . a simple and effective generalization of the present model can be formulated along the following lines . consider the case of a system in which different kinds of polar elements , characterized by their dipolar moments , @xmath273 , are present . in this case , the probability distribution function will depend on several orientation vectors , @xmath274 , associated to each type of polar element . for example , in the case @xmath275 , the distribution function @xmath276 , will satisfy a bivariate conservation equation entirely similar to eq . ( [ mnet001 ] ) , but having the corresponding divergence term in the @xmath277-space . as a consequence , a bivariate fokker - planck equation similar to eq . ( [ mnet009 ] ) can be obtained from the mesoscopic thermodynamics analysis . this equation can in turn be used to formulate the corresponding evolution equations for the moments of the distribution , in which @xmath278 , @xmath279 , are the dipole moments , see eq . ( [ multi001 ] ) . here , @xmath280 is the number of polar elements of class @xmath281 , and the operator @xmath282 , evaluates the value of the field @xmath283 in the continuous position @xmath6 , see for example @xcite . the definitions of the quadrupolar and octupolar moments are therefore given by entirely similar relations as those of eqs . ( [ multi002 ] ) and ( [ multi003 ] ) . if we assume that these polar elements are independent , a simple and powerful model can be formulated to account for experiments . under the independence assumption , the evolution equations associated with @xmath284 and @xmath285 , are also independent and therefore , the total relaxation process results from the sum of both contributions . taking into account only the time dependent field , the relation between the polarization and the applied electric field can be written as @xmath286 where @xmath287 , since in the present approximation the response function is independent of the applied field . as an example , consider that @xmath275 and that the diffusion of polarization effect is subdiffusive anomalous . thus , by applying a similar procedure to the one yielding eq . ( [ homo006 ] ) , we obtain @xmath288 where the subscripts for the parallel and perpendicular components have been suppressed in order to simplify the notation . in this case , the susceptibility is also asymmetric but it may present two broad peaks when the characteristic relaxation times associated to @xmath289 and @xmath290 are sufficiently separated . the imaginary part of eq . ( [ many002 ] ) can be used to account for the dielectric relaxation spectra of glassy materials . this is shown in fig . [ figure005 ] , where a comparison between experiments ( symbols ) and theory ( lines ) is presented in a normalized representation of the dielectric loss , @xmath291 , see ref . the experiments were performed with 50/50 ( wt / wt ) glycerol / water samples at 160 k ( solid ) , in the absence and the presence of myoglobin at different levels of hydration . the symbol @xmath292 stands for the level of hydration , where @xmath293 corresponds to the case when no myoglobin was added . the dielectric relaxation spectroscopy experiments were performed by introducing the samples in a capacitor and applying an oscillatory electric field @xcite . the presence of the hydrated myoglobin broadens the spectrum and shifts the maximum towards low frequencies . in fig . [ figure005 ] we have included two cases : ( a ) without considering diffusion of polarization , i.e. @xmath294 , in eq . ( [ many002 ] ) . ( b ) considering the diffusion of polarization , i.e. @xmath295 . it can be observed that the diffusion of polarization improves the agreement between theoretical and experimental results . in presence of myoglobin with different levels of hydration . the symbols represent experimental results taken from ref . @xcite whereas the lines are the fits from eq . ( [ many002 ] ) . our results suggest that the structural @xmath247-relaxation process affects the observed @xmath272-relaxation spectrum by introducing an asymmetry in the amplitudes of the spectrum at low frequencies . , title="fig : " ] in presence of myoglobin with different levels of hydration . the symbols represent experimental results taken from ref . @xcite whereas the lines are the fits from eq . ( [ many002 ] ) . our results suggest that the structural @xmath247-relaxation process affects the observed @xmath272-relaxation spectrum by introducing an asymmetry in the amplitudes of the spectrum at low frequencies . , title="fig : " ] it is interesting to notice that the diffusion of polarization appears to have most dramatic effects in the absence of myoglobin , while its effects are small in the presence of myoglobin . these effects can be understood by considering that the reduction of the amplitude of the dielectric loss is due to the presence of both the myoglobin and the hydration shell , whereas the addition of new modes at low and high frequencies can be attributed to the presence of the hydration shell @xcite . then , it is plausible to assume that the hydration shell induces long - ranged orientational correlations between solvent particles , which reduce the relative importance of the diffusion of polarization term with respect to the non - diffusive terms in all the evolution equations for the multipolar moments . the orientational correlations explain why in fig . [ figure005]a the agreement for the larger value of the hydration number ( @xmath296 ) is better at low frequencies than in the case of low hydration numbers ( @xmath297 ) . for low @xmath292 , the hydration shell is less important and therefore the amplitude of the dielectric loss increases as well as the importance of the diffusion term in the evolution equations for the multipoles . in this specific application , the superposition of two independent dielectric responses given by eq . ( [ many002 ] ) , accounts very well for the complicated shape of the dielectric loss found experimentally . according to our model , this function should consist of a peak centered at small frequencies , associated with the so called @xmath247-relaxation processes , superposed to a peak located at larger frequencies , which is associated with slower or @xmath272-relaxation processes . the latter peak is the one observed in figs . [ figure005 ] ( a ) and ( b ) . it can be noticed that the diffusion of polarization allow for fitting the experimental data in the range of small frequencies . notice that , as @xmath292 increases , the exponent @xmath272 increases as well , implying that @xmath298 and indicating that hydration favors a transition from anomalous to normal diffusion of polarization , see the tables in appendix b. a physical insight of the mechanism underlying the power - law behavior of the memory kernel in eq . ( [ frac002 ] ) can be elucidated by decomposing the corresponding spectrum in terms of elementary debye processes , that is , by appealing to the theory of distribution of relaxation times @xcite . for simplicity , our analysis will be focused on the description leading to cole - cole parallel susceptibility ( [ limiting004 ] ) with the use of the power law memory function , eq . ( [ frac002 ] ) . we will proceed by assuming the existence of a distribution of elementary processes with proper relaxation times and amplitudes , whose superposition generates a wider spectrum such that corresponding to the cole - cole function , eq . ( [ limiting004 ] ) . in this form , the cole - cole function can be recovered from a superposition of elementary debye processes through the relation @xcite @xmath299 where for simplicity in notation we have introduced the variable @xmath300 and used the definition @xmath301 in the second equality . the function @xmath302 is a continuous distribution of relaxation times having the explicit form @xmath303 in fact , the analysis can be simplified by considering a discrete expansion of eqs . ( [ limiting008 ] ) and ( [ limiting009 ] ) , that is , by approximating the cole - cole response law by a superposition of independent debye - like processes @xmath304 , distributed over a discrete space in the form @xmath305 where the amplitudes , @xmath306 , and relaxation times , @xmath307 , of the @xmath308-th process are given , respectively , by @xmath309 notice that the index @xmath308 in the summation of eq . ( [ limiting011 ] ) , quantifies the time - scale separation between the @xmath308-th elementary process and the process with the relaxation time @xmath310 , which defines the location of the maximum of the dielectric response function . those processes labeled by an index @xmath311 , evolve toward equilibrium in a characteristic time which is much larger than @xmath310 , while those with @xmath312 , decay much faster than the processes with relaxation time of the order of @xmath310 . [ figure006 ] illustrates the approximation of both the real and imaginary parts of the normalized cole - cole function , @xmath313^{2 } + \left ( \tau_{cc } \omega \right)^{2\alpha } \sin^{2 } \left ( \alpha \pi / 2\right ) } , \label{limiting011a}\ ] ] @xmath314^{2 } + \left ( \tau_{cc } \omega \right)^{2\alpha } \sin^{2 } \left ( \alpha \pi / 2\right ) } , \label{limiting011b}\ ] ] by the superposition of debye components over the discrete distribution of relaxation times given by eq . ( [ limiting011 ] ) with @xmath315 , and where the infinite sum in eq . ( [ limiting011 ] ) has been approximated by a sum over a finite interval ranging from @xmath316 to @xmath317 . ( [ limiting011 ] ) constitutes a very good approximation to the cole - cole function , reproducing very well the behavior of the latter over the whole frequency range considered . it is interesting to notice that a partial superposition of independent debye contributions in which elementary processes within a given range are suppressed , yields a two step relaxation dynamics manifested through the dielectric loss with well marked peaks as those observed experimentally @xcite . this elimination of elementary processes can be represented by non - uniform distributions of characteristic times @xcite . ) . the real and imaginary part of the cole - cole function , eq . ( [ limiting004 ] ) , are represented by the continuous curve . they correspond to @xmath318 . the superposition of debey components , eqs . ( [ limiting010 ] ) and ( [ limiting011 ] ) , is represented by the square symbols and it can be observed that it fits the cole - cole function over the whole range of frequencies . some of the debye components , those with indexes @xmath319 , are explicitly shown . ] these results can be used to give an insight of the physical mechanisms underlying the relaxation process if we consider eqs . ( [ limiting002 ] ) and ( [ limiting003 ] ) , and the definition of the static susceptibility @xmath320 . the amplitude of each elementary debye contribution in eq . ( [ limiting010 ] ) , can be written in terms of an effective numerical density of polar elements @xmath321 participating in the process with relaxation time @xmath307 , that is @xmath322 comparing eqs . ( [ limiting011 ] ) and ( [ limiting013 ] ) we obtain @xmath323 where @xmath324 , is the effective fraction of polar elements which contribute to process @xmath308 . in the limit of slow and fast processes , i.e. @xmath311 and @xmath312 , respectively , and in terms of the relaxation time @xmath307 , eq . ( [ limiting014 ] ) , @xmath325 can be recast into @xmath326 this result implies that the power law memory kernel ( [ frac002 ] ) and consequently the cole - cole function ( [ limiting004 ] ) , can be suitable for describing the dynamics of systems in which a distribution of relaxation times exist , such that the fraction of polar elements of the system participating in the processes with relaxation time @xmath307 , scales as @xmath327 in the range of fast processes , while in the range of slow processes it scales as @xmath328 . in addition , it should be noticed that the fraction of polar elements participating in a given relaxation process depends on the dimensionless parameter @xmath265 which in turn depends on the static applied field @xmath181 . starting from purely thermodynamic grounds , we have formulated a very general framework , which is useful for describing non - debye dielectric relaxation in diverse materials including solids , glass - forming liquids and polymer melts . the framework is based on the derivation of a non - markovian fokker - planck like equation that incorporates memory effects for both rotational diffusion and diffusion of polarization and therefore provides a very wide range of possible behaviors as analyzed through section v. we have shown that fractional fokker - planck descriptions can be derived as particular cases of our model when a specific form of the memory kernel is chosen . the formulation of the model on thermodynamic grounds simplifies its generalization to cases in which external temperature or pressure gradients , among others , are applied on the system . the evolution equations for the polarization vector and higher order multipoles can be derived systematically . we analyzed the evolution equations for the first two multipoles by closing the corresponding hierarchy assuming that the higher order multipoles may be considered near to equilibrium . this assumption yields previously known results for the dependence of the susceptibilities and the relaxation times when the system evolves in the presence of a time dependent externally applied electric field consisting of a static and a harmonic dependent component . we analyzed in detail the effect that each mechanism considered has in the behavior of the susceptibilities and shown how empirical formulas can be obtained from the general formalism . in addition , we outlined how the present approach can be generalized to consider the non - markovian dynamics of several polar elements and the effect of this on the total susceptibility . in this case , our results were used to fit with recent experiments performed with glassy materials damped with myoglobin proteins at different levels of hydration . the agreement between theory and experiment was very good . the analysis of the susceptibilities by means of the theory of distribution of times lead us to propose a physical interpretation of each relaxation process in terms of the fraction of polar elements that participate in the relaxation process for a given frequency of the external field . we found that this fraction of polar elements , related to a particular characteristic relaxation time , depends on the static externally applied field @xmath181 . this result indicates that the application of a static field may be used to control the fraction of polar elements that participate in a given relaxation process ( possibly including multistep relaxation ) and thus can be useful in both the analysis and the design of materials with specific absorption windows . hh acknowledges unam - dgapa for financial support . jgmb thanks conacyt for financial support and ish thanks unam - dgapa for partial financial support of grant no . in102609 . 9 y. yin , s. v. shiyanovskii , a. b. golovin , and o. d. lavrentovich , phys . rev . lett . * 95 * , 087801 ( 2005 ) . s. cerveny , r. bergman , g. a. schwartz , and p. jacobsson , macromolecules * 35 * , 4337 ( 2002 ) . a. gadomski , eur . j. b * 9 * , 569 ( 1999 ) . m. h. vainstein , i. v. l. costa , r. morgado , and f. a. oliveira , europhys . lett . * 73 * , 726 ( 2006 ) . r. hernndez , and f. l. somer , j. phys . b , * 103 * , 1064 ( 1999 ) . e. hershkovitz , and r. hernndez , j. chem . phys . * 122 * , 014509 ( 2005 ) . djardin , phys . e * 68 * , 031108 ( 2003 ) . djardin , and j. jadzyn , j. chem . phys . * 122 * , 074502 ( 2005 ) . k. funke , s. brckner , c. cramer , d. wilmer , j. non - cryst . solids * 921 * , 307 ( 2002 ) . s. kumar , and k. j. rao , chemical physics letters * 387 * , 91 ( 2004 ) . w. kob , j. phys . matter * 11 * , r85 ( 1999 ) . m. f. shlesinger , and e. w. montroll , proc . usa * 81 * , 1280 ( 1984 ) . f. lvarez , a. alegra , and j. colmenero , phys . b * 44 * , 7306 ( 1991 ) . a. k. jonscher , j. mat . * 16 * , 2037 ( 1981 ) . d. reguera , j. m. rubi , and j. m. g. vilar , j. phys . b * 109 * , 21502 ( 2005 ) . i. santamara - holek , a. prez - madrid , and j. m. rubi , j. chem . phys . * 120 * , 2818 ( 2004 ) . a. prez - madrid , j. chem . phys . * 122 * , 214914 ( 2005 ) . j. m. rubi , and a. prez - madrid , physica a * 298 * , 177 ( 2001 ) . w. t. coffey , y. p. kalmykov , and j. t. waldron , _ the langevin equation _ ( world scientific , singapore , 1996 ) . p. talkner , e. hershkovitz , e. pollak , and p. hnggi , surf . sci . * 437 * , 198 ( 1999 ) . k. s. cole , and r. h. cole , j. chem . * 9 * , 341 ( 1941 ) . m. krger , and p. ilg , j. chem . phys . * 127 * , 034903 ( 2007 ) . m. krger , and h. s. sellers , complex fluids , lecture notes in physics vol . 415 ( springer , new york , 1992 ) , pp . . m. doi , and s. f. edwards , _ the theory of polymer dynamics _ , ( oxford university , new york , 1998 ) . r. kubo , m. toda , and n. hashitsume , _ statistical physics ii . nonequilibrium statistical mechanics _ , second edition , ( springer , berlin , 1991 ) . i. gyarmati , _ non - equilibrium thermodynamics : field theory and variational principles _ , ( springer , berlin , 1970 ) . r. zwanzig , phys . rev . * 124 * , 983 ( 1961 ) . i. santamara - holek , j. m. rubi , and a. gadomski , j. phys . b * 111 * , 2293 ( 2007 ) . j. g. mndez - bermudez , i. santamara - holek , physica a ( 2010 ) , doi:10.1016/j.physa.2009.12.058 r. metzler , and j. klafter , phys . * 339 * , 1 ( 2000 ) . r. metzler , e. barkai , and j. klafter , phys . rev . lett . * 82 * , 3563 ( 1999 ) . h. risken , _ the fokker - planck equation _ , ( springer , berlin , 1996 ) . r. zwanzig , nonequilibrium statistical mechanics , ( university press , oxford , 2001 ) . h. block , and e. f. hayes , trans . faraday soc . * 66 * , 2512 ( 1970 ) . i. santamara - holek , g. barrios , and j. m. rubi , phys . rev . e * 79 * , 031201 ( 2009 ) . h. frauenfelder , g. chen , j. berendzen , p. w. fenimore , h. jansson , b. h. mcmahon , i. r. stroe , j. swenson , and r. d. young , proc . sci . * 106 * , 5129 ( 2009 ) . r. balescu , _ statistical dynamics _ , ( imperial college , singapore , 1997 ) . a. bello , e. laredo , and m. grimau , phys . b * 60 * , 12764 ( 1999 ) . k. adachi , and t. kotaka , prog . * 58 * , 585 ( 1993 ) . h. hjar , j. g. mndez - bermdez , and i. santamara - holek , 2010 ( in progress ) in section v , we have discussed how the uniaxial symmetry of the problem reduces the matrices @xmath154 , @xmath155 and @xmath171 to diagonal forms that contain two contributions , one parallel and one perpendicular to the static applied field . making use of eqs . ( [ perturbation011 ] ) , ( [ perturbation012 ] ) and ( [ illustration001])-([illustration003 ] ) , it can be shown that the explicit form of the elements @xmath173 , @xmath174 , @xmath175 and @xmath176 are given by the expressions : @xmath329 @xmath330 @xmath331 \right\ } , \label{illustration007}\ ] ] @xmath332 \right\ } . \label{illustration008}\ ] ] from these equations , the explicit form of the susceptibilities eqs . ( [ illustration010 ] ) and ( [ illustration011 ] ) was calculated . the parameters used for fitting the experimental data with eq . ( [ many002 ] ) and represented in figs . [ figure005 ] ( a ) and ( b ) , are given in the following table .

we evaluate the majorana phases for a general complex symmetric neutrino mass matrix on the basis of mohapatra - rodejohann s phase convention using the three rephasing invariant quantities , and proposed by sarkar and singh . we find them interesting as they allow us to evaluate each majorana phase in a model independent way even if one eigenvalue is zero . utilizing the solution of a general complex symmetric mass matrix for eigenvalues and mixing angles we determine the majorana phases for both the hierarchies , normal and inverted , taking into account the constraints from neutrino oscillation global fit data as well as bound on the sum of the three light neutrino masses ( ) and the neutrinoless double beta decay ( ) parameter . this methodology of finding the majorana phases is applied thereafter in some predictive models for both the hierarchical cases ( normal and inverted ) to evaluate the corresponding majorana phases and it is shown that all the sub cases presented in inverted hierarchy section can be realized in a model with texture zeros and scaling ansatz within the framework of inverse seesaw although one of the sub case following the normal hierarchy is yet to be established . except the case of quasi degenerate neutrinos , the methodology obtained in this work is able to evaluate the corresponding majorana phases , given any model of neutrino masses . = 1

1502.06508

apart from hierarchical structure of massive neutrinos a fundamental qualitative nature of these elusive particles whether they are dirac or majorana type is yet unknown . neutrinoless double beta decay ( @xmath7 ) mode is able to discriminate between the two different types . positive evidence of the above experimental search will be able to determine the majorana nature of neutrinos assuming the above decay is mediated due to light neutrino . several @xmath7 experiments are ongoing and planned . in ref . @xcite a brief discussion about some of the important experiments is presented . among them , exo-200@xcite experiment puts an upper limit on the relevant neutrino mass matrix element @xmath6 within a range as @xmath8 ( 0.14 - 0.35 ev ) . further , next-100 @xcite experiment will be able to bring down the above value of the order of 0.1 ev . thus in an optimistic point of view such property of neutrino could be testified by the next generation experiments . however , even if it is possible to pin down the value of @xmath6 , it is still difficult to predict the values of the majorana phases until we can fix the absolute neutrino mass scale . it is shown in ref.@xcite that in addition to the @xmath7 decay experiments , lepton number violating processes in which the majorana phases show up are also corroborative to determine the individual majorana phases . another interesting physical aspect such as contribution of the majorana phases to the generation of @xmath9 within the present @xmath10 range of neutrino oscillation global fit data is also studied in the literature@xcite . @xcite discusses how to constrain the majorana phases using the results from cosmology and double beta decay . thus it is worthwhile to study the calculability of the majorana phases in terms of a general neutrino mass matrix ( @xmath11 ) parameters . in the present work we evaluate individual majorana phases in terms of the parameters of a general @xmath11 using three rephasing invariants @xmath1 , @xmath2 and @xmath3 presented in ref.@xcite on the basis of mohapatra - rodejohann s phase convention@xcite . although there are several papers which discusses the general procedure for calculating the majorana phases , motivation behind taking the rephasing invariants is that the methodology we present here is capable of calculating the majorana phase in a model independent way even if one of the eigenvalue is zero which is still allowed as far as the present neutrino oscillation global fit data is concerned . moreover as one of the rephasing invariant ( @xmath3 ) is directly proportional to @xmath12 , therefore it vanishes if @xmath13 and hence shows a strong dependency of the majorana phases with the light neutrino masses . in the present work we evaluate the majorana phases for a general complex symmetric neutrino mass matrix ( @xmath11 ) taking into account the global fit oscillation data and the upper bound on the sum of the three light neutrino masses ( @xmath4 ) along with the @xmath5 decay parameter for both the hierarchical cases . we then conclude except the case of quasi degeneracy ] , the methodology presented in this work is able to calculate the majorana phases , given any model of neutrino masses and for convenience , we further numerically estimate the ranges of each majorana phase for both types of hierarchies , in the context of a cyclic symmetric model as well as a model with scaling ansatz property . it is also shown that all the sub cases we present in inverted hierarchy section of the general discussion can be realized through the choice of a model with scaling ansatz with texture zeros within the framework of inverse seesaw while one of the phenomenologically viable sub case of the normal hierarchy section is yet to be identified . the plan of the paper is as follows . [ [ section ] ] in section [ s2 ] we briefly discuss the basic formalism to set the convention of the majorana phase representation within the framework of neutrino oscillation phenomena . cp violating rephasing invariants are presented in section [ s3 ] . section [ s4 ] contains explicit calculation of the majorana phases for both types of neutrino mass hierarchies along with phenomenologically viable different sub cases . numerical estimation of the majorana phases , their connection to the physical observables and discussions about their testability for the general case taking into account the constraints from the extant data for both types of neutrino mass hierarchies are presented in section [ s5 ] . in section [ s6 ] application of the above methodology in the context of cyclic symmetric and scaling ansatz invariant models is presented . section [ s7 ] contains summary of the present work . experimental observation of neutrino flavour oscillation constitutes a robust evidence in favour of nonzero neutrino masses . the flavour transition process is basically a quantum mechanical interference phenomena with the explicit relationship between the left handed quantum fields ( @xmath14 ) of the flavour basis and the mass basis ( @xmath15 ) as _ l&=&_i u^*_i_il where @xmath16 corresponds to the flavour index and @xmath17 implies the mass index and the matrix @xmath18 is the corresponding neutrino mixing matrix . for three generation of fermions , i.e , for @xmath19 , the weak lagrangian containing charged lepton fields and the neutrino fields can be written in the mass basis as -^cc&= & |l_l^(u_l^u^*_)_i _ ilw_^-+ h.c . where @xmath20 is the unitary mixing matrix in the charged lepton sector . the matrix @xmath21 is the leptonic mixing matrix and is known as the @xmath22 mixing matrix ( @xmath23 ) which contains 3 mixing angles and 6 phases in general . it is useful to redefine the mixing matrix by absorbing the unphysical phases into the charged lepton fields and the neutrino fields ( dirac type ) . if the neutrinos are majorana type , they break the global @xmath24 symmetry and hence , redefinition of the neutrino fields is not possible . therefore , out of 6 phases 3 unphysical phases can be absorbed by redefining only the charged lepton fields and thus the @xmath23 matrix is parametrized as u_pmns&=&u_ckmp_m[a ] where @xmath25 is the usual ckm type matrix and is given by u_ckm= c_1 2c_1 3 & s_1 2c_1 3 & s_1 3e^-i + -s_1 2c_2 3-c_1 2s_2 3s_1 3 e^i & c_1 2c_2 3-s_1 2s_1 3 s_2 3 e^i & c_1 3s_2 3 + s_1 2s_2 3-c_1 2s_1 3c_2 3e^i & -c_1 2s_2 3-s_1 2s_1 3c_2 3e^i & c_1 3c_2 3 where @xmath26,@xmath27 and @xmath28 is the dirac cp phase . @xmath29 is a @xmath0 diagonal phase matrix and following mohapatra - rodejohann s convention @xcite it is given by p_m&=&(1,e^i , e^i(+ ) ) where @xmath30 and @xmath31 are the majorana phases which do not appear in the @xmath32 oscillation experiments @xcite . regarding the structure of @xmath29 matrix we would like to mention the following : the advantage of using the above majorana phase convention is that for @xmath13 it is possible to calculate the single existing majorana phase @xmath30 while , for @xmath33 , only the phase difference @xmath34 is calculable . the result will be reversed if we utilize the pdg@xcite convention . explicitly , with pdg convention , if @xmath13 , only the phase difference is calculable , however if @xmath35 is vanishing it is possible to calculate the existing majorana phase . based on pdg convention two of the authors presented a detailed calculation @xcite for both the majorana phases in context of a general @xmath11 , however , if one of the eigenvalue is zero which is still allowed by the present neutrino experimental data , it is not possible to calculate individual phases in that case . the above mentioned problem is successfully resolved in the present work . + cp violating effect of majorana phases in @xmath36 oscillation@xcite and some lepton number violating ( lnv ) processes are studied in detail in ref.@xcite . in this work , using the rephasing invariants constructed out of the neutrino mass matrix elements @xcite we determine the majorana phases for two different hierarchical cases . considering neutrinos as the majorana fermions in extended standard model one can parametrize @xmath23 with the cp violating phases as given in eqn.([a ] ) where we redefine the charged lepton fields absorbing the unphysical phases of total mixing matrix @xmath37 . hence , in principle the mixing matrix @xmath37 can be defined as up_u_pmns[uckm ] where @xmath38 is a @xmath39 diagonal phase ( unphysical ) matrix and is given by p_&=&diag ( e^i_1,e^i_2,e^i_3 ) . now , as the low energy neutrino mass matrix is complex symmetric it can be diagonalized as u^m_u^*&= & d_[diag ] whered_&=&diag(m_1,m_2,m_3 ) . substituting eqn.([uckm ] ) in eqn.([diag ] ) we get u_pmns^p_^m_p_^ * u_pmns^*&= & d _ alternately p_^m_p_^ * & = & u_pmns d_u_pmns^t [ m1 ] . thus @xmath38 rotates the mass matrix @xmath11 in phase space . therefore , the rephasing invariants ( remain invariant under phase rotation ) of @xmath11 contain the informations about the cp violating phases . it has been shown explicitly in ref.@xcite that for three generations of neutrinos there are three independent rephasing invariants and are given by i_12&=&im[m_11m_22m_12^*m_21^ * ] + i_23&=&im[m_22m_33m_23^*m_32^ * ] + i_13&=&im[m_11m_33m_13^*m_31^*][pi2 ] where @xmath40 is the element of @xmath11 at @xmath41 position with @xmath42 . now since the invariants of eqn.([pi2 ] ) are independent of phase rotation of @xmath11 , therefore to evaluate them in terms of mixing angles , cp violating phases and the eigenvalues we can rewrite eqn.([m1 ] ) as m_&= & u_pmns d_u_pmns^t [ m1m]where without any loss of generality we assume @xmath43 which corresponds to the structure of @xmath38 as @xmath44 . now writing down eqn.([m1 m ] ) explicitly one can find the mass matrix elements as m_11 & = & c_12 ^ 2 c_13 ^ 2m_1 + s_12 ^ 2 c_13 ^ 2m_2 e^2 i+m_3s_13 ^ 2e^-2i+2i(+)[m2 ] + m_12 & = & c_13-m_1(c_12s_12c_23+c_12 ^ 2s_13s_23 e^i ) + & & + m_2 e^2i(c_12s_12c_23-s_12 ^ 2s_13s_23 e^i)+m_3c_13s_13s_23e^-i+2i(+ ) + m_13&= & c_13m_1 ( c_12s_12s_23-c_12 ^ 2s_13c_23 e^i ) + & & -m_2e^2 i(c_12s_12s_23+s_12 ^ 2s_13c_23 e^i)+m_3c_13s_13c_23e^-i+2i(+ ) + m_22 & = & m_1 ( s_12c_23+c_12s_23s_13e^i ) ^2 + & & + m_2 e^2 i(c_12c_23-s_12s_23s_13e^i ) ^2+m_3c_13 ^ 2s_23 ^ 2e^2i(+ ) + m_23 & = & m_1c_12s_12s_13(c_23 ^ 2-s_23 ^ 2)e^i + c_12 ^ 2c_23s_23s_13 ^ 2e^2i - s_12 ^ 2s_23c_23 + & & -m_2 e^2i c_12s_12s_13(c_23 ^ 2-s_23 ^ 2)e^i -s_12 ^ 2c_23s_23s_13 ^ 2e^2i+c_12 ^ 2s_23c_23 + & & + m_3c_23s_23c_13 ^ 2e^2i(+ ) + m_33 & = & m_1(c_12c_23s_13e^i - s_12s_23)^2 + & & + m_2 e^2i(s_12c_23s_13e^i+c_12s_23)^2+m_3c_23 ^ 2c_13 ^ 2e^2i(+).[m3 ] it is now straightforward to calculate @xmath1 and @xmath2 using eqn.([m2 ] ) to eqn.([m3 ] ) . neglecting terms @xmath45 and higher order we obtain @xmath1 and @xmath2 as i_12&=&a c_23 ^ 3[bc_23 - 2s_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 ] + & & + m_3 ^ 2c_12c_13 ^ 4s_12c_23[-2c_12 ^ 2s_23 ^ 3c_13 ^ 2m_1s_13a_1 + 2s_23 ^ 3c_13 ^ 2s_12 ^ 2m_2s_13a_2 ] + & & + m_3c_12c_13 ^ 4s_12c_23[s_12s_23 ^ 2c_13 ^ 2m_1c_12 ^ 3c_23a_3+s_12 ^ 3s_23 ^ 2c_13 ^ 2m_2c_12c_23a_4 + & & -2c_12 ^ 4s_23m_1m_2c_23 ^ 2s_13a_5 - 2s_23m_1m_2s_12 ^ 4c_23 ^ 2s_13a_5 + 2c_12 ^ 2s_23s_12 ^ 2s_13c_23 ^ 2 a_6 + & & + 2c_12 ^ 4s_23 ^ 3c_13 ^ 2m_1 ^ 2s_13a_7 + 4(2)c_12 ^ 2s_23 ^ 2c_13 ^ 2m_1m_2s_12 ^ 2s_13a_7 + 2s_23 ^ 3c_13 ^ 2m_2 ^ 2s_12 ^ 4s_13a_7][i1 ] i_13&=&a s_23 ^ 3[bs_23 + 2c_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 ] + & & -m_3 ^ 2c_12c_13 ^ 4s_12s_23[-2c_12 ^ 2c_23 ^ 3c_13 ^ 2m_1s_13a_1 + 2c_23 ^ 3c_13 ^ 2m_2s_12 ^ 2s_13a_2 ] + & & + m_3c_12c_13 ^ 4s_12s_23[c_12 ^ 3c_23 ^ 2c_13 ^ 2m_1s_12s_23a_3+c_12c_23 ^ 2c_13 ^ 2m_2s_12 ^ 3s_23a_4 + & & + 2c_12 ^ 4c_23m_1m_2s_23 ^ 2s_13a_5 + 2c_23m_1m_2s_12 ^ 4s_23 ^ 2s_13a_5 - 2c_12 ^ 2c_23s_12 ^ 2s_13s_23 ^ 2 a_6 + & & -2c_12 ^ 4c_23 ^ 3c_13 ^ 2m_1 ^ 2s_13a_7 -4(2)c_12 ^ 2c_23 ^ 3c_13 ^ 2m_1m_2s_12 ^ 2s_13a_7 - 2c_23 ^ 3c_13 ^ 2m_2 ^ 2s_12 ^ 4s_13a_7][i2 ] where a & = & - c_12s_12m_1m_2c_13 ^ 4[a ] + b&=&(2)c_12s_12(m_2 ^ 2-m_1 ^ 2)[b ] + _ 1 & = & ( 2-)m_1+m_2 + _ 2 & = & ( 2+)m_2-m_1and a_1&=&()m_1+(2-)m_2 + a_2&=&()m_2-(2+)m_1 + a_3&=&2(+)m_1 ^ 2 + 2(2 - 2 - 2)m_1m_2-(4 - 2 - 2)m_2 ^ 2 + a_4&=&2(++)m_1 ^ 2 - 22(+)m_1m_2-(2 - 2 - 2)m_2 ^ 2 + a_5&=&(2 - 2-)m_1+(2+)m_2 + a_6&=&(2+)m_1 ^ 3-(2 + 2+)m_1 ^ 2m_2 -(4 - 2-)m_1m_2 ^ 2+(2 - 2- ) + a_7&=&(2+)m_1+(2 - 2-)m_2 .a careful inspection reveals that the invariants are expressed in a tricky way . to be more precise , they are written as i_ij&=&_1 + s_13_2+m_3 ^ 2_3+m_3_4 where ` @xmath46 ' is some parameter dictated by eqn.([i1 ] ) and eqn.([i2 ] ) . the reason behind such a way to write down the invariants are the following : firstly , the popular paradigm in the neutrino mass models is to generate vanishing @xmath9 at the leading order and thereafter nonzero value of the same is generated by the means of some perturbation to the mass matrix and finally as the oscillation data dictates the mass square differences only , there is also a possibility of a vanishing neutrino mass ( e.g , models with scaling ansatz , zee - babu model etc . ) . therefore one can see the direct impact of their presence or absence in the measures of cp violation . + now the remaining invariant ( @xmath3 ) has a special character that it vanishes for @xmath13 and it comes out as i_23&=&m_3 ^ 3c_23s_23c_13 ^ 2b_1+m_3 ^ 2c_23s_23c_13 ^ 2[2c_12 ^ 3c_13 ^ 2m_2s_12s_13(c_23 ^ 2-s_23 ^ 2)b_2 + & & -2c_12c_13 ^ 2m_1s_12 ^ 3s_13(c_23 ^ 2-s_23 ^ 2)b_3]+m_3c_23s_23c_13 ^ 2[c_12 ^ 6c_23m_2 ^ 3s_23b_4 + & & + c_12 ^ 4c_23m_1m_2 ^ 2s_12 ^ 2s_23b_5+c_12 ^ 2c_23 ^ 2m_1 ^ 2m_2s_12 ^ 4s_23b_6+c_23m_1 ^ 3s_12 ^ 6s_23b_7 + & & -2c_12 ^ 5m_2 ^ 2s_12s_13(c_23 ^ 2-s_23 ^ 2)b_8 - 4(2)c_12 ^ 3m_1m_2s_12 ^ 3s_13(c_23 ^ 2-s_23 ^ 2)b_8 ] + & & -2c_12m_1 ^ 2s_12 ^ 5s_13(c_23 ^ 2-s_23 ^ 2)b_8][i3 ] with b_1&=&2(--)c_12 ^ 2c_23 ^ 2c_13 ^ 4m_2s_2 - 2(+)c_23c_13 ^ 4m_1s_12 ^ 2s_23 + & = & _ 1 + b_2&=&(2 -)m_1+()m_2 + & = & -_2 + b_3&= & ( ) m_1-(2 + ) m_2 + b_4&=&-(2 -2 - 2 ) + b_5&=&22(+)-(4 - 2 - 2 ) + b_6&=&-2(2 --)+2(2 + + ) + b_7&=&2(+ ) + b_8&=&(2+)m_1+(2- 2 - ) m_2 . at the outset , first , we would like to mention that the three independent invariants @xmath1 , @xmath2 and @xmath3 stand for the three cp violating phases @xmath30 , @xmath47 and @xmath28 , however in this section we solve the invariants only for the majorana phases ( @xmath48 ) while the dirac cp phase @xmath28 is calculable from the usual jarlskog measure of cp violation . next , for a general @xmath11 where all the parameters are present and all the eigenvalues and mixing angles are nonzero , all the invariants are independent and in principle one can extract the @xmath30 and @xmath47 phases without any specific hierarchical assumption which is also useful for the quasi - degenerate case . however , the calculation is too cumbersome in this general situation . in the present work we consider a simplified approach assuming hierarchical structure of neutrino masses and calculate the majorana phases utilizing the invariants @xmath1 , @xmath2 and @xmath3 for both , normal and inverted hierarchical cases . * case i : * @xmath50 , @xmath51 : @xmath52 @xmath53 @xmath54 . + in this case utilizing eqn.([i1 ] ) and ( [ i2])the majorana phase @xmath30 comes out as & = & ^-1- [ 2.1]where @xmath55 and we neglect the terms containing @xmath56 in both the invariants ( @xmath1 and @xmath2 ) . another equivalent expression of @xmath30 can also be obtained from eqn.([m2])(neglecting the term containing @xmath57 ) showing explicit relationship with @xmath5 decay parameter @xmath6 as & = & ^-1- . [ alph ] in principle we can use any of the equation ( eqn.([2.1 ] ) or eqn.([alph ] ) ) to find @xmath30 . the first one depends upon the explicit construction of @xmath1 and @xmath2 in terms of the neutrino mass matrix ( @xmath11 ) elements while the second one requires the knowledge of @xmath5 decay parameter @xmath6 . + in order to calculate @xmath47 from eqn.([i3 ] ) the terms involving @xmath58 can be neglected . therefore , assuming inverted hierarchy @xmath3 can be approximated with dominant term as i_23&=&m_2 ^ 3m_3c_23s_23c_13 ^ 2c_12 ^ 6c_23s_23b_4 + & = & -m_2 ^ 3m_3c_23 ^ 2s_23 ^ 2c_13 ^ 2c_12 ^ 6(2 - 2[+ ] ) . reverting the above equation the majorana phase @xmath47 is expressed as + & = & -^-1-+.[c1 ] * case ii : * @xmath59 , @xmath13 : @xmath60 @xmath53 @xmath54 . + in this case utilizing eqn.([i1]),([i2 ] ) and ( [ i3 ] ) the three rephasing invariants @xmath61 and @xmath3 come out as i_12&=&a c_23 ^ 3[bc_23 - 2s_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 ] [ m4 ] + & = & i_12 ^ 0 - 2ac_23 ^ 3s_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 + i_13&=&a s_23 ^ 3[bs_23 + 2c_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 ] [ m5 ] + & = & i_13 ^ 0 + 2as_23 ^ 3c_23s_13c_12 ^ 2m_1_1 + s_12 ^ 2m_2_2 + i_23&=&0[m6 ] where i_12 ^ 0&=&abc_23 ^ 4 + i_13 ^ 0&=&abs_23 ^ 4 with a , b already defined in eqn.([a ] ) and ( [ b ] ) respectively . as one of the invariant vanishes due to the condition @xmath13 , therefore , the three independent cp phases can not be solved from the above invariants and thus the two non zero invariants corresponds to one majorana phase ( @xmath30 ) and the dirac cp phase ( @xmath28 ) as @xmath47 vanishes for @xmath13 . proceeding as previous we get the same expression for the majorana phase @xmath30 as given in eqn.([alph ] ) . furthermore , solving eqn.([m4 ] ) to eqn.([m5 ] ) an equivalent expression of @xmath30 , same as eqn.([2.1 ] ) is also obtained . + * case iii : * @xmath62 , @xmath63 : @xmath64 @xmath53 @xmath65 . + in this case the invariants given in eqn.([m4]),([m5 ] ) and ( [ m6 ] ) become i_12&=&i_12 ^ 0 + i_13&=&i_13 ^ 0 and i_23&=&0 it is amply clear that the first two invariants @xmath1 and @xmath2 are not independent@xcite of each other and their correlated relationship leads to the estimation of only one majorana phase @xmath30 while the information about the dirac cp phase is lost . * case i : * @xmath50 , @xmath51 : @xmath52 @xmath53 @xmath54 . + in this case since @xmath67 and @xmath66 , we simplify @xmath1 and @xmath2 as i_12&=&(2 - 2[+])+s_13s_23 ^ 2 + i_13&=&(2 - 2[+])-s_13c_23 ^ 2[3.1 ] where the parameters @xmath68 and @xmath69 are defined through & = & -c_12 ^ 2c_23 ^ 2c_13 ^ 6m_2 ^ 3m_3s_12 ^ 4s_23 ^ 2&=&2c_12c_23c_13 ^ 6m_2 ^ 2m_3 ^ 2s_12 ^ 3s_23.now from eqn.([3.1 ] ) we get ( 2 - 2[+])&= & + & = & [ 2.2 ] . again due to the hierarchical condition @xmath66 , @xmath3 can be approximated as i_23&&m_3 ^ 3c_23s_23b_1 + & = & m_3 ^ 3c_23s_23[(2 - 2[+])c_12 ^ 2c_23c_13 ^ 4m_2s_2 - 2(+)c_23c_13 ^ 4m_1s_12 ^ 2s_23].[2.3 ] inserting eqn.([2.2 ] ) in ( [ 2.3 ] ) we get + & = & ^-1ct_12 ^ 2-where @xmath70 . + it is now straight forward to calculate the other majorana phase @xmath30 from eqn.([2.2 ] ) and it comes out as & = & .[c2 ] * case ii : * @xmath71 , @xmath33 : @xmath60 @xmath53 @xmath54 . + in this case neglecting terms like @xmath72 and @xmath58 in @xmath3 only the majorana phase difference @xmath73)$ ] is calculable and is given by -[+]&=&^-1along with an explicit relationship between the three invariants as - .therefore , essentially we get two independent invariants corresponding to the majorana phase difference and the dirac cp phase . + * case iii : * @xmath74 , @xmath35 , @xmath75 : @xmath64 @xmath53 @xmath65 . + in such a condition the three invariants are coming out in a correlated manner as i_12&=&2(-[+ ] ) + & = & -2(-[+])c_12 ^ 2c_23 ^ 2s_23 ^ 2c_13 ^ 6s_12 ^ 4m_2 ^ 3m_3 + & = & i_13 + i_23&=&2(-[+])c_12 ^ 2c_23 ^ 2s_23 ^ 2c_13 ^ 6m_2m_3 ^ 3 + & = & ( -s_12 ^ -4)i_12 + & = & ( -s_12 ^ -4)i_13 and in this case only independent invariant @xmath1 is connected to the majorana phase difference @xmath73)$ ] . although in the present work we are not discussing the quasi - degenerate case which is relevant in the cosmological context @xcite , however , one can calculate the majorana phases in a model independent way by directly solving the invariants as mentioned at the beginning of sec . [ s4 ] . to be precise , using eqn . ( [ i1 ] ) , ( [ i2 ] ) and eqn . ( [ i3 ] ) one can extract all the cp violating phases without any hierarchical assumption . however , the calculation is tedious and will be studied elsewhere . another alternative way is to follow the calculations presented in ref . however , in that case the phase convention is different . utilizing the phase convention presented in this work one can calculate all the phases in the second method also . a general solution for a three generation complex symmetric majorana mass matrix is given in ref.@xcite . in order to estimate the majorana phases obtained in the present work we utilize the expressions of the three eigenvalues and the three mixing angles . we also use the global fit data of neutrino oscillation experiments shown in table [ tl ] and the upper limits on the sum of the neutrino masses ( @xmath76 ev)@xcite and the @xmath5 parameter ( @xmath77 ev)@xcite to obtain model independent ranges of the majorana phases . + .input experimental values@xcite [ cols="^,^",options="header " , ] + as previously mentioned , unlike the dirac cp phase @xmath28 , the majorana phases do not appear in the neutrino@xmath78neutrino oscillation . therefore , a natural question arises how and where these phases can be measured . as a direct detection , in ref . @xcite xing suggested a thought experiment ( neutrino@xmath78 antineutrino oscillation ) in which he pointed out these phases may appear in the probability expression of the flavour oscillation and thus also in the expression of the cp asymmetry parameter @xmath79 which is the measure of cp violation . however , this kind of experiment is purely academic at this moment and practically difficult to design as the oscillation probability is highly suppressed by the factor @xmath80 , where @xmath81 is the mass of the light neutrino and @xmath82 is the beam energy . now considering @xmath82 @xmath83 mev and the masses of the neutrinos to be less than 1 ev , one can calculate @xmath80 to be @xmath84 . to improve @xmath85 , a novel suggestion@xcite is to lower the value of @xmath82 , however , in that case the estimated size of the base line length and the detector are beyond the reach of the present experimental facilities . but as an optimistic point of view we expect these kind of experiments will be designed in future and thus the prediction of the majorana phases will be tested . beside neutrino @xmath78 antineutrino oscillation there are several lnv processes like @xmath5 decay , @xmath86 ( in type ii seesaw model)@xcite etc . , which play a crucial role for the indirect measurement of the majorana phases . now coming into our work , we present a table in the appendix which shows the ranges of the obtained majorana phases for some typical values of @xmath6 and for convenience , in figure [ obs ] we present variation of the majorana phases with @xmath6 for the best fit value of @xmath87 and taking all the other constraints in their 3@xmath88 ranges for both the hierarchies . we would like to mention that even if we take the 3@xmath88 range of @xmath87 , over all ranges of the majorana phases do not differ much , however , unlike the plots of figure [ obs ] , the plots in that case become more wider for the higher values of @xmath6 ( @xmath89 ev ) . although , the present experimental upper bound on @xmath6 is 0.35 ev , next will be able to bring down the value to 0.1 ev and thus the approximate ranges of the majorana phases can be predicted . , @xmath47 ) vs @xmath6 for normal ( left ) and inverted ( right ) hierarchies for best fit values of @xmath87.,title="fig : " ] , @xmath47 ) vs @xmath6 for normal ( left ) and inverted ( right ) hierarchies for best fit values of @xmath87.,title="fig : " ] + , @xmath47 ) vs @xmath6 for normal ( left ) and inverted ( right ) hierarchies for best fit values of @xmath87.,title="fig : " ] , @xmath47 ) vs @xmath6 for normal ( left ) and inverted ( right ) hierarchies for best fit values of @xmath87.,title="fig : " ] thus far we have estimated the majorana phases in a general context . latter , we apply the expressions obtained for @xmath30 and @xmath47 for few testable flavour models ( models with lesser number of parameters ) as an application of the general result although our analysis is true for any hierarchical model of neutrino masses . the reason we discuss this section is to make certain whether the results obtain in the general case are consistent with the other models or not . moreover the models with certain flavour symmetries are highly predictive in nature . therefore , precise measurement of the cp violating phases may act as the important tools to verify the testability of the flavour models@xcite . in inverted hierarchy section we present a model with scaling ansatz and texture zeros within the framework of inverse seesaw through which all the sub cases presented in sec . [ invh ] can be realized while in the normal hierarchy section we present a model with cyclic symmetry within the framework of type i seesaw . obviously the choices are for illustration . one can also consider inverse or linear seesaw for normal hierarchy@xcite and type i seesaw for inverted hierarchy@xcite . in principle one can use the technique in any hierarchical flavour models . the above numerical results are obtained for the general @xmath11 where all the 9 independent parameters are present . however , as previously said , one can reduce the number of parameters by invoking some symmetry or ansatz in the lagrangian which is more predictive in nature and thus testable in the experiments . in this section we provide applications of the general results in few typical cases for both the hierarchies , normal and inverted . in this case we explore a model that corresponds to * case i * of the normal hierarchical scenario mentioned in section ( [ nrmlh ] ) . the model is based on cyclic symmetry with type i seesaw mechanism to accommodate the neutrino oscillation data . in the fundamental level the symmetry exists in the neutrino sector of the lagrangian and due to the symmetry a degeneracy in masses occurs removal of which therefore requires breaking of the symmetry . it is shown that a minimal breaking in the majorana mass matrix is sufficient to explain the extant data . in this model the low energy broken symmetric mass matrix @xmath90 originated from type i seesaw mechanism is given by m_= m_0 p^2e^2i++&pe^i++ & + pqe^i(+)+ + pe^i++&1++&++qe^i + + pqe^i(+)+&++qe^i&+q^2e^2i+ where m_d = y_1&y_2&y_3 + y_3&y_1&y_2 + y_2&y_3&y_1 , m_r = diag(m+ _ 1 , m+_2 , m ) with @xmath91 and @xmath92 as the breaking parameters and m_0=-,p e^i=,q e^i=. vs @xmath47 . ] + for numerical analysis we choose the mass scale of @xmath93 to be of the order of @xmath94 gev and @xmath95 to be at electroweak scale . further redefining the breaking parameters as @xmath96 and @xmath97 we allow them to vary as @xmath98 to keep the breaking effect small . we then constrain the parameter spaces taking into account the 3@xmath88 ranges of neutrino oscillation global fit data and explicitly evaluate both the majorana phases . from figure [ fig1 ] the ranges read as @xmath99 and @xmath100 . note that the ranges of both the phases are embedded within the values obtained for the general case . similar to the general case , in figure [ obsn ] we also present the variation of the majorana phases with the @xmath5 parameter . one can see the upper limit of @xmath6 is @xmath83 0.07 ev which is well within the reach of the future planned experiments . + and @xmath47 with @xmath6 for cyclic symmetric case ( normal hierarchy).,title="fig : " ] and @xmath47 with @xmath6 for cyclic symmetric case ( normal hierarchy).,title="fig : " ] + as the model consists of lesser number of parameters , we also expect a significant correlation between the phase invariants and are depicted in figure [ incor1 ] . vs @xmath3 ( left ) and @xmath2 vs @xmath3 ( right).,title="fig : " ] vs @xmath3 ( left ) and @xmath2 vs @xmath3 ( right).,title="fig : " ] + in this case , we explore a model based on scaling ansatz with inverse seesaw mechanism@xcite . in this mechanism @xmath11 is given by m_&= & m_d m_rs^-1 ( m_d m_rs^-1)^t + where @xmath95 is the usual dirac type matrix and the other two matrices @xmath101 ( majorana type ) and @xmath102 ( dirac type ) arise due to the interaction between the additional singlet fermion and right handed neutrino considered in this type of seesaw mechanism . to further reduce the number of parameters texture zeros@xcite are assumed in the constituent @xmath95 and @xmath101 matrices . scaling ansatz invariance dictates @xmath13 and @xmath103 and this case corresponds to * case iii * of sec . [ invh ] . thus to generate non zero @xmath9 breaking of the ansatz is necessary . incorporating breaking in @xmath95 through a small parameter @xmath104 , there are two different phenomenologically survived textures which are given by m_^1&= & m_0 1 & k_1p & p + k_1p & k_1 ^ 2(q^2e^i+p^2 ) & k_1(q^2e^i+p^2 ) + p & k_1(q^2e^i+p^2)&(q^2e^i+p^2 ) + m_0 0&0&0 + 0&2k_1 ^ 2q^2e^i&k_1q^2e^i + 0&k_1q^2e^i&0 + [ mab ] and m_^2 = m_0 1&k_1(p+qe^i)&p+qe^i + k_1(p+qe^i)&k_1 ^ 2(2pqe^i+p^2)&k_1(2pqe^i+p^2 ) + p+qe^i&k_1(2pqe^i+p^2)&(2pqe^i+p^2 ) + + + m_0 0&k_1qe^i&0 + k_1qe^i&2k_1 ^ 2pqe^i&k_1pqe^i + 0&k_1pqe^i&0 + [ mc ] where all the parameters are complex @xcite . in both the cases @xmath105 however , @xmath13 due to singular nature of @xmath101 matrix and this case corresponds to * case ii * of sec . + we further consider the most general version of the above case through the breaking of the ansatz in both @xmath95 and @xmath101 matrices through two small parameters @xmath104 and @xmath106 respectively and the neutrino mass matrix @xmath107 comes out as m_^3= m_0 1 & k_1p & p + k_1p & k_1 ^ 2(q^2e^i+p^2 ) & k_1(q^2e^i+p^2 ) + p & k_1(q^2e^i+p^2)&(q^2e^i+p^2 ) + m_0 0&0&0 + 0&2k_1 ^ 2q^2e^i&k_1q^2e^i + 0&k_1q^2e^i&0 + + + m_0 ^ 0&k_1p&p + k_1p&0&0 + p&0&0 [ mnu3 ] and in this situation both @xmath9 and @xmath12 are nonzero corresponding to * case i * of sec . thus the whole inverted hierarchical sector is generated through the choice of the above model . + now , with the explicit construction of rephasing invariants we calculate the majorana phases in each case . interestingly , for all the cases , the value of @xmath108 comes out very small due to smallness of the dirac cp phase @xmath28 , or more precisely , due to almost real nature of the mass matrices . therefore , such typical nature of the mass matrices also constrains the majorana phases approximately as @xmath109 for the first two matrices ( @xmath110 and @xmath111 ) and @xmath112,@xmath113 for the matrix @xmath114 along with an approximate range of @xmath5 parameter @xmath6 as 0.01 ev @xmath115 0.0148 ev and 0.01 ev @xmath115 0.0152 ev respectively . for illustration , in figure [ npb ] we plot @xmath30 and @xmath47 with @xmath6 for @xmath116 . for other two matrices ( @xmath117 and @xmath118 ) the variations of @xmath30 with @xmath6 are almost same as that of the extreme left plot of the lower panel of figure [ npb ] . the model is highly predictive and hence , if significant cp violation is observed , the model will be ruled out . vs @xmath47 ( upper panel ) and variation of @xmath30 and @xmath47 with @xmath6 for inverted hierarchy : scaling ansatz case ( lower panel).,title="fig : " ] + vs @xmath47 ( upper panel ) and variation of @xmath30 and @xmath47 with @xmath6 for inverted hierarchy : scaling ansatz case ( lower panel).,title="fig : " ] vs @xmath47 ( upper panel ) and variation of @xmath30 and @xmath47 with @xmath6 for inverted hierarchy : scaling ansatz case ( lower panel).,title="fig : " ] we plot the correlation between the invariants in figure [ incor2 ] . + vs @xmath3 ( left ) and @xmath2 vs @xmath3 ( right).,title="fig : " ] vs @xmath3 ( left ) and @xmath2 vs @xmath3 ( right).,title="fig : " ] finally , we summarize our results in table [ t2 ] that shows the ranges of the majorana phases for all the cases . & @xmath119 & @xmath120 & @xmath119 & @xmath120 & @xmath120 + & @xmath121 @xmath51 & @xmath122 @xmath51 & @xmath121 @xmath51 & @xmath122 @xmath51 & @xmath123 @xmath51 + @xmath30 ( deg . ) & @xmath124 & @xmath125 & @xmath126 & @xmath127 & @xmath128 + @xmath47 ( deg.)&@xmath129&@xmath130&@xmath131&@xmath132&@xmath133 + in the present work we calculate the majorana phases of a general complex symmetric @xmath0 neutrino mass matrix utilizing the three rephasing invariant quantities @xmath1 , @xmath2 and @xmath3 proposed by sarkar and singh for both the hierarchical structures of neutrinos using mohapatra - rodejohann s phase convention . motivation behind the usage of the invariants to calculate the majorana phases is that such methodology enables us to evaluate the existing majorana phase even if one of the eigenvalue ( @xmath12 ) is zero in a model independent way . however , if @xmath33 , this methodology will only enable us to calculate the difference of the majorana phases therefore it is needed to change the phase convention in that case . after the presentation of the generalized prescription , we further present the maximal allowed ranges of the majorana phases in general context for both the viable hierarchical structures of neutrino masses and address our methodology to be true for any model except the case of quasi degeneracy . we then talk about the connection of the majorana phases with physical observables like @xmath5 parameter @xmath6 and the branching ratios of charged higgs ( @xmath134 ) decay where the phases show up . as a direct measurement of the majorana phases we give the example of neutrino@xmath78 antineutrino oscillation which is a thought experiment right now , however , well studied in literature . although the presented methodology can be used in any hierarchical models in neutrino physics , after discussing the general case we further exemplify our methodology in few typical testable models ( models with lesser number of parameters ) leading to normal and inverted hierarchy and with their significant predictions on @xmath6 and the majorana phases . for normal hierarchical case we give an example of a model based on cyclic symmetry with type i seesaw mechanism . we estimate the majorana phases for the broken symmetric case , since cyclic symmetry dictates a degeneracy in the mass eigenvalues . as an example of inverted hierarchy , we cite a model with high predictability and comprised of scaling ansatz , texture zeros and inverse seesaw mechanism . it is seen that all the sub cases of inverted hierarchy mentioned in section [ invh ] can be obtained depending upon the scheme of incorporation of ansatz breaking mechanism while a phenomenologically viable sub case ( @xmath33 ) of the normal hierarchy is yet to be established through the choice of a suitable model . & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142 & @xmath143&@xmath144 + @xmath30 ( deg . ) & @xmath124&@xmath124 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath150 & @xmath151 + @xmath152 ( deg . ) & @xmath129 & @xmath129 & @xmath129 & @xmath130 & @xmath130 & @xmath153 & @xmath154 & @xmath155 & @xmath155 & @xmath156 + + & @xmath137 & @xmath157 & @xmath138 & @xmath158 & @xmath159 & @xmath140 & @xmath141 & @xmath142 & @xmath143&@xmath144 + @xmath30 ( deg . ) & @xmath130 & @xmath130 & @xmath130 & @xmath153 & @xmath153 & @xmath153 & @xmath155 & @xmath155 & @xmath160 & @xmath160 + @xmath152 ( deg . ) & @xmath161 & @xmath161 & @xmath161 & @xmath125 & @xmath125 & @xmath162 & @xmath163 & @xmath164 & @xmath164 & @xmath165 + l. serra _ et al . _ [ next collaboration ] , arxiv:1410.6700 [ physics.ins-det ] . m. agostini _ et al . _ [ gerda collaboration ] , phys . lett . * 111 * , no . 12 , 122503 ( 2013 ) doi:10.1103/physrevlett.111.122503 [ arxiv:1307.4720 [ nucl - ex ] ] . b. majorovits [ gerda collaboration ] , aip conf . proc . * 1672 * , 110003 ( 2015 ) doi:10.1063/1.4928005 [ arxiv:1506.00415 [ hep - ex ] ] . k. asakura _ et al . _ [ kamland - zen collaboration ] , arxiv:1509.03724 [ hep - ex ] . m. agostini _ et al . _ , phys . procedia * 61 * , 828 ( 2015 ) . doi:10.1016/j.phpro.2015.06.002 w. xu _ et al . _ [ majorana collaboration ] , arxiv:1501.03089 [ nucl - ex ] . f. simkovic , s. m. bilenky , a. faessler and t. gutsche , phys . d * 87 * , no . 7 , 073002 ( 2013 ) [ arxiv:1210.1306 [ hep - ph ] ] . j. j. gomez - cadenas and j. martin - albo , arxiv:1502.00581 [ hep - ex ] . w. tornow , arxiv:1412.0734 [ nucl - ex ] . s. m. bilenky and c. giunti , arxiv:1411.4791 [ hep - ph ] . e. w. hennecke , o. k. manuel and d. d. sabu , phys . c * 11 * , 1378 ( 1975 ) . k. zuber , j. phys . ser . * 578 * , no . 1 , 012007 ( 2015 ) . b. schwingenheuer , annalen phys . * 525 * , 269 ( 2013 ) doi:10.1002/andp.201200222 [ arxiv:1210.7432 [ hep - ex ] ] . m. auger _ [ exo collaboration ] , phys . * 109 * , 032505 ( 2012 ) [ arxiv:1205.5608 [ hep - ex ] ] . [ david lorca for the next collaboration ] , arxiv:1411.0475 [ physics.ins-det ] . z. z. xing and y. l. zhou , phys . rev . d * 88 * , 033002 ( 2013 ) [ arxiv:1305.5718 [ hep - ph ] ] . s. gupta , s. k. kang and c. s. kim , arxiv:1406.7476 [ hep - ph ] . h. minakata , h. nunokawa and a. a. quiroga , ptep * 2015 * , 033b03 ( 2015 ) doi:10.1093/ptep / ptv010 [ arxiv:1402.6014 [ hep - ph ] ] . u. sarkar and s. k. singh , nucl . b * 771 * , 28 ( 2007 ) [ hep - ph/0608030 ] . r. n. mohapatra and w. rodejohann , phys . b * 644 * ( 2007 ) 59 [ hep - ph/0608111 ] . s. m. bilenky , j. hosek and s. t. petcov , phys . b * 94 * , 495 ( 1980 ) . c. giunti , phys . b * 686 * , 41 ( 2010 ) [ arxiv:1001.0760 [ hep - ph ] ] . j. beringer _ [ particle data group collaboration ] , phys . d * 86 * , 010001 ( 2012 ) . b. adhikary , m. chakraborty and a. ghosal , jhep * 1310 * , 043 ( 2013 ) [ erratum - ibid . * 1409 * , 180 ( 2014 ) ] [ arxiv:1307.0988 [ hep - ph ] ] . z. z. xing , phys . d * 87 * , no . 5 , 053019 ( 2013 ) [ arxiv:1301.7654 [ hep - ph ] ] . z. z. xing , int . j. mod . a * 29 * , 1444003 ( 2014 ) . d. delepine , v. g. macias , s. khalil and g. l. castro , phys . b * 693 * , 438 ( 2010 ) . f. beutler _ et al . _ [ boss collaboration ] , mon . not . soc . * 444 * , 3501 ( 2014 ) doi:10.1093/mnras / stu1702 [ arxiv:1403.4599 [ astro-ph.co ] ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1502.01589 [ astro-ph.co ] . d. v. forero , m. tortola and j. w. f. valle , phys . rev . * d 86 * ( 2012 ) 073012 [ arxiv:1205.4018 [ hep - ph ] ] . y. shimizu and m. tanimoto , arxiv:1507.06221 [ hep - ph ] . b. adhikary , a. ghosal and p. roy , indian j. phys . * 88 * , 979 ( 2014 ) [ arxiv:1311.6746 [ hep - ph ] ] . m. chakraborty , h. z. devi and a. ghosal , phys . b * 741 * , 210 ( 2015 ) arxiv:1410.3276 [ hep - ph ] . b. adhikary , m. chakraborty and a. ghosal , phys . d * 86 * , 013015 ( 2012 ) [ arxiv:1205.1355 [ hep - ph ] ] . r. n. mohapatra and j. w. f. valle , phys . d * 34 * , 1642 ( 1986 ) . j. bernabeu , a. santamaria , j. vidal , a. mendez and j. w. f. valle , phys . b * 187 * , 303 ( 1987 ) . r. n. mohapatra , phys . lett . * 56 * , 561 ( 1986 ) . j. schechter and j. w. f. valle , phys . d * 25 * , 774 ( 1982 ) . p. s. b. dev and r. n. mohapatra , phys . d * 81 * , 013001 ( 2010 ) doi:10.1103/physrevd.81.013001 [ arxiv:0910.3924 [ hep - ph ] ] . r. l. awasthi , m. k. parida and s. patra , jhep * 1308 * , 122 ( 2013 ) doi:10.1007/jhep08(2013)122 [ arxiv:1302.0672 [ hep - ph ] ] . a. abada and m. lucente , nucl . b * 885 * , 651 ( 2014 ) doi:10.1016/j.nuclphysb.2014.06.003 [ arxiv:1401.1507 [ hep - ph ] ] . a. palcu , arxiv:1408.6518 [ hep - ph ] . j. schechter and j. w. f. valle , phys . d * 22 * , 2227 ( 1980 ) . s. fraser , e. ma and o. popov , phys . b * 737 * , 280 ( 2014 ) [ arxiv:1408.4785 [ hep - ph ] ] . h. hettmansperger , m. lindner and w. rodejohann , jhep * 1104 * , 123 ( 2011 ) [ arxiv:1102.3432 [ hep - ph ] ] . s. s. c. law and k. l. mcdonald , phys . d * 87 * , no . 11 , 113003 ( 2013 ) [ arxiv:1303.4887 [ hep - ph ] ] . a. blum , r. n. mohapatra and w. rodejohann , phys . d * 76 * , 053003 ( 2007 ) [ arxiv:0706.3801 [ hep - ph ] ] . m. obara , arxiv:0712.2628 [ hep - ph ] . a. damanik , m. satriawan , muslim and p. anggraita , arxiv:0705.3290 [ hep - ph ] . w. grimus and l. lavoura , j. phys . g g * 31 * , 683 ( 2005 ) [ hep - ph/0410279 ] . s. dev , r. r. gautam and l. singh , phys . d * 89 * , no . 1 , 013006 ( 2014 ) [ arxiv:1309.4219 [ hep - ph ] ] . p. h. frampton , s. l. glashow and d. marfatia , phys . b * 536 * , 79 ( 2002 ) [ hep - ph/0201008 ] . k. whisnant , j. liao and d. marfatia , aip conf . * 1604 * , 273 ( 2014 ) . p. o. ludl and w. grimus , jhep * 1407 * , 090 ( 2014 ) [ arxiv:1406.3546 [ hep - ph ] ] . p. o. ludl and w. grimus , arxiv:1501.04942 [ hep - ph ] . w. grimus and p. o. ludl , pos eps * -hep2013 * , 075 ( 2013 ) [ arxiv:1309.7883 [ hep - ph ] ] . l. lavoura , arxiv:1502.03008 [ hep - ph ] . p. m. ferreira and l. lavoura , nucl . b * 891 * , 378 ( 2015 ) [ arxiv:1411.0693 [ hep - ph ] ] . z. z. xing and z. h. zhao , arxiv:1501.06346 [ hep - ph ] . z. z. xing , phys . b * 530 * , 159 ( 2002 ) [ hep - ph/0201151 ] . j. liao , d. marfatia and k. whisnant , jhep * 1409 * , 013 ( 2014 ) [ arxiv:1311.2639 [ hep - ph ] ] . h. fritzsch , z. z. xing and s. zhou , jhep * 1109 * , 083 ( 2011 ) [ arxiv:1108.4534 [ hep - ph ] ] . a. merle and w. rodejohann , phys . d * 73 * , 073012 ( 2006 ) [ hep - ph/0603111 ] . w. wang , eur . j. c * 73 * , 2551 ( 2013 ) [ arxiv:1306.3556 [ hep - ph ] ] . w. wang , phys . b * 733 * , 320 ( 2014 ) [ erratum - ibid . b * 738 * , 524 ( 2014 ) ] [ arxiv:1401.3949 [ hep - ph ] ] . l. lavoura , phys . b * 609 * , 317 ( 2005 ) [ hep - ph/0411232 ] . a. kageyama , s. kaneko , n. shimoyama and m. tanimoto , phys . b * 538 * , 96 ( 2002 ) [ hep - ph/0204291 ] . w. wang , phys . d * 90 * , 033014 ( 2014 ) [ arxiv:1402.6808 [ hep - ph ] ] . g. c. branco , d. emmanuel - costa , m. n. rebelo and p. roy , phys . d * 77 * ( 2008 ) 053011 [ arxiv:0712.0774 [ hep - ph ] ] . s. choubey , w. rodejohann and p. roy , nucl . b * 808 * , 272 ( 2009 ) [ erratum - ibid . * 818 * , 136 ( 2009 ) ] [ arxiv:0807.4289 [ hep - ph ] ] . b. adhikary , a. ghosal and p. roy , jhep * 0910 * ( 2009 ) 040 [ arxiv:0908.2686 [ hep - ph ] ] . b. adhikary , a. ghosal and p. roy , jcap * 1101 * ( 2011 ) 025 [ arxiv:1009.2635 [ hep - ph ] ] . b. adhikary , a. ghosal and p. roy , mod . a * 26 * ( 2011 ) 2427 [ arxiv:1103.0665 [ hep - ph ] ] a. ghosal and r. samanta , jhep * 1505 * , 077 ( 2015 ) , [ arxiv:1501.00916 [ hep - ph ] ] .

we prove the boundedness of the directional hilbert transform in the plane relative to measurable vector fields which are constant on suitable lipschitz curves .

1401.2890

consider the directional hilbert transform in the plane defined for a fixed direction @xmath1 as h_v f(x , y):=p.v . _ f(x -t , y- u t)t for any test function @xmath2 . by the dilation symmetry , the length of the vector @xmath3 is irrelevant for the value of @xmath4 , which explains our normalization of the first component . by an application of fubini s theorem and the @xmath5 bounds for the classical hilbert transform one obtains a priori @xmath5 bounds for @xmath4 . on the other hand , the corresponding maximal operator @xmath6 for varying directions is well known to not satisfy any a priori @xmath5 bounds . + in @xcite , bateman and thiele proved that @xmath7 for the range @xmath8 . this result is half of a maximal bound since the supremum is squeezed between the two @xmath5 norms on the left hand side , compared to being completely inside or outside as in our previous remarks . the case @xmath9 of ( [ bt ] ) goes back to coifman and el kohen ( see @xcite for the detailed discussion ) , who noticed that a fourier transform in the @xmath10 direction makes @xmath11 for @xmath9 equivalent to @xmath0 bounds for carleson s operator . estimate ( [ bt ] ) highlights a bi - parameter structure of the directional hilbert transform . the bi - parameter structure arises since the kernel is a tensor product between a hilbert kernel in direction v and a dirac delta distribution in the perpendicular direction . if one considers the linearized maximal operator @xmath12 for some function @xmath13 , then inequality ( [ bt ] ) can be rephrased as a bound for the linearized maximal operator under the assumption that @xmath14 is constant on every vertical line @xmath15 for all @xmath16 . such vector fields @xmath3 of the form @xmath17 for some measurable function @xmath18 are call one - variable vector fields in @xcite . + the purpose of the present paper is to relax this rigid assumption on @xmath14 , and prove an analogue of ( [ bt ] ) for vector fields which are constant along suitable families of lipschitz curves . to formulate such a result , we perturb ( [ bt ] ) by a bi - lipschitz horizontal distortion , that is @xmath19 with @xmath20 for every @xmath21 and every @xmath10 , such that the transformation ( [ bilip ] ) maps vertical lines to near vertical lipschitz curves : latexmath:[\[\label{liplines } two conditions can be rephrased as 1/a_0_1ga_0 |_2g|b_0 under these assumptions , @xmath5 norms are distorted boundedly under the transformation ( [ bilip ] ) . namely , ( [ hordis ] ) implies for every @xmath10 that [ mm1.8 ] a_0 ^ -1f(x , y)_l^p(x)^p f(g(x , y),y)_l^p(x)^pa_0 f(x , y)_l^p(x)^p and we may integrate this in @xmath10 direction to obtain equivalence of @xmath5 norms in the plane . hence the change of measure is not the main point of the following theorem , but rather the effect of the transformation on the linearizing function @xmath14 , which is now constant along the family of lipschitz curves which are the images of the lines @xmath15 under the map ( [ bilip ] ) . let @xmath23 satisfy assumption ( [ hordis ] ) for some @xmath24 and assumption ( [ liplines ] ) for some @xmath25 . then for any @xmath26 , we have @xmath27 here @xmath28 is a constant depending only on @xmath24 and @xmath29 . the constant @xmath28 is independent of @xmath25 due to the unisotropic scaling symmetry @xmath30 . in view of the implicit function theorem ( see @xcite for recent developments ) , our result covers a large class of vector fields which are of the critical lipschitz regularity . indeed , it implies the following [ corollary1.2 ] for a measurable unit vector field @xmath31 , suppose that \i ) there exists a bi - lipschitz map @xmath32 s.t . [ e1.8 ] v_0(g_0(x , y ) ) y ; \ii ) there exists @xmath33 s.t . @xmath34 , [ scaling ] ( _ 2 g_0(x , y ) , v_0(g_0(x , y ) ) ) d_0 ya.e . . then the associated hilbert transform , which is defined as h_v_0f(x , y):=_f((x , y)-t v_0(x , y))dt / t , is bounded in @xmath0 , with the operator norm depending only on @xmath35 and the bi - lipschitz norm of @xmath36 . [ nnremark1.4 ] the structure theorem for lipschitz functions by azzam and schul in @xcite states exactly that any lipschitz function @xmath37 ( any lipschitz unit vector field @xmath38 in our case ) can be precomposed with a bi - lipschitz function @xmath39 such that @xmath40 is lipschitz in the first coordinate and constant in the second coordinate , when restricted to a `` large '' portion of the domain . without the assumption that @xmath33 , the operator @xmath41 might be unbounded in @xmath5 for any @xmath42 . the counter - example is based on the besicovitch - kakeya set construction , which will be discussed at the end of the proof of the corollary . to our knowledge , this is the first result in the context of the directional hilbert transform with a lipschitz assumption in the hypothesis . lipschitz regularity is critical for the directional hilbert transform as we will elaborate shortly . to use the assumption that @xmath3 is constant along lipschitz curves , we apply an adapted littlewood - paley theory along the level lines of @xmath3 . this is a refinement of the analysis of coifman and el kohen who use a fourier transform in the @xmath10 variable and the analysis of bateman and thiele who use a classical littlewood - paley theory in the @xmath10 variable . this adapted littlewood - paley theory is the main novelty of the present paper . it is in the spririt of prior work on the cauchy integral on lipschitz curves , for example @xcite , but it differs from this classical theme in that it is more of bi - parameter type as it is governed by a whole fibration into lipschitz curves . we crucially use jones beta numbers as a tool to control the adapted littlewood - paley theory . to our knowledge this is also the first use of jones beta numbers in the context of the directional hilbert transform . + in this paper we focus on the case @xmath0 , since our goal here is to highlight the use of the adapted littlewood - paley theory and jones beta numbers in the technically most simple case . we expect to address the more general case @xmath5 with a range of @xmath43 as in the bateman - thiele theorem in forthcoming work . while coifman and el kohen use the difficult bounds on carleson s operator as a black box , bateman and thiele have to unravel this black box following the work of lacey and li @xcite-@xcite and use time - frequency analysis to prove bounds for a suitable generalization of carleson s operator . luckily , in the present work we do not have to delve into time - frequency analysis as we can largely recycle the work of bateman and thiele for this aspect of the argument . an upper bound such as @xmath44 is necessary in our theorem . by a limiting argument we may recover the theorem of bateman and thiele , using the scaling to tighten the lipschitz constant @xmath25 at the same time as relaxing the condition @xmath44 . an interesting open question remains whether the same holds true for @xmath45 . we do not know of a soft argument to achieve this relaxation , as the norm bounds arising from our argument grow unlimited as @xmath29 approaches @xmath46 . this question suggests itself for further study . + part of our motivation is a long history of studies of the linearized maximal operator ( [ linmax ] ) under various assumptions on the linearizing function @xmath14 . if one truncates @xmath47 as @xmath48 then it is reasonable to ask for pure regularity assumptions on @xmath14 to obtain boundedness of @xmath49 . it is known that lipschitz regularity of @xmath14 is critical , since a counterexample in @xcite based on a construction of the besicovitch - kakeya set shows that no bounds are possible for @xmath50 regularity with @xmath51 . however , it remains open whether lipschitz regularity suffices for bounds for @xmath49 . on the regularity scale , the only known result is for real analytic vector fields @xmath3 by stein and street in @xcite . a prior partial result in this direction appears in @xcite . it is our hope that our result corners some of the difficulties of approaching lipschitz regularity in the classical problem . further substantial progress ( including the case @xmath45 ) is likely to use lipschitz regularity not only of the level curves of @xmath14 but also of @xmath14 itself across the level curves . for example , one possibility would be to cut the plane into different pieces by the theorem of azzam and schul stated in remark [ nnremark1.4 ] , and to analyze each piece separately by the main theorem in the present paper . we leave this for the future study . + related to the directional hilbert transform and thus additional motivation for the present work is the directional maximal operator @xmath52 which arises for example in lebesgue type differentiation questions and has an even longer history of interest than the directional hilbert transform . hilbert transforms and maximal operators share many features , in particular they have the same scaling and thus share the same potential @xmath5 bounds . the maximal operator is in some ways easier as it is positive and does not have a singular kernel . for example , bounds for the maximal operator under the assumption of real analytic vector fields were proved much earlier by bourgain in @xcite . an instance of bounds satisfied by the maximal operator but not the hilbert transform arises when one restricts the range of the function @xmath14 instead of the regularity . for certain sets of directions characterized by bateman in @xcite there are bounds for the maximal operator ( for example for the set of lacunary directions ) , while karagulyan proves in @xcite that no such bounds are possible for the hilbert transform . on the other hand , the hilbert transform is easier in some other aspects , most notably it is a linear operator . for example , bounds for the bilinear hilbert transform mapping into @xmath53 were known ( lacey and thiele @xcite , @xcite ) before the corresponding maximal operator bounds ( lacey @xcite ) , due to the fact that orthogonality between different tiles is preserved under the hilbert transform but not the maximal operator . in particular we do not know at the moment whether the analogue of our main theorem holds for the directional maximal operator . this may be an interesting subject for further investigation . + * outline of paper : * in section 2 we will prove corollary [ corollary1.2 ] by reducing it to the main theorem . the reduction will also be used later in the proof of the main theorem . in section 3 we will state the strategy of the proof for the main theorem . as it appears that our result is a lipschitz perturbation of the one by bateman and thiele , this turns out also to be the case for the proof : if we denote by @xmath54 a littlewood - paley operator in the @xmath10-variable , the main observation in bateman and thiele s proof is that @xmath4 commutes with @xmath54 . in our case , this is no longer true . however , we can make use of an adapted version of the littlewood - paley projection operator @xmath55 ( see definition [ definition1 ] ) to partially recover the orthogonality . we split the operator @xmath4 into a main term and a commutator term _ k h_v p_k(f)=_k(h_v p_k(f)-_k h_v p_k(f)+_k h_v p_k(f ) ) . the boundedness of the main term @xmath56 is essentially due to lacey and li @xcite , with conditionality on certain maximal operator estimate . in section 4 we modify bateman s argument in @xcite and @xcite to the case of vector fields constant on lipschitz curves and remove the conditionality on that maximal operator . the main novelty is the boundedness of the commutator term _ k(h_v p_k(f)-_k h_v p_k(f ) ) , which will be presented in section 5 . to achieve this , we will view lipschitz curves as perturbations of straight lines and use jones beta number condition for lipschitz curves and the carleson embedding theorem to control the commutator . here we shall emphasis again that the commutator estimate is free of time - frequency analysis . + * notations : * throughout this paper , we will write @xmath57 to mean that @xmath58 , @xmath59 to mean that there exists a universal constant @xmath28 s.t . @xmath60 , and @xmath61 to mean that @xmath59 and @xmath62 . @xmath63 will always denote the characteristic function of the set @xmath64 . + * acknowledgements . * this work is done under the supervision of prof . christoph thiele , to whom the author would like to express his most sincere gratitude , for suggesting him such an interesting topic , for sharing with him his deep insight into this problem , and for numerous suggestions on the exposition of this paper . in this section we prove corollary [ corollary1.2 ] , by reducing it to the main theorem . some parts of the reduction will also be used in the proof of the main theorem in the rest of the paper . + we first divide the unit circle @xmath65 into @xmath66 arcs of equal length , with the angle of each arc being @xmath67 . choose n > 6/d_0 s.t . @xmath68 . denote these arcs as @xmath69 . for each @xmath70 , define @xmath71 if we were able to prove that @xmath72 is bounded by a constant @xmath28 which is independent of @xmath73 , then we conclude that h_v_0_22c n(d_0 ) . now fix one @xmath70 , we want to show the boundedness of @xmath74 . choose a new coordinate such that the @xmath75-axis passes through @xmath70 and bisects it . then all the vectors in @xmath70 form an angle less than @xmath76 with the @xmath75-axis . as we assume that ( _ 2 g_0 , v_0(g_0))d_0>0 , we see that the vector @xmath77 forms an angle less than @xmath78 with the @xmath10-axis . renormalize the unit vector @xmath38 such that the first component is 1 , i.e. write @xmath79 , then by the fact that @xmath38 forms an angle less than @xmath76 with the @xmath75-axis , we obtain |u_0|(d_0/6 ) . next we construct the lipschitz function @xmath80 in the main theorem from the bi - lipschitz map @xmath36 , and the coordinate we will use here is still the one associated to @xmath70 as above . under this linear change of variables , we know that @xmath36 is still bi - lipschitz . we renormalize the bi - lipschitz map in such a way that g_0(x,0)=(x,0 ) , x. fix @xmath81 , the map @xmath36 , when restricted on the vertical line @xmath82 , is still bi - lipschitz . we denote by @xmath83 the image of this bi - lipschitz map , i.e. _ x:=\{g_0(x , y ) : y}. define the function @xmath80 by the following relation ( g(x , y),y)=g_0(x , y ) , for some @xmath84 . by the fact that @xmath36 is bi - lipschitz , we know that such @xmath84 exists and is unique . + from the above construction and the fact that @xmath77 forms an angle less than @xmath78 with the @xmath10-axis , we see easily that [ ee2.9 ] |g(x , y_1)-g(x , y_2)|(d_0/2 ) |y_1-y_2| , x , y_1 , y_2 . hence what is left is to show that condition is also satisfied with a constant @xmath24 depending only on @xmath35 and the bi - lipschitz constant of @xmath36 . one side of the equivalence @xmath85 is quite clear from the picture below : the bi - lipschitz map @xmath36 sends the points @xmath86 to @xmath87 separately , then by definition of bi - lipschitz map , there exists constant @xmath24 s.t . g(x_1 , y)-g(x_2 , y)|p - q|(x_1-x_2 ) . for the other side , we argue by contradiction . if for any @xmath88 large , there exists @xmath89 s.t . g(x_1 , y)-g(x_2 , y)m(x_1-x_2 ) , then together with , this implies that dist(k , _ x_1)m(d_0/2)(x_1-x_2 ) . but this is not allowed as by the definition of the bi - lipschitz map @xmath36 and the lipschitz function @xmath80 , @xmath90 must be comparable to @xmath91 . so far , we have verified all the conditions in the main theorem with b_0=(d_0/2 ) c_0=(d_0/6)/(d_0/2)<1 . hence we can apply the main theorem to obtain the boundedness of @xmath92 . + in the end , as claimed in the corollary , we still need to show the blowing up of the operator norm without the assumption that @xmath33 . indeed , the blowing up happens in @xmath5 not only for @xmath9 but also for all @xmath42 . for the range @xmath93 , the counter example is simply a knapp example : take the function @xmath94 , let @xmath95 be the upper cone which forms an angle less than @xmath96 with the vertical axis . first define the vector field @xmath97 for @xmath98 , then extend the definition to the whole plane properly such that @xmath3 satisfies the condition . it is then easy to see that |h_v f(x)|~ , xb_1(0 ) , which does not belong to @xmath99 for @xmath100 for the range @xmath101 , the counter example is given by the standard besicovitch - kakeya set construction , which can be found in @xcite and @xcite . if we linearize the maximal operator in the main theorem , what we need to prove turns to be the following [ mm3.1 ] _ f(g(x , y)-t , y - t u(x))dt / t_2f_2 , where @xmath18 is a measurable function such that @xmath102 . the change of coordinates ( x , y)(g(x , y ) , y ) in also changes the measure on the plane . however , we still want to use the original lebesgue measure for the littlewood - paley decomposition . hence we invert and denote the inversion by [ mm3.3 ] ( x , y)(p(x , y ) , y ) , where `` p '' stands for `` projection '' , and the reason of calling so can be illustrated by the following picture : the change of coordinates in turns the estimate into the following equivalent form _ f(x - t , y - t u(p(x , y)))dt / t_2 f_2 . moreover , we will denote [ mm3.5 ] h_v f(x , y):=_f(x - t , y - t u(p(x , y)))dt / t . in the rest of the paper , we want to make the convention that whenever @xmath4 appears , it denotes the hilbert transform along the vector field @xmath103 , which is the above , just to distinct it from the various @xmath4 that have appeared in the introduction . + to prove the above estimate , we first make several reductions : by the unisotropic scaling xx , yy , we can w.l.o.g . assume that @xmath104 . by a similar cutting and pasting argument to that in the proof of corollary [ corollary1.2 ] , we can assume that @xmath105 , i.e. the vector field @xmath3 is of the form @xmath106 with @xmath107 . + now we start the detail of the proof . it was already observed in bateman @xmath108 that under the assumption @xmath107 , we can w.l.o.g . assume that supp @xmath109 lies in a two - ended cone which forms an angle less than @xmath110 with the vertical axis , as for functions @xmath2 with frequency supported on the cone near the horizontal axis , we have that h_vf(x , y)=h_(1 , 0)f(x , y ) , which is the hilbert transform along the constant vector field @xmath111 . but @xmath112 is bounded by fubini s theorem and the @xmath0 boundedness of the hilbert transform . for the frequencies outside the cone near the horizontal axis , the proof consists of two steps . in the first step we will prove the boundedness of @xmath4 when acting on functions with frequency supported in one single annulus . to be precise , let @xmath95 be the cone which forms an angle less than @xmath110 with the vertical axis , @xmath113 be the projection operator on @xmath95 , i.e. _ f:=^-1 _ f , where @xmath114 stands for the fourier transform and @xmath115 the inverse transform . let @xmath54 be the @xmath116-th littlewood - paley projection operator in the vertical direction , as we are always concerned with the frequency in @xmath95 , later for simplicity we will just write @xmath54 instead of @xmath117 for short . then what we will prove first is [ thm1 ] under the same assumptions as in the main theorem , we have for @xmath118 that h_v p_k ( f)_pp_k(f)_p , with the constant being independent of @xmath119 . in order to prove the boundedness of @xmath4 , we need to put all the frequency pieces together . in the case of @xmath120 vector fields for any @xmath121 , lacey and li s idea in @xcite is to prove the almost orthogonality between different frequency annuli . in the case where the vector field is constant along vertical lines , an important observation in the paper of bateman and thiele is that @xmath4 and @xmath54 commute , which then makes it possible to apply a littlewood - paley square function estimate . in our case bateman and thiele s observation is no longer true . we need to take into account that the vector field is constant along lipschitz curves , which gives rise to an adapted littlewood - paley projection operator ( the following definition [ definition1 ] ) . before defining this operator , we first need to make some preparation . fix one @xmath122 , take the curve @xmath123 which passes through @xmath124 , recall that @xmath123 is given by the set @xmath125 , where @xmath80 is the lipschitz function in the main theorem . by the definition of the operator @xmath4 we know that the vector field @xmath3 is equal to the constant vector @xmath126 along @xmath123 . change the coordinate s.t . the horizontal @xmath127-axis is parallel to @xmath126 . the following lemma says that in the new coordinate , the curve @xmath123 can still be realized as the graph of a lipschitz function . [ lem3.2 ] for any fixed @xmath122 , there exists a lipschitz function @xmath128 s.t . @xmath123 can be reparametrized as @xmath129 . moreover , we have that @xmath130 , where @xmath25 is the constant in the main theorem . * proof of lemma [ lem3.2 ] : * denote by @xmath131 the angle between the vector @xmath126 and the @xmath75-axis as in the picture below . + the new coordinate of the point @xmath132 will be given by [ e3.4 ] ( x^ , y^)= ( + g ( , ) , -g ( , ) ) . look at the identity for the second component y^= -g ( , ) , we want to solve @xmath133 by @xmath134 by using the implicit function theorem . as = - , by the fact that @xmath107 and @xmath135 , we obtain that , from which it is clear that the implicit function theorem is applicable . after solving @xmath133 by @xmath134 , we just need to substitute @xmath133 into the identity for the first component in , which is x^= + g ( , ) , to get an implicit expression of @xmath127 in terms of @xmath134 , which we will denote as @xmath128 . to estimate the lipschitz norm of the function @xmath136 , we just need to observe that when doing the above change of variables , we have rotated the axis by an angle @xmath131 which satisfies @xmath137 . together with the fact that @xmath138 , we can then derive that || , which finishes the proof of lemma [ lem3.2 ] . @xmath139 ( adapted littlewood - paley projection)[definition1 ] select a schwartz function @xmath140 with support on @xmath141\cup [ -\frac{5}{2},-\frac{1}{2}]$ ] such that _ k _ 0(2 ^ -kt)=1 , t0 . for @xmath142 , for every fixed @xmath122 , define the adapted ( one dimensional ) littlewood - paley projection on @xmath123 by _ k(f)(x^ , y^):=_f(g_(z ) , z ) _ k(y^-z)dz = p_k()(y^ ) , where @xmath143 denotes one point in @xmath123 , @xmath144 and we use @xmath145 to denote the function @xmath146 , and @xmath54 the one dimensional littlewood - paley projection operator . now it is instructive to regard the lipschitz curves as perturbation of the straight lines , or equivalently , to think that @xmath147 still has frequency supported near the @xmath116-th frequency band , which has already been used by lacey and li in their almost orthogonality estimate for @xmath120 vector fields in @xcite . we then subtract the term @xmath148 from @xmath149 , and estimate the commutator . to be precise , we first write [ e3.5 ] _ k h_v p_k(f)=_k(h_v p_k(f)-_k h_v p_k(f)+_k h_v p_k(f ) ) , then by the triangle inequality , we have _ k h_v p_k(f)_2_k(h_v p_k(f)-_k h_v p_k(f))_2+_k_k h_v p_k(f)_2 . we call the second term the main term , and the first term the commutator term . the @xmath0 boundedness of the main term will follow from orthogonality argument , which is the following adapted littlewood - paley theorem . [ lemma3 ] for @xmath150 , we have the following variants of the littlewood - paley estimates : [ ee3.9 ] ( _ k|_k(f)|^2)^1/2_p~f_p , [ e3.10 ] ( _ k|_k^*(f)|^2)^1/2_p~f_p , with constants depending only on @xmath24 . * proof of lemma [ lemma3 ] : * in from the introduction , we have already explained the following coarea formula : [ e3.11 ] _ ^2|f(x , y)|dxdy~ _ [ _ _ [ e3.9 ] ( _ k|_k(f)|^2)^1/2_p^p ~ _ _ _ ( _ k |_k ( f)|^2)^p/2ds _ d. for every fixed @xmath151 , by definition [ definition1 ] , the right hand side of turns to _ [ _ ( _ k |p_k(_)(y^)|^2)^p/2dy^]d , where @xmath152 . then the classical littlewood - paley theory applies and we can bound the last expression by _ f_l^p(_)^p d f_l^p^p . for the boundedness of the adjoint operator , it suffices to prove that [ e3.12 ] _ k _ k^*(f ) , f_kf_l^p ( _ k |f_k|^2)^1/2_l^p^. first by linearity and hlder s inequality , we derive _ k _ k^*(f ) , f_k = f , _ k _ k ( f_k)f_l^p _ k _ k ( f_k)_l^p^. applying the coarea formula , we obtain _ k _ k ( f_k)_l^p^~(_(__|_k_k(f_k)|^pds_)d)^1/p. by the definition [ definition1 ] , for every fixed @xmath122 , the inner integration in the last expression turns to [ e3.25 ] _ |_kp_k(_k , ) ( y^)|^pdy^ , where @xmath153 . now the classical littlewood - paley theory applies and we bound the term in by [ e3.26 ] _ ( _ k|_k , ( y^)|^2)^p/2dy^__(_k|f_k|^2)^p/2ds_(_k|f_k|^2)^1/2_l^p(_)^p. then to prove , we just need to integrate @xmath154 in and apply the coarea formula to derive @xmath155 thus we have finished the proof of lemma [ lemma3].@xmath139 + now we will show how to prove the @xmath0 boundedness of the main term by lemma [ lemma3 ] and proposition [ thm1 ] : first by duality , we have @xmath156 applying the cauchy - schwartz inequality and hlder s inequality , we can bound the last term by _ g_2=1 ( _ k|h_v p_k(f)|^2)^1/2_2 ( _ k @xmath157 for the latter term , lemma [ lemma3 ] implies that ( _ k |_k^*(g)|^2)^1/2_2 g_2 . thus we have proved the @xmath0 boundedness the main term , modulo proposition [ thm1 ] . + as the second step , we will prove the @xmath0 boundedness of the commutator , which is _ k(h_v p_k(f)-_k h_v p_k(f))_2 f_2 . to do this , we first split the operator @xmath4 into a dyadic sum : select a schwartz function @xmath140 such that @xmath140 is supported on @xmath141 $ ] , let _ l(t):=_0(2 ^ -l t ) , by choosing @xmath140 properly , we can construct a partition of unity for @xmath158 , i.e. _ ( 0,)=_l _ l. let [ e4.24 ] h_lh(x , y):= _ l(t)h(x - t , y - tu(p(x , y)))dt , then the operator @xmath4 can be decomposed into the sum h_v=-+ 2 _ l h_l . hence to bound the commutator , it is equivalent to bound the following _ k _ l(h_l p_k f- _ k h_l p_k f ) . notice that by definition , @xmath159 vanishes for @xmath160 , which simplifies the last expression to _ l0 _ k ( h_k - l p_k f- _ k h_k - l p_k f ) . by the triangle inequality , it suffices to prove [ thm2 ] under the same assumption as in the main theorem , there exists @xmath161 such that _ k ( h_k - l p_k f- _ k h_k - l p_k f)_22 ^ -lf_2 , with the constant independent of @xmath162 so far , we have reduced the proof of the main theorem to that of proposition [ thm1 ] and proposition [ thm2 ] , which we will present separately in the following sections . lacey and li in their prominent work @xcite have reduced the @xmath0 boundedness of the operator @xmath163 to the boundedness of one new operator they introduced , which is the so called lipschitz - kakeya maximal operator . as soon as this operator is bounded , we can then repeat the argument in @xcite to obtain proposition [ thm1 ] as a corollary . here we follow @xcite , where a slightly different version of the lipschitz - kakeya maximal operator is used , see the following lemma [ lem4.8 ] . the only place in @xcite where the one - variable vector field plays a special role is lemma 6.2 in page 1037 . hence to prove proposition [ thm1 ] , we just need to replace this lemma by lemma [ lem4.8 ] , and leave the rest of the argument unchanged . + in this section we make an observation that both the boundedness of the lipschitz - kakeya maximal operator ( corollary [ lemma1 ] ) and its variant ( lemma [ lem4.8 ] ) can be proved by adapting bateman s argument in @xcite to our case where the vector fields are constant only on lipschitz curves . before defining the lipschitz - kakeya maximal operator , we first need to introduce several definitions . ( popularity ) for a rectangle @xmath164 , with @xmath165 its length , @xmath166 its width , we define its uncertainty interval @xmath167 to be the interval of width @xmath168 and centered at slope(@xmath169 ) . then the popularity of the rectangle @xmath169 is defined to be pop_r:=|\{(x , y)^2 : u(p(x , y))ex(r)}|/|r| . given two rectangles @xmath170 and @xmath171 in @xmath172 , we write @xmath173 whenever @xmath174 and @xmath175 , where @xmath28 is some properly chosed large constant , and @xmath176 is the rectangle with the same center as @xmath171 but dilated by the factor @xmath28 . denote @xmath177 , pop_r\ge \delta , w(r)=\omega\}$ ] , where @xmath178 is the collection of all the rectangles in @xmath172 . then the lipschitz - kakeya maximal function is defined as [ maxoperator ] m _ _ , ( f)(x):=_xr _ , _ r |f| [ lem4.8 ] let @xmath14 and @xmath179 be the functions given in the definition of the operator @xmath4 in . suppose @xmath180 is a collection of pairwise incomparable ( under `` @xmath181 '' ) rectangles of uniform width such that for each @xmath182 , we have , ( pop_r ) and _ r _ f . then for each @xmath42 , _ r_0|r| . the same covering lemma argument as in lemma 3.1 @xcite shows the boundedness of lacey and li s lipschitz - kakeya maximal operator as a corollary of lemma [ lem4.8 ] . [ lemma1 ] for all @xmath183 we have the following bound m _ _ , _ l^pl^pc(p , a_0 ) * proof of lemma [ lem4.8 ] : * the proof is essentially due to bateman @xcite . most of the argument in @xcite remains , with just one minor modification in order to adapt to the family of lipschitz curves on with the vector field is constant . [ definition4](rectangles adapted to the vector field ) for a rectangle @xmath184 , with its two long sides lying on the parallel lines @xmath185 and @xmath186 for some @xmath187 $ ] and @xmath188 , define @xmath189 to be the adapted version of @xmath169 , which is given by the set \{(x , y ) : p(x , y)p(r)}\{(x , kx+b):x , b } , where @xmath179 is the projection operator in . what we need to do is just to replace the rectangles @xmath169 in @xcite by @xmath189 , and observe that the two key quantities length and popularity of rectangles are both preserved under the projection operator @xmath179 up to a constant depending on the constant @xmath24 in the main theorem . hence we leave out the details and refer to @xcite.@xmath139 this section consists of three subsections . in the first subsection we will introduce some notations , most of which we adopt from bateman s paper @xcite , with minor changes for our purpose . in the second we will use jones beta numbers and the carleson embedding theorem to prove proposition [ thm2 ] , modulo one crucial lemma which will be presented afterwards in the third subsection . the content of this subsection is basically taken from bateman s paper @xcite , with minor changes as we are now dealing with all frequencies instead of one single frequency annulus . + * discretizing the functions : * fix @xmath190 , we write @xmath191 as the collection of the dyadic intervals of length @xmath192 contained in @xmath193 $ ] . fix a smooth positive function @xmath194 s.t . ( x)=1 , |x|1 ; ( x)=0 , |x|2 . also choose @xmath195 such that @xmath196 is a smooth function . then fix an integer @xmath197(whose exact value is unimportant ) , for each @xmath198 , define _ ( x)= ( 2^l+c(x - c__1 ) ) , where @xmath199 is the right half of @xmath200 and @xmath201 is its center . define _ l(x)=__l _ ( x ) , note that _ l(x+2 ^ -l)=_l(x ) , x. define _ l=_-1 ^ 1 _ l(x+t)dt , because of the above periodicity , we know that @xmath202 is constant for @xmath203 $ ] , independent of @xmath204 . say @xmath205 , hence _ l(x ) _ [ -1,1](x)=_[-1,1](x ) . define another multiplier @xmath206 with support in @xmath207 $ ] and @xmath208 for @xmath209 $ ] . we define the corresponding multiplier on @xmath172 : @xmath210 then what we need to bound can be written as @xmath211 where the terms @xmath212 for @xmath160 in the sum vanish as explained before . * constructing the tiles : * for each @xmath119 and @xmath198 with @xmath190 , let @xmath215 be a partition of @xmath172 by rectangles of width @xmath216 and length @xmath217 , whose long side has slope @xmath131 , where @xmath218 , which is the center of the interval @xmath200 . if @xmath219 , we will write @xmath220 , and @xmath221 to be the right half of @xmath200 , @xmath222 the left half . this subsection is devoted to the proof of proposition [ thm2 ] , which is motivated in large by the proof of the @xmath234 theorem and the boundedness of the paraproduct , see @xcite and @xcite for example . in our case , unlike bateman and thiele s proof for the one - variable vector fields , it s no longer true that @xmath147 still has frequency in the @xmath116-th annulus . in order to get enough orthogonality for the term @xmath147 to apply the littlewood - paley theory , we need to subtract the term @xmath235 , which should be viewed as a family of paraproducts . + we proceed with the details of the proof . if we expand the summation on the left hand side of proposition [ thm2 ] with , what we need to bound can be rewritten as _ k _ _ l_s_k , f , _ s(_s-_k _ s)_22 ^ -lf_2 . in order to use the orthogonality of different wave packets , we will prove the @xmath0 bound for the dual operator , which is _ k _ _ l_s_k , h , _ s-_k _ notice that for @xmath236 and @xmath237 with @xmath238 , we have [ orthogonality ] _ s_1 , _ s_2=0 by the definition of the wavelet function @xmath227 in . also if we know that @xmath239 are in the same @xmath215 , for some @xmath116 and @xmath200 , then we can find @xmath240 s.t . c(s_2)=c(s_1)+(ml(s_1 ) , nw(s_1 ) ) where @xmath225 is the center of the tile @xmath226 , @xmath241 its length and @xmath242 its width . then by the non - stationary phase method we know for any @xmath243 , there exists a constant @xmath244 depending only on @xmath66 s.t . [ almostorthogonality ] |_s_1 , _ s_2| . here we want to make a remark that the exact value of @xmath66 is not important , it just denotes some large number which might vary from line to line if we use the same notation later . + applying the above two estimates , we obtain @xmath245 as we know for any @xmath246 there exists @xmath240 s.t . [ e1.66 ] c(s_2)=c(s_1)+(ml(s_1 ) , nw(s_1 ) ) , the above sum can be rewritten as _ m , n _ k__l_s_1_k , h , _ s_1-_k _ s_1_s_1,_s_2h , _ s_2-_k _ s_2with @xmath247 satisfying the relation . + now fix @xmath240 , by the estimate in , we know that @xmath248 by the cauchy - schwartz inequality , the last term is bounded by _ k__l_s_k,|h , _ s|^2 , then it suffices to prove that [ e5.23 ] _ k _ _ l_s_k , h , _ s-_k _ s^2 2 ^ -lh_2 ^ 2 . first to estimate every single term @xmath249 for a fixed tile @xmath226 : denote @xmath250 to be the shift of @xmath226 by @xmath251 units , i.e. s_m , n:=\{(x , y)^2 : ( x - ml(s ) , y - nw(s))s } , then by the triangle inequality we know that |h , _ s-_k _ s|_m , n|_s_m , n h(_s-_k _ s)dydx| . recall that in definition [ definition4 ] we use @xmath189 to denote the adapted version of the rectangle @xmath169 to the family of lipschitz curves , then clearly @xmath252 . thus now for the inner integration along the curve @xmath83 , we do the same change of coordinates and the same parametrization of @xmath83 as in definition [ definition1 ] , i.e. we choose the coordinates s.t . the horizontal axis is parallel to @xmath255 , and represent the curve @xmath83 by the lipschitz function @xmath256 . if we let @xmath257 denote the projection of @xmath258 on the new vertical axis , the last expression becomes [ ee5.27 ] _ m , n _ p(s_m , n)_j(x , s_m , n)|h(g_x(y ) , y ) ( _ s(g_x(y ) , y ) - p_k [ _ s(g_x(y ) , y)])|dy dx . ( @xcite ) for a lipschitz function @xmath259 , we first take the caldern decomposition of @xmath260 , which yields the representation a(x)=_i dyadic a_i _ i(x ) , where @xmath261 is some mean zero function supported on @xmath262 , @xmath263 for each dyadic interval @xmath264 , let _ i=_|j||i| a_i _ j(c_i ) , where @xmath265 stands for the center of @xmath264 , denote the `` average slope '' of the lipschitz curve near @xmath264 , and define the beta number @xmath266 and the @xmath267-th beta number @xmath268 [ claim1 ] for @xmath271 , we have the following estimate : @xmath272)|dy\\ & \lesim \sum_{j_0\in \n}\frac{2^{-3l/2}}{<|j_0|+|m|+|n|>^n } \beta_{j_0}(x , s_{m , n } ) [ h]_{x , s_{m , n } } \mathbbm{1}_{\{-u(x)\in \omega_{s,2}\}}(x)\end{aligned}\ ] ] where @xmath273 is the @xmath267-th beta number for the lipschitz curve @xmath256 on the interval @xmath257 , @xmath274_{x , s_{m , n}}$ ] is the average of the function @xmath275 on the interval @xmath257 , i.e. _ x , s_m , n:=_j(x , s_m , n)|h(g_x(y ) , y)|dy . the proof of lemma [ claim1 ] will be postponed to the next subsection . substitute the estimate in lemma [ claim1 ] into the estimate for the term @xmath249 , we then have that @xmath276)|dy dx\\ & \lesim \sum_{m , n}\int_{p(s_{m , n})}\sum_{j_0\in \n}\frac{2^{-3l/2}}{<|j_0|+|m|+|n|>^n } \beta_{j_0}(x , s_{m , n } ) [ h]_{x , s_{m , n}}\mathbbm{1}_{\{-u(x)\in \omega_{s,2}\}}(x)dx\end{aligned}\ ] ] hence @xmath277_{x , s_{m , n}}\mathbbm{1}_{\{-u(x)\in \omega_{s,2}\}}(x)dx|^2\\ & \lesim \sum_{m , n , j_0}\frac{2^{-2l}}{<|j_0|+|m|+|n|>^n } ... \\ & \text { } \text { } \text { } \text { } \text { } \text { } ... \sum_k \sum_{\omega\in \mathcal{d}_{l}}\sum_{s\in \mathcal{u}_{k,\omega } } w(s ) \int_{p(s_{m , n } ) } \beta_{j_0}^2(x , s_{m , n } ) [ h]^2_{x , s_{m , n}}\mathbbm{1}_{\{-u(x)\in \omega_{s,2}\}}(x)dx\end{aligned}\ ] ] * proof of lemma [ lemma7 ] : * this lemma is akin to the carleson embedding theorem , as we have the following carleson type condition _ s_m , n_s^_m , n : j(x , s^_m , n)j(x , s_m , n ) _ j_0 ^ 2(j(x , s^_m , n ) ) w(s^_m , n)j_0 ^ 3 lip^2(_x ) , where the term @xmath279 plays such a role that , originally there are @xmath280 groups of dyadic rectangles _ k _ _ l_s_k , \{s_m , n } in the summation @xmath281 , which means that there are also @xmath280 groups of dyadic intervals _ k _ _ l_s_k , \{j(x , s_m , n ) } which are the projections of the intersection of the dyadic rectangles with @xmath83 on the vertical axis , the term @xmath279 just guarantees that there is just one such collection which has contribution , i.e. which has the right orientation in the sense of lemma [ lemma5 ] . we assume that @xmath283 , which means the vector @xmath255 is roughly parallel to the long side of @xmath250 , otherwise the left hand side in lemma [ claim1 ] will also vanish due to lemma [ lemma5 ] . after the change of variables in , the vector @xmath255 turns to @xmath111 . + * proof by ignoring the tails : * in order to explain how jones @xmath195-number appears , we first sketch the proof by ignoring the tails of the wavelet functions and the tail of the kernel of the littlewood - paley projection operator @xmath54 . s ) ) [ h]_x , s. for fixed @xmath75 , we denote by @xmath285 the line of `` average slope '' we picked in the definition of the beta number for the lipschitz curve @xmath256 on the interval @xmath286 , for the sake of simplicity we assume @xmath287 . moreover , as both @xmath75 and @xmath226 are fixed , we will also just write @xmath288 instead of @xmath289 . then we make the crucial observation that [ aa5.38 ] p_k[_s^x(y , y)]=_s^x(y , y ) , where _ s^x(y , y):=__s(t)_s(y - t , y)dt , due to the fact that for any function @xmath227 with frequency supported on the @xmath116-th annulus , if we restrict the function to a straight line , it will still have frequency supported on the @xmath116-th annulus ( with one dimension less ) . in comparison with the definition of @xmath290 in , @xmath291 is defined as the hilbert transform along the vector @xmath255 ( which is ( 1 , 0 ) after the change of the variables we made in lemma [ lem3.2 ] and in the expression ) instead of the direction of the vector field @xmath3 at the point @xmath292 . as we have also ignored the tails of the kernel of @xmath54 , it is easy to see that the former and the latter terms in the last expression can essentially be handled in the same way . hence in the following we will only consider the former term , which corresponds to the term [ aa5.41 ] & |_s(g_x(y),y)-_s^x(y , y)| + & = | _ _ k - l(t)_s(g_x(y)-t , y)dt- _ _ k - l(t)_s(y - t , y)dt| + & = 2^k - l |__0(2^k - lt)_s(g_x(y)-t , z)dt- _ _ 0(2^k - lt)_s(y - t , y)dt| + & = 2^k - l |_[_0(2^k - l(t+g_x(y)-y))-_0(2^k - lt)]_s(y - t , z)dt| . by the definition of the beta numbers , we have that |g_x(y)-y|_0(x , s ) 2 ^ -k , which implies that |_0(2^k - l(t+g_x(y)-y))-_0(2^k - lt)|2 ^ -l _ 0(x , s ) by the fundamental theorem . in the end , by substituting the above estimate into and we obtain the desired estimate . the full proof : * the main idea is still the same , and the difference is that we need to be more careful with the tails of the wavelet functions and the kernel of @xmath54 . for fixed @xmath75 , fixed @xmath294 and @xmath295 , denote @xmath296 as the line of `` average slope '' for the lipschitz curve @xmath256 on the interval @xmath257 , for the sake of simplicity we assume @xmath287 . then the crucial observation becomes [ ee5.37 ] p_k[_s^x((x , s_m , n)y , y)]=_s^x((x , s_m , n)y , y ) . hence similar to , we obtain from that @xmath297\\ & = \phi_s(g_x(y),y)-\phi_s^x(\tau(x , s_{m , n})y , y)-p_k[\phi_s(g_x(y),y)-\phi_s^x(\tau(x , s_{m , n})y , y)].\end{aligned}\ ] ] denote i_s_m , n=|_j(x , s_m , n)h(g_x(y),y ) ( _ s(g_x(y),y)-_s^x((x , s_m , n)y , y))dy| and also ii_s_m , n= |_j(x , s_m , n)h(g_x(y),y ) p_k[_s(g_x(y),y)-_s^x((x , s_m , n)y , y)]dy| . let us first continue the proof of lemma [ claim1 ] : for the first term @xmath300 , we take @xmath267 in lemma [ lemma9 ] to be zero , then |_s(g_x(z ) , z)-_s^x((x , s_m , n)z , z)| , which implies that i_s_m , n _ 0(x , s_m , n ) [ h]_x , s_m , n . for the second term @xmath301 , by the definition of @xmath54 , @xmath302|\\ & = |\int_{\r } ( \phi_s(g_x(z),z)-\phi_s^x(\tau(x , s_{m , n})z , z))2^k \check{\psi}_0(2^k(y - z))dz|\\ & \le |\sum_{j_0\in \z } \int_{j(x , s_{m , n})+j_0 2^{-k } } ( \phi_s(g_x(z),z)-\phi_s^x(\tau(x , s_{m , n})z , z))2^k \check{\psi}_0(2^k(y - z))dz|.\end{aligned}\ ] ] for @xmath303 and @xmath304 , by the non - stationary phase method , we have that |_0(2^k(y - z))| , together with the estimate in lemma [ lemma9 ] , we arrive at @xmath302|\\ & \lesim \sum_{j_0\in \z } \frac{\beta_{|j_0|}(x , s_{m , n } ) 2^{k } 2^{-3l/2}}{<\min\{|m|+|n| , |m|+|n|-|j_0|\}>^n } \frac{1}{<j_0>^n}\\ & \lesim \sum_{j_0\in \z } \frac{\beta_{|j_0|}(x , s_{m , n } ) 2^{k } 2^{-3l/2}}{<|m|+|n|+|j_0|>^n}.\end{aligned}\ ] ] substitute the last expression into the estimate for @xmath301 , we get the desired estimate . so far we have finished the proof of lemma [ claim1 ] except the lemma [ lemma9 ] , which we will do now . + * proof of lemma [ lemma9 ] : * as @xmath75 and @xmath250 are fixed now , later for simplicity we will just write @xmath288 instead of @xmath305 . notice that in the new coordinate we chose for @xmath83 , the vector field along @xmath83 points in the direction of @xmath111 . then by the definition of @xmath290 and @xmath293 , we have @xmath306\varphi_{s}(\tau z - t , z)dt|.\end{aligned}\ ] ] by the definition of the beta numbers , we have that |g_x(z)-z|_|j_0|(x , s_m , n ) 2 ^ -k , which implies that |_0(2^k - l(t+g_x(z)-z))-_0(2^k - lt)|2 ^ -l _ |j_0|(x , s_m , n ) by the fundamental theorem . in the end , non - stationary phase method leads to the final estimate : @xmath307\varphi_{s}(\tau z - t , z)dt|\\ & \lesim \frac{2^{-l } \beta_{|j_0|}(x , s_{m , n } ) 2^{\frac{k}{2 } } 2^{\frac{k - l}{2}}}{<\min\{|m|+|n| , |m|+|n|-|j_0|\}>^n } .\end{aligned}\ ] ] thus we have finished the proof of lemma [ lemma9 ] and hence lemma [ claim1 ] . 20 p. auscher , s. hofmann , c. muscalu , t. tao and c. thiele : carleson measures , trees , extrapolation , and @xmath234 theorems . 46 ( 2002 ) , no . 2 , 257 - 325 . j. azzam and r. schul : hard sard : quantitative implicit function and extension theorems for lipschitz maps . geom . funct . 22 ( 2012 ) , no . 5 , 1062 - 1123 . m. bateman : @xmath5 estimates for maximal averages along one - variable vector fields in @xmath172 . ( 2009 ) , no . 3 , 955 - 963 . m. bateman : single annulus @xmath308 estimates for hilbert transforms along vector fields . rev . mat . iberoam . 29 ( 2013 ) , no . 3 , 1021 - 1069 . m. bateman : kakeya sets and directional maximal operators in the plane . duke math . j. 147 ( 2009 ) , no . 1 , 55 - 77 . m. bateman and c. thiele : @xmath309 estimates for the hilbert transforms along a one - variable vector field . to appear in anal . pde , arxiv : 1109.6396 ( 2011 ) . j. bourgain : a remark on the maximal function associated to an analytic vector field . analysis at urbana . i ( urbana , il , 1986 - 1987 ) , 111 - 132 . m , christ , a. nagel , e. stein and s. wainger : singular and maximal radon transforms : analysis and geometry . ann . of math . ( 2 ) 150 ( 1999 ) , no . 2 , 489 - 577 . r. coifman , p. jones and s. semmes : two elementary proofs of the @xmath0 boundedness of cauchy integrals on lipschitz curves . soc . 2 ( 1989 ) , no . 3 , 553 - 564 . a. cordoba : the kakeya maximal function and the spherical summation multipliers . 99 ( 1977 ) , no . 1 , 1 - 22 . jones : square functions , cauchy integrals , analytic capacity , and harmonic measure . harmonic analysis and partial differential equations ( el escorial , 1987 ) , 24 - 68 , lecture notes in math . , 1384 , springer , berlin , 1989 . g. karagulyan : on unboundedness of maximal operators for directional hilbert transforms . ( 2007 ) , no . 10 , 3133 - 3141 ( electronic ) . m. lacey : the bilinear maximal functions map into @xmath5 for @xmath310 . ann . of math . ( 2 ) 151 ( 2000 ) , no . 1 , 35 - 57 . m. lacey and x. li : maximal theorems for the directional hilbert transform on the plane 358 ( 2006 ) , no . 9 , 4099 - 4117 . m. lacey and x. li : on a conjecture of e. m. stein on the hilbert transform on vector fields . ( 2010 ) , no . 965 , viii+72 pp . isbn : 978 - 0 - 8218 - 4540 - 0 m. lacey and c. thiele : @xmath5 estimates for the bilinear hilbert transform . 94 ( 1997 ) , no . 1 , 33 - 35 . m. lacey and c. thiele : on caldern s conjecture for the bilinear hilbert transform usa 95 ( 1998 ) , no . 9 , 4828 - 4830 ( electronic ) . o. perron : ber einen satz von besicovitch . z. 28 ( 1928 ) , no . 1 , 383 - 386 . e. m. stein and b. street : multi - parameter singular radon transforms iii : real analytic surfaces . ( 2012 ) , no . 4 , 2210 - 2238 .

as a promising candidate system to realize topological superconductivity , the system of a 3d topological insulator ( ti ) grown on top of the-wave superconductor has been extensively studied . to access the topological superconductivity experimentally , the 3d ti sample must be thin enough to allow for cooper pair tunneling to the exposed surface of ti . the use of magnetically ordered dopants to break time - reversal symmetry may allow the surface of a ti to host majorana fermion , which are believed to be a signature of topological superconductivity . in this work , we study a magnetically - doped thin film ti - superconductor hybrid systems . considering the proximity induced order parameter in thin film of ti , we analyze the gap closing points of the hamiltonian and draw the phase diagram as a function of relevant parameters : the hybridization gap , zeeman energy , and chemical potential of the ti system . our findings provide a useful guide in choosing relevant parameters to facilitate the observation of topological superconductivity in thin film ti - superconductor hybrid systems . in addition , we further perform numerical analysis on a ti proximity coupled to a-wave superconductor and find that , due to the spin - momentum locked nature of the surface states in ti , the induced-wave order parameter of the surface states persists even at large magnitude of the zeeman energy .

1609.04129

a particularly interesting direction within the landscape of topological systems is to realize topological superconductivity . within condensed matter physics there has been a significant effort to find systems that exhibit topological superconductivity as such systems are predicted to harbor the , heretofore , elusive majorana fermions@xcite . there have been a number of proposals that have been predicted to realize topological superconductivity , and these proposals may be grouped into two : ( i ) unconventional superconducting materials such as sr@xmath1ruo@xmath2@xcite or doped superconducting materials such as cu@xmath3bi@xmath4se@xmath5@xcite and ( ii ) proximity - coupled system comprised of a conventional superconductor and a system such as strongly spin - orbit coupled semiconductors@xcite , magnetic adatoms@xcite , or 3d time - reversal invariant ( tri ) topological insulators ( ti)@xcite . while much work has taken place on both groups of proposals , there have been few unambiguous signs of topological superconductivity observed experimentally . in this endeavor , the most promising experimental signatures have come from the second class of proposals , in particular the spin - orbit coupled semiconductors@xcite and magnetic adatoms proximity - coupled with @xmath0-wave superconductors@xcite . nonetheless , it is clear that within each of the proposals to observe topological superconductivity , there is a trend in the components required to produce the unconventional superconductivity : non - zero berry curvature induced by spin - orbit coupling and broken time reversal symmetry by magnetism . of the available platforms within which one may combine these ingredients , one of the well known routes to generate topological superconductivity is via the superconducting proximity effect in a heterostructure sample of a conventional @xmath0-wave superconductor and 3d tri ti@xcite . in the pioneering work of fu and kane , cooper pairs from @xmath0-wave superconductors that are proximity - coupled to 3d tri ti tunnel from the superconductor into the ti resulting in the acquisition of an topological superconductivity that behaves an effective spinless , chiral @xmath6 superconductor without breaking time - reversal symmetry . as compared to proposals using non - ti heterostructures@xcite or intrinsic superconductors@xcite , the fu - kane proposal is attractive as it does not require further assumptions on any of the physical parameters such as cooper pairing amplitudes between different orbitals@xcite or the position of the chemical potential@xcite . to facilitate the generation of chiral edge states , a zeeman field may be introduced to open a gap in the energy spectrum and thereby form a boundary at the surface of the ti@xcite . to this end , topological insulators with magnetic dopants that break time - reversal symmetry are of great interest@xcite as a platform to observe topological superconductivity and chiral edge states . in this work , we study how introducing magnetic dopants affects the proximity induced superconductivity of the 3d ti system . unlike a similar study that has been performed on magnetically - doped ti whose zeeman field is randomly oriented within the ti@xcite , we consider magnetically ordered dopants through the addition of a uniform zeeman splitting term in a thin 3d tri ti sample such as bi@xmath1se@xmath7 to form a magnetic domain via the net exchange field@xcite . as experimental ti samples must be thin for superconductivity to be observed on the surface , we focus on the `` ultrathin '' limit of the ti where the surface states are not well - isolated but hybridized resulting in a gapped surface state spectrum . . the induced order parameter at top ( bottom ) surface is indicated as @xmath8 ( @xmath9 ) . ( b ) after we apply the proper rotation to the system , we obtain two decoupled systems in hybridization basis of the top and bottom surfaces . two individual sectors are referred as to a sector 1 and sector 2 . ( c ) total induced superconducting order parameter is equivalent to a combination of ( spatially ) symmetric and anti - symmetric superconducting order parameter.,scaledwidth=50.0% ] we seek to understand the physics of magnetically - doped , ultrathin ti and its topological phase by analyzing the superconducting order parameter using both analytical and numerical techniques . in section [ sec : phenomenological ] , we introduce a 2d continuum model for the surface states of an ultrathin ti that accounts for the hybridization gap . by applying a series of unitary transformations , we note that the model can be separated into independent sectors , whose relevant pairing potential have a symmetric and anti - symmetric spatial form when superconductivity is added . we then analyze the symmetric and anti - symmetric @xmath0-wave pairing potential at the phenomenological level by assuming a constant induced order parameter , and find that anti - symmetric pairing is dominant for experimentally relevant strengths of the zeeman field . simplifying the hamiltonian with assumed anti - symmetric pairing potential , we find that gap closing points exist and are controlled by three parameters that can be tuned in experiment : the chemical potential , hybridization gap , and zeeman energy . as our system is in @xmath10 class within the altland - zirnbauer classification@xcite , it is characterized by a @xmath11 topological invariant@xcite . thus we analyze the gap closing points and corresponding topological phase by evaluating the chern number to obtain the resulting phase diagram . in section [ sec : results3d ] , we model a more realistic lattice system by self - consistently solving for the superconducting order parameter in a heterostructure of a @xmath0-wave superconductor coupled to a ti using the bogoliubov - de gennes ( bdg ) formalism . our numerical simulation accurately captures the effects of bulk and surface bands that are present in ti and includes the spin dynamics of these bands when magnetism is introduced . an induced superconducting order parameter obtained from bulk states of the ti shows a rapid decay in magnitude with increasing magnetic impurity concentration as the zeeman energy splits the band and suppresses the @xmath0-wave pairing . in contrast , the surface states show an induced order parameter that persists over an experimentally relevant range of the zeeman energy due to their spin - momentum locked nature and non - zero projection of electron pairs into the @xmath0-wave pairing component . moreover , self - consistent calculation shows that the anti - symmetric pairing potential is dominant at non - zero zeeman energy , thereby , we confirm our phenomenological analysis . lastly , in section [ sec : conclusion ] , we summarize our results and provide our concluding remarks . [ fig : system](a ) shows the system under consideration , in which a ti is placed on top of a conventional @xmath0-wave superconducting metal . we first consider the surface state hamiltonian of ti , @xmath12 , where @xmath13^t$ ] . we define @xmath14 ( @xmath15 ) for electron annihilation operator at top ( bottom ) surface with up ( down ) spin states at @xmath16 . then the hamiltonian @xmath17 is defined as @xmath18 where @xmath19 are the pauli matrices for the spin degree of freedom , @xmath20 is a @xmath21 identity matrix , and @xmath22 is the hybridization gap . we use @xmath23 and @xmath24 for the low energy description of top and bottom surface state hamiltonians , respectively@xcite , to form a hybridized hamiltonian that preserves time reversal symmetry . as the focus of this work is to understand the influence of magnetic dopants , we add the effect of a uniform perpendicular magnetization to the proximity - coupled ti surface via addition of the zeeman energy splitting term within the hamiltonian in eq . ( [ eq : hsurf_bare ] ) . the zeeman term whose principle axis is aligned in @xmath25 takes the form @xmath26 , where @xmath27 is the energetic splitting of spin states due to the magnetization arising from the dopants included within the ti film , and the identity matrix acts on top and bottom surface ( or pseudospin ) degree of freedom . previous work considering the effects of magnetic dopants on the surface physics of tri ti indicates that the addition of magnetically ordered impurities allows for the development of a net ferromagnetic order,@xcite and thus a uniform magnetization can provide a simple but accurate picture of the ti with magnetic dopants . within this work , we ignore any orbital effects as those resulting from magnetic dopants are negligible@xcite . including the zeeman term , we may write eq . ( [ eq : hsurf_bare ] ) in @xmath28 matrix form as @xmath29 eq . ( [ eq : hsurf ] ) is further simplified by applying proper rotational matrices . without loss of generality , we remove @xmath30 by applying an @xmath31 rotation , @xmath32 , where @xmath33 . in addition , we apply another @xmath31 rotation on the pseudo - spin degree of freedom ( top / bottom layer ) , @xmath34 , where @xmath35 is the pauli matrix for the pseudospin degree of freedom , and rewrite the hamiltonian in eq . ( [ eq : hsurf ] ) using a hybridized basis of top and bottom layer . the resultant hamiltonian is @xmath36 we may write eq . ( [ eq : hsurf2 ] ) in @xmath28 matrix form : @xmath37 where @xmath38 . in eq . ( [ eq : hsurf22 ] ) , we define effective zeeman energy @xmath39 and @xmath40 . note that the hamiltonian eq . ( [ eq : hsurf22 ] ) is now decoupled into two sectors : @xmath41 where @xmath42 is the pauli matrix in pseudospin space whose basis is now in linear combination of spin and layer degree of freedoms . for the following arguments , @xmath43 and @xmath44 refers to the hamiltonian in sector 1 and sector 2 , respectively , as shown in fig . [ fig : system ] ( b ) . the rotated basis for eq . ( [ eq : hsurf2 ] ) is @xmath45 where we omit in - plane momentum index @xmath46 for simplicity . eq . ( [ eq : u2c ] ) shows that the electron states at top and bottom surface form bonding and anti - bonding like hybridized states , whose annihilation operator is defined by @xmath47 . due to the opposite helicity of top and bottom surfaces , the hybridized basis satisfies following unitary transformation @xmath48 which exchanges the particles at top and bottom surfaces , namely , @xmath49 and @xmath50 . consequently , a band having a basis of @xmath51 has @xmath52 as another basis to satisfy eq . ( [ eq : helicity ] ) and this particular combination of basis forms the bands in sector 1 . the other set of basis states forms the bands in sector 2 . having two well separated sectors in our hamiltonian , we may analyze possible superconductor pairing order parameters . when a surface state is coupled to a superconducting system , one may observe an induced superconducting order parameter in the surface states of the ti as cooper pairs tunnel from the superconductor system@xcite . in our system , the induced order parameters in the top and bottom surface differ in their magnitude as the cooper pair tunneling probability decays as a function of a spatial separation from the interface between ti and @xmath0-wave superconductor to the other surface of the ti@xcite . due to this gradient in order parameter magnitude , we may decompose the induced order parameter into two distinct components , whose individual @xmath53 phases are symmetric and anti - symmetric in @xmath25 direction . [ fig : system](c ) shows the simplest example , e.g. @xmath54 where @xmath55 and @xmath56 . having the basis representation in eq . ( [ eq : u2c ] ) , we may examine the induced pair correlation function defined within each of the sectors or across different sectors . a pair correlation function of the intra - band @xmath0-wave pairing ( within sector 1 or sector 2 ) is , for example , @xmath57 where we suppress the in - plane momentum index @xmath46 for brevity . ( [ eq : intra ] ) shows an odd parity under the exchange of the layer degree of freedom , or , @xmath58 . by defining the on - site @xmath0-wave order parameter@xcite as @xmath59 , we find that the intra - band @xmath0-wave order parameter is anti - symmetric in the @xmath25 direction . meanwhile , a pair correlation function of the inter - band @xmath0-wave pairing ( between sector 1 and sector 2 ) is , for example , @xmath60 which shows an even parity under the exchange of the layer degree of freedom , or , @xmath61 and , thus , the resultant inter - band @xmath0-wave order parameter , @xmath62 , is symmetric in @xmath25 direction . therefore , eq . ( [ eq : intra ] ) shows that the anti - symmetric order parameter is responsible for superconducting pairing that occurs within the same sectors , whereas eq . ( [ eq : inter ] ) shows that the symmetric order parameter results in superconducting pairing across different sectors . the above argument is explicitly shown by constructing the bogoliubov - de gennes ( bdg ) hamiltonian . we add an induced order parameter to the system as a constant value , @xmath63 , at the phenomenological level@xcite . then the bdg hamiltonian is @xmath64 where the @xmath0-wave pairing hamiltonian is @xmath65 where @xmath66 and @xmath67 are the @xmath0-wave order parameter at top and bottom surface , respectively . we then decompose the order parameter in eq . ( [ eq : hpair ] ) into two components , namely , the symmetric and anti - symmetric components . fig . [ fig : system](c ) depicts symmetric and anti - symmetric components of superconducting order parameter written as @xmath68 respectively , where @xmath69 and @xmath70 are corresponding decomposed order parameters . by defining @xmath71 , we obtain a simplified hamiltonian in bdg form , @xmath72 where the symmetric and anti - symmetric component of pairing hamiltonian becomes @xmath73 and @xmath74 respectively . the pairing hamiltonian in eq . ( [ eq : sym ] ) explicitly shows that an electron ( hole ) band in sector 1 is coupled with the hole ( electron ) band in sector 2 , which we refer to as an inter - band pairing . likewise , eq . ( [ eq : asym ] ) shows that the anti - symmetric superconducting order parameter couples electron bands and hole bands in the same sector , which we refer to as intra - band pairing . although the bands are initially degenerate having contributions from both the symmetric and anti - symmetric superconducting pairing potential , introducing zeeman term splits the bands in energy and changes their relative contributions to the resultant superconductivity . to examine the relative contribution of the symmetric and anti - symmetric parts of the pairing order parameter , we calculate the ground state energy of the bdg hamiltonian in eq . ( [ eq : bdg ] ) . to utilize the insight on the order parameter analysis from section [ sec:2dop ] , we adopt a low energy description of the hamiltonian of the ultrathin ti surface states in eq . ( [ eq : hsurf ] ) and compute the correlation function , @xmath75 . for calculation details on the correlation function , see appendix [ app : op ] . then the @xmath0-wave order parameter is obtained from @xmath76,\ ] ] where @xmath77 is an on - site attractive interaction potential , and the summation is performed until a momentum cut - off @xmath78 . although the attractive interaction is not a prerequisite for the induced order parameter in the ultrathin ti , we use an phenomenological interaction potential , @xmath79 , in eq . ( [ eq : deltastb ] ) to obtain non - zero pairing hamiltonian to obtain self - consistently calculated ground state energy . once the order parameter in eq . ( [ eq : deltastb ] ) is determined , the pairing hamiltonian is constructed using eq . ( [ eq : hpair ] ) to obtain the bdg hamiltonian @xmath80 in eq . ( [ eq : bdg ] ) . the process continues until the calculation reaches self - consistency in the order parameter magnitude . then the total energy of the superconducting system is determined by@xcite @xmath81 where @xmath82 is energy of the system and the summation runs over all occupied states . the total energy of the normal system , @xmath83 , is obtained similarily by setting @xmath84 in eq . ( [ eq : deltastb ] ) and ignoring second term in eq . ( [ eq : es ] ) . to further examine the total energy of the system whose order parameter has a symmetric and anti - symmetric form , we define the symmetric and anti - symmetric order parameter , @xmath85 using eq . ( [ eq : deltasyasy ] ) , we decompose the pairing hamiltonian in eq . ( [ eq : hpair ] ) into symmetric and anti - symmetric components and construct corresponding pairing hamiltoinan using eq . ( [ eq : bcssy ] ) and ( [ eq : bcsasy ] ) , respectively . then we obtain the bdg hamiltonian , @xmath86 which is used to compute the order parameter self - consistently . we plot the total energy , @xmath87 and @xmath88 , in fig . [ fig:2dselfconsist](a ) and corresponding order parameters , @xmath69 and @xmath70 , in fig . [ fig:2dselfconsist](b ) as a function of the zeeman energy . -wave order parameter . the solid line of the inset shows that the critical zeeman energy is proportional to the induced superconducting gap , @xmath89 , which results in a good agreement with a linear fit indicated as a dashed line . the hybridization gap of the system is set to @xmath90 ev . we set @xmath91 and cut - off momentum is set so that corresponding cut - off energy is @xmath92 . we set @xmath93 to obtain a magnitude of the order parameter that is smaller in energy than the hybridization gap.,scaledwidth=50.0% ] in the absence of the zeeman energy , the system with symmetric pairing potential shows the lowest energy in fig . [ fig:2dselfconsist](a ) . although we may conclude that symmetric pairing potential determines the system order parameter at the zero or low zeeman energy regime , fig . [ fig:2dselfconsist](b ) shows that the energetically favored order parameter changes its form from symmetric ( blue dashed line ) to anti - symmetric ( red solid line ) as we increase the zeeman energy . this particular transition of the dominant pairing term from symmetric to anti - symmetric form may be understood in terms of the inter - band and intra - band pairing described in eq . ( [ eq : sym ] ) and ( [ eq : asym ] ) , which is governed by the symmetric and anti - symmetric pairing potential , respectively . the cooper pairs formed by the inter - band pairing experience a fermi surface mismatch as the zeeman energy splits the bands in energy . although fig . [ fig:2dselfconsist](b ) shows that the superconducting gap is largely unaffected for small zeeman energy , further increase in the zeeman energy results in a larger mismatch in fermi surface that costs more energy to form cooper pairs . consequently , the total energy in fig . [ fig:2dselfconsist](a ) gradually increases and the system eventually undergoes a first - order phase transition from superconducting state to normal state when the zeeman energy is as large as the superconducting gap at zero field , or @xmath94 at zero temperature limit@xcite . consequently , the order parameter obtained from symmetric pairing potential in fig . [ fig:2dselfconsist](b ) exhibits a significant drop in its magnitude due to this phase transition . the critical zeeman energy found in our self - consistent calculation is proportional to the induced superconducting gap at the zero field condition , @xmath95 , as it is shown in the inset of fig . [ fig:2dselfconsist](b ) , which confirms the above analysis regarding the first order phase transition point . in contrast , the intra - band pairing still persists for the zeeman fields above the critical value . in the absence of the zeeman field , intra - band pairing potential results in a non - zero pair correlation function as one can find @xmath0-wave electron pairing within each band due to the spin - momentum locked nature of the surface states . in the presence of the zeeman field , the ti surface states exhibit tilted `` hedgehog '' spin texture@xcite and , consequently , the projected @xmath0-wave pairing magnitude may decrease due to the out - of - plane tilting induced by the zeeman field . however , each band still possesses non - zero @xmath0-wave pairing while the paired electrons experience no fermi surface mismatch unlike those with inter - band pairing potential in the same band . therefore , the intra - band pairing potential results in a decreasing but non - vanishing @xmath0-wave order parameter in fig . [ fig:2dselfconsist](b ) . the above analysis shows that anti - symmetric pairing potential is the dominant factor to form induced superconductivity for zeeman energies larger than the zero - field superconducting gap , which is experimentally relevant regime considering that the typical superconducting gap is few mev for 4-quintuple - layer bi@xmath1se@xmath7@xcite while the zeeman energy may be as large as @xmath96 mev in @xmath97% mn doped bi@xmath1se@xmath7@xcite system . in this regard , the anti - symmetric pairing potential may be more relevant than the symmetric one for magnetically - doped ultrathin ti system . our analysis in section [ sec:2dop ] shows that the anti - symmetric pairing potential plays a major role in superconductivity in the magnetically - doped ultrathin ti system when the induced zeeman energy is larger than the induced superconducting gap . taking advantage of the property of the anti - symmetric pairing potential discussed in eq . ( [ eq : asym ] ) , the bdg hamiltonian in eq . ( [ eq : bdgp ] ) is now decoupled into two independent sectors . with the reduced bdg hamiltonian , we now obtain the analytic form of the quasi - particle spectrum and use the gap closing points to identify topologically non - trivial phases . from eqs . ( [ eq : hsurf22 ] , [ eq : bdgp ] , [ eq : asym ] ) , we obtain a block diagonal form of the bdg hamiltonian , which we rewrite as two decoupled bdg hamiltonian for sector 1 and sector 2 , @xmath98 in eq . ( [ eq : bdgasym ] ) , @xmath35 is the pauli matrix whose representation is in the hybridized basis shown in eq . ( [ eq : u2c ] ) , and @xmath99 is the sector @xmath100 hamiltonian in eq . ( [ eq : hsurf12 ] ) . from this , we obtain the quasi - particle spectrum @xmath101 where @xmath102 and @xmath40 is the effective magnetic field defined in eq . ( [ eq : hsurf22 ] ) for each of the sectors . for non - zero @xmath103 , we find that a gap closing occurs only at @xmath104 when @xmath105 . the system undergoes a phase transition at the identified gap closing point@xcite with the zeeman energy @xmath106 and , therefore , the gap closing point is determined by the system parameters : the hybridization gap ( @xmath107 ) , the position of the chemical potential ( @xmath108 ) , and the induced superconducting gap ( @xmath103 ) . as a topological classification of our system is in @xmath10 class@xcite , the relevant phase is classified by the @xmath11 topological invariant . in this regard , we numerically evaluate the chern number by varying the chemical potential and zeeman energy , whose values are normalized by the hybridization gap . prior to our analysis on the system chern number , we first define a lattice regularized model for the top surface hamiltonian@xcite @xmath109 \\ & + ( d / a^2)[2-\cos ( k_xa)-\cos ( k_ya)]\sigma_z , \end{split}\ ] ] and bottom surface hamiltonian is @xmath110 , where @xmath111 is a lattice constant , @xmath112 is the fermi velocity , @xmath10 is a parameter that controls the quadratic term at higher energy . without loss of generality , we set @xmath113 and @xmath114 . the second line of the eq . ( [ eq : h2d ] ) gaps out the extra dirac points at @xmath115 . although this term breaks time - reversal symmetry , the system under consideration has broken time - reversal symmetry due to the magnetic dopants . therefore , eq . ( [ eq : h2d ] ) correctly captures relevant low energy physics in our system . using the eqs . ( [ eq : bdg ] , [ eq : h2d ] ) , we construct the bdg hamiltonian to obtain eigenstates of the @xmath116th band to calculate corresponding chern number@xcite , @xmath117 . for calculation details of the chern number , see appendix [ app : chern ] . finally , the total chern number of the system is obtained using @xmath118 , where @xmath116 runs over all of the occupied bands . we return to the discussion below eq . ( [ eq : bdgasym ] ) to examine the chern number of the bdg hamiltonian for the sector 1 with the assumed anti - symmetric pairing potential . assuming the hybridization gap satisfies @xmath119 , the proximity coupled ultrathin ti system is initially gapped and in the trivial regime for @xmath120 . as we increase the magnetic dopant concentration and , as a result , the zeeman energy @xmath121 , the chern number of the system becomes -1 after the first gap closing point at @xmath122 and becomes -2 after the second gap closing point at @xmath106 . other gap closing points exist at @xmath123 and @xmath124 for @xmath125 and at @xmath126 for @xmath127 . as a result , the chern number becomes @xmath128 in the large @xmath121 limit . however , we only consider low - energy physics in this work , thus we only consider the gap closing point at @xmath129 . a similar analysis may be applied for @xmath130 using the hamiltonian in sector 2 . , and the zeeman energy @xmath131 . both axises are normalized by the hybridization gap , @xmath107 . for illustration purposes , we draw schematics that describe the motion of the surface state bands in different sectors and the corresponding chern numbers . note that solid curve ( dashed curve ) represents surface state band in sector 1 ( sector 2 ) , and black ( red ) color represents conduction ( valence ) band when @xmath132 . the horizontal line in each band diagram corresponds to the position of the chemical potential . the @xmath0-wave order parameter at top and bottom surface is @xmath133 and @xmath134 , respectively.,scaledwidth=50.0% ] to obtain a generic phase diagram , we assume that both the symmetric and anti - symmetric pairing potential are present . specifically , we assume that the magnitude of the induced superconducting gap satisfies the experimentally relevant regime for the ultrathin ti system@xcite , namely , @xmath135 . in this regime , the specific form of the pairing potential is insignificant as we construct the phase diagram with other parameters relative to @xmath107 . thus , we set the pairing hamiltonian to satisfy @xmath136 . then we compute the chern number and construct the phase diagram of the ultrathin ti in fig . [ fig : phase ] by varying the chemical potential , @xmath108 , and the zeeman energy , @xmath121 , normalized by the hybridization gap , @xmath107 . each region in the phase diagram is illustrated by a schematic of the surface state band diagram in a normal phase ti with a specified location of the chemical potential . specifically , solid and dashed curves are the bands in sector 1 and sector 2 , respectively , and the horizontal line indicates a location of the chemical potential . note that the normal phase of the ultrathin ti undergoes the quantum phase transition at @xmath137@xcite due to the crossing of the conduction and valance band , which are initially separated by the hybridization gap . in order to capture this quantum phase transition point in our schematics , the initial valence and conduction bands at @xmath132 are indicated by red and black color , respectively . to illustrate the system behavior in more detail , we follow , for example , the @xmath138 cut in fig . [ fig : phase ] . at @xmath139 , the chemical potential is located within the hybridization gap and the system is in trivial phase ( the chern number is 0 ) . as we increase the zeeman energy , the conduction band is shifted and touches the chemical potential at @xmath104 , which corresponds to a gap closing point in the quasi - particle spectrum of bdg hamiltonian . then , the band acquires a non - trivial chern number of -1 . note that the valence band and conduction band are shifted further due to the zeeman energy until they touch at @xmath140 . however , this particular band crossing of the conduction and valence band has no effect on the chern number , as there is no gap closing in quasi - particle spectrum of the bdg hamiltonian . once the zeeman energy passes the second gap closing point , the additional band crossing at the chemical potential results in a chern number of -2 . therefore , it is important to place the relevant parameters such as chemical potential and zeeman energy in the right position to observe the topological superconductivity in the system . further increase in zeeman energy , @xmath121 , will result in a gap closing at other high symmetry points , such as @xmath123 and @xmath141 , which eventually results in the chern number becoming @xmath128 in the @xmath142 limit . however , this is not important as we present the phase diagram within the experimentally relevant region where the zeeman energy of the magnetic dopants is comparable to the hybridization gap . the physics of the non - trivial phase in the system may be further described by mapping our system to one of the well known topological superconducting systems . the hamiltonian in eq . ( [ eq : bdgasym ] ) has a close analogy to the surface states of the 3d ti ( for example , a top surface ) proximity coupled to the interface of a ferromagnet and @xmath0-wave superconductor@xcite , where one may find chiral majorana edge modes at the domain wall of the ferromagnet and superconductor region . likewise , by properly choosing the concentration of the magnetic dopants and chemical potential of the ultrathin ti , we may also find the chiral edge states at the domain wall of the magnetically doped and undoped region of our system . our motivation in this section is to examine the induced order parameter beyond the phenomenological model presented in section [ sec : phenomenological ] . instead of assuming the induced order parameter is a constant value , we examine the self consistently calculated induced order parameter by adopting a four band 3d ti model coupled to an @xmath0-wave superconductor . -wave superconductor3d ti heterostructure that we consider in this work . in our 3d self - consistent simulation , we include a non - zero interaction strength within the @xmath0-wave superconductor , @xmath77 , but , the interaction strength is initially set to zero in the ti as it contains no natural pairing . the self - consistent procedure to solve for the order parameter then allows for pairing amplitudes from the superconductor to enter the ti and induce superconductivity via the proximity effect . [ fig : schematic ] ] in fig . [ fig : schematic ] , we show a schematic of the @xmath0-wave superconductor and 3d ti system that we consider in this work . in this heterostructure , we assume the presence of a non - zero attractive interaction , parametrized by the pairing strength @xmath77 , within the superconducting layer and no explicit pairing strength within the ti film . we begin the construction of our model by first considering the non - superconducting hamiltonian for our heterostructure including the contributions from both the superconductor and the ti . as we wish to accurately model spatially large structures , we write the in - plane directions ( @xmath143 and @xmath144 ) within the momentum - space representation while we write the out - of - plane direction ( @xmath145 ) in real space . in order to maintain notational simplicity , @xmath46 is used to represent the in - plane momentum @xmath146 . the mixed - representation , or that which is composed of both real- and momentum - space components , non - superconducting , nearest - neighbor - hopping hamiltonian is @xmath147 , \end{split}\ ] ] where the annihilation operator @xmath148 is defined by a basis containing two orbitals ( @xmath149 , @xmath150 ) and two spin degrees of freedom ( @xmath151 , @xmath152 ) . the ti hamiltonian is represented by a minimal bulk model for 3d topological insulator which consists of two spin and two orbital bases@xcite , @xmath153 with a momentum space lattice description of @xmath154 \\ & - 3b / a^2+\mathbb m , \end{split}\ ] ] where @xmath112 is the fermi velocity , @xmath111 is a lattice constant , and @xmath155 is a material parameter used to fit to a specific material band structure . in eq . ( [ eq : dm ] ) , @xmath156 is a parameter that controls a topological phase of the system and the system is a trivial ( topological ) insulator when @xmath157 ( @xmath158)@xcite . the gamma matrices in eq . ( [ eq : hti ] ) are defined as @xmath159 and @xmath160 , where @xmath161 are @xmath162 identity matrices , @xmath163 and @xmath164 are the pauli matrices for spin and orbital degrees of freedom , respectively . the hamiltonian in eq . ( [ eq : hti ] ) captures the low - energy characteristics of 3d tri tis with the @xmath149 and @xmath150 orbitals , for example , that correspond to the linear combination of the @xmath165 orbitals of bi@xmath166 and se@xmath167 of bi@xmath1se@xmath7 , respectively@xcite . in addition , we introduce the zeeman energy , @xmath121 , with the corresponding gamma matrix @xmath168 in the ti hamiltonian to capture the magnetization arising from the magnetic dopants in the ti@xcite . to model the metallic superconductor - ti interface of our heterostructure , we represent the metal portion of our heterostructure by a simple four - fold degenerate hamiltonian @xmath169 i_4,\ ] ] where @xmath170 is the hopping parameter of the metallic system . without a loss of generality , we choose @xmath171 , @xmath172 , and @xmath173 as our main focus of this work is a qualitative analysis on the order parameter rather than an analysis of an order parameter for a specific material . to examine the strong topological insulator regime , we set the mass parameter @xmath174@xcite . is plotted at ( a ) the metallic surface ( @xmath175 ) , ( b ) the bottom surface ( @xmath176 ) , and ( c ) the top surface ( @xmath177 ) for @xmath178 at @xmath179 . similarly , @xmath180 is plotted at ( d ) the metallic surface , ( e ) the bottom surface , and ( f ) the top surface for @xmath181 . the obtained @xmath182 is normalized by its maximum magnitude at each layer . , scaledwidth=50.0% ] in our model , we select the first 10 grid points within our solution space in the @xmath145-direction to be the metal and the subsequent 4 points are chosen to be 3d ti in order to represent the possible experimental setup in fig . [ fig : schematic ] . the disparate metal and ti models are incorporated into the full mixed - space hamiltonian by making the following definitions for the matrices given in eq . ( [ eq : tb - ham ] ) : @xmath183 where @xmath184 and @xmath185 are the chemical potential of the metallic and ti system , respectively , and @xmath186 is a coupling constant that controls a coupling strength between the metallic and ultrathin ti system . unlike the isolated ti system , the presence of the metallic states modifies the surface states of the ti when the coupling constant , @xmath187 , is introduced . to examine how this coupling modifies the surface states of the ti , we compute the momentum- and real - space resolved local density of states , @xmath188 , computed by the system green s function@xcite using eqs . ( [ eq : h_0 ] , [ eq : h_z ] ) . figs . [ fig : dos ] ( a , b , c ) depict the @xmath188 at the metallic surface , the bottom layer , and the top layer of the ultrathin ti system , respectively , when a coupling constant is @xmath178 . due to the connecting hopping constant , @xmath189 , whose magnitude is comparable to that of the hopping constant in ti , the surface states of the ti exhibit strong imprint of the electronic states from the metallic states and , as a result , the top surface of the ti shows a reduced hybridization gap . in addition , charge accumulation at the interface may occur and cause a structural inversion asymmetry ( sia ) potential , which further modifies the surface states band structure of ti@xcite . such substrate effect and the resultant discrepancy of the band structure from that of the ideal ultrathin ti surface states are observed in , for example , bi@xmath1se@xmath7 system grown on si(111 ) substrate@xcite or aln(0001 ) substrate@xcite . on the other hand , no obvious substrate effect is observed in bi@xmath1se@xmath7 system epitaxially grown on @xmath190al@xmath1o@xmath7@xcite , gaas(111)a@xcite , and conventional @xmath0-wave superconductor 2h - nbse@xmath1@xcite . in this work , we focus on the moderate to weak coupling regime where the generic band structure of the ti system is minimally affected by the metallic system and we may exclude substrate effects from subsequent analysis . in figs . [ fig : dos ] ( d , e , f ) , we shows the @xmath188 at @xmath181 , where we find the ti surface states are close to that of the ideal ultrathin ti system with a well defined hybridization gap . for this reason , we use the coupling constant @xmath181 for the remainder of this work . with the non - superconducting hamiltonian defined , we may now incorporate the superconductivity into the system . in this work , we include superconductivity at the mean - field level , using an intra - orbital on - site interaction of the form @xcite @xmath191 where @xmath77 is the interaction strength , @xmath192 is the singlet pair annihilation operator , and @xmath193 is the @xmath0-wave order parameter . here we define the order parameter as a unitless quantity by separating the interaction strength , @xmath77 , from the order parameter in order to examine the induced order parameter in the ti where we set @xmath84 . for definitions and numerical calculation procedures for the order parameter in eq . ( [ eq : h_int ] ) , see appendix [ app : op2 ] and eq . ( [ eq : deltasintra ] ) . to incorporate superconductivity into the non - interacting hamiltonian of eq . , we expand it into a bdg hamiltonian with corresponding eight - component nambu spinor @xmath194^t $ ] as@xcite @xmath195 the intra - orbital @xmath0-wave interaction term in eq . is incorporated in the metallic portion of the hamiltonian through off - diagonal components @xmath196 , which are written as @xmath197 using the material parameters defined in section [ sec:3dti ] , we obtain a hybridization gap of @xmath198 mev and bulk gap of @xmath199 ev for the 4 layer thick ti hamiltonian in eqs . ( [ eq : h_0 ] , [ eq : h_z ] ) in normal phase . the chemical potential of the metal is set to @xmath200 to obtain sufficient density of states near the chemical potential to induce superconductivity within the metallic system . the interaction strength of the metal is set to @xmath201 ev to obtain a converged superconducting gap whose magnitude in energy is @xmath202 . from the self - consistent solution of the bdg hamiltonian given in eq . ( [ eq : hbdg3d ] ) , we compute the induced intra - orbital @xmath0-wave pair correlation function , @xmath203 , and corresponding @xmath0-wave order parameter , @xmath204/2 $ ] , within the ultrathin ti system . for detailed numerical calculation procedure , see appendix [ app : op2 ] . although similar analysis may be done for inter - orbital pairing , we focus on the intra - orbital induced order parameter for this study as an inclusion of the inter - orbital pairing does not differ from the qualitative trends presented here . -wave order parameter magnitude as a function of zeeman splitting energy at the chemical potential @xmath205 mev , whose location crosses both bulk and surface bands of the ti . black circles indicate the bulk state contribution and red squares indicate the surface state contribution for the obtained order parameter at the interface of the metal - ti ( @xmath176 ) . ( b ) a plot of the induced @xmath0-wave order parameter as a function of zeeman splitting energy at the chemical potential of @xmath206 , whose location only crosses the surface bands of the ti . unlike ( a ) , the induced order parameter is computed solely from the surface states of ti . black circles indicate the order parameter obtained at bottom layer ( @xmath176 ) , whereas the red squares indicate the order parameter at top layer ( @xmath177).,scaledwidth=50.0% ] we first examine the induced order parameter for surface and bulk states in the ti and compare its relative magnitude , as well as its evolution as a function of the zeeman energy . to this end , we locate the chemical potential of the ti at @xmath205 mev , which crosses both bulk and surface states of the ti . we calculate the magnitude of the intra - orbital @xmath0-wave order parameter , @xmath207 , by considering the surface and bulk state contributions separately as a function of the zeeman splitting energy . the details of how the bulk and surface contributions are separated are discussed in appendix [ app:3dkspace ] . the induced order parameter is maximized at the interface ( @xmath176 ) as the ti layer is directly coupled to the metallic @xmath0-wave superconducting system and decays as a function of depth as we move toward the other ti surface ( @xmath177 ) . to see a clear trend of the induced order parameter as a function of the zeeman splitting energy , we examine the magnitude of the induced order parameter at the interface ( @xmath176 ) . considering that a typical range of the hybridization gap of @xmath208 quintuple layer bi@xmath1se@xmath7 is @xmath209 mev@xcite and magnetic doping induced gap size is @xmath210 mev for @xmath211 % mn doping in bi@xmath1se@xmath7@xcite , we sweep the zeeman splitting energy up to @xmath212 . as we increase @xmath121 , fig . [ fig : op_bs](a ) shows a rapid decrease in induced order parameter of the bulk states ( black circle ) , whereas the induced order parameter of the surface states ( red square ) shows nearly constant magnitude . the seemingly different trend of the bulk and surface states may be explained by understanding the changes in band structure and the spin polarization near the fermi surface . to explain this , we plot the band structure of a 4 layer thick ti with corrsponding spin texture in fig . [ fig : bandstructure ] , specifically focusing on the states that cross the solid line representing a chemical potential of @xmath205 mev . when @xmath213 in fig . [ fig : bandstructure](a ) , we see that the surface states exhibit their spin - locked nature . as the zeeman field is increased in figs . [ fig : bandstructure](b - d ) , the bulk and surface states show qualitatively different pictures . the bulk states split into strictly up and down spin states , making a spin singlet state energetically unfavorable to form and causing @xmath214 to drop precipitously . the spin of the surface states is markedly different . because the spin is already locked to momentum at @xmath213 , the addition of the zeeman interaction has the effect of canting the spin out of the @xmath215-@xmath216 plane . however , as there is always a projection of spin that is anti - aligned in @xmath215-@xmath216 plane , @xmath0-wave pairing is always allowed , explaining why the induced @xmath0-wave pairing in the surface states decreases much smaller rate than in the bulk . we now narrow down our scope to the surface states and examine the induced order parameter at the top and bottom surfaces . particularly , we change our chemical potential and place it to the bottom of the surface band of the ti . as the introduced zeeman energy lifts the band degeneracy , this particular position of the chemical potential allows us to focus on a single , non - degenerate surface band and the resultant induced order parameter to elucidate and validate our analysis of the 2d surface model in section [ sec : phenomenological ] . to this end , we locate the chemical potential of the ti at @xmath206 indicated as a dashed line in fig [ fig : bandstructure](a ) . in fig . [ fig : op_bs](b ) , we find a purely real @xmath0-wave order parameter from our self - consistent calculation at the top ( @xmath177 , red square ) and bottom ( @xmath176 , black circle ) layer of the ultrathin ti . both top and bottom layer order parameter exhibit positive sign for zero or small zeeman splitting energy , implying that symmetric part of the pairing is the dominant factor . as we increase zeeman energy , we lift the degeneracy to separate the surface bands beyond the superconducting gap and observe a sign change in the induced @xmath0-wave order parameter at the top layer , implying that the anti - symmetric pairing potential now plays a dominant role over the symmetric pairing potential . although the overall behavior agrees with the phenomenological model analysis , fig . [ fig : op_bs](b ) shows that the spatial distribution of the top and bottom layer induced order parameter is always a mixture of the symmetric and anti - symmetric form , whereas we observe a sharp transition from the ground states possessing symmetric to anti - symmetric pairing potential in the 2d surface model in fig . [ fig:2dselfconsist ] . this discrepancy is due to the fact that the 3d model hamiltonian takes account the effect of the hopping from the metallic system to the ti . for clarity , we adopt 2d surface model to write an analytic expression of the self - energy term of the hopping from metallic system to bottom surface state of the ti as@xcite @xmath217 in eq . ( [ eq : selfenergy ] ) , the pauli matrices @xmath218 act on the nambu space , @xmath219 is an assumed order parameter of the @xmath0-wave superconducting system , and @xmath220 is an estimation of the electron life - time at the metal - ti interface which is determined by the hopping constants , @xmath187 and @xmath170 , and the chemical potential of the metal . the phenomenological model in eq . ( [ eq : bdg ] ) takes account second term of eq . ( [ eq : selfenergy ] ) , however , the first term has been ignored . assuming that the above self - energy is applied only to the bottom layer of the ultrathin ti due to its direct proximity to the metallic system , we may insert the first term as an on - site potential within the bottom layer hamiltonian . specifically , we first define the first term of the self - energy in eq . ( [ eq : selfenergy ] ) as @xmath221 . utilizing the symbols defined in eq . ( [ eq : hsurf ] ) and below , we incorporate the first term of the self - energy in eq . ( [ eq : selfenergy ] ) with the 2d surface hamiltonian in eq . ( [ eq : hsurf ] ) as an on - site potential at bottom layer , @xmath222 . following the discussion in eq . ( [ eq : hsurf2 ] ) , we apply the relevant rotational matrices to obtain a surface hamiltonian @xmath223u_2^\dagger u_1^\dagger,\ ] ] whose matrix form is @xmath224 where @xmath225 . note that the off - diagonal term , @xmath226 , in eq . ( [ eq : hsurfomega ] ) couples the sectors 1 and 2 which were previously decoupled in eq . ( [ eq : hsurf22 ] ) . considering the correlation function near the chemical potential , @xmath227 , the additional coupling term @xmath226 has a small but non - zero value . therefore , the presence of the coupling between sector 1 and sector 2 leads the co - existence of the symmetric and anti - symmetric components of the induced order parameter , as we no longer can decouple sector 1 and sector 2 completely . as a result , fig . [ fig : op_bs](b ) exhibits the spatial distribution of the induced order parameter as a mixture of the symmetric and anti - symmetric form and exhibits a smooth transition of the pairing potential from symmetric to anti - symmetric form . nevertheless , we clearly observe a transition of the induced @xmath0-wave order parameter from a symmetric dominant to an anti - symmetric dominant spatial distribution and thereby confirm that qualitative behavior of the induced order parameter is sufficiently captured at the phenomenological level . in summary , we studied magnetically - doped ultrathin ti system that is proximity coupled to a conventional @xmath0-wave superconductor in order to identify possible topological phases . we find that the system is described in two individual sectors comprised of the hybridized basis with top and bottom surfaces of ti . using a simplified picture for proximity induced order parameter , we identify that a symmetric and anti - symmetric spatial configuration of the induced order parameter pair electrons in the individual sectors differently . our subsequent analysis of the total energy of this system reveals that the anti - symmetric spatial distribution of the induced order parameter is energetically preferred in the presence of a zeeman energy larger than the magnitude of the induced order parameter . as the choice of an anti - symmetric order parameter greatly simplifies the analysis , we perform analytic analysis on the quasi - particle spectrum of the bdg hamiltonian . we then find the gap closing points at @xmath228 and identify that the gap closing points are modified by the hybridization gap , zeeman energy , and chemical potential of the ultra - thin ti . taking account both symmetric and anti - symmetric pairing potential , we draw a generic phase diagram to identify a relevant region in parameter space for topological superconductivity . to enhance our understanding on proximity coupled ultrathin ti system , we study a more realistic model with a four band 3d ti hamiltonian directly coupled with metallic superconducting system . in this system , the induced order parameter is determined self - consistently . we confirm that our phenomenological model captures the correct behavior of the induced order parameter as the self - consistently determined order parameter shows a clear transition from symmetric to anti - symmetric dominant spatial distribution form . furthermore , the results show that the surface state induced @xmath0-wave order parameter survives even at a relatively high magnitude of zeeman energy , whereas the bulk state induced @xmath0-wave order parameter exhibits a rapid decay for the increased zeeman energy . we believe that this work not only explains why topological superconductivity yet to be observed in this system , but also provides useful experimental insight into the proper choice of parameters where it may be found . , * * , ( ) , issn , http://iopscience.iop.org/article/10.1088/0953-8984/21/16/164210 http://stacks.iop.org/0953-8984/21/i=16/a=164210?key=crossref.d86dbfbb1c5ef88ec2a1069587bb717e[http://iopscience.iop.org/article/10.1088/0953-8984/21/16/164210 http://stacks.iop.org/0953-8984/21/i=16/a=164210?key=crossref.d86dbfbb1c5ef88ec2a1069587bb717e ] . , , , , , , , , , , , * * , ( ) , issn , http://www.sciencemag.org/content/329/5992/659.abstract http://www.sciencemag.org/cgi/doi/10.1126/science.1189924[http://www.sciencemag.org/content/329/5992/659.abstract http://www.sciencemag.org/cgi/doi/10.1126/science.1189924 ] . , , , , , , , , , , , * * , ( ) , issn , http://dx.doi.org/10.1038/nphys3139 http://www.nature.com/doifinder/10.1038/nphys3139[http://dx.doi.org/10.1038/nphys3139 http://www.nature.com/doifinder/10.1038/nphys3139 ] . , , , , , , * * , ( ) , issn , http://link.aps.org/doi/10.1103/physrevb.82.045122 http://journals.aps.org/prb/pdf/10.1103/physrevb.82.045122[http://link.aps.org/doi/10.1103/physrevb.82.045122 http://journals.aps.org/prb/pdf/10.1103/physrevb.82.045122 ] . in this section , we describe how the induced order parameter in 2d surface state model of the ti is calculated . within the phenomonological model for the induced order parameter , we consider the @xmath229 2d bdg hamiltonian defined in eq . ( [ eq : bdg ] ) in the main text . the particle operator is defined as @xmath230^{t}$ ] where @xmath231 $ ] . we diagonalize the hamiltonian in eq . ( [ eq : bdg ] ) and obtain eigenfunctions , @xmath232 , and eigenvalues , @xmath10 , which satisfies @xmath233 . then we define the quasi - particle operator @xmath234 from @xmath235 , where @xmath236^t$ ] with the quasi - particle spectrum index ranging from @xmath237 . the particle operator may be written as bogoliubov quasi - particle operators , namely , @xmath238 , where each element is expressed as @xmath239_i=\sum_j [ u]_{ij}[\gamma_{{{\mathbf k}}}]_{j}$ ] . for example , the electron annihilation operator of the top surface is @xmath240_1=\sum_i[u]_{1i}[\gamma]_{i } = \sum_iu_{1i}\gamma_{i,{{\mathbf k}}},\ ] ] where we define @xmath241_{ij}$ ] . as a result , we may write the correlation function of the top surface as @xmath242_{16 } , \end{split}\ ] ] where @xmath243 is fermi - dirac distribution at energy @xmath244 , and we utilize the quasi - particle operator properties : @xmath245 and @xmath246 . finally , the correlation function of the s - wave order parameter at the top surface is defined as @xmath247_{16 } - [ u\cdot f(d ) \cdot u^\dagger]_{25}. \\ \end{split}\ ] ] the correlation function of the bottom surface is determined similarily . in this section , we describe how the chern number is calculated in a 2d square lattice . we begin with the generic hamiltonian having its eigenfucntion of the @xmath116th band , @xmath248 , where the momentum is defined in a square lattice brillouin zone as @xmath249 with the momentum index @xmath250 and @xmath251 . note that the eigenfunction is periodic in momentum space and satisfies @xmath252 , where @xmath253 is a vector in the @xmath254 direction . to compute the field strength through a lattice plaquette , we first define a link variable@xcite that corresponds to the berry connection in reciprocal space in the continuum as @xmath255 assuming that @xmath256 is well - defined . then the field strength is evaluated as@xcite @xmath257 which satisfies @xmath258 due to the logarithm applied to the link variables . by summing over all the plaquettes , we obtain the chern number of @xmath116th band as @xmath259 in this section , we describe how the induced order parameter of the 3d ti model is calculated . in order to self - consistently solve the system of equations , we diagonalize the bdg hamiltonian to determine the spectrum of the proximity - coupled system . to simplify this procedure , the bogoliubov transform is taken on eq . to form the eigenvalue equation @xmath260 where we suppress spin and orbital index for simplicity and the energy eigenvalues @xmath83 correspond to the eigenvector @xmath261 . then a pair correlation function is defined and may be computed numerically from this diagonalization procedure as following : @xmath262,\ ] ] where @xmath263 are orbital index , @xmath264 are spin index , @xmath243 is fermi - dirac distribution . we note that the selection of a temperature in our system is of little relevance to the simulations so long as it is smaller than the size of the superconducting gap . threfore , we set the temperature to @xmath265 k. in this work , we consider hubbard type on - site interaction for spin - singlet pairing . to consider relevant electron - electron pairing , we consider an on - site interaction for spin - singlet pairing channel : @xmath266 where @xmath267 represents interaction strength , @xmath268 is the density operator to describe the occupation number on @xmath269 orbital with spin state @xmath270 at site @xmath271 . assuming a constant interaction strength , @xmath272 , the mean field decomposition of the spin - singlet pairing hamiltonian in eq . ( [ eq : intst ] ) results in @xmath273 where we define the pair correlation function in real space as @xmath274 and a spin - singlet order parameter in real space , @xmath275.\ ] ] we perform a fourier transform of eq . ( [ eq : hsreal ] ) to momentum space in @xmath276 and @xmath277 direction . as a result , we obtain @xmath278 where we drop the constant term from eq . ( [ eq : hsreal ] ) . then the intra - orbital singlet order parameter is defined as @xmath279,\ ] ] and the inter - orbital singlet order parameter is defined as @xmath280,\ ] ] where @xmath281 , and each of the fourier transformed pair correlation functions are computed numerically using eq . ( [ eq : delta ] ) . note that eqs . ( [ eq : deltasintra ] ) and ( [ eq : deltasinter ] ) are even under momentum exchange and odd under spin exchange . ) of a non - superconducting 4-unit - cell ti at @xmath282 and @xmath283 with the chemical potential of 850 mev as indicated by the solid line . spin is indicated with blue arrows , and for visual clarity , the horizontal axis for the arrows corresponds to the @xmath144-component of spin , while the vertical axis is for @xmath145-component of spin . ( b ) the magnitude of the momentum resolved @xmath0-wave order parameter magnitude @xmath284 defined in eq . ( [ eq : kresolveddelta ] ) . the result displays a clear bulk contribution in the brillouin zone center and a surface state contribution in the outer ring . ( c ) band structure of the top surface of a non - superconducting 4-unit - cell ti at @xmath121= 80 mev and @xmath283 with the chemical potential of 850 mev as indicated by the solid line . we observe that the spin - degenerate bulk states split due to the presence of the magnetization . in the surface states , the zeeman energy instead cants the spin out of the plane into the @xmath285-direction . ( d ) the bulk contribution to the @xmath0-wave component of the order parameter is strongly suppressed due to the presence of @xmath25-directed magnetism while the surface contribution is resilient as the canted spins still have a significant in - plane projection that allows for persistent @xmath0-wave pairing . , scaledwidth=70.0% ] a wealth of information can be gathered from analyzing the momentum resolution we have in our self - consistent 3d simulation by defining the momentum - resolved intra - orbital @xmath0-wave order parameter @xmath286.\ ] ] we first consider these features when time - reversal symmetry is preserved with @xmath213 in figs . [ fig : bands_op ] ( a - b ) . in fig . [ fig : bands_op](a ) , we plot the band structure of the top - most layer of a 4-unit - cell ti without proximity coupling or superconductivity . the @xmath144 and @xmath145 directional spin texture are illustrated as the horizontal and vertical vector components , respectively , of the superimposed arrows , whose length is proportional to the relative magnitude of the individual components . in the surface band structure , we can clearly observe both the hybridization gap and the spin - momentum locking of the topologically non - trivial surface bands . the bulk states are spin - degenerate , therefore , we see no net spin texture in bulk bands . the chemical potential indicated by the solid , horizontal line shows that both bulk and surface states are available for pairing at the fermi surface . by noting the intersection of the chemical potential with the energy bands of the non - superconducting ti , we can identify the bulk and surface components of the @xmath0-wave order parameter of the proximity - coupled heterostructure in fig . [ fig : bands_op](b ) . the large contribution in the inner ring corresponds to the bulk component of @xmath0-wave pairing . similarly , the large outer ring in @xmath287 corresponds to the wave vectors associated with the surface states and their associated @xmath0-wave contribution . this knowledge allows us to unambiguously identify the surface and bulk contributions analyzed in the main text . specifically , we separate surface states contribution from bulk states contribution by computing eq . ( [ eq : deltasintra ] ) using the correlation function obtained at momentum space points that satisfy @xmath288 . in this work , we set the momentum cut - off as @xmath289 in the determination of the @xmath0-wave order parameters for bulk and surface states . with the physics of the order parameter understood when time - reversal symmetry is preserved , figs . [ fig : bands_op ] ( c - d ) now show the effects of zeeman energy on the resulting superconducting order parameter when @xmath290 mev , which explicitly breaks time - reversal symmetry . the effect of zeeman energy is evident in the non - superconducting ti surface band structure , shown in fig . [ fig : bands_op](c ) . when @xmath291 , we observe that the spin - degenerate bulk bands split with spin oriented completely in the @xmath145 direction with the up - spin states are pushed to higher energies and the down - spin states to lower . in the surface bands , we see a qualitatively different picture . because of the initial in - plane spin configuration of the hybridized surface bands , @xmath292 has the effect of canting the spin texture towards the @xmath293 , or , out - of - plane direction . this canted spin texture is characteristic of the zeeman splitting of surface bands that has been observed experimentally in the presence of magnetically ordered impurities in ti samples@xcite . due to the hybridization of the surface state wave functions in the system , the surface bands also split in energy with the energy change dictated by the sign of @xmath294-directed ca nt : up - canted spins rise in energy and down - canted spins lower . the presence of magnetism has immediate consequences to the different components of the superconducting order parameter . in fig . [ fig : bands_op](d ) , we observe that because the up- and down - spin states in the bulk are no longer present at opposing momenta , the bulk contribution to @xmath0-wave pairing is strongly suppressed . the surfaces states however , still have strong @xmath0-wave pairing despite the zeeman splitting from the magnetism . due to the fact that the spins of the surface states are only canted , a projection of spin is still anti - parallel across the brillouin zone and @xmath0-wave pairing is still energetically favorable . this results in a persisting magnitude of the @xmath0-wave order parameter in the surface states despite the energetic separation of surface bands .

we show that grand unified theories based on generate naturally the next to leading baryon number violating operators of dimension seven . these operators , which violate , lead to unconventional decays of the nucleon such as and . in two step breaking schemes of non - supersymmetric , nucleon lifetime for decays into these modes is found to be within reach of experiments . we also identify supersymmetric scenarios where these decays may be accessible , consistent with gauge coupling unification . further , we show that theasymmetry generated in the decays of gut scale scalar bosons and/or gauge bosons can explain consistently the observed baryon asymmetry of the universe . the induced asymmetry is sphaleron proof , and survives down to the weak scale without being erased by the electroweak interactions . this mechanism works efficiently in a large class of non susy and susy models , with either a or a higgs field employed for rank reduction . in minimal models the induced baryon asymmetry is tightly connected to the masses of quarks , leptons and neutrinos and is found to be compatible with observations . osu - hep-12 - 04 + umd - pp-012 - 004 * violating nucleon decay and + gut scale baryogenesis in * + * k.s . babu* and * r.n . mohapatra* _department of physics , oklahoma state university , stillwater , ok 74078 , usa _ _maryland center for fundamental physics , department of physics , + university of maryland , college park , md 20742 , usa _

1203.5544

baryon number violation is a very sensitive probe of physics beyond the standard model ( sm ) . interactions which violate baryon number ( @xmath8 ) are not present in the renormalizable part of the sm lagrangian , but they can arise as effective higher dimensional operators . the lowest @xmath8violating operators @xcite have @xmath9 and are suppressed by two powers of an inverse mass scale . these operators are realized naturally when sm is embedded in a grand unified theory ( gut ) such as @xmath10 and @xmath0 upon integrating out the heavy vector gauge bosons and colored scalar bosons . they lead to the decay of the nucleon into modes such as @xmath11 and @xmath12 . present experimental limits on nucleon lifetime constrain the masses of the mediators ( vector gauge boson or scalar bosons ) to be larger than about @xmath13 gev , which is close to the unification scale determined from the approximate meeting of the three gauge couplings when extrapolated to higher energies . an interesting feature of the @xmath9 baryon number violating operators is that they conserve baryon number minus lepton number ( @xmath14 ) symmetry , leading to the selection rule @xmath15 for nucleon decay@xcite . thus , observation of any decays which violate @xmath15 rule would hint at new dynamics different from those responsible for the @xmath9 operators . decay modes in this category include @xmath3 , @xmath16 etc , which all obey the selection rule @xmath17 . observation of these decay modes would thus furnish evidence against the simple gut picture with one step breaking to the sm . in this paper we study the next to leading @xmath18 operators , which obey the selection rule @xmath19 for nucleon decay @xcite , and show that they arise naturally within @xmath0 grand unified theories . in non supersymmetric @xmath0 models with an intermediate scale we find the nucleon lifetime for decay modes such as @xmath20 to be within reach of ongoing and proposed experiments . we also identify susy @xmath0 models where these decays may be within reach , consistent with gauge coupling unification . while we focus mainly on renormalizable @xmath0 models with @xmath4 of higgs bosons employed for rank reduction , we show that our results also hold for models with @xmath5 used for this purpose . the second main result of this paper is a mechanism for generating the baryon asymmetry of the universe at the gut epoch . the way it comes about is as follows . there are heavy scalar bosons and gauge bosons in @xmath0 theories which generate the @xmath18 operators . these particles have @xmath1violating two body decays , which can generate the observed baryon asymmetry of the universe naturally , as we show here . this would not be possible in the case of @xmath1preserving decays of gut scale particles such as the ones in @xmath10 . although grand unified theories were thought to be the natural stage for implementing the sakharov s conditions for baryogenesis @xcite up until the mid-1980 s @xcite , this idea was practically abandoned after the realization that the sphalerons @xcite , which violate @xmath21 symmetry , would erase any baryon asymmetry that obeyed the @xmath22 selection rule . this is because the effective interactions generated by sphalerons , the non - perturbative configuration of the weak interactions , are in thermal equilibrium for temperatures in the range @xmath23 gev , and violate @xmath24 symmetry . however , if baryon asymmetry was generated by @xmath1violating decays of gut scale particles , they would be immune to sphaleron destruction . we show that this mechanism of baryogenesis , which also induces the @xmath18 @xmath8violating operators , is very efficient and occurs quite generically in @xmath0 models . in minimal models there is a tight connection between the induced baryon asymmetry and the masses of quarks , leptons and the neutrinos . we also note that the minimal renormalizable versions of these models @xcite have been extremely successful in describing neutrino masses and mixings @xcite , and in particular predicted relatively large value for the neutrino mixing angle @xmath25 , which is consistent with recent results from daya bay , t2k , double - chooz and minos experiments @xcite . the results of the present paper show that these models can also explain the observed baryon asymmetry in a manner closely connected to the neutrino oscillation parameters . this paper is organized as follows . in sec . 2 we discuss the @xmath18 @xmath8 and @xmath1violating operators . in sec . 3 we show how these operators arise in unified @xmath0 theories , both in the non - supersymmetric version and in the susy version . in sec . 4 we address nucleon decay lifetime for the @xmath1violating modes in @xmath0 models with an intermediate scale . 5 is devoted to gut scale baryogenesis mechanism tied to the @xmath18 operators . here we show the close connection between baryon asymmetry and fermion masses and mixings . finally , we conclude in sec . we begin by recalling the leading baryon number violating operators in the standard model which have @xmath9 . there are five such operators with baryon number @xmath26 @xcite : @xmath27 here we have not shown the color contractions ( which is unique in each term via @xmath28 ) , and we have suppressed the flavor indices . we have followed the standard notation for fermion fields with all fields being left handed . thus @xmath29 stands for the left handed antiparticle of @xmath30 . the spinor indices are contracted via the charge conjugation matrix between fields in parentheses . @xmath31 are the @xmath32 indices . the complex conjugate operators of eq . ( [ dim6 ] ) would of course carry @xmath33 . an interesting feature of the @xmath9 baryon number violating operators on eq . ( [ dim6 ] ) is that they all carry lepton number @xmath34 along with @xmath35 . consequently these operators preserve @xmath1 . thus nucleon decay mediated by these operators would obey the selection rule @xmath36 . the decays @xmath37 and @xmath38 would be allowed by this selection rule , while decays such as @xmath39 and @xmath20 , which require @xmath17 would be forbidden . grand unified theories based on @xmath10 and @xmath0 gauge symmetries generate the operators of eq . ( [ dim6 ] ) suppressed by two inverse powers of gut scale masses . with the gut scale near @xmath13 gev , as suggested by the approximate unification of the three gauge couplings , these operators lead to nucleon lifetimes of order @xmath40 years for @xmath1 conserving modes , which are in the range that is currently being probed by experiments . discovery of nucleon decay into @xmath17 channels such as @xmath41 would however suggest that the underlying dynamics is quite different from that of the @xmath9 effective operators of eq . ( [ dim6 ] ) . as already noted in the introduction , while gut scale particles can generate a baryon asymmetry in their @xmath8violating decays , it was realized that interactions of the electroweak sphalerons would wash out any such asymmetry that conserves @xmath1 . gut scale baryogenesis thus went out of fashion after the discovery of sphalerons . this was also in part due to the leptogenesis mechanism @xcite discovered soon thereafter , which can elegantly explain the observed baryon asymmetry with a connection to the small neutrino masses induced via the seesaw mechanism@xcite . now we turn to the next to leading @xmath8violating operators beyond those of eq . ( [ dim6 ] ) , which are of dimension seven . these operators are interesting in that they carry @xmath42 @xcite . while they are suppressed by one additional power of a heavy mass scale , they can naturally lead to sphaleron proof baryogenesis , as we show here . in several instances we also find that these operators may lead to observable @xmath1 violating nucleon decay . there are nine @xmath18 baryon number violation operators with @xmath43 listed below @xcite : @xmath44 we have used the same notation as in eq . ( [ dim6 ] ) . here @xmath45 is the standard model higgs doublet transforming under @xmath46 as @xmath47 . @xmath48 stands for the covariant derivative with respect to the @xmath46 gauge symmetry . note that @xmath49 , @xmath50 and @xmath51 must be antisymmetric in the down flavor indices . operators of the type @xmath52 are not written , since they are related to those listed in eq . ( [ dim7 ] ) by the equations of motion . all vector and tensor operators can be fierz transformed into the set of operators in eq . ( [ dim7 ] ) . note that all operators of eq . ( [ dim7 ] ) carry @xmath35 and @xmath53 , and thus @xmath54 , with the complex conjugates operators carrying @xmath55 . it is these operators which can mediate nucleon decay of the type @xmath56 and @xmath3 . the higher dimensionality of these operators would suggest naively that the nucleon decay lifetime would be much longer than the ones obtained from eq . ( [ dim6 ] ) . however , as we show below , in unified theories based on @xmath0 with an intermediate scale , these decays may be accessible to experiments . most interestingly , these operators can naturally generate baryon asymmetry of the universe at the gut scale , which is facilitated by the fact that the electroweak sphaleron interactions do not wash out a @xmath1 asymmetry generated at such a scale . in the supersymmetric version of the standard model , baryon number violation can arise through operators in the superpotential analogous to eq . ( [ dim7 ] ) . these superpotential operators would have dimension six . holomorphicity of the superpotential would however constrain the allowed operators . there is a single operator of dimension six given by the superpotential coupling @xmath57 which carries @xmath33 and @xmath58 . this operator must be antisymmetric in the down - flavor indices owing to bose symmetry . @xmath59 here is the up type higgs doublet of mssm . in the superpotential , this operator will appear with two inverse powers of a heavy mass scale . terms in the lagrangian resulting from eq . ( [ w ] ) would have two fermion fields , one higgs field and two superpartner scalar fields , for example . when the superpartner scalar fields are converted to standard model fermions by a gaugino loop , effective @xmath18 operators of eq . ( [ dim7 ] ) would be generated , suppressed by a factor @xmath60 , rather than @xmath61 that occurs for eq . ( [ dim7 ] ) without susy , where @xmath62 is the heavy mass scale . therefore , potentially these susy contributions can be more significant for nucleon decay . we can now present the complete list of @xmath55 effective operators through @xmath18 in the sm by adding to eq . ( [ dim7 ] ) operators with @xmath63 . these operators have been classified in ref . @xcite . while not directly related to nucleon decay , these operators arise along with the @xmath18 operators of eq . ( [ dim7 ] ) in @xmath0 unified theories , and they are also relevant for gut scale baryogenesis . here we collect the linearly independent set of these operators through @xmath18 . the leading operator of course is the well - known @xmath64 seesaw operator @xcite @xmath65 the next to leading operators are of @xmath18 , and there are ten of them , as listed below . @xmath66 we shall see the appearance of some of these operators in the embedding of eq . ( [ dim7 ] ) in @xmath0 models . in the supersymmetric standard model , the four holomorphic operators @xmath67 would be allowed in the superpotential ( with no significance attributed to the spinor contractions of eq . ( [ lep ] ) when applied to the superfields ) . in this section , we show that the @xmath18 baryon number violating operators of eq . ( [ dim7 ] ) arise naturally in the context of @xmath0 unified theories after the spontaneous breaking of @xmath1 , which is a part of the gauge symmetry . the @xmath1 symmetry may break at the gut scale so that @xmath0 breaks directly to the standard model gauge symmetry , or it may break at an intermediate scale @xmath68 below the gut scale . in the latter case the intermediate symmetry could be one among several possibilities : @xmath69 ; @xmath70 ; @xmath71 ; @xmath72 ; or @xmath73 , with or without left right parity symmetry . in the non supersymmetric version an intermediate scale is necessary to be compatible with gauge coupling unification@xcite , while with supersymmetry the direct breaking of @xmath0 down to the mssm is preferable . even in the latter case , there is room for intermediate scale particles , provided that they form complete multiplets of the @xmath10 subgroup , since such particles do not spoil the unification of gauge couplings observed with the mssm spectrum . to see how the @xmath18 operators of eq . ( [ dim7 ] ) arise within @xmath0 , we first focus on the scalar mediated operators and write down the yukawa couplings in the most general setup . the higgs fields which can couple to the fermion bi - linears @xmath74 are @xmath75 , @xmath76 and @xmath77 , with the couplings of the @xmath75 and @xmath76 being symmetric in flavor indices @xmath78 and those of the @xmath77 being antisymmetric . the terms in these yukawa couplings that are relevant to the generation of the @xmath18 operators are given below @xcite . @xmath79 , \label{yuk10}\end{aligned}\ ] ] @xmath80 , \label{yuk126}\end{aligned}\ ] ] @xmath81 . \label{yuk120}\end{aligned}\ ] ] these terms are written in terms of the standard model decomposition of the sub - multiplets . we have followed the phase convention of ref . @xmath82 stands for the @xmath83 tensor @xmath28 . here we have not displayed terms that are irrelevant for inducing the @xmath18 baryon number violating operators . ( specifically , we have omitted color singlet , color octet , and color sextet couplings . ) the yukawa couplings obey @xmath84 , @xmath85 and @xmath86 . the @xmath46 quantum numbers of the various sub - multiplets are given as follows . @xmath87 different fields with the same sm quantum numbers appear in some couplings , they are distinguished by subscripts @xmath88 etc . we have used the same notation for fields with the same sm quantum numbers in @xmath75 , @xmath76 and @xmath77 , but it should be understood that these are distinct fields . after gut symmetry breaking various subfields with the same sm quantum number would mix . some of these mixings would involve the vacuum expectation value of the sm singlet field from the @xmath76 , denoted by @xmath89 carrying @xmath55 . it is this field that supplies large majorana mass for the right handed neutrino through the coupling @xmath90 . with @xmath91 , trilinear scalar couplings of the type @xmath92 , @xmath93 , @xmath94 and @xmath95 , will develop . this issue will be addressed in more detail below , but we note that such couplings are invariant under the unbroken sm gauge symmetry . when combined with the yukawa couplings of eqs . ( [ yuk10])-([yuk120 ] ) , they would induce the @xmath18 baryon number violating operators of eq . ( [ dim7 ] ) . operators induced by the symmetric yukawa couplings of @xmath75 and @xmath76 of @xmath0 . here the sm quantum numbers of the various fields are @xmath96 , and @xmath97 . ] to see how the @xmath18 operators arise in more detail , let us focus of the flavor symmetric yukawa couplings of eqs . ( [ yuk10])-([yuk126 ] ) . these couplings generate two of the @xmath18 operators as shown in fig . [ sym ] . here @xmath97 is the sm higgs doublet , which is a linear combination of the @xmath98 and @xmath99 fields from @xmath75 , @xmath76 as well as any other higgs sub - multiplet with the quantum number of @xmath100 in the theory with which these fields mix . similarly , @xmath101 generically stands for any linear combination of @xmath102 and @xmath103 from the @xmath75 , @xmath104 and @xmath105 from the @xmath76 , etc . before estimating the strength of these operators , let us examine the origin of the trilinear scalar couplings that appear in these diagrams in @xmath0 . to see the origin of @xmath92 and similar vertices , let us recall first the decomposition of various @xmath0 fields under the subgroups @xmath69 and @xmath73 . under @xmath106 , we have the following decomposition : @xmath107 under the @xmath108 subgroup various fields decompose as follows . @xmath109 now , the quartic coupling @xmath110 , which is invariant ( there is a single such coupling ) , contains the term @xmath111 under @xmath112 . the @xmath113 field is a subset of @xmath114 fragment . the @xmath97 field is part of @xmath114 , while @xmath115 and one such field is also part of @xmath116 . thus one sees that the coupling @xmath111 would contain the term @xmath117 , where @xmath118 denotes the sm singlet field from @xmath4 that acquires a gut scale vev . in the @xmath119 decomposition , @xmath120 , @xmath121 , and @xmath122 . thus @xmath110 contains the term @xmath123 , which has the piece @xmath117 . there are three non - trivial invariants of he type @xmath124 . these couplings also contain the term @xmath125 , as can be seen by examining the decomposition under @xmath112 and separately under @xmath119 . coupling would also contain a term @xmath126 under @xmath112 , which has a @xmath117 term in it . however , the @xmath102 appearing here is from the @xmath127 , which is @xmath128 of eq . ( [ yuk126 ] ) . @xmath128 coupling by itself does not violate baryon number @xcite , as is evident from eq . ( [ yuk126 ] ) . ] in an analogous fashion one sees that the coupling @xmath129 contains @xmath125 , with the @xmath130 arising from the @xmath131 . to complete this discussion we also note that there are three invariants of the type @xmath132 , which contain @xmath133 , with the @xmath134 and one @xmath114 taken from the @xmath135 . the @xmath134 fragment contains the @xmath136 of eq . ( [ higgs ] ) , which would enter the @xmath18 operators arising by integrating out the flavor antisymmetric @xmath135 fragments . in particular , this term would induce the needed @xmath137 vertex . the @xmath133 term also contains @xmath95 vertex , with @xmath138 , @xmath139 from the @xmath135 . the @xmath132 invariant also contains the term @xmath140 , with the @xmath141 and one @xmath114 taken from the @xmath135 . this piece generates the terms @xmath142 vertex , with @xmath143 . finally , there are two invariants of the type @xmath144 , which also contain the term @xmath133 . in order to complete @xmath0 symmetry breaking , additional higgs fields such as a @xmath145 , @xmath146 or a @xmath147 is needed . the interactions of these fields with @xmath148 and @xmath135 can generate further trilinear and quadrilinear couplings . take for example the case of a @xmath146 employed for completing the symmetry breaking . the trilinear couplings @xmath149 and @xmath150 are then invariant . noting that under @xmath73 subgroup , @xmath151 , we see that the latter coupling would contain a term @xmath152 . this has a piece @xmath153 , which would mix the two @xmath154 fields once @xmath155 develops , breaking @xmath1 by two units . ( note that the @xmath156 and @xmath157 carry different @xmath1 charges . ) the cubic coupling @xmath150 also contains the terms @xmath158 and @xmath159 under @xmath73 . these terms contain the couplings @xmath160 and @xmath161 respectively . these are the desired trilinear couplings for the generation of the @xmath18 operators . in this case , the mixing of @xmath162 with the @xmath163 is utilized in order to connect the @xmath154 field with the fermion fields in fig . similar results follow from the quartic couplings @xmath164 and @xmath165 . while trilinear scalar couplings of the type @xmath93 and @xmath166 do arise for the @xmath76 sub - multiplets , these couplings do not directly lead to baryon number violation . the @xmath167 field from @xmath76 has the coupling @xmath168 , while @xmath169 has the coupling @xmath170 ( see eq . ( [ yuk126 ] ) ) . the exchange of @xmath171 from @xmath76 would lead to an effective operator @xmath172 , while that of @xmath173 would generate the operator @xmath174 , which is operator @xmath175 of eq . ( [ lep ] ) . the @xmath93 and @xmath166 couplings of the @xmath76 sub - multiplets would however be relevant for gut scale baryogenesis , since they also violate @xmath1 symmetry . baryon number violating operators obtained by integrating our fields from the @xmath135 . here the sm quantum numbers of the scalar fields are : @xmath176 and @xmath177 in fig . [ antisym ] we display the effective @xmath18 operators obtained by integrating out the flavor antisymmetric @xmath135 coupling to fermions . we see that four operators are induced this way . as already noted , the required trilinear vertices to complete these diagrams arise from quartic couplings . it is also possible to replace the @xmath154 fields in fig . [ antisym ] by a @xmath154 field from the @xmath178 , in which case the sum of such diagrams would have no definite symmetry property in two of the flavor indices . note that all of the @xmath18 operators arising from fig . [ sym ] and [ antisym ] respect @xmath14 symmetry , as can be seen by assigning @xmath179 . it should be noted that operators @xmath180 of eq . ( [ dim7 ] ) do not arise at tree level by integrating out superheavy particles . they can arise via loops , with suppressed strength . in the estimate of nucleon lifetime these operators play a subleading role , since the amplitude for the decay would be further suppressed by the nucleon momentum ( rather than the electroweak vev for operators @xmath181 ) . what if the higgs field employed for reducing the rank of @xmath0 is a @xmath5 rather than a @xmath4 ? in this case the @xmath18 diagrams of fig . [ sym ] and [ antisym ] would still arise , albeit in a slightly different way . the @xmath1 quantum numbers of the higgs doublets @xmath98 from the @xmath75 , @xmath4 and @xmath77 fields are all zero . the @xmath5 contains a sm singlet filed with @xmath182 which acquires a gut scale vev . it also contains a @xmath183 field with @xmath184 . similarly , @xmath185 contains a sm singlet field with @xmath186 and a @xmath98 field with @xmath187 . the trilinear scalar couplings of the type @xmath188 and @xmath189 would mix the @xmath190 higgs doublet @xmath98 from the @xmath75 and the @xmath98 higgs from the @xmath185 which has @xmath182 . the light sm higgs doublet then would have no definite @xmath14 quantum number . the @xmath191 component of @xmath185 under @xmath112 contains the field @xmath113 , and the @xmath192 of @xmath185 under @xmath112 contains @xmath193 , and thus the coupling @xmath92 is generated via the @xmath194 coupling . one could also take @xmath101 from the @xmath75 , @xmath113 from the @xmath185 and @xmath97 from the @xmath185 to generate the @xmath92 coupling from @xmath189 . it should be noted that the @xmath195 field from the @xmath185 does have yukawa couplings to fermions , since in this type of models the heavy majorana masses of the @xmath196 fields arise from the couplings @xmath197 , and upon insertion of one vev for the sm singlet field here , the coupling of @xmath198 to fermions is realized . we point out that the @xmath199 field is partly in the goldstone mode , associated with the breaking of @xmath0 to @xmath10 . however , since other fields such as @xmath145 , @xmath146 or @xmath147 should be employed to complete the symmetry breaking down to the sm , one such physical @xmath199 will remain in the spectrum . this is because the @xmath145 , @xmath146 and @xmath147 all contain a @xmath199 field , and only one @xmath154 is absorbed by the gauge multiplet . baryon number violating operators via the exchange of vector gauge bosons @xmath200 and @xmath201 of @xmath0 . ] the @xmath18 operators of eq . ( [ dim7 ] ) can also arise by integrating out the gauge bosons of @xmath0 . these vector operators are related by fierz identities to the scalar operators displayed in eq . ( [ dim7 ] ) . the relevant diagrams are shown in fig . [ gauge ] . [ gauge ] ( d ) conserves @xmath8 , but violates @xmath202 and @xmath14 @xcite . the effective operator from this diagram , after a fierz rearrangement can be identified with @xmath203 of eq . ( [ lep ] ) . ) in these diagrams @xmath204 denotes the gauge boson with the sm quantum numbers @xmath205 ( same as the quark doublet @xmath206 ) , and @xmath207 is the gauge boson that transforms as @xmath208 , the same as that of @xmath29 fermion . in fig . [ gauge ] , each vertex conserves @xmath14 as it should , which can be seen by assigning @xmath209 to @xmath210 . the covariant derivative for the rank reducing field @xmath4 would contain the term @xmath211 which enters the @xmath18 operators . when @xmath5 is used instead of the @xmath4 , the covariant derivative would contain a similar term , but now the @xmath45 and @xmath89 fields would carry @xmath187 and @xmath212 respectively . baryon number violating operator of eq . ( [ w ] ) in @xmath0 . diagram ( a ) arises from integrating out colored fields from @xmath75 and @xmath76 , while diagram ( b ) arises from @xmath77 . ] we have identified in eq . ( [ w ] ) a single holomorphic operator of dimension six in the superpotential which violates @xmath8 and @xmath1 . in fig . [ susy ] we show how this operator can arise in susy @xmath0 . fig . [ susy ] ( a ) is obtained by integrating out colored fields in the @xmath75 and @xmath76 , while fig . [ susy ] ( b ) is obtained by integrating out such fields from the @xmath77 . these are superfield diagrams , effectively generating @xmath213terms in the lagrangian . note that the effective superpotential is antisymmetric in down - quark flavor in both diagrams ( a ) and ( b ) . this can be explicitly verified , by making use of the @xmath214 vertex in fig . [ susy ] ( a ) which is color antisymmetric . in fig . [ susy ] ( a ) , the vertex @xmath215 is contained in the yukawa coupling of eq . ( [ yuk126 ] ) and the vertex @xmath216 can be either from eq . ( [ yuk10 ] ) or from eq . ( [ yuk126 ] ) . the @xmath217 and @xmath218 transitions occur through @xmath219 ( or @xmath220 ) mass term , for example . as for the vertex @xmath221 here is a concrete example . the coupling @xmath222 is invariant , and contains such a term , where @xmath223 , @xmath224 and @xmath225 . note that @xmath225 carries @xmath226 , since it is part of the fragment @xmath227 under @xmath112 . of course , the mssm field @xmath59 in this case is an admixture of @xmath98 from @xmath75 , @xmath4 , @xmath76 and @xmath228 and carries no definite @xmath1 charge . as for fig . [ susy ] ( b ) , the vertices @xmath229 and @xmath230 are contained in the @xmath77 yukawa coupling of eq . ( [ yuk120 ] ) . the vertex @xmath231 can arise from the superpotential coupling @xmath232 , which is invariant . suporpotential operators of eq . ( [ w ] ) can also arise in susy @xmath0 models which utilize low dimensional higgs representations , for e.g. , @xmath233 . such models have been widely discussed in the literature @xcite . while @xmath234parity is no longer automatic in these models ( unlike in the case of @xmath4 models ) , it can be ensured by a discrete @xmath235 symmetry which distinguishes @xmath5 from the chiral fermions @xmath236 and which remains unbroken . the doublet triplet splitting problem can be addressed without fine - tuning in these models via the dimopoulos wilczek mechanism @xcite . generation of fermion masses in these models relies on higher dimensional operators such as @xmath197 which induce heavy majorana masses for the @xmath196 , and @xmath237 which induce lighter family masses and ckm mixings . ( @xmath5 and @xmath185 acquire gut scale vevs along their sm singlet components , while @xmath238 and @xmath239 do not . the @xmath32 doublet components of @xmath5 and @xmath238 acquire weak scale vevs , see for e.g. , discussions in ref . @xcite . ) to see the origin of eq . ( [ w ] ) in such models , consider generation of fig . [ susy ] ( a ) . the vertex @xmath240 arises from the yukawa coupling @xmath237 after a gut scale vev is inserted for the @xmath5 ( note that @xmath238 contains a @xmath154 field ) , and the vertex @xmath241 is contained in the coupling @xmath242 . now , @xmath243 can convert itself into @xmath244 via the coupling @xmath245 when the @xmath1preserving vev of @xmath246 is inserted , while the @xmath154 can transition into @xmath247 via the coupling @xmath248 with the insertion of a @xmath185 vev . the @xmath249 term is contained in the coupling @xmath250 , which completes the diagram . explicit models @xcite contain all these terms necessary for generating the @xmath9 superpotential operator . before discussing the rates for @xmath1violating nucleon decay , let us note that certain scalar bosons and certain gauge bosons would induce the more dominant @xmath9 baryon number violating operators of eq . ( [ dim6 ] ) . scalar bosons with sm quantum numbers @xmath101 , @xmath251 and @xmath252 and vector gauge bosons with sm quantum numbers @xmath253 and @xmath200 can can induce these @xmath1 preserving @xmath9 operators . current nucleon lifetime limits restrict the masses of these gauge bosons to be larger than about @xmath13 gev , and those for the scalar bosons to be heavier than about @xmath254 gev ( for yukawa couplings of order @xmath255 ) . since some of the @xmath18 operators arise in @xmath0 via the exchange of these particles , the above limits have to be met in our estimate of @xmath18 decay rates . we proceed to estimate the lifetime of the nucleon for its @xmath1violating decays in @xmath0 a straightforward way.preserving nucleon decay has been studied in the context of @xmath234parity breaking susy in ref . the diagrams of fig . [ sym ] lead to the estimate @xmath256 here we have defined the yukawa couplings of @xmath102 and @xmath154 fields appearing in fig . [ sym ] to be @xmath257 and @xmath258 . these couplings are linear combinations of the yukawa coupling matrices @xmath259 , @xmath260 and @xmath261 of eqs . ( [ yuk10])-([yuk120 ] ) with flavor indices corresponding to the first family fermions . the factors @xmath262 and @xmath213 are chiral lagrangian factors , @xmath263 and @xmath264 . @xmath265 is the nucleon decay matrix element @xcite , @xmath266 , and @xmath267 gev . we have defined the trilinear coupling of fig . [ sym ] to have a coefficient @xmath268 . as expected , the rate is suppressed by six powers of inverse mass , owing to the higher dimensionality of the effective operator . the mass of @xmath101 is constrained to be relatively large , as it mediates @xmath9 nucleon decay . for @xmath269 , @xmath270 gev must be met from the @xmath9 decays . as an illustration , choose @xmath271 , @xmath272 gev , @xmath273 gev , and @xmath274 gev in eq . ( [ tau ] ) . this choice would result in @xmath275 yrs . such a spectrum is motivated by the intermediate symmetry @xmath276 ( without discrete parity ) , which is found to be realized at @xmath277 gev from gauge coupling unification @xcite . as a second example , take @xmath278 gev , @xmath279 gev , @xmath280 gev , @xmath281 . this choice of spectrum leads to @xmath282 yrs . and two @xmath283 scalar multiplets at @xmath284 gev ( right panel ) . the left panel corresponds to having one @xmath199 and one @xmath283 scalar fields at the weak scale.,title="fig : " ] and two @xmath283 scalar multiplets at @xmath284 gev ( right panel ) . the left panel corresponds to having one @xmath199 and one @xmath283 scalar fields at the weak scale.,title="fig : " ] this second choice for the spectrum can be motivated as follows . the unification of gauge couplings may occur without any particular intermediate symmetry , but with certain particles surviving to an intermediate scale . we have found two examples of this type where the @xmath18 nucleon decay is within observable range . suppose the @xmath199 particle , along with a pair of @xmath283 scalar particles ( contained in the @xmath246 , @xmath285 or @xmath228 needed for symmetry breaking ) survive down to @xmath286 gev . the three sm gauge couplings are found to unify at a scale @xmath287 gev in this case . this is shown in fig . [ unif ] right panel . this figure is obtained by the one - loop renormalization group evolution of the sm gauge couplings above @xmath68 with beta function coefficients @xmath288 , where @xmath289 . since the @xmath9 nucleon decay lifetime would also be near @xmath290 yrs with such a unification scale , this scenario would predict observable rates for both the @xmath1conserving and @xmath1violating nucleon decay modes . another possible scenario for consistent gauge coupling unification is to assume that @xmath199 and one @xmath283 scalar multiplet survive down to the weak scale . again , the gauge couplings unify around @xmath287 gev , as shown in the left panel of fig . [ unif ] . here the @xmath291function coefficients used are @xmath292 , appropriate for this spectrum . the estimate @xmath293 yrs would follow , as in the second example above , but now with @xmath294 . nucleon decay rates arising from diagrams of fig . [ antisym ] are similar , but with a significant difference . take for example fig . [ antisym ] ( b ) with the @xmath295 trilinear vertex . neither the @xmath136 nor the @xmath199 field would mediate @xmath9 nucleon decay , and may be considerably lighter than the gut scale . in non supersymmetric @xmath0 models it is natural that some scalars survive down to an intermediate scale . typically the intermediate scale is of order @xmath296 gev if the associated symmetry contains @xmath297 . thus it is possible that both @xmath167 and @xmath154 have masses of order @xmath298 gev . nucleon decay rate is now estimated to be @xmath299 here @xmath268 is defined as the coefficient of the trilinear scalar vertex . for @xmath300 gev , @xmath301 gev , @xmath302 , @xmath303 yrs , which is in the observable range . the gauge boson exchange diagrams of fig . [ gauge ] would also induce @xmath304)violating nucleon decay . however , we find the rates for these decays to be suppressed . [ gauge ] ( a ) for example . the vector gauge boson @xmath204 must have mass of order @xmath13 gev , since it mediates @xmath9 nucleon decay , while @xmath207 could be much lighter as it does not have @xmath8violating interaction by itself . the @xmath305 vertex necessary for connecting the @xmath18 diagram has a coefficient of order @xmath306 , while the mass of @xmath207 is of order @xmath307 . the amplitude for @xmath18 nucleon decay arising from fig . [ gauge ] is then given by @xmath308 . this amplitude is a factor @xmath309 smaller compared to the standard @xmath9 nucleon decay amplitude originating from gut scale gauge bosons . unless the @xmath1breaking scale @xmath310 is close to the weak scale @xmath311 , not a likely scenario based on gauge coupling unification , nucleon lifetime for @xmath1violating modes from these diagrams would be beyond the reach of experiments . however , as we shall see in the next section , the @xmath1violating decays of the gauge boson can naturally explain the observed baryon asymmetry of the universe . as for the supersymmetric diagram of fig . [ susy ] , the @xmath9 operator in the superpotential eq . ( [ w ] ) has a strength @xmath312 . the lagrangian would contain a term @xmath313 , upon insertion of the vev of @xmath59 . the scalars can be converted to @xmath314 and @xmath315 quarks via a gluino loop , and would result in the following estimate for processes such as @xmath316 : @xmath317 here @xmath318 is a typical susy breaking mass scale . for @xmath319 , @xmath320 , @xmath321 gev , @xmath322 gev , @xmath323 tev , one obtains @xmath324 yrs . as for the consistency of such an intermediate scale mass for @xmath154 with gauge coupling unification in susy models , we note that @xmath199 forms a complete @xmath10 @xmath131plet along with a @xmath208 and a @xmath325 fields . if these fields also have an intermediate scale mass , unification of gauge coupling would work as in the mssm . it should be noted that in susy models , the decay @xmath326 would be suppressed , since the vev of @xmath59 picks a neutrino field in eq . ( [ w ] ) . discovery of @xmath326 decay would thus hint at a deeper non supersymmetric dynamics . we now proceed to the computation of the baryon asymmetry of the universe induced at the gut epoch . the @xmath1violating decays of the scalars @xmath101 and @xmath136 and of the vector gauge boson @xmath200 will be used to illustrate the mechanism . we shall see that in each case , the out of equilibrium condition can be satisfied , and that there is enough cp violation . these decays generate an asymmetry in @xmath1 , which is not destroyed by the effective interactions induced by the electroweak sphalerons , and would survive to low temperatures . this is in contrast with the induced baryon asymmetry in the @xmath1preserving decays of gut scale scalars and gauge bosons in unified models such as @xmath10 , which is washed out by the spharleron interactions . we begin with the @xmath1violating decay of the scalar @xmath101 which is assumed to have a mass of order the gut scale . ( @xmath14 asymmetry in decays of specific heavy particles has recently been discussed in ref . @xcite . ) to be concrete , we shall work in the framework of non supersymmetric @xmath0 , although our results would hold for susy @xmath0 as well , with some minor modifications . we identify @xmath102 to be the lightest of the various @xmath327 scalar fields in the @xmath0 theory . the yukawa couplings of eq . ( [ yuk10])-([yuk120 ] ) imply that @xmath102 ( which is in general a linear combination of @xmath328 from the @xmath75 and @xmath76 and @xmath77 fields ) has two body decays into fermions of the type @xmath329 . these decays preserve @xmath1 , as can be seen by assigning @xmath330 . now , @xmath102 also has a two body scalar decay , @xmath331 as shown in fig . [ baryo1 ] ( a ) , which uses the @xmath1 breaking vev of @xmath89 . the scalar field @xmath154 has two body fermionic decays of the type @xmath332 ( the latter if kinematically allowed ) , which define @xmath1 charge of @xmath154 to be @xmath333 . thus the decay @xmath331 would violate @xmath1 by @xmath334 ( recall that @xmath45 has zero @xmath1 charge ) . asymmetry in @xmath102 decay . ] focussing on the @xmath1violating decay @xmath331 , we define a @xmath1 asymmetry parameter @xmath335 as follows . let the branching ratio for @xmath331 be @xmath336 which produces a net @xmath1 number of @xmath337 , and that for @xmath338 be @xmath339 , with net @xmath340 . the branching ratio for the two fermion decays @xmath341 is then @xmath342 which has @xmath343 , and that for @xmath344 is @xmath345 which has @xmath346 . thus in the decay of a @xmath347 pair , a net @xmath1 number , defined as @xmath335 , is induced , with @xmath348 note that an interplay between the @xmath1conserving decays and the @xmath1preserving decays of @xmath102 is necessary for inducing this asymmetry . in addition , cp violation is required , otherwise @xmath349 and thus @xmath350 . nonzero @xmath335 also requires a loop diagram which has an absorptive part . all these conditions are realized in @xmath0 models . the loop diagrams for @xmath351 are shown in fig . [ baryo1 ] ( b)-(d ) , which involve the exchange of fermions . since @xmath102 can also decay to two on shell fermions , these loop diagrams have absorptive parts . note that fig . [ baryo1 ] ( b)-(c ) are proportional to the majorana masses for the @xmath196 fields , while fig [ baryo1 ] ( d ) is not . these diagrams have cp violating phases , ensuring a nonzero @xmath335 . we evaluate fig . [ baryo1 ] in a basis where the majorana mass matrix of the @xmath196 fields is diagonal and real . the contribution of fig . [ baryo1 ] ( b ) to @xmath335 is found to be @xmath352 { \rm br}. \label{asymb}\ ] ] here we have defined the trilinear scalar vertex of fig . [ baryo1 ] ( a ) to have a coefficient @xmath268 in the lagrangian . @xmath353 is the yukawa coupling matrix corresponding to the coupling @xmath354 , etc . @xmath355 is the diagonal and real mass matrix of @xmath196 fields . @xmath356 stands for the branching ratio @xmath357 . a factor of @xmath358 has been included here for the two @xmath32 final states in the decay . the function @xmath359 is defined as @xmath360 with @xmath361 denoting the mass of @xmath362 . here @xmath363 stands for the step function , signalling additional ways of cutting the diagram when @xmath364 in fig . [ baryo1 ] ( b ) . [ baryo1 ] ( c ) yields the following contribution to @xmath335 : @xmath365{\rm br } , \label{asymc}\ ] ] where @xmath366 is defined as @xmath367 fig . [ baryo1 ] ( d ) arises because in any realistic @xmath0 model there are at least two @xmath102 fields . the heavier @xmath102 field is denoted as @xmath368 . in principle one can sum over all such @xmath368 contributions , but here we have kept only one such @xmath368 field . its contribution to @xmath335 is found to be @xmath369 { \rm br } , \label{asymd}\ ] ] where @xmath370 is defined as @xmath371 this contribution , which is non - vanishing even in the limit of vanishing @xmath361 , requires @xmath372 . here we have defined the trilinear coupling @xmath373 to have a coefficient @xmath374 in the lagrangian . we have also assumed that @xmath375 , so that there is no resonant enhancement for the decay . ( when @xmath102 and @xmath368 are nearly degenerate in mass , such a resonant enhancement is possible . in this case , the expression for the decay rate will be smoothened by the width of these particles . appropriate expressions in this case can be found in ref . @xcite . ) the branching ratio factor @xmath376 appearing in eqs . ( [ asymb ] ) , ( [ asymc ] ) , ( [ asymd ] ) can be estimated as follows . for this purpose let us assume that @xmath102 is the field @xmath102 from @xmath75 with yukawa couplings as given in eq . ( [ yuk10 ] ) . the partial widths for the decays @xmath377 and @xmath378 are then given by @xmath379 in the expression for @xmath380 we have assumed that @xmath196 is much lighter than @xmath102 . in terms of these partial widths , the branching ratio that appears in eqs . ( [ asymb ] ) , ( [ asymb ] ) , ( [ asymd ] ) is given as @xmath381 . to get a feeling for numbers , let us choose a realistic set of parameters : @xmath279 gev , @xmath382 ( corresponding to the top quark yukawa coupling at gut scale ) with other @xmath383 negligible , and @xmath384 gev . this would correspond to @xmath385 , which shows a strong dependence on @xmath268 . the total @xmath1 asymmetry in @xmath331 and its conjugate decay is given by @xmath386 this will result in the baryon to entropy ratio @xmath387 given by @xmath388 where @xmath389 is the total number of relativistic degrees of freedom at the epoch when these decays occur . in our present example @xmath390 which includes the sm particles and the @xmath154 and @xmath102 scalar fields . the factor @xmath314 in eq . ( [ eta ] ) is the dilution factor which takes into account back reactions that would partially wash out the induced baryon asymmetry . @xmath314 is determined by solving the boltzmann equations numerically , but simple analytic approximations are available suitable to the present setup . defining a ratio @xmath391 where @xmath392 is the hubble expansion rate , @xmath393 the dilution factor can be written as @xcite @xmath394 these approximations work well for @xmath395 or so , beyond which @xmath314 would be exponentially suppressed . for @xmath279 gev , @xmath384 gev , we find @xmath396 , with the corresponding dilution factors being @xmath397 . thus we see that there is not much dilution with this choice of parameters , although @xmath385 can become small for smaller values of @xmath268 . if we choose @xmath398 gev instead , and vary @xmath384 gev , then we find @xmath399 and the corresponding dilution factors to be @xmath400 , with @xmath401 . although the electroweak sphaleron interactions would not wash away the gut scale induced @xmath1 asymmetry , partial wash out can occur via the @xmath1violating interactions of the right handed neutrinos . this is possible because the @xmath196 fields acquire @xmath1violating majorana masses , and their interactions with the higgs field and the lepton fields can erase part of the gut induced asymmetry . in some cases it may be desirable to have partial wash out . one can also prevent any wash out by decoupling the @xmath196 fields at the same temperature as the @xmath102 field ( which would happen if @xmath402 ) . for the surviving light @xmath196 fields , the condition @xmath403 would guarantee no further wash out , where @xmath404 is the dirac yukawa coupling of the light @xmath196 field . while we do not present the calculation of @xmath387 in the supersymmetric version of the @xmath0 model , results in that case would be similar to the non susy case . more diagrams contribute to the generation of @xmath1 asymmetry with superparticle decays included . the number of relativistic degrees of freedom @xmath389 would also double in this case . the @xmath1 asymmetry induced in the decay @xmath405 is similar to the one induced in the decay @xmath331 . the tree level @xmath1violating decay and the one loop corrections are shown in fig . [ baryo2 ] . @xmath167 has a fermionic decay mode @xmath406 shown in fig . [ baryo2 ] ( a ) , which can be used to define its @xmath1 quantum number as @xmath407 . the decay @xmath408 ( fig . [ baryo2 ] ( b ) ) would then violate @xmath1 by @xmath409 units , since @xmath410 obtained from the decays @xmath411 . the one loop correction to this decay is shown in fig . [ baryo2 ] ( c ) , which is evaluated in analogy to fig . [ baryo1 ] ( c ) to be @xmath412{\rm br}. \label{asymeta}\ ] ] here the function @xmath413 is defined as @xmath414 and @xmath268 identified as the coefficient of the trilinear vertex @xmath137 . asymmetry in @xmath415 decay . ] a noteworthy feature of baryon asymmetry generated in the decay @xmath415 is that both the scalars @xmath167 and @xmath154 can be relatively light , since they do not induce @xmath9 baryon number violating operators of eq . ( [ dim6 ] ) . in supersymmetric extensions of @xmath0 , this allows for the possibility that the gravitino abundance problem of supergravity models can be evaded . typically in supergravity models , the reheat temperature after inflation is required to be @xmath416 gev , in order to sufficiently dilute the gravitino abundance in the universe . if @xmath417 gev , this requirement would be compatible with the @xmath1 asymmetry generation . the @xmath18 baryon number violating operators of eq . ( [ dim7 ] ) can arise by integrating out the @xmath200 and @xmath418 gauge bosons of @xmath0 ( see fig . [ gauge ] ) , which lie outside of the @xmath10 subgroup . the decay @xmath419 and the conjugate decay @xmath420 can produce a primordial @xmath1 asymmetry at the gut scale , which would survive down to low temperatures without being washed out by the sphaleron interactions . level decay diagram and the one loop correction are shown in fig . [ baryo3 ] . the vector gauge boson @xmath204 has two fermion decays into the following channels : @xmath421 for the charge @xmath422 component , and @xmath423 for the charge @xmath424 component . these decays conserve @xmath1 , as can be seen by assigning @xmath425 . the gauge boson @xmath426 has the fermionic decays @xmath427 , suggesting that @xmath428 . the decay @xmath419 would then change @xmath1 by @xmath409 , as in the case of the scalar decay @xmath331 . the @xmath1 asymmetry arising from fig . [ baryo3 ] is found to be @xmath429{\rm br } , \label{asymv}\ ] ] with the function @xmath430 defined as @xmath431 here we have defined the lagrangian coefficient of the @xmath305 vertex to be @xmath432 . this is a consistent definition , since the lighter gauge boson @xmath207 would acquire a mass of order @xmath307 . here @xmath433 is a clebsch factor of order unity , with its value depending on the higgs representation used for rank reduction ( @xmath4 or @xmath5 ) . asymmetry in the decay of vector gauge boson @xmath204 . ] the factor @xmath356 that appears in eq . ( [ asymv ] ) is the branching ratio @xmath434 . it is determined in terms of the two partial widths as @xmath435 , where @xmath436 if we choose @xmath437 gev , @xmath438 ( with @xmath439 and @xmath440 ) , so that the dilution factor is @xmath441 . there is a modest suppression in @xmath442 arising from the branching ratio , since @xmath443 with this choice . we now show how the gut scale induced asymmetry in @xmath331 decay can consistently explain the observed value of @xmath444 , in a class of minimal @xmath0 models . in this class of models , a single @xmath75 and a single @xmath76 couple to fermions , as in eqs . ( [ yuk10])-([yuk126 ] ) . in susy models this is automatic with a single @xmath75 and @xmath76 employed . in non susy models the @xmath445 can also couple ( @xmath75 must be complexified to generate realistic fermion masses ) , however if a peccei quinn symmetry is assumed , the @xmath445 coupling would be absent . this class of models is highly constrained due to the small number of parameters that describe the fermion masses and mixings , and leads to predictions for the neutrino oscillation parameters @xcite . in addition to generating large mixing angles for solar neutrino oscillations and for atmospheric neutrino oscillations , these models predict a relatively large value of @xmath25 , viz . , @xmath446 , both in the non supersymmetric and the supersymmetric versions , consistent with recent results from daya bay and other experiments @xcite . to illustrate how realistic choice of parameters generate acceptable @xmath387 , we choose the @xmath102 field to be almost entirely in the @xmath75 . we also choose @xmath447 that appears in fig . [ baryo1 ] ( d ) to be small , so that the leading contribution to @xmath335 is from fig . [ baryo1 ] ( c ) , as given in eq . ( [ asymc ] ) . in this limit , we find @xmath448 here we have kept only the third family yukawa couplings , which is the leading contribution , and we have defined @xmath449 . choosing @xmath450 ( the top quark yukawa coupling at the gut scale ) , and @xmath451 , @xmath452 gev , @xmath453 ( so that @xmath454 gev , consistent with the light @xmath455 mass arising via the seesaw mechanism ) , @xmath456 , we find @xmath457 . if @xmath398 gev , then @xmath458 , @xmath459 so that the dilution factor is @xmath460 . this results in a net @xmath461 , consistent with observations . while natural choices of parameters can generate acceptable @xmath387 , due to the high sensitivity of dilution factor on the masses of the heavy particles , precise predictions are difficult to make . for @xmath419 decay we find the process to be typically out of equilibrium so that @xmath462 for @xmath463 gev . natural values of the asymmetry parameter in this case is @xmath464 . some dilution effects from the @xmath196 interactions would be welcome in this case . it should be mentioned that the @xmath18 operators of eq . ( [ lep ] ) also arise naturally in @xmath0 models , as already noted . the @xmath93 and the @xmath166 vertices arising from the @xmath4 couplings can be used for gut scale @xmath1genesis without generating @xmath18 nucleon decay operators . the decays @xmath465 has already been analyzed , but if these particles arise from @xmath4 they do not lead to @xmath1violating nucleon decay . in conclusion , we have pointed out that the complete set of @xmath18 baryon number violating operators that lead to the selection rule @xmath466 in nucleon decay can emerge as effective low energy operators in @xmath0 unified theories with either a @xmath4 or a @xmath5 higgs field used for breaking the @xmath14 gauge symmetry . the strength of these operators is unobservable in single step models where @xmath0 breaks directly down to the standard model . in non supersymmetric @xmath0 models , an intermediate symmetry is required in order for the gauge couplings to unify correctly . we have shown that in several instances with such an intermediate scale , the @xmath18 baryon number violating operators can lead to observable nucleon decay rates . the decay modes are distinct from the conventional gut motivated modes , and include @xmath467 , etc . we have also identified supersymmetric scenarios where such modes may be within reach of experiments , consistent with gauge coupling unification . a second major result of this paper is a new way of generating @xmath1 asymmetry in the early universe by the decay of gut mass particles . it is these particles which also induce the @xmath18 nucleon decay operators . such an asymmetry is sphaleron proof , in that it does not get erased by the effective interactions of the electroweak sphalerons . we present several examples where consistent asymmetry can be generated with the gut scale decays of particles obeying the @xmath468 selection rule . further , we show that in minimal @xmath0 models which explain the large neutrino mixing angles and predict relative large value for @xmath25 , consistent with recent experimental results , that the induced baryon asymmetry via the proposed gut scale mechanism is compatible with observations . there is thus a strong connection between neutrino oscillation parameters and baryon asymmetry in this class of models . the work of ksb is supported in part the us department of energy , grant numbers de - fg02 - 04er41306 and that of rnm is supported in part by the national science foundation grant number phy-0968854 . ksb acknowledges helpful discussions with j. julio . a. d. sakharov , pisma zh . fiz . * 5 * , 32 ( 1967 ) [ jetp lett . * 5 * , 24 ( 1967 ) ] . for reviews on gut scale baryogenesis before the discovery of the sphaleron and for original references see : p. langacker , phys . rept . * 72 * , 185 ( 1981 ) ; e. w. kolb and m. s. turner , ann . nucl . part . sci . * 33 * , 645 ( 1983 ) ; _ the early universe _ , by e. kolb and m. turner , ( frontiers in physics ) ( 1986 ) . k. s. babu , r. n. mohapatra , phys . lett . * 70 * , 2845 ( 1993 ) ; b. bajc , g. senjanovic and f. vissani , hep - ph/0110310 ; phys . rev . lett . * 90 * , 051802 ( 2003 ) ; t. fukuyama and n. okada , jhep * 0211 * , 011 ( 2002 ) ; h. s. goh , r. n. mohapatra , s. p. ng , phys b570 * , 215 ( 2003 ) ; k.s . babu , c. macesanu , phys . rev . * d72 * , 115003 ( 2005 ) ; s. bertolini , t. schwetz , m. malinsky , phys . rev . * d73 * , 115012 ( 2006 ) ; a. s. joshipura , k. m. patel , arxiv:1105.5943 [ hep - ph ] . c. s. aulakh , b. bajc , a. melfo , g. senjanovic , f. vissani , phys . * b588 * , 196 - 202 ( 2004 ) ; t. fukuyama , a. ilakovac , t. kikuchi , s. meljanac , n. okada , phys . * d72 * , 051701 ( 2005 ) ; c. s. aulakh and s. k. garg , nucl . b * 857 * , 101 ( 2012 ) . f. p. an _ et al . _ [ daya - bay collaboration ] , arxiv:1203.1669 [ hep - ex ] . k. abe _ et al . _ [ t2k collaboration ] , phys . 107 * , 041801 ( 2011 ) ; p. adamson _ [ minos collaboration ] , phys . lett . * 107 * , 181802 ( 2011 ) ; h. de . kerrect [ double chooz collaboration ] , talk at the lownu conference in seoul , korea ( 2011 ) , http://workshop.kias.re.kr/lownu11/?program . p. minkowski , phys . b * 67 * , 421 ( 1977 ) ; m. gell - mann , p. ramond , and r. slansky , in _ supergravity _ , eds . d. freedman _ et al . _ , ( north - holland , amsterdam , 1980 ) ; t. yanagida , in _ proceedings of the workshop on baryon number in the universe _ , eds . o. sawada and a. sugamoto , ( kek , 1979 ) ; r. mohapatra and g. senjanovi , phys . lett . * 44 * , 912 ( 1980 ) . d. chang , r. n. mohapatra , j. gipson , r. e. marshak and m. k. parida , phys . d * 31 * , 1718 ( 1985 ) ; r. n. mohapatra and m. k. parida , phys . d * 47 * , 264 ( 1993 ) ; n. g. deshpande , e. keith and p. b. pal , phys . rev . d * 46 * , 2261 ( 1993 ) ; s. bertolini , l. di luzio and m. malinsky , phys . d * 80 * , 015013 ( 2009 ) ; s. bertolini , l. di luzio and m. malinsky , arxiv:1202.0807 [ hep - ph ] . c. h. albright and s. m. barr , phys . d * 58 * , 013002 ( 1998 ) ; c. h. albright , k. s. babu and s. m. barr , phys . lett . * 81 * , 1167 ( 1998 ) ; c. h. albright and s. m. barr , phys . * 85 * , 244 ( 2000 ) .

we have performed detailed x - ray scattering measurements on single crystals of the spin - peierls compound tiobr in order to study the critical properties of the transition between the incommensurate spin - peierls state and the paramagnetic state at t 48 k. we have determined a value of the critical exponent which is consistent with the conventional 3d universality classes , in contrast with earlier results reported for tiobr and tiocl . using a simple power law fit function we demonstrate that the asymptotic critical regime in tiobr is quite narrow , and obtain a value of = 0.32 0.03 in the asymptotic limit . a power law fit function which includes the first order correction - to - scaling confluent singularity term can be used to account for data outside the asymptotic regime , yielding a more robust value of = 0.39 0.05 . we observe no evidence of commensurate fluctuations above t in tiobr , unlike its isostructural sister compound tiocl . in addition , we find that the incommensurate structure between t and t is shifted in * q*-space relative to the commensurate structure below t .

0908.2103

the physics of low dimensional quantum magnets is a very interesting and topical area of research . these systems exhibit a broad range of novel physical properties and exotic quantum ground states , ranging from spin - peierls and haldane gap behavior in quasi - one - dimensional systems to high temperature superconductivity in quasi - two - dimensional systems . spin - peierls systems in particular have undergone a recent renaissance due to the discovery of two new , and somewhat unconventional , inorganic spin - peierls compounds , the titanium oxyhalides tiobr and tiocl . spin - peierls behavior arises in quasi - one - dimensional ( 1d ) systems which possess a combination of spin 1/2 magnetic moments , short - range antiferromagnetic interactions , and strong magneto - elastic coupling . at low temperatures these systems undergo a spin - peierls phase transition which is characterized by dimerization of the lattice and the development of a non - magnetic singlet ground state . as the presence of interchain interactions typically drives quasi-1d systems towards long - range magnetic order , only a select family of compounds with particularly strong spin - phonon coupling have been found to exhibit a spin - peierls transition . the first experimental realization of the spin - peierls transition was discovered in organic charge transfer compounds , such as ttf - cubdt@xcite and mem-(tcnq)@xmath7@xcite . it was considerably later that spin - peierls behavior was observed in cugeo@xmath8 , the first inorganic spin - peierls compound ( t@xmath9 @xmath1 14 k)@xcite . tiobr and tiocl , the recently discovered inorganic spin - peierls compounds based on ti@xmath10 , are often referred to as unconventional " spin - peierls systems . unlike standard spin - peierls systems , which undergo a single continuous phase transition at t@xmath9 , the titanium oxyhalides undergo two successive phase transitions upon warming - a first - order transition at t@xmath6 between commensurate and incommensurate spin - peierls states , and a higher - order transition at t@xmath0 between the incommensurate spin - peierls state and a disordered pseudogap state which extends up to t * @xmath1 135 k@xcite . in addition , the titanium oxyhalides are unique among spin - peierls compounds because of their unusually high transition temperatures ( t@xmath6/t@xmath0 @xmath1 65 k/92 k in tiocl@xcite and 27 k/48 k in tiobr@xcite ) and surprisingly large singlet - triplet energy gap ( e@xmath11 @xmath1 430 to 440 k in tiocl@xcite and 150 k in tiobr@xcite ) . the origins and properties of the incommensurate spin - peierls state between t@xmath6 and t@xmath0 have attracted particular attention to these compounds . both tiobr and tiocl crystallize into a feocl - type structure , with an orthorhombic space group of _ pmmn _ at room temperature@xcite . this structure consists of buckled ti - o bilayers in the _ ab_-plane , which are separated by double layers of either br@xmath12 or cl@xmath12 ions and stacked along the _ c_-axis . the additional reduction of effective dimensionality within the ti - o bilayers , from two to one , is believed to arise from ordering of the ti@xmath10 orbital degrees of freedom@xcite . spin - peierls dimerization occurs along the crystallographic _ b_-axis , and is driven by the direct exchange of ti _ _ d__@xmath13 orbitals@xcite . high temperature susceptibility measurements suggest that both tiobr and tiocl can be fit to a s = 1/2 heisenberg chain model , with nearest - neighbor exchange couplings of @xmath1 660 k and 360 k respectively@xcite . in general , tiobr has attracted considerably less attention than its sister compound tiocl . this is largely due to the fact that it is more difficult to grow single crystals of tiobr than tiocl , although it is also more challenging to work with tiobr because it is extremely hygroscopic ( even more so than its isostructural counterpart ) and has substantially lower transition temperatures . all measurements to date seem to suggest that apart from a slight difference in energy scales , the two compounds behave almost identically - they have matching low temperature phase diagrams , and appear to be isostructural in each of the major phases . however , there are subtle differences between the lattice constants of the two materials , perhaps most notably in the ratio of the _ a _ and _ b _ lattice constants which describe the ti - o bilayers . while the room temperature lattice parameters reported for tiobr are _ a _ = 3.78 , _ b _ = 3.49 , and _ c _ = 8.53 @xcite , those of tiocl are _ a _ = 3.79 , _ b _ = 3.38 , and _ c _ = 8.03 @xcite . thus , while the _ a_-axis ( i.e. interchain ) spacing of the two compounds is almost identical , the _ b_-axis ( i.e. intrachain ) spacing is @xmath1 3 to 4% smaller in tiocl . as a result , both the relative and absolute strength of the intrachain interactions are expected to be greater in tiocl , giving rise to physical properties which are more strongly one - dimensional in nature . this appears to be reflected in both the magnitude of the nearest - neighbor exchange coupling and the size of the singlet - triplet energy gap . the critical behavior of spin - peierls systems is a subject which has generated considerable interest and confusion over the past thirty years . widely varying and often contradictory critical exponents have been reported for a number of spin - peierls compounds . early studies of cugeo@xmath8 , the first inorganic spin - peierls system , led to conflicting claims of tricritical@xcite , three - dimensional@xcite , and mean - field - like@xcite critical behavior . subsequent work@xcite has revealed that much of this discrepancy can be attributed to the narrowness of the asymptotic critical region , and it has been shown that the system belongs to a conventional 3d universality class . similarly , while the organic compounds mem-(tcnq)@xmath14 and ttf - cubdt were initially believed to exhibit mean - field - like critical behavior@xcite , more detailed measurements@xcite suggest that these compounds correspond to 3d universality classes as well . the critical behavior of the recently discovered ti - based spin - peierls compounds , tiobr and tiocl , has been less extensively studied but appears similarly enigmatic . to date there has been only one report of critical exponents for these systems , which placed the value of @xmath2 for the continuous transition at t@xmath0 ( between the disordered pseudogap state and the incommensurate spin - peierls state ) between 0.10 @xmath4 0.04 and 0.26 @xmath4 0.09 in tiobr@xcite . similar values of @xmath2 = 0.15@xcite and @xmath2 = 0.25@xcite can be extracted from other x - ray diffraction measurements on tiocl and tiobr , although these results have not been explicitly quoted as critical exponents by the authors . this range of @xmath2 values suggests some form of either quasi-2d or tricritical behavior at t@xmath0 , a result which is somewhat surprising for a phase transition which is known to be driven by inherently three - dimensional spin - phonon interactions . for two compounds which have already been shown to exhibit behavior which significantly differs from the standard spin - peierls scenario , this poses a very interesting question - is the critical behavior of these compounds truly distinct from that of cugeo@xmath8 and the organic spin - peierls systems or is 3d universality one of the fundamental properties of any spin - peierls transition ? the need for a detailed study of the critical behavior of these systems has been further underlined by a recent claim based on symmetry analysis@xcite that the transition at t@xmath0 is intrinsically first order in nature , despite the rather substantial weight of experimental evidence to the contrary@xcite . in this paper we report the first detailed study of the critical properties of tiobr . using an approach which incorporates both standard power law fits and modified power law fits which include a first order correction - to - scaling term , we find that the transition at t@xmath0 can be well described by conventional 3d universality classes . we offer a qualitative comparison which suggests that tiocl also corresponds to a similar universality class , and discuss the potential change in criticality which results when tiocl is doped with non - magnetic sc ions . our results indicate that the asymptotic critical regime in tiobr is quite narrow , providing a natural explanation for the anomalously low values of @xmath2 which have previously been reported in the literature . the ease with which our data can be fit to models of a continuous phase transition clearly contradicts arguments for the first order nature of the spin - peierls transition at t@xmath0@xcite . in addition , we offer evidence that commensurate fluctuations , which inhabit the incommensurate spin - peierls and pseudogap phases of tiocl , are absent in tiobr . we propose that this may reflect the weaker one - dimensionality of the system , and the greater importance of interchain interactions . finally , we observe that the incommensurate and commensurate spin - peierls states in tiobr appear to be shifted in * q*-space with respect to each other . this may be an indication that the structure of the incommensurate spin - peierls state is more complex than previously assumed . single crystal samples of tiobr were prepared using the chemical vapor transport method . a more detailed description of the crystal growth procedure has been reported elsewhere@xcite . the dimensions of the sample used in this experiment were approximately 4.0 @xmath15 3.0 @xmath15 0.5 mm . the sample was mounted on the cold finger of a closed cycle refrigerator and aligned within a huber four circle diffractometer . this sample environment provided a base temperature of @xmath1 7 k , with temperature stability of better than @xmath4 0.005 k. x - ray scattering measurements were performed using cu - k@xmath16 radiation ( @xmath17 = 1.54 ) produced by an 18 kw rotating anode x - ray source with a vertically focussing pyrolytic graphite monochromator . x - ray scattering scans were performed through several equivalent commensurate ordering wave vectors of the form ( h , k+1/2 , l ) , where superlattice bragg peaks are expected to arise as the result of spin - peierls dimerization . due to the thickness of our sample and the relatively short penetration depth of the wavelength being employed ( @xmath1 20 @xmath18 m ) , these measurements were largely restricted to reflection geometry . ultimately , our measurements focused around the ( 0 , 0.5 , 6 ) position in reciprocal space , as it was here that the intensity of the superlattice bragg peaks was found to be strongest . scans were carried out along the h , k , and l directions in reciprocal space in order to characterize all of the relevant commensurate and incommensurate superlattice peaks . these measurements were repeated at several temperatures within each of the four major regions of the tiobr phase diagram - below t@xmath6 in the commensurate spin - peierls phase , between t@xmath6 and t@xmath0 in the incommensurate spin - peierls phase , between t@xmath0 and t * in the pseudogap phase , and above t * in the paramagnetic phase . once the general features of the scattering at the superlattice peak positions had been well characterized , a more detailed study of the incommensurate scattering at ( h @xmath4 @xmath19 , k+1/2 , l ) was performed close to t@xmath0 in order to investigate the critical behavior of the system . in particular , a series of scans were taken through the incommensurate ordering wave vector at ( -@xmath19 , 0.5 , 6 ) to follow the temperature evolution of the incommensurate scattering intensity . to ensure that these measurements captured all of the relevant incommensurate scattering , two long scans were performed at every temperature - one scan in the h - direction to locate the center of the incommensurate satellite peak , followed by a second scan in the k - direction to obtain a measure of the integrated scattering intensity . measurements were performed over a range of temperatures extending from @xmath1 11 k below the phase transition to @xmath1 3 k above the transition . the size of the temperature steps varied from as large as 0.25 k ( more than 6 k away from t@xmath0 ) to as small as 0.05 k ( within 1 k of t@xmath0 ) . in order to check for the presence of thermal hysteresis at t@xmath0 , two separate data sets were collected while warming and cooling through the transition temperature . as no observable difference could be found between the measurements obtained on warming and cooling , both sets of measurements will be treated as a single data set in the analysis which follows in section iii.a . figure 1 shows two color contour plots , consisting of h - scans collected for tiobr ( left panel ) and tiocl ( right panel ) . each of these maps is composed of a series of h - scans performed through one of the superlattice peak positions - the ( 0 , 0.5 , 6 ) position in the case of tiobr , and the ( 2 , 1.5 , 1 ) position in the case of tiocl . note that in the case of tiocl a high temperature background data set , collected at t = 150 k , has been subtracted from each scan . this correction is necessary in order to eliminate weak @xmath17/2 contamination present in the incident x - ray beam , which can give rise to bragg scattering at the superlattice peak positions . the @xmath17/2 scattering at ( 2 , 1.5 , 1 ) in tiocl is comparable in strength to the weak @xmath17 scattering in the pseudogap phase ( @xmath1 0.1 counts / s ) , but it can be eliminated with a simple background subtraction due to its lack of temperature dependence . no higher order wavelength contamination was observed at the ( 0 , 0.5 , 6 ) position in tiobr , so a corresponding high temperature background subtraction has not been performed . the color contour maps in fig . 1 offer a qualitative summary of the main features of the low temperature phase diagram of the titanium oxyhalides . below t@xmath6 there is a dimerized , commensurate two - fold superstructure , identified by a single superlattice peak at ( h , k+1/2 , l ) . between t@xmath6 and t@xmath0 two incommensurate satellite peaks appear at the ( h @xmath4 @xmath19 , k+1/2 , l ) positions , signifying the development of an incommensurately modulated version of the dimerized structure . the primary component of the incommensurate modulation lies along the h - direction in reciprocal space , or the _ a_-axis in real space , which means that it runs perpendicular to the dimerized chains of ti atoms in the spin - peierls state . note that additional commensurate scattering , which has been attributed to commensurate fluctuations@xcite , is visible between t@xmath6 and t * in tiocl but not in tiobr . this result will be discussed in greater detail in section iii.c . for the present time it is worth noting that , if anything , the physical picture for tiobr is simpler than that of its isostructural counterpart . , in the commensurate sp phase ] and ( @xmath20 , 0.5 , 6 ) [ above t@xmath6 , in the incommensurate sp phase ] in tiobr . the discontinuous drop in scattering intensity at t@xmath6 @xmath1 27 k is indicative of a first order phase transition , while the gradual decrease towards t@xmath0 is consistent with a continuous phase transition . ] tiobr is known to undergo two successive phase transitions at low temperatures - a continuous transition into an incommensurate spin - peierls state at t@xmath0 @xmath1 48 k , followed by a first order transition into a commensurate spin - peierls state at t@xmath6 @xmath1 27 k. these two transitions are clearly illustrated in fig . 2 , which shows the temperature dependence of the integrated scattering intensity in the vicinity of the ( 0 , 0.5 , 6 ) superlattice peak position . below t@xmath6 the scattering intensity arises from the commensurate superlattice peak at ( 0 , 0.5 , 6 ) , while between t@xmath6 and t@xmath0 it originates from the incommensurate satellite peaks at ( @xmath4 @xmath19 , 0.5 , 6 ) . all of the scattering intensities provided in fig . 2 were obtained by integrating over h - scans of the form shown in fig . it should be noted that the intensity of the scattering which results from spin - peierls dimerization is proportional to the square of the relevant atomic displacements , and hence can be directly related to the square of the order parameters associated with the phase transitions at t@xmath6 and t@xmath0 . for this reason , the discontinuous jump in scattering intensity at t@xmath6 is evidence for a first order phase transition , while the gradual decrease towards t@xmath0 is indicative of a continuous phase transition . both of these results are consistent with the generally accepted nature of the transitions at t@xmath6 and t@xmath0 . , 0.5 , 6 ) as a function of temperature . these fits were performed using all data collected above t@xmath21 = 43 k , employing both a straight power law fit function ( blue , @xmath3 = 0.32 ) and a power law including the first order correction - to - scaling confluent singularity term ( red , @xmath5 = 0.39 ) . in the inset the quality of these fits is shown on a logarithmic scale . ] since the intensity of the dimerization scattering is proportional to the square of the order parameter , it is possible to extract a value for the critical exponent @xmath2 from our x - ray diffraction data . as only the phase transition at t@xmath0 is continuous , we will focus our attention on the critical behavior associated with the transition between the incommensurate spin - peierls phase and the disordered pseudogap phase . in this case , the order parameter for the phase transition will be related to the intensity of the incommensurate satellite peaks which appear at the ( h @xmath4 @xmath19 , k + 1/2 , l ) wave - vectors . the integrated scattering intensity of the ( -@xmath19 , 0.5 , 6 ) incommensurate peak , which was found to be the strongest of the experimentally accessible incommensurate peaks , was carefully determined by performing a series of h and k - scans as described in section ii . the integrated scattering intensity was then fit to two different functional forms , a straight power law ( eq . 1 ) and a modified power law which includes the first order correction - to - scaling confluent singularity term ( eq . 2 ) . @xmath22 @xmath23 here @xmath24 is the reduced temperature , while the value of the exponent @xmath25 has been determined to be @xmath1 0.5@xcite . representative fits to the integrated intensity data using eqs . 1 and 2 are provided in fig . the choice of two fit functions was motivated by previous studies on the critical behavior of cugeo@xmath8@xcite which showed that the asymptotic critical region was extremely narrow , extending less than 0.5 k ( or @xmath1 4% ) below t@xmath9 . in this region the length scale associated with fluctuations in the order parameter is greater than any of the other relevant length scales that occur in the system . within the limit of the asymptotic critical regime it is expected that the order parameter should be accurately described by the simple power law provided in eq . 1 . farther away from the transition temperature however , outside the asymptotic critical regime , other length scales in the system become significant and the behavior of the order parameter begins to deviate from this ideal form . it is this effect which was responsible for the wide variation between the initially reported critical exponents for cugeo@xmath8 , and we propose that the same holds true for tiobr . in order to account for deviations from standard power law behavior farther away from t@xmath0 , we can modify our original fit function to include the first order correction - to - scaling confluent singularity term , as shown in eq . it should be noted that this correction is only the first term in a larger series expansion which describes the behavior of the order parameter beyond the limits of the asymptotic critical regime@xcite . and goodness of fit parameter @xmath26 on the transition temperature t@xmath0 . the bottom panel contains fit parameters obtained using a straight power law fit function ( as described by eq . 1 ) , while the top panel shows parameters obtained from fits to a power law which includes the first order correction - to - scaling term ( as described by eq . both sets of fits were performed using data collected above t@xmath21 = 43 k. ] one of the factors that complicates the accurate determination of critical exponents is the uncertainty inherent in simultaneously fitting both the critical exponent @xmath2 and the transition temperature t@xmath0 . as shown in fig . 4 , these two parameters are intimately related , and the optimal value of @xmath2 is strongly dependent on the appropriate choice of t@xmath0 . this situation can be even further complicated by the presence of critical fluctuations , which are typically found in the vicinity of a continuous phase transition and can obscure the precise value of the transition temperature . fortunately , as in the case of cugeo@xmath8@xcite , the critical fluctuations around t@xmath0 in tiobr are quite difficult to observe . this is illustrated by the data provided in fig . 3 , which clearly shows a sharp drop in scattering intensity at t@xmath0 with very little broadening or rounding near the transition . this makes tiobr an excellent candidate for studies of critical behavior , and should allow for a particularly accurate determination of @xmath2 and t@xmath0 . as a function of t@xmath21 , the start of the fitting range . the black circles represent values obtained from fits to a straight power law ( eq . 1 ) , while the red squares represent values obtained from fits to a power law which includes the first order correction - to - scaling term ( eq . the predicted values of @xmath2 for the tricritical and conventional 3d universality classes have been provided as a guide to the eye . ] the importance of the correction - to - scaling term is illustrated by the data provided in fig . 5 , which shows the effect of fitting range on the value of the critical exponents extracted from eqs . 1 and 2 . here the fitting range is described in terms of the initial temperature , t@xmath21 , which represents the start of the fitting range and the lower temperature bound for the data points included in the fit . since the upper bound for each fitting range was fixed at a final temperature of t@xmath27 = 51 k , it follows that t@xmath21 can be taken as both a measure of the size of a given fitting range and an indication of how far a particular data set extends below the transition temperature t@xmath0 . thus higher values of t@xmath21 correspond to smaller fitting ranges , and describe fits to data which falls within ( or at least close to ) the asymptotic critical regime . the exponents determined from fits to the straight power law of eq . 1 are shown by the black circles in fig . 5 . note that these exponents follow a clear trend , gradually increasing as t@xmath21 rises and the fitting range is reduced . closer to the transition temperature ( i.e. above t@xmath21 @xmath1 43 k ) the values of @xmath2 appear to level off , approaching an asymptotic limit of @xmath3 = 0.32 @xmath4 0.03 . this behavior suggests that the asymptotic critical regime in tiobr is quite narrow , extending less than 5 k ( or @xmath1 9% ) below t@xmath0 . within the critical region , where the critical behavior of the system can accurately be described by a standard power law , eq . 1 provides consistent and reliable @xmath2 values . over larger fitting ranges , for data sets which extend beyond the critical region , deviations from simple power law behavior become increasingly significant and fits to eq . 1 yield progressively worse values of @xmath2 . hence @xmath3 represents the most accurate value of @xmath2 that can be obtained from eq . 1 , taken in the appropriate asymptotic limit close to t@xmath0 . the critical exponents obtained from fits to the modified power law of eq . 2 are illustrated by the red squares in fig . 5 . in contrast to the results of the straight power law fits , the values of @xmath2 obtained from fits to eq . 2 follow a relatively constant trend , with slow variation around @xmath5 = 0.39 @xmath4 0.05 . thus we find that the introduction of the correction - to - scaling term compensates for deviations from standard power law behavior beyond the asymptotic critical regime , dramatically reducing the effect of fitting range on the resulting critical exponents . ideally , one might expect that sufficiently close to t@xmath0 the values of @xmath2 from the two fit functions should converge . in practice , the quality and reliability of fits performed when t@xmath21 is within @xmath1 2 k of the phase transition are somewhat limited by the finite size of the data sets , which grow progressively smaller as t@xmath21 increases . however , bearing this restriction in mind , it is very encouraging to note that the values obtained for @xmath3 from eq . 1 and @xmath5 from eq . 2 are fully consistent within the bounds of experimental uncertainty . from these results , we can conclude that the critical behavior of tiobr can be described by one of the conventional 3d universality classes - 3d ising ( @xmath2 = 0.326 ) , 3d xy ( @xmath2 = 0.345 ) or 3d heisenberg ( @xmath2 = 0.367 ) . both @xmath3 , the result obtained from straight power law fits performed in the appropriate limit , and @xmath5 , the result obtained from modified power law fits which included a correction - to - scaling term , are consistent with these values . due to the relatively large size of our error bars , it is difficult to narrow down the list of potential universality classes to suggest a specific symmetry for the order parameter . one might only note that the best fits to the data are provided by the modified power law of eq . 2 , which leads to a critical exponent @xmath5 indicative of a 3d universality class with continuous symmetry , either xy or heisenberg . this would be consistent with previous findings for conventional spin - peierls systems such as cugeo@xmath8@xcite and mem-(tcnq)@xmath7@xcite . it should be noted that the success of our analysis of the critical behavior in tiobr provides extremely compelling evidence that the phase transition at t@xmath0 is continuous , rather than first order , in nature . the fact that the integrated intensity of the superlattice peaks can be fit so well using a continuous model strongly contradicts the recent claims of schonleber et al@xcite , while lending support to earlier findings based on magnetic susceptibility@xcite , nmr@xcite , specific heat@xcite , thermal expansion@xcite , and x - ray diffraction measurements@xcite . there can be no confusing the results of the present study with the previously reported critical exponents of @xmath2 = 0.10 to 0.26@xcite , 0.15@xcite , and 0.25@xcite for tiobr and tiocl . however , our examination of the relationship between the fitting range employed and the resulting value of @xmath2 provides a simple and straightforward explanation for how these anomalously low critical exponents may have come about - they are the result of fits performed to measurements collected outside the asymptotic critical regime . as a function of t@xmath21 , the start of the fitting range . the black circles correspond to fits performed to a straight power law fit function ( eq . 1 ) , while the red squares correspond to fits to a power law which includes the first order correction - to - scaling term ( eq . the goodness of fit parameter , @xmath26 , is plotted as a function of fitting range in fig . similar to fig . 5 , this plot serves to emphasize the importance of considering correction - to - scaling terms when dealing with data sets which extend beyond the asymptotic critical regime . the most striking difference between the values of @xmath26 obtained from eqs . 1 and 2 emerges over large fitting ranges ( i.e. for low values of t@xmath21 ) . for t@xmath21 less than 43 k , as the fitting range grows larger one can observe a gradual monotonic increase in the @xmath26 values from the straight power law fits . again , this reflects the fact that outside the asymptotic critical regime a simple power law form is no longer sufficient to describe the behavior of the order parameter . in contrast , the @xmath26 values from fits to the modified power law which incorporates the first order correction - to - scaling term remain small , and relatively constant , down to the lowest values of t@xmath21 . for small fitting ranges ( i.e. for t@xmath21 greater than 44 k ) , the quality of fit decreases markedly for both fit functions . this is simply a result of the reduced number of data points available for fits at higher t@xmath21 , which ultimately becomes the limiting factor for the accuracy of the fits in this regime . more generally , it is important to note that for any value of t@xmath21 the goodness of fit is always better when the data is fit to eq . 2 rather than eq . 1 . in other words , the correction - to - scaling fits consistently offer a better description of the data than a straight power law , regardless of the fitting range chosen . @xmath19 , 1.5 , 1 ) superlattice peak position in tiocl ( top ) and ti@xmath28sc@xmath29ocl ( bottom ) . these intensities have been obtained by integrating over h - scans of the form shown in the colour contour maps of figure 1 . the tiocl data set can be well described by a straight power law fit which is consistent with conventional 3d universality ( @xmath2 = 0.34 , shown in blue ) , while the ti@xmath28sc@xmath29ocl data set is much better fit by a power law function consistent with mean - field - like behavior ( @xmath2 = 0.50 , shown in red ) . ] having completed our analysis of the critical properties of tiobr , it is natural that we next turn our attention to what our results can suggest about the critical behavior of tiocl . while we lack sufficient data to perform the analysis necessary to precisely determine the value of @xmath2 for tiocl , we can clearly demonstrate that the critical exponents determined in the previous section provide a very convincing description of recently reported measurements on tiocl@xcite . as illustrated by the colour contour maps provided in fig . 1 , the low temperature phase diagram of tiocl is considerably more complicated than that of tiobr . in particular , tiocl is distinguished by the presence of commensurate fluctuations which persist from t@xmath6 up to t * , throughout both the incommensurate spin - peierls and pseudogap phases . however , since the order parameter associated with the transition at t@xmath0 is strictly related to the intensity of the incommensurate satellite peaks , rather than the sum of all dimerization scattering , it is only the incommensurate scattering at ( h @xmath4 @xmath19 , k+1/2 , l ) which must be considered when studying the critical behavior of tiocl . the upper panel of fig . 7 shows the temperature dependence of the integrated scattering intensity of the ( 2 @xmath4 @xmath19 , 1.5 , 1 ) incommensurate satellite peaks in tiocl . these intensities were obtained by integrating over small * q*-ranges around the incommensurate ordering wave vectors in h - scans similar to those used to construct the colour contour maps of fig . the data were fit to a standard power law , as described by eq . 1 , although due to the low density of points near t@xmath0 the value of the transition temperature was fixed at 92 k. this fixed value of t@xmath0 is consistent with previously reported susceptibility@xcite , heat capacity@xcite , nmr@xcite , and x - ray diffraction measurements@xcite for tiocl . the resulting line of best fit ( indicated by the solid blue line in the upper panel of fig . 7 ) corresponds to a critical exponent of @xmath2 = 0.34 @xmath4 0.04 . it should be noted that this value of @xmath2 is consistent with the theoretical predictions of the 3d ising , 3d xy and 3d heisenberg models , indicating that the transition in tiocl can be well described by any of the conventional 3d universality classes . furthermore , there is excellent agreement between the value of @xmath2 = 0.34 @xmath4 0.04 for tiocl and the value of @xmath3 = 0.32 @xmath4 0.03 obtained earlier from power law fits to the tiobr data . in fact , this relatively simple estimate for @xmath2 also agrees quite well with the value of @xmath5 = 0.39 @xmath4 0.05 obtained for tiobr using the modified power law of eq . 2 , in spite of the fact that the large fitting range used for the tiocl data set almost certainly extends well beyond the asymptotical critical regime . unfortunately , due to the small size and low point density of the tiocl data set , it is difficult to extract any meaningful parameters from fits performed using the modified power law described by eq . 2 . on a more qualitative level however , it is worth noting that the function in eq . 2 can fit the tiocl data very well using a fixed transition temperature of t@xmath0 = 92 k and the critical exponent of @xmath5 = 0.39 determined for tiobr . by eye , such a fit is almost indistinguishable from the solid blue curve which represents the best fit line for eq . 1 , hence it has been omitted from the upper panel of fig . a very interesting comparison can also be made using previous x - ray data collected on ti@xmath28sc@xmath29ocl , a compound in which sc - doping has been shown to suppress commensurate spin - peierls order to below at least t @xmath1 7 k@xcite . the temperature dependence of the integrated intensity of the incommensurate scattering at ( 2 @xmath4 @xmath19 , 1.5 , 1 ) in ti@xmath28sc@xmath29ocl is provided in the bottom panel of fig . 7 . if we attempt to fit this data using the simple power law fit function described by eq . 1 we find that the critical behavior of sc - doped tiocl is clearly different from that of either pure tiocl or tiobr . the best power law fit to the ti@xmath28sc@xmath29ocl data ( illustrated by the solid red line in the bottom panel of fig . 7 ) corresponds to a critical exponent of @xmath2 = 0.50 @xmath4 0.07 and a transition temperature of t@xmath0 = 109 @xmath4 6 k. this value of @xmath2 is consistent with the exponent predicted by mean field theory ( @xmath2 = 0.5 ) , and is incompatible with the theoretical predictions for any of the conventional 3d universality classes . in addition , the value of @xmath2 for ti@xmath28sc@xmath29ocl is dramatically different from the critical exponents that we determined for tiocl and tiobr . this difference in critical behavior is evident even from the raw data provided in fig . the intensity of the incommensurate scattering in ti@xmath28sc@xmath29ocl appears to grow linearly with decreasing temperature , while the scattering in tiocl clearly follows a sublinear power law trend . it is also instructive to compare these results to previous studies on the critical properties of cugeo@xmath30@xcite . it has been well established that in both pure cugeo@xmath30 and cugeo@xmath30 doped with ions of similar ionic radius the critical behavior of the spin - peierls transition can be described by one of the conventional 3d universality classes with continuous symmetry ( such as 3d xy or 3d heisenberg ) . however , when cugeo@xmath30 is doped with ions of larger ionic radius , by substituting cd@xmath31 for cu@xmath31 or ga@xmath32 for ge@xmath32 for example , the observed universality class is found to change towards mean - field - like behavior . this effect appears to have little or no concentration dependence , and is believed to be related to the presence of local strain fields induced by the larger dopant ions . the significance of local strain effects on the behavior of the order parameter is reasonably intuitive given both the magneto - elastic nature of the spin - peierls transition and the underlying importance of lattice properties in the spin - peierls materials . hence , if the dopant ion is smaller or of comparable size to the ion it is replacing ( i.e. if @xmath33 is less than roughly + 3% of the ionic radius ) the universality class of the transition remains unchanged . alternatively , for large , positive changes in ionic radius ( i.e. @xmath33 is on the order of + 9% or more ) there appears to be a distinct change in critical properties . the difference in ionic radius between sc@xmath10 and ti@xmath10 corresponds to @xmath33 = + 11% . thus , the change in critical behavior we observe between tiocl and ti@xmath28sc@xmath29ocl appears fully consistent with lessons learned from studies of doped cugeo@xmath8@xcite . although further study is still required to provide a more detailed understanding of this phenomenon , on the basis of these results it is certainly tempting to conclude that doping - induced changes in criticality are a fundamental property associated with the spin - peierls transition . a second significant feature which arises from the comparison of the tiocl and ti@xmath28sc@xmath29ocl data sets is the change in transition temperature , t@xmath0 , which is induced by sc - doping . as shown by fig . 7 , the value of t@xmath0 is considerably higher in ti@xmath28sc@xmath29ocl than in tiocl , with the fitted value of t@xmath0 = 109 @xmath4 6 k for the doped sample lying well above the accepted value of 92 @xmath4 1 k for the pure sample . this result is quite surprising , particularly when one considers that doping has been universally found to suppress the spin - peierls transition temperature in cugeo@xmath8 . one possible explanation for this effect is that since no commensurate fluctuations are observed below t * @xmath1 135 k in ti@xmath28sc@xmath29ocl , the t @xmath1 109 k temperature scale actually corresponds to the appropriately lowered equivalent of t * from tiocl . this argument is supported by the fact that the incommensurate scattering in ti@xmath28sc@xmath29ocl is very broad , implying that the length scale for correlations in the doped compound is even shorter than that of the commensurate fluctuations in pure tiocl . in addition , the @xmath1 20 % drop between the 135 k and 109 k temperature scales is quite consistent with the 15 to 45 % suppression of t@xmath9 per 1 % concentration of dopants which has been reported in the literature for doped cugeo@xmath8@xcite . as discussed earlier , there are several respects in which the low temperature scattering of tiobr differs from that of its isostructural counterpart tiocl . the chief feature which distinguishes these two compounds from each other is the presence of commensurate fluctuations , which are clearly evident in tiocl@xcite but appear to be absent in tiobr . in tiocl , these commensurate fluctuations can be observed throughout both the incommensurate spin - peierls phase ( from t@xmath6 to t@xmath0 ) and the pseudogap phase ( from t@xmath0 to t * ) , as shown in the right panel of fig . the commensurate scattering is not resolution - limited above t@xmath6 , indicating that it does not correspond to static long - range order . however , the width of the scattering implies the presence of relatively long correlation lengths , on the order of @xmath1 100 or more . interestingly , between t@xmath6 and t@xmath0 the commensurate fluctuations appear to both coexist with and compete against the incommensurate scattering at ( h @xmath4 @xmath19 , k + 1/2 , l ) . the commensurate fluctuations also appear to be shifted in * q*-space with respect to both the accompanying incommensurate scattering and the long - range ordered commensurate state below t@xmath6 , with observed peak positions displaced along both the h and k directions . in tiobr , we find no indication of commensurate fluctuations in either the incommensurate spin - peierls state ( between t@xmath6 and t@xmath0 ) or the pseudogap state ( between t@xmath0 and t * ) . this was the case at each of the 7 different commensurate ordering wave - vectors investigated in this study . one possible explanation for the absence of commensurate fluctuations in tiobr may arise from the subtle differences between the structures of tiocl and tiobr . while the distance between neighboring dimerized chains of ti atoms ( i.e. the interchain spacing , related to the _ a _ lattice parameter ) is almost identical in both two compounds , the distance between neighbouring ti atoms on the same dimerized chain ( i.e. the intrachain spacing , given by the _ b _ lattice parameter ) is approximately 4 % larger in tiobr . as a result of this change in lattice parameters , tiobr appears to be less one - dimensional than its isostructural counterpart . the difference is clearly evident from the magnetic susceptibility measurements reported for the two compounds@xcite , both in terms of the magnitude of the nearest - neighbor exchange coupling ( which is estimated to be @xmath1 45 % smaller in tiobr ) and the observed deviations from the idealized bonner - fisher curve ( which suggest significantly stronger interchain couplings in tiobr ) . doping studies on tiocl@xcite indicate that the relative strengths of the intrachain and interchain interactions are intimately connected with the stability of the incommensurate structure . thus , if the interchain coupling is stronger with respect to the intrachain coupling in tiobr than it is in tiocl then it could easily lead to suppression of commensurate fluctuations ( as in ti@xmath28sc@xmath29ocl ) , stabilization of the incommensurate structure , and a loss of phase coexistence between t@xmath6 and t@xmath0 . it should be emphasized that the lack of observable commensurate fluctuations in the present study does not necessarily imply that such scattering is completely absent in tiobr . in fact , the difficulty in detecting such a weak , broad feature - particularly one which may be slightly shifted in * q*-space - is clearly underscored by the number of x - ray diffraction studies which previously failed to observe such commensurate fluctuations in tiocl . it should also be noted that the ( 2 , 1.5 , 1 ) wave - vector , where the commensurate scattering was found to be strongest in tiocl , was experimentally inaccessible in this study . thus , there remains a possibility that stronger commensurate fluctuations could still be found in tiobr at other , as of yet unstudied , positions in reciprocal space . further work is needed to provide a more thorough understanding of this intriguing difference between the tiox spin - peierls systems . , 0.5 , 6 ) superlattice peak in tiobr ( top ) . the black circles correspond to data collected within the commensurate spin - peierls state ( t = 20 k , @xmath19 = 0 ) , while the red squares correspond to data collected within the incommensurate spin - peierls state ( t = 45 k , @xmath19 @xmath1 0.08 ) . for illustrative purposes the intensity of the t = 45 k data set has been scaled by a factor of four . the solid lines represent fits to a simple two - gaussian function . note that the peaks in the incommensurate phase are shifted in * q*-space relative to those of the commensurate phase . the temperature dependence of the shift , @xmath34 , is shown in the bottom panel , along with the h - component of the incommensurate modulation wave vector , @xmath19 . ] our measurements in the incommensurate spin - peierls phase also reveal an interesting property of the incommensurate modulation in tiobr . in addition to the primary component of the incommensurate modulation wave - vector which runs along the h direction ( perpendicular to the direction of the dimerized chains ) , a smaller secondary component of the modulation has been reported along the k direction ( parallel to the dimerized chains ) in both tiocl and tiobr . the magnitude of the k - component of the incommensurate modulation , @xmath34 , increases monotonically between t@xmath6 and t@xmath0 , reaching a maximum value of @xmath1 0.015 r.l.u . in both compounds@xcite . even though the experimental resolution of our present study is too broad to fully resolve the distinct incommensurate satellite peaks which should arise from the k - component of the incommensurate modulation , we would anticipate that the effects of this secondary incommensuration should be manifested as a significant broadening of the superlattice peaks along the k - direction . however , as the representative k - scans provided in the upper panel of fig . 8 clearly demonstrate , there is no observable change in the width of the ( -@xmath19 , 0.5 , 6 ) superlattice peak between the commensurate and incommensurate spin - peierls phases . what these k - scans do reveal is an obvious shift in the position of the incommensurate satellite peaks above t@xmath6 relative to the commensurate superlattice peaks below t@xmath6 . the temperature dependence of this shift in * q*-space , which we will call @xmath34 , is illustrated in the bottom panel of fig . note that superlattice reflections are observed at the ( @xmath4 @xmath19 , 0.5 + @xmath34 , 6 ) positions , but not at ( @xmath4 @xmath19 , 0.5 - @xmath34 , 6 ) . in this respect , our k - shift clearly differs from the previously reported k - incommensuration , which when coupled with the incommensuration along h should produce a set of 4 incommensurate peaks at ( @xmath4 @xmath19 , 0.5 @xmath4 @xmath34 , 6 ) . furthermore , there are indications that this k - shift exhibits some form of systematic * q*-dependence . at the ( 0 , 0.5 , 6 ) , ( 0 , -0.5 , 6 ) , and ( 1 , 0.5 , 6 ) positions in reciprocal space we find the incommensurate peaks shift to higher * q * positions ( i.e. to k + @xmath35 ) , while for ( 0 , 0.5 , 5 ) and ( 1 , 0.5 , 5 ) we find the peaks shift to lower q ( i.e. to k - @xmath35 ) . at the ( 0 , 1.5 , 6 ) wave - vector no discernable k - shift could be observed . based on alignment scans performed through several nearby structural bragg peaks we can conclude that this shift in * q*-space is not simply the result of a change in the lattice constants of the material . interestingly , the shift in * q*-space appears to have a different temperature dependence than the h - component of the incommensurate modulation wave - vector , as shown in the bottom panel of fig . 8 . while @xmath19 and @xmath34 both increase steadily with temperature above t@xmath6 , @xmath19 approaches saturation by @xmath1 35 k , while @xmath34 continues to grow until the system reaches t@xmath0 . if the observed k - shift was simply equivalent to the k - component of the incommensurate modulation wave - vector it seems surprising that it would not follow the same temperature evolution as @xmath36 . this effect has also been observed in recent x - ray measurements performed by sasaki et al.@xcite , which show that the incommensurate scattering at ( @xmath4 @xmath19 , 2.5 , 0 ) is shifted in the k - direction ( towards lower * q * ) with respect to the commensurate scattering at ( 0 , 2.5 , 0 ) . furthermore , the magnitude and temperature dependence of this shift in * q*-space are almost identical to that reported here . curiously , in spite of the absence of additional satellite peaks at ( @xmath4 @xmath19 , 2.5 + @xmath34 , 0 ) , sasaki et al . treat this behavior as equivalent to the k - incommensuration reported by van smaalen et al.@xcite . this assumption is not completely unjustified , particularly since recent x - ray diffraction measurements on tiocl@xcite have shown that in some cases there can be dramatic differences between the observed scattering intensities of neighbouring pairs of incommensurate satellite peaks . these intensity variations have been attributed to changes in structure factor resulting from phase shifts between the modulations of neighboring chains of ti atoms in the unit cell . in tiocl these intensity differences have been found to be up to one to two orders of magnitude in size@xcite . if the scattering at the ( @xmath4 @xmath19 , 0.5 - @xmath34 , 6 ) position in tiobr is suppressed by a similar factor then it could be too weak to be detectable with our current experimental configuration . however , while it is possible that weak superlattice scattering could be obscured by experimental background in the present study , it is less likely that this same problem would also have affected the results of sasaki et al . due to the much higher signal - to - noise ratio available at synchrotron x - ray sources . an alternative possibility , which naturally arises from the parallels between tiocl and tiobr , is that the k - shift between the commensurate and incommensurate scattering in tiobr is an analog of the * q*-shift between the commensurate fluctuations and long - range ordered commensurate / incommensurate scattering in tiocl . the magnitude of the * q*-shift reported for the commensurate fluctuations in tiocl@xcite is very similar to the size of the k - shift we observe in tiobr , and it is clear that both of these features are intimately connected with the first - order phase transition at t@xmath6 . thus , the shift in * q*-space between the commensurate and incommensurate scattering in tiobr could represent a distinctly different effect from the k - component of the incommensurate modulation . in this case , the k - shift could reflect changes in the balance of interchain and intrachain interactions at t@xmath6 . hence , below and above t@xmath6 the ratio of relevant interaction strengths favors commensurate and incommensurate modulations of the dimerized structure with two slightly different wave - vectors . it is difficult to comment on the specific implications of these * q*-shifts without further , and more thorough , study of additional superlattice peak positions . however , this result may provide an indication that the structure of the incommensurate spin - peierls state of tiobr may be more complex than previously assumed . in conclusion , we have performed the first detailed study of the critical properties of tiobr . while the spin - peierls behavior of this system appears to be unconventional in a number of respects , the critical properties of the system can be very well described by conventional 3d universality classes . this finding is consistent with previous results obtained from studies performed on other spin - peierls systems such as mem-(tcnq)@xmath7@xcite and cugeo@xmath8@xcite . by analyzing our data using a simple power law fit function we determined that the critical exponent associated with the transition at t@xmath0 approaches a value of @xmath3 = 0.32 @xmath4 0.03 in the appropriate asymptotic critical regime . the strong dependence of @xmath2 on the size of the fitting range suggests that the asymptotic critical region in tiobr is quite narrow , as in cugeo@xmath8@xcite . this provides an explanation for the apparently contradictory critical exponents that have previously been reported for this compound . the narrowness of the critical regime can be accounted for by employing a modified power law fit function which includes the first order correction - to - scaling confluent singularity term . using this approach we obtain a value of @xmath5 = 0.39 @xmath4 0.05 , a result which is consistent with both the 3d xy and 3d heisenberg universality classes . based on a simple , primarily qualitative comparison , we argue that tiocl , the isostructural cousin of tiobr , can also be well described by one of the conventional 3d universality classes . more detailed studies would still be required in order to offer a more quantitative description of the critical behavior of tiocl , or to further narrow down the list of potential universality classes for the transition . however , this result appears to add growing weight to the notion that the spin - peierls transition is by nature a conventional three - dimensional phase transition , whether it is realized in organic or inorganic systems or even the unconventional " titanium oxyhalides . applying the same approach to measurements performed on ti@xmath28sc@xmath29ocl we find evidence that sc - doping induces a fundamental change in the critical properties of the transition at t@xmath0 . we suggest that this effect is likely caused by the presence of local lattice strains which arise due to the significantly larger ionic radius of the sc@xmath10 dopant ions . in addition , we describe several interesting features of the incommensurate structure of tiobr between t@xmath6 and t@xmath0 . our measurements reveal no evidence for commensurate fluctuations of the type observed in the incommensurate spin - peierls and pseudogap phases of tiocl . this result may reflect the weaker one - dimensional properties of tiobr , and in particular the substantially smaller size of the intrachain interaction . we also observe that the commensurate and incommensurate structures are shifted in * q*-space with respect to one another , with a displacement along the k - direction that steadily grows between t@xmath6 and t@xmath0 . we hope that these measurements will help to guide and inform future theoretical and experimental studies of these novel spin - peierls systems . bray , l.v . interrante , i.s . jacobs , and j.c . bonner in _ extended linear chain compounds _ , edited by j.s . miller ( plenum press , new york , 1983 ) , vol . 3 , p. 353 , and references therein . b. van bodegom , b.c . larson , and h.a . mook , phys . b * 24 * , 1520 ( 1981 ) ; m.d . lumsden and b.d . gaulin , phys . b * 59 * , 9372 ( 1999 ) and references contained therein . m. hase , i. terasaki , and k. uchinokura , phys . 70 * , 3651 ( 1993 ) ; j.p . pouget , l.p . regnault , m. ain , b. hennion , j.p . renard , p. veillet , g. dhalenne , and a. revcolevschi , phys . 72 * , 4037 ( 1994 ) . t. imai and f.c . chou , cond - mat/0301425 ( unpublished ) ; s.r . saha , s. golin , t. imai , and f.c . chou , j. phys . solids * 68 * , 2044 ( 2007 ) . a. seidel , c.a . marianetti , f.c . chou , g. ceder , and p.a . lee , phys . b * 67 * , 020405(r ) ( 2003 ) . a. krimmel , j. strempfer , b. bohnenbuck , b. keimer , m. hoinkis , m. klemm , s. horn , a. loidl , m. sing , r. claessen , and m. v. zimmermann , phys . b * 73 * , 172413 ( 2006 ) . r. ruckamp , j. baier , m. kriener , m.w . haverkort , t. lorenz , g.s . uhrig , l. jongen , a. moller , g. meyer , and m. gruninger , phys . lett . * 95 * , 097203 ( 2005 ) . j. hemberger , m. hoinkis , m. klemm , m. sing , r. claessen , s. horn , and a. loidl , phys . b * 72 * , 012420 ( 2005 ) . p. lemmens , k.y . choi , r. valenti , t. saha - dasgupta , e. abel , y.s . lee , and f.c . chou , new j. phys . * 7 * , 74 ( 2005 ) . t. sasaki , m. mizumaki , t. nagai , t. asaka , k. kato , m. takata , y. matsui , and j. akimitsu , j. phys . . jpn . * 74 * , 2185 ( 2005 ) . s. van smaalen , l. palatinus , and a. schonleber , phys . b * 72 * , 020105(r ) ( 2005 ) . p.j . baker , s.j . blundell , f.l . pratt , t. lancaster , m.l . brooks , w. hayes , m. isobe , y. ueda , m. hoinkis , m. sing , m. klemm , s. horn , and r. claessen , phys . b * 75 * , 094404 ( 2007 ) . h. schafer , f. wartenpfuhl , and e. weise , z. anorg . * 295 * , 268 ( 1958 ) . von schnering , m. collin , and m. hassheider , z. anorg . allg . chem . * 387 * , 137 ( 1972 ) . c. kato , y. kobayashi , and m. sato , j. phys . jpn * 74 * , 473 ( 2005 ) . harris , q. feng , r.j . birgeneau , k. hirota , k. kakurai , j.e . lorenzo , g. shirane , m. hase , k. uchinokura , k. kojima , i. tanaka , and y. shibuya , phys . b * 50 * , 12606 ( 1994 ) . m. fujita , m. arai , k. ubukata , h. ohta , m. mino , m. motokawa , k. knight , r. ibberson , j.b . forsyth , s.m . bennington , o. akimitsu , and j. fujita , physica b * 213 & 214 * , 288 ( 1995 ) . m. saint paul , p. monceau , and a. revolevschi , solid state commun . * 93 * , 7 ( 1995 ) . harris , q. feng , r.j . birgeneau , k. hirota , g. shirane , m. hase , and k. uchinokura , phys . rev . b * 52 * , 15420 ( 1995 ) . h. winkelmann , e. gamper , b. buchner , m. braden , a. revcolevschi , and g. dhalenne , phys . b * 51 * , 12884 ( 1995 ) . t. lorenz , u. ammerahl , t. auweiler , b. buchner , a. revcolevschi , and g. dhalenne , phys . b * 55 * , 5914 ( 1997 ) . r.j . birgeneau , v. kiryukhin , and y.j . wang , phys . b * 60 * , 14816 ( 1999 ) . lumsden , b.d . gaulin , h. dabkowska , and m.l . plumer , phys . . lett . * 76 * , 4919 ( 1996 ) ; m.d . lumsden , b.d . gaulin , and h. dabkowska , phys . b * 57 * , 14097 ( 1998 ) . s. huizinga , j. kommandeur , g.a . sawatzky , b.t . thole , k. kopinga , w.j.m . de jonge , and j. roos , phys . b * 19 * , 4723 ( 1979 ) . moncton , r.j . birgeneau , l.v . interrante , and f. wudl , phys . lett . * 39 * , 507 ( 1977 ) . lumsden and b.d . gaulin , phys . b * 59 * , 9372 ( 1999 ) . t. sasaki , m. mizumaki , t. nagai , t. asaka , k. kato , m. takata , y. matsui , and j. akimitsu , cond - mat/0509358 ( unpublished ) . abel , k. matan , f.c . chou , e.d . isaacs , d.e . moncton , h. sinn , a. alatas , and y.s . lee , phys . b * 76 * , 214304 ( 2007 ) . a. schonleber , g. shcheka , and s. van smaalen , phys . b * 77 * , 094117 ( 2008 ) . m. shaz , s. van smaalen , l. palatinus , m. hoinkis , m. klemm , s. horn , and r. claessen , phys . b * 71 * , 100405(r ) ( 2005 ) . clancy , b.d . gaulin , k.c . rule , j.p . castellan , and f.c . chou , phys . b * 75 * , 100401(r ) ( 2007 ) . a. aharony and g. ahlers , phys . * 44 * , 782 ( 1980 ) . j.c . le guillou and j. zinn - justin , phys . 39 * , 95 ( 1977 ) . clancy , b.d . gaulin , j.p . castellan , k.c . rule , and f.c . chou , phys . b * 78 * , 014433 ( 2008 ) lumsden , b.d . gaulin , and h. dabkowska , phys . b * 58 * , 12252 ( 1998 ) . m. hase , i. terasaki , y. sasago , k. uchinokura , and h. obara , phys . lett . * 71 * , 4059 ( 1993 ) . oseroff , s.w . cheong , b. aktas , m.f . hundley , z. fisk , and l.w . rupp , jr . lett . * 74 * , 1450 ( 1995 ) . schoeffel , j.p . pouget , g. dhalenne , and a. revcolevschi , phys . b * 53 * , 14971 ( 1996 ) . martin , m. hase , k. hirota , g. shirane , y. sasago , n. koide , and k. uchinokura , phys . b * 56 * , 3173 ( 1997 ) . b. grenier , j .- renard , p. veillet , c. paulsen , g. dhalenne , and a. revcolevschi , phys . b * 58 * , 8202 ( 1998 ) . k. manabe , h. ishimoto , n. koide , y. sasago , k. uchinokura , phys . b * 58 * , r575 ( 1998 ) anderson , j.z . liu , and r.n . shelton , phys . b * 56 * , 11014 ( 1997 ) . t. masuda , a. fujioka , y. uchiyama , i. tsukada , and k. uchinokura , phys . * 80 * , 4566 ( 1998 ) ; y.j . wang , v. kiryukhin , r.j . birgeneau , t. masuda , i. tsukada , and k. uchinokura , phys . 83 * , 1676 ( 1999 ) .

for kernels which are positive and integrable we show that the operator on a finite time interval enjoys a regularizing effect when applied to hlder continuous and lebesgue functions and a `` contractive '' effect when applied to sobolev functions . for hlder continuous functions , we establish that the improvement of the regularity of the modulus of continuity is given by the integral of the kernel , namely by the factor . for functions in lebesgue spaces , we prove that an improvement always exists , and it can be expressed in terms of orlicz integrability . finally , for functions in sobolev spaces , we show that the operator `` shrinks '' the norm of the argument by a factor that , as in the hlder case , depends on the function ( whereas no regularization result can be obtained ) . these results can be applied , for instance , to abel kernels and to the volterra function , the latter being relevant for instance in the analysis of the schrdinger equation with concentrated nonlinearities in . _ keywords : _ volterra functions , singular kernels , volterra integral equations , sonine kernels , orlicz integrability . _ msc 2010 : _ 26a33 , 47g10 , 45e99 , 44a99 , 46e30 .

1611.08503

many mathematical models of physical phenomena deal with systems of volterra integral equations with singular kernels ( e.g. @xcite ) . in this paper , motivated by some nonlinear volterra integral equations arising in quantum mechanics , we investigate the properties of convolution operators with kernels possibly more singular than the more known abel ones . namely , given a generic positive , locally integrable function @xmath0 , we study the action of the operator @xmath7 defined by @xmath8 on intervals @xmath9 $ ] , with @xmath10 ( this assumption being understood in the whole paper ) . precisely , we prove its regularizing effect in hlder and lebesgue spaces and its `` contractive '' effect in sobolev spaces ( where with `` contractive '' we mean that the sobolev norm of @xmath11 on @xmath9 $ ] can be estimated by the norm of @xmath12 times a constant that gets smaller as @xmath13 ) . it is also worth highlighting that the assumption of local integrability of @xmath0 is the minimum requirement so that definition make sense in general . in fact , the aim of the paper ( even though some results will require additional hypothesis ) is to work with the least set of assumptions that are necessary in order to detect remarkable effects from the application of the operator @xmath3 . a particular relevance in applications is acquired by the case @xmath14 ( figure [ fig : nu ] ) , where the operator @xmath3 reads @xmath15 we observe that , if we denote by @xmath16 the _ volterra functions _ defined by @xmath17 then @xmath18 coincides with @xmath19 , which is the so - called volterra function _ of order _ @xmath20 ( see @xcite , section 18.3 ) , that is discussed in section [ sec - volt ] . in addition , recalling that a kernel @xmath21 is said a _ sonine kernel _ if it is a divisor of the unit with respect to the convolution operation , that is , if there exists another kernel @xmath22 such that @xmath23,\ ] ] then , one can prove that @xmath24 is a sonine kernel , with @xmath25 , @xmath26 representing the euler - mascheroni constant ( see eq . ) . the class of sonine kernels is wide and there are many papers ( see e.g. @xcite and references therein ) , starting with the pioneering one by sonine ( @xcite ) , where embedding theorems for integral operators with kernels displaying singularities at the origin of the type @xmath27 are discussed . however , we stress that the results proved in the present paper are more general since they take into account also kernels that are more singular in a neighborhood of the origin , such as , indeed , the volterra function @xmath24 , whose asymptotyc expansion near @xmath28 is given by @xmath29 ( see ) . it is also worth mentioning that a first discussion on the operator @xmath30 is present in @xcite , whereas similar integral operators , but with more regular kernels , have been investigated more recently by @xcite . more in detail , in @xcite it is analyzed the case of a certain class of _ almost decreasing _ sonine kernels in terms of _ weighted generalized _ hlder spaces , while in @xcite an `` inverse '' operator is discussed within the framework of @xmath31 spaces . we also recall that in @xcite some relevant features of volterra functions are pointed out , such as asymptotic expansions and some striking relations with the ramanujan integrals . the interest of the operator @xmath30 is mainly due to its applications in quantum mechanics , and precisely in the study of the schrdinger equation with nonlinear point interactions in @xmath32 . we recall briefly that a schrdinger equation with a _ linear _ point interaction with strength @xmath33 , placed at @xmath34 , is @xmath35 where @xmath36 is a differential operator with domain @xmath37 \displaystyle \hspace{7 cm } \lim_{{\mathbf{x}}\to { \mathbf{y } } } \phi_\lambda({\mathbf{x } } ) = \left(\alpha+\tfrac{1}{2\pi } \log \tfrac{\sqrt{\lambda}}{2 } + \tfrac{\gamma}{2\pi}\right)q\bigg\ } \end{array}\ ] ] ( @xmath38 denoting the green s function of @xmath39 in @xmath32 ) and action @xmath40 for a complete discussion on the solution of this equation through the theory of self - adjoint extension , we refer the reader to @xcite . in addition , it is well known that , given an initial datum @xmath41 , the solution of the associated cauchy problem reads @xmath42 where @xmath43 is the the _ propagator _ of the free schrdinger equation in @xmath32 ( with integral kernel @xmath44 ) and @xmath45 ( with a little abuse of notation that is usual in the literature ) is a complex - valued function satisfying the so - called _ charge _ equation @xmath46 now , a _ nonlinear _ point interaction arises when one assumes that the strength of the interaction depends in fact on the function @xmath45 , and in particular when one sets @xmath47 ( @xmath48 , @xmath49 ) in , thus obtaining @xmath50 ( see @xcite ) . since in the nonlinear case no theory of self - adjoint extensions is available , the relevance of the operator @xmath30 is clear : the well - posedness of the associated cauchy problem is strictly related to the study of the existence and uniqueness of solutions of , which strongly depends on the properties of @xmath30 . even though the application presented above concerns complex - valued functions , this papers only manages real - valued functions . however , one can check that the results of section [ sob ! ] ( which are actually required in @xcite ) can be easily generalized to complex - valued functions . another topical example of integral kernels that are included in our general framework are the well known _ abel kernels _ , which correspond to the choice @xmath51 in . these ones are very important in the theory of _ fractional integration _ and _ generalized differentiation _ ( @xcite ) and , again , in quantum mechanics . in the study of nonlinear point interactions in @xmath52 and @xmath53 , indeed , the resulting integral equations present the kernel , with @xmath54 , in place of @xmath24 ( see @xcite ) . finally , we describe briefly the main results of the paper . they concern , as we told at the beginning , the properties of the operator @xmath3 in hlder spaces , @xmath31 spaces and sobolev spaces . preliminarily , since it is crucial in the following , we define the integral function of the kernel @xmath0 @xmath55 since @xmath0 is always supposed positive and locally integrable , it turns out that @xmath4 is a positive , increasing and absolutely continuous function with @xmath56 as @xmath57 . in the case of hlder spaces , it is well known ( @xcite , theorem 4.2.1 p. 70 ) that when the kernel is @xmath58 , @xmath59 , the operator @xmath3 transforms @xmath60 functions into @xmath61 ones , improving this way the regularity of the modulus of continuity . as a consequence of our main result of section [ sectionhold ] ( theorem [ casoholder ] ) , we will see that more generally the improvement is at least given by the integral function of the kernel : the phenomenon that the power @xmath62 gives as improvement the exponent @xmath63 is therefore true also for any locally integrable kernel which is assumed just equivalent to a decreasing function in a neighborhood of the origin and not blowing too much ( derivative bounded above , for instance ) in its domain . in the case of @xmath31 spaces , it is well known ( @xcite , theorem 4.1.4 p. 67 ) that when the kernel is @xmath58 , the operator @xmath3 transforms @xmath64 functions , @xmath65 , into @xmath66 functions . to a minor integrability of the kernel corresponds a minor gain of integrability for @xmath67 , and apparently the gain disappears when the kernel is just @xmath68 . as a consequence of our main result of section [ sectionellepi ] ( theorem [ casolp ] ) , we will show that _ any kernel locally integrable ( again , we assume that it is equivalent to a decreasing function in a neighborhood of the origin ) gives an improvement of integrability , measured in terms of orlicz spaces . the improvement is strictly linked to the orlicz integrability of the kernel , hence it always exists : a classical , remarkable theorem in orlicz spaces theory ( see e.g. @xcite , p. 60 ) tells that any function @xmath68 is always in some orlicz space strictly contained in @xmath68 . furthermore , in the case @xmath69 , we show ( proposition [ pro : infinity ] ) , under the unique assumption of local integrability , that @xmath3 transforms @xmath70 functions in continuous functions and that the @xmath70 norm of @xmath11 on @xmath9 $ ] is controlled by the norm of @xmath12 times @xmath71 . _ finally , in the case of sobolev spaces , it is well known ( @xcite , theorem 4.2.2 p. 73 ) that when the kernel is @xmath58 , the operator @xmath3 transforms @xmath72 functions , with @xmath73 , in @xmath74 functions . analogous results for @xmath75 functions are discussed in @xcite . in this case , the minor integrability of the kernel yields a minor gain in the sobolev index , which disappears when the kernel is just @xmath68 ( also the preservation of the index is not straightforward ) . as a consequence of our main result of section [ sob ! ] ( theorem [ contr_lemma ] ) , we will show that when @xmath76 , provided @xmath77 , the sobolev index is in fact preserved and the sobolev norm of @xmath11 is bounded , up to a multiplicative constant , by the norm of @xmath12 times @xmath71 . furthermore , we will prove that ( almost ) the same result holds for @xmath78 functions , but just in the case @xmath79 ( theorem [ contr_lemma_2 ] ) , and for @xmath80 functions ( theorem [ contr_lemma_3 ] ) . since the case of a kernel equal to the volterra function @xmath24 ( defined by , figure [ fig : nu ] ) is the most relevant in the applications , it is worth stressing some basic features of @xmath24 . in this way one can easily see that the abstract results established in the following sections can be actually applied to this kernel . first , we recall ( see @xcite ) that @xmath24 is analytic for @xmath82 and that @xmath83 , & \quad\text{as } x\to 0\\[1 cm ] \displaystyle { \mathcal{i}}(x ) = e^{x}+\mathcal{o}(x^{-1 } ) , & \quad\text{as } x\to + \infty . \end{array}\ ] ] consequently , the first expansion shows that @xmath84 and that @xmath85 , for any @xmath86 . is in black , the plot of the first order of the asymptotic expansion of @xmath18 around 0 is dotted.,scaledwidth=40.0% ] also the derivative and the integral function of @xmath24 will play a crucial role in the sequel . hence , recalling ( see again @xcite ) @xmath87 and @xmath88 , we stress that @xmath89\ ] ] and that @xmath90 furthermore , we can state the following lemma . [ lem : conv ] the function @xmath24 is convex on @xmath91 and admits a positive minimum . the second part is immediate since @xmath24 is continuous , positive and coercive by . on the other hand , in order to prove the second part , it is sufficient to show @xmath92 ; namely , by , that @xmath93 . now , following @xcite ( eq . ( 3.1 ) ) and @xcite , we find that @xmath94 denotes the _ ramanujan _ function ( figure [ fig : erre ] ) . hence , @xmath95 and , since @xmath96 is _ completely monotonic _ ( i.e. , for every @xmath97 , @xmath98 does exist and @xmath99 ) , this entails that @xmath93 and thus that @xmath24 is convex . .,scaledwidth=40.0% ] it is also convenient to introduce the function @xmath100 ( figure [ fig : enne ] ) . by the properties of @xmath81 , we see that @xmath101 is positive , increasing and absolutely continuous on bounded intervals intervals . moreover , @xmath102 as @xmath57 , and precisely @xmath103 around zero , scaledwidth=50.0% ] one easily sees that , as one sets @xmath79 in , @xmath4 is equal to @xmath101 . on the other hand , from one also notes that @xmath101 coincides , up to an additive constant , with @xmath104 . finally , we point out a relevant property of the operator @xmath30 defined by , which is strictly connected to the fact that @xmath24 is a sonine kernel . first , define the integral operator @xmath105 then , one notes that , as @xmath106 , @xmath107 is well defined for each function @xmath108 . in addition , one can prove the following result . [ pro : i_identity ] if @xmath109 , then @xmath110.\ ] ] we first observe that one has @xmath111 in @xcite , lemma 32.1 , it is indeed claimed that ( setting @xmath112 therein ) @xmath113 with @xmath114 , and since from @xmath115 , is proved . now , in the expression @xmath116 we exchange the order of the integration , since @xmath117 \displaystyle = \int_0^{x } \int_0^{x } { \mathcal{i}}(s ) \phi(|x-\sigma - s| ) g(\sigma)ds\,d\sigma = 2 \int_0^{x } \int_0^{x - s } { \mathcal{i}}(s ) \phi(|x-\sigma - s| ) g(\sigma)d\sigma\,ds , \end{array}\ ] ] and using , we conclude that @xmath118 finally , given that an easy change of variable shows @xmath119 , the proof is complete . we note that , in view of theorem [ contr_lemma_3 ] , if one also assumes @xmath120 in proposition [ pro : i_identity ] , then @xmath121 let @xmath122 and let @xmath0 , @xmath123 be absolutely continuous and positive in @xmath1240,\tau_0]$ ] . we say that @xmath0 , @xmath123 are @xmath125 if there exist two positive constants @xmath126 such that @xmath1270,\tau_0].\ ] ] of course any function equivalent to @xmath0 in @xmath1240,\tau_0]$ ] is of type @xmath128 , where @xmath129 is absolutely continuous in @xmath1240,\tau_0]$ ] and such that @xmath1300,\tau_0].\ ] ] the statements of this section hold for certain functions @xmath0 which are decreasing in intervals of the type @xmath131 and , more generally , they hold for functions equivalent to decreasing functions in intervals of the type @xmath131 . for the sake of simplicity , the functions @xmath0 in the statements will be always assumed equivalent to decreasing functions in intervals of the type @xmath131 , and the corresponding decreasing functions will be written as products @xmath128 , where @xmath129 is an absolutely continuous function in @xmath1240,\tau_0]$ ] satisfying . [ primolemma ] let @xmath132 , and let @xmath1330,t])\cap l^1(0,t)$ ] be positive and equivalent to a decreasing function in @xmath131 for some @xmath134 . if @xmath135 then @xmath136 where @xmath4 is defined by . inequalities coming from assumptions of monotonicity of ratios between functions and powers are very well known among researchers working in orlicz spaces . some proofs of such inequalities work also without the assumption of convexity ( the reader may compare this lemma e.g. with theorem 3 in @xcite or with the results in section 3 of @xcite ) , however , the main feature of is that it has been obtained from assumptions of monotonicity which hold only in a neighborhood of the origin and not in the whole domain of the functions involved ( where , however , at least a boundedness is required ; this assumption appears implicitly in the hypothesis of continuity which holds until the endpoint @xmath137 ) . we preliminarly note that , since @xmath128 is decreasing in @xmath131 , @xmath138\ , .\ ] ] let @xmath139 , @xmath140 . if @xmath141 , by and respectively , @xmath142 and , if @xmath143 , @xmath144 then again by , @xmath145 and therefore , setting @xmath146}\nu\ , , $ ] @xmath147 finally , if @xmath143 , @xmath148 then , since @xmath149 implies that @xmath4 is increasing , @xmath150 from , , , we get . the following statement is an immediate consequence of lemma [ primolemma ] . [ usatodavvero ] in the same assumptions of lemma [ primolemma ] , for any @xmath151 it is @xmath152 uniformly in @xmath153 , @xmath139 . the next lemma is trivially true for decreasing functions @xmath0 ( see the case ( i ) of the proof ) , and it provides a version of the inequality in case of functions which are decreasing only in a neighborhood of the origin ( however , as in the remark above , one can see that , again , an assumption of boundedness has been made implicitly ) . [ secondolemma ] if @xmath1330,t])\cap l^1(0,t)$ ] is positive and equivalent to a decreasing function in @xmath131 for some @xmath134 , then @xmath154 we examine the three cases \(i ) @xmath155 \(ii ) @xmath156 \(iii ) @xmath157 case ( i ) : since @xmath128 is decreasing in @xmath158 , @xmath159 case ( ii ) : we have @xmath160 where @xmath161}\nu$ ] . there are two possibilities : @xmath162 @xmath163 @xmath164 @xmath165 in the case @xmath162 , since @xmath128 is decreasing in @xmath158 , @xmath166 and therefore from @xmath167 in the case @xmath164 , setting @xmath1680,t]}\nu>0 $ ] , @xmath169 and then , using @xmath170 , from we get @xmath171 case ( iii ) : we have @xmath172 and applying case ( i ) with @xmath173 replaced by @xmath174 and case ( ii ) with @xmath175 replaced by @xmath174 , @xmath176 since in our case @xmath177 , it is @xmath178 , hence the first term can be estimated by the right hand side of ; similarly , since @xmath179 , it is @xmath180 and the same conclusion holds for the second term . in next theorem we are going to consider an assumption on @xmath0 stronger ( as we are going to see ) with respect to that one of lemma [ primolemma ] : in the case @xmath181 ( a similar digression can be done in the general case , replacing @xmath0 by @xmath128 ) we will assume that the positive function @xmath1330,t])\cap l^1(0,t)$ ] is such that @xmath182 is decreasing in @xmath183 for some @xmath134 , @xmath184 , where @xmath63 is a given number in @xmath185 . it is easy to verify that this latter assumption implies that the function @xmath4 , defined by , is such that @xmath186 and also that @xmath0 is decreasing in @xmath183 ( because @xmath187 is product of positive decreasing functions ) . let us verify the first assertion . since @xmath188 decreasing in @xmath183 , it is ( note that , since also @xmath0 is absolutely continuous , their derivatives exist a.e . ) @xmath189 hence , integrating the above inequality in @xmath190 , where @xmath191 , and noting that @xmath192 is decreasing too , we have @xmath193 @xmath194 if we let @xmath195 , since @xmath0 is decreasing , it is @xmath196 and therefore we get @xmath197 from which the assertion follows . on the other hand , the fact that the assumption is really stronger is shown by the following let @xmath198 , and let @xmath199 so that @xmath200 then @xmath201 is satisfied for @xmath202 , while for the same @xmath203 the function @xmath182 is not decreasing in @xmath183 ( because it is not decreasing in @xmath204 ) . before the statement of the main theorem of this section , we observe that the kernels @xmath0 of our interest are such that their difference quotients are bounded above , i.e. there exists a constant @xmath205 such that @xmath206 this property ( which holds automatically , in particular , for all the kernels @xmath0 which are decreasing in the whole @xmath207 ) is expressed in an equivalent way in the assumption below . [ casoholder ] if @xmath208)$ ] , @xmath59 , @xmath209 , @xmath210 , and if @xmath1330,t])\cap l^1(0,t)$ ] , @xmath211 , is such that @xmath212 @xmath2130,\tau_0[)\ , , \ , 0<b^-\le b(x)\le b^+<\infty\ , , \ ] ] @xmath214 then , setting @xmath215 it is @xmath216 let @xmath217 . it is @xmath218ds\\ & -\int_{y - x}^y \nu(s - y+x)g(y)ds+\int_{y - x}^y \nu(s - y+x)[g(y)-g(y - s)]ds\\ & = g(y)\int_0^y \nu(s)ds-\int_0^{y - x } \nu(s)[g(y)-g(y - s)]ds-\int_{y - x}^y \nu(s)[g(y)-g(y - s)]ds\\ & -g(y)\int_{y - x}^y \nu(s - y+x)ds+\int_{y - x}^y \nu(s - y+x)[g(y)-g(y - s)]ds\\ & = g(y)[n(y)-n(x)]-\int_0^{y - x } \nu(s)[g(y)-g(y - s)]ds\\ & + \int_{y - x}^y [ \nu(s - y+x)-\nu(s)][g(y)-g(y - s)]ds . \end{split}\ ] ] therefore @xmath219|+\left\vert\int_0^{y - x } \nu(s)[g(y)-g(y - s)]ds\right\vert\\ & + \left\vert\int_{y - x}^y [ \nu(s - y+x)-\nu(s)][g(y)-g(y - s)]ds\right\vert:=a+b+c . \end{split}\ ] ] we estimate each term in turn . since @xmath208)$ ] , @xmath210 , and since @xmath4 is increasing and @xmath59 , @xmath220|\\ & \le [ g]_\alpha y^\alpha [ n(y)-n(x)]\\ & = [ g]_\alpha ( y^\alpha - x^\alpha ) [ n(y)-n(x)]+[g]_\alpha x^\alpha [ n(y)-n(x)]\\ & \le [ g]_\alpha ( y - x)^\alpha [ n(y)-n(x)]+[g]_\alpha x^\alpha [ n(y)-n(x ) ] . \end{split}\ ] ] by the function @xmath128 is decreasing in @xmath183 , hence by @xmath221 @xmath222 and therefore @xmath223_\alpha ( y - x)^\alpha n(y - x)+[g]_\alpha x^\alpha [ n(y)-n(x)].\ ] ] the first term can be estimated by the right hand side of . as to the second term , we begin observing that @xmath224=x^\alpha \frac{n(y)-n(x)}{y - x}(y - x ) = x^\alpha \nu(\xi)(y - x),\ ] ] for some @xmath225 . we consider first the case @xmath226 , so that @xmath227 ( because @xmath128 is decreasing in @xmath183 ) . since @xmath151 , by @xmath228 and therefore , since @xmath229 , @xmath230 @xmath231 @xmath232 @xmath233 now we consider the case @xmath234 . setting @xmath235}x^\alpha\nu ( x ) $ ] , @xmath2360,t]}x^\alpha\nu ( x)>0 $ ] , it is @xmath237 now there are two possibilities : @xmath238 , @xmath239 . in the first possibility , @xmath240 and therefore @xmath241 @xmath242 in the second possibility , @xmath243 and therefore , using @xmath244 , @xmath245 @xmath246 we have therefore shown that also the second term can be estimated by the right hand side of . on the other hand , it is quite easy to check that also @xmath247 can be estimated in the same way , in fact @xmath248ds\right\vert\le \int_0^{y - x } \nu(s)[g]_\alpha s^\alpha ds\ ] ] @xmath249_\alpha\int_0^{y - x } \nu(s ) ( y - x)^\alpha ds= [ g]_\alpha ( y - x)^\alpha n(y - x).\ ] ] it remains to estimate @xmath250 . it is @xmath251[g(y)-g(y - s)]ds\right|\le [ g]_\alpha \int_{y - x}^y |\nu(s - y+x)-\nu(s)|s^\alpha ds\ ] ] we now use the assumption , so that @xmath252_\alpha \int_{y - x}^y \left[\nu(s - y+x)-\nu(s)\right]s^\alpha ds+[g]_\alpha\int_{y - x}^y k_0(y - x)s^\alpha ds\ , .\ ] ] making the change of variables @xmath253 in the first term , we have @xmath254_\alpha ( y - x)^{\alpha+1 } \int_{1}^{\frac{y}{y - x}}\left [ \nu((y - x)(\sigma-1 ) ) - \nu((y - x)\sigma)\right]\sigma^\alpha d\sigma+[g]_\alpha\int_{y - x}^y k_0(y - x)s^\alpha ds\ ] ] we observe that @xmath255 implies @xmath256 , therefore @xmath257_\alpha ( y - x)^{\alpha+1}\left\ { \int_{1}^{2}\left [ \nu((y - x)(\sigma-1 ) ) - \nu((y - x)\sigma)\right]\sigma^\alpha d\sigma+\right.\\&+ \left.\int_{2}^{\frac{y}{y - x}}\left [ \nu((y - x)(\sigma-1 ) ) - \nu((y - x)\sigma)\right]\sigma^\alpha d\sigma \right\}+[g]_\alpha\int_{y - x}^y k_0(y - x)s^\alpha ds \\ & : = c_{1}+c_{2}+c_{3}\end{split}\ ] ] we observe that the sign of @xmath258 and the sign of @xmath259 are not necessarily positive ; both @xmath258 and @xmath259 will be splitted into more terms , each of them being not necessarily positive ; however , all of them will be shown to be smaller than the right hand side of . the first term can be estimated as follows : @xmath260_\alpha ( y - x)^{\alpha+1 } \int_{1}^{2}\left [ \nu((y - x)(\sigma-1 ) ) - \nu((y - x)\sigma)\right ] \sigma^\alpha d\sigma\\ & \leqslant 2^{\alpha}[g]_\alpha ( y - x)^{\alpha+1 } \int_{1}^{2}\nu((y - x)(\sigma-1 ) ) d\sigma \\ & = 2^{\alpha}[g]_\alpha ( y - x)^{\alpha } \int_{0}^{y - x}\nu(\tau ) d\tau \\ & = 2^{\alpha}[g]_\alpha ( y - x)^{\alpha } n(y - x)\ , . \end{split}\ ] ] in order to estimate @xmath259 , we need to use the following inequality , which , as we are going to see , follows easily from the fact that the positive @xmath2610,t])$ ] is decreasing in @xmath183 : @xmath262 where @xmath161}\nu$ ] and @xmath263}\nu>0 $ ] . in order to show , let us fix @xmath264 . if @xmath265 , using that @xmath128 is decreasing in @xmath183 , we have @xmath266 if @xmath267 , @xmath268 the last case is @xmath269 , where the following inequalities hold : @xmath270 @xmath271 using the change of variable @xmath272 , the integration by parts formula , @xmath211 , and lemma [ secondolemma ] , we now estimate @xmath259 as follows : @xmath273_\alpha ( y - x)^{\alpha+1 } \int_{2}^{\frac{y}{y - x}}\left [ \nu((y - x)(\sigma-1 ) ) - \nu((y - x)\sigma)\right]\sigma^\alpha d\sigma\\ & = [ g]_\alpha ( y - x)^{\alpha+1 } \int_{1}^{\frac{x}{y - x}}\left [ \nu((y - x)\tau ) - \nu((y - x)(\tau+1))\right](\tau+1)^\alpha d\tau\\ & = [ g]_\alpha ( y - x)^{\alpha+1 } \int_{1}^{\frac{x}{y - x}}(\tau+1)^\alpha d\left(\frac{1}{y - x}\int_{(y - x)(\tau+1)}^{(y - x)\tau}\nu ( s)ds\right)\\ & = [ g]_\alpha ( y - x)^{\alpha+1 } \left\ { \left . \frac{(\tau+1)^\alpha}{y - x}\int_{(y - x)(\tau+1)}^{(y - x)\tau}\nu ( s)ds \right\vert^{\tau=\frac{x}{y - x}}_{\tau=1}\right.\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad-\left . \frac{1}{y - x } \int_{1}^{\frac{x}{y - x}}\left(\int_{(y - x)(\tau+1)}^{(y - x)\tau}\nu ( s)ds\right)\alpha ( \tau+1)^{\alpha-1}d\tau \right\}\\ & = [ g]_\alpha ( y - x)^{\alpha } \left\ { \left . ( \tau+1)^\alpha\int_{(y - x)(\tau+1)}^{(y - x)\tau}\nu ( s)ds \right\vert^{\tau=\frac{x}{y - x}}_{\tau=1}\right.\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad+\left . \int_{1}^{\frac{x}{y - x}}\left(\int_{(y - x)\tau}^{(y - x)(\tau+1)}\nu ( s)ds\right)\alpha ( \tau+1)^{\alpha-1}d\tau \right\}\\ & = [ g]_\alpha ( y - x)^{\alpha } \left\ { \left(\frac{y}{y - x}\right)^{\alpha}\int_{y}^{x}\nu(s)ds-2^{\alpha}\int_{2(y - x)}^{y - x}\nu(s)ds\right.\\ & \qquad\qquad\qquad\qquad\qquad\qquad\qquad+\left . \int_{1}^{\frac{x}{y - x}}\left[\int_{(y - x)\tau}^{(y - x)(\tau+1)}\nu ( s)ds\right]\alpha ( \tau+1)^{\alpha-1}d\tau \right\}\\ & \leqslant [ g]_\alpha ( y - x)^{\alpha}\left\{2^{\alpha}\int_{y - x}^{2(y - x)}\nu(s)ds + c\left(b(\cdot),\nu(\cdot)\right ) \int_{1}^{\frac{x}{y - x}}\left [ ( y - x)\nu ( ( y - x)\tau ) \right]\alpha ( \tau+1)^{\alpha-1}d\tau \right\}\\ & \leqslant [ g]_\alpha ( y - x)^{\alpha}\left\{2^{\alpha}c\left(b(\cdot),\nu(\cdot),t\right)n(y - x)\right.\\ & \qquad\qquad\qquad+\left . c\left(b(\cdot),\nu(\cdot)\right)(y - x)\alpha\frac{n(y - x)}{y - x}\int_{1}^{\frac{y}{y - x}}(\tau+1)^{\alpha-1}\frac{\nu((y - x)\tau)\,(y - x)}{n(y - x)}d\tau\right\ } \end{split}\ ] ] by corollary [ usatodavvero ] the integral inside the parenthesis is bounded by a constant independent of @xmath274 , depending only on @xmath275 . hence the estimate becomes @xmath276_\alpha ( y - x)^{\alpha}\left\{2^{\alpha}c\left(b(\cdot),\nu(\cdot),t\right)n(y - x ) + \alpha \textit{c}\left(b(\cdot),\nu(\cdot),t,\alpha+\beta\right ) n(y - x ) \right\}\ ] ] and therefore also @xmath259 is estimated by the right hand side of . finally , we need to estimate @xmath277 . we have @xmath278_\alpha\int_{y - x}^y k_0(y - x)s^\alpha ds\le [ g]_\alpha k_0 \int_0^t s^\alpha ds ( y - x)\ ] ] @xmath279_\alpha(y - x)=\textit{c}\left(\nu(\cdot),t,\alpha\right)[g]_\alpha \frac{(y - x)^{1-\alpha}}{n(y - x)}(y - x)^{\alpha}n(y - x)\ ] ] @xmath280_\alpha(y - x)^{\alpha}n(y - x)\ ] ] where the last inequality follows from the fact that from and from @xmath281 it follows that @xmath282 hence @xmath283 is bounded below in @xmath207 by a positive constant , i.e. its reciprocal is bounded above by a positive constant ( depending only on @xmath284 , @xmath0 and @xmath63 ) . in the case @xmath181 , @xmath285 , @xmath286 , theorem [ casoholder ] gives back theorem 4.2.1 in @xcite . another interesting case is @xmath287 which satisfies the assumption of theorem [ casoholder ] for _ any @xmath59 , for _ any @xmath184 , with @xmath181 . of course those positive functions @xmath1330,t])\cap l^1(0,t)$ ] , which are just _ equivalent to the right hand side of only in a neighborhood of the origin , and then not blowing up `` too much '' ( as , for instance , the function @xmath288 in , whose derivative see is again a volterra function which is bounded above ) , are examples for theorem [ casoholder ] and in such cases the resulting regularity for @xmath3 is the same as that one given for . _ _ _ from the proof of theorem [ casoholder ] it is clear that the assumption can be weakened as follows : @xmath289 for a given @xmath0 satisfying the assumptions of theorem [ casoholder ] one may look for the _ best regularity action for @xmath3 , i.e. one may look for the greatest @xmath63 satisfying the assumptions of the theorem or , equivalently , for the smallest @xmath203 satisfying ( in the case of the classical spaces of hlder continuous functions , the inclusions between the spaces are easy and well known ; for a recent book on this topic see @xcite ) : this problem is linked to the notion of boyd indices ( see e.g. @xcite ) . for a short survey including a bibliography on this topic , and for a `` concrete '' way to compute them for explicit examples , see e.g. @xcite,@xcite . _ we begin some background on young s functions and orlicz spaces . in the following a convex function @xmath290 defined on @xmath291 is said to be a young s function if it is convex and such that @xmath292 , @xmath293 for @xmath294 . this assumption implies that young s functions are strictly increasing and invertible , so that it makes sense to consider its inverse @xmath295 , defined in @xmath291 . the orlicz space @xmath296 ( here @xmath137 is a fixed positive real number ) is the banach function space of all real - valued ( lebesgue ) measurable functions @xmath297 on @xmath207 such that @xmath298 ( here we use the convention @xmath299 ) . in the special case @xmath300 , @xmath301 , the orlicz space reduces to the familiar lebesgue space . for essentials about orlicz spaces and banach function spaces the reader may refer to @xcite , sections 2.10.2 and 2.10.3 ( and references therein for extensive treatments ) . a well known result of the theory is that if two young s functions @xmath290 , @xmath247 are such that @xmath302 then , in spite @xmath303 , @xmath304 may be different , the spaces themselves ( namely , the set of the functions such that the norms are finite ) coincide . in particular , the spaces are completely determined by the values of the young s functions assumed for @xmath175 large . for measurable functions @xmath305 on @xmath207 , the decreasing rearrangement @xmath306 is defined by the right continuous inverse of @xmath307 , i.e. @xmath308 . orlicz spaces @xmath296 are rearrangement - invariant : this means , in particular , that the norm is not affected after the action of the decreasing rearrangement operator : @xmath309 . we may state the following [ casolp ] let @xmath310 , @xmath209 , and let @xmath290 be a young s function . if @xmath311 , @xmath211 , is such that @xmath312 @xmath313 @xmath314 where @xmath315 then , setting @xmath215 it is @xmath316 where @xmath250 is the young s function ( proof of lemma 4.2 in @xcite ) defined by @xmath317 before giving the proof of the theorem , which is a quite easy consequence ( in fact , an application ) of a classical result about fractional integration in orlicz spaces , we highlight a couple of examples which are relevant for this paper . let @xmath310 and @xmath318 , and let @xmath319 , for all @xmath320 . then @xmath321 satisfies , hence @xmath250 given by is @xmath322 , and @xmath323 for all @xmath324 . this special case gives back the refined version of the continuity property for the abel operator in @xmath64 spaces , see theorem 4.1.3 in @xcite . notice that when @xmath63 approaches @xmath28 , the exponent @xmath325 of integrability of @xmath0 approaches @xmath326 , and the exponent @xmath327 of integrability of @xmath328 approaches @xmath329 , which means no gain of integrability : in the framework of the lebesgue spaces , kernels in @xmath68 which do not possess the higher integrability property ( see e.g. next two examples and , in particular , the volterra function @xmath288 ) are not able to improve the integrability through the operator @xmath67 . let @xmath310 and let @xmath330 , and let @xmath331 . then @xmath332 hence @xmath333 satisfies . the young s function @xmath250 given by is @xmath334 and @xmath335 for all @xmath324 . it is interesting to note that the kernel does not belong to any lebesgue space @xmath64 with @xmath86 , and that the logarithm in the expression of @xmath250 ( which has a positive power and therefore it is divergent at infinity ) represents an orlicz gain of integrability for @xmath328 . for any kernel considered in theorem [ casolp ] , the existence of a young function @xmath290 satisfying can be easily established ; moreover , for any @xmath0 the function @xmath328 always enjoys an orlicz gain of integrability with respect to @xmath12 : this is the heart of the following simple result , which is consequence of standard statements of orlicz spaces theory , namely , of the fact that any function @xmath336 is always in some orlicz space @xmath337 strictly contained in @xmath336 ( see e.g. @xcite , p.60 ) , of the hlder s inequality in orlicz spaces ( see e.g. @xcite , 8.11 p. 234 ) @xmath338 where @xmath339 is the young function defined by @xmath340 , of the equivalences ( see e.g. @xcite , ( 7 ) p. 230 and @xcite , respectively ) @xmath341 @xmath342 and finally of @xmath343 in the assumptions of theorem [ casolp ] , for every @xmath0 there exists a young function @xmath290 satisfying , and therefore @xmath344 is strictly contained in @xmath345 . let @xmath290 be a young function such that @xmath346 , @xmath347 being increasing and divergent at infinity . by , , respectively , @xmath348 as to the second part of the statement , from we get that setting , for large @xmath349 , @xmath350 and @xmath351 ( note that @xmath175 and @xmath352 are large as well ) @xmath353 and using both implications in , we get the assertion . it is immediate to realize that the statement of theorem [ casolp ] remains true if @xmath0 is replaced by any function equivalent to @xmath0 in a neighborhood of the origin . hence all the previous remark still holds if @xmath0 is replaced by the function @xmath288 in . the proof of theorem [ casolp ] will follow as consequence of the following result appeared in sharpley ( ( * ? ? ? * theorem 3.8 ) ) , in the more abstract setting of general convolution operators , defined in @xcite . in this latter paper our operator @xmath3 , which goes back to @xcite , is explicitly mentioned as example ( see the end of the section iv therein ) . here we state it in a more convenient form , and using our notation : [ bysharpley ] suppose that @xmath290 , @xmath247 are young s functions such that @xmath354 and @xmath355 where @xmath250 is the young s function defined by @xmath356 then , @xmath357 where ( @xmath3 is defined by and ) @xmath358 denotes the averaged rearrangement of @xmath0 , defined by @xmath359 _ proof of theorem [ casolp ] . _ setting @xmath360 , is obviously satisfied with equality and @xmath361 ; from we get that if @xmath250 is defined by @xmath362 i.e. if holds , then is satisfied : in fact , from the convexity of the young function @xmath290 , the function @xmath363 is decreasing , hence from we deduce that also the following ones are decreasing : @xmath364 and therefore @xmath365 is increasing , from which is satisfied with @xmath366 . we are therefore allowed to apply theorem [ bysharpley ] . from and it follows that for small values of @xmath352 it is @xmath367 , hence , by , for @xmath175 small it is @xmath368 and of course the same conclusion holds ( for a possibly different @xmath369 ) for all @xmath370 . hence the supremum in the right hand side of is a finite constant @xmath371 , from which follows . finally , the case @xmath69 is considered in the next result , where the regularizing effect of @xmath67 is expressed through continuity . [ pro : infinity ] if @xmath149 and @xmath311 , then for any @xmath372 there results @xmath373)\ ] ] and @xmath374 where @xmath4 is the integral function of @xmath0 defined by . recalling and , is immediate . then , it is left to prove that @xmath11 is continuous . to this aim , fix @xmath375 and @xmath376x_0,t]$ ] . easy computations yield @xmath377}(s)g(s)ds+\\[.2 cm ] & \ , -\int_0^t\big(\nu(x_0-s)-\nu(x - s)\big){\mathds{1}}_{[0,x_0]}(s)g(s)ds \end{aligned}\ ] ] and hence @xmath378}(s)|g(s)|ds+ \\ & \,+\int_0^t|\nu(x_0-s)-\nu(x - s)|{\mathds{1}}_{[0,x_0]}(s)|g(s)|ds \\ \leq & \,n(x - x_{0})\|g\|_{l^\infty(0,t)}+\|g\|_{l^\infty(0,t ) } \int_\mathbb{r}|\nu_e(x_0-s)-\nu_e(x - s)|ds , \end{aligned}\ ] ] where @xmath3790,t[\,,\\[.2 cm ] 0 , & \text{if } s\in{\mathbb{r}}\backslash\;]0,t[\ , . \end{array } \right.\ ] ] therefore , the first term converges to zero by the continuity of @xmath4 and the second term converges to zero by the _ mean continuity property _ ( see @xcite ) . since the same holds if @xmath380 , one has @xmath381 , which concludes the proof . we start with some basics on sobolev spaces with _ fractional _ index . let @xmath382 and @xmath383 . we denote by @xmath384 the sobolev space defined by @xmath385_{\dot{h}^\theta(a , b)}^2<\infty\},\ ] ] where @xmath386_{\dot{h}^\theta(a , b)}^2 : = \int_{[a , b]^2}\frac{|g(x)-g(y)|^2}{|x - y|^{1 + 2\theta}}dydx.\ ] ] this is a hilbert space with the natural norm @xmath387_{\dot{h}^\theta(a , b)}^2.\ ] ] when @xmath388 and @xmath389 , @xmath390 can be equivalently defined using the fourier transform ( see @xcite ) ; that is , if we define the fourier transform as @xmath391 then there exist two constants @xmath392 such that @xmath393 [ lem - ext ] let @xmath399 , @xmath400 , and set @xmath401,\\[.3 cm ] g(2t - x ) , & \text{if } x\in\;]t,2t],\\[.3 cm ] 0 , & \text{if } x\in{\mathbb{r}}\backslash[0,2 t ] . \end{array } \right.\ ] ] the following holds : the proof is a straightforward application of lemma 2.1 in @xcite . cases @xmath407 are trivial . consider , then , an arbitrary @xmath408 . first , we can easily check that @xmath409 and that , with some change of variables , @xmath410_{\dot{h}^\theta(0,2t)}^2 = 2[g]_{\dot{h}^\theta(0,t)}^2 + 2\int_{[0,t]^2}\frac{|g(x)-g(y)|^2}{|x+y-2t|^{1 + 2\theta}}dydx.\ ] ] moreover , since @xmath411 for every @xmath412 ^ 2 $ ] , @xmath413 ^ 2}\frac{|g(x)-g(y)|^2}{|x+y-2t|^{1 + 2\theta}}dydx \leq [ g]_{\dot{h}^\theta(0,t)}^2.\ ] ] hence @xmath414_{\dot{h}^\theta(0,2t)}^2 \leq 4[g]_{\dot{h}^\theta(0,t)}^2 $ ] , so that @xmath415 now , from lemma 2.1 of @xcite , we know that if @xmath4040,1/2[\,$ ] , then there exists @xmath416 such that @xmath417 on the other hand , the same lemma shows that holds even if @xmath4041/2,1[\,$ ] , provided that @xmath210 ( since this entails by definition @xmath418 ) . combining and , the proof is complete . note that , when @xmath419 , the assumption @xmath210 is meaningful as @xmath12 is continuous by sobolev embeddings ( @xcite ) . what is more , this requirement is mandatory , since otherwise @xmath420 might not preserve continuity on @xmath52 . we also stress that the case @xmath421 is not managed by lemma [ lem - ext ] since lemma 2.1 in @xcite is not valid in general for this choice of @xmath422 , due to the failure of hardy inequality ( see e.g. @xcite ) . fix @xmath423\backslash\{1/2\}$ ] . then , let again @xmath4260,t[\,,\\[.3 cm ] 0 , & \text{if } x\in{\mathbb{r}}\backslash\;]0,t [ \end{array } \right.\ ] ] and , for any @xmath399 , define @xmath427 where @xmath420 is the extension of @xmath12 obtained via lemma [ lem - ext ] . note that applies if either @xmath428 or @xmath419 with , in this second case , the further assumption that @xmath210 . as @xmath429 for all @xmath430 $ ] , @xmath431 now , by definition @xmath432}(s)ds\,dx.\ ] ] since @xmath433}(s ) = h(s ) - h(s - x)$ ] , where @xmath434 denotes the heaviside function , @xmath435 with @xmath436 . consequently , by well known properties of the fourier transform , @xmath437 thus , noting that @xmath438 and @xmath439 , @xmath440 then , combining , and , @xmath441 moreover , we observe that @xmath442 and , plugging into , that @xmath443 combining with , follows . we observe that the `` contractive '' effect of @xmath3 , pointed out in the introduction , is in the fact that @xmath444 , as @xmath13 . it entails that on small intervals the operator `` shrinks '' the norm of the argument function by a factor that gets smaller whenever @xmath137 gets smaller . note that in the previous theorem , when @xmath419 the assumption @xmath210 can not be removed , since otherwise the result is false . if one dropped this requirement , indeed , then the statement would imply that @xmath445 belongs to @xmath398 , which can not hold in general . a remarkable counterexample is given by the case @xmath79 , where one can prove that @xmath446 does not belong to @xmath398 for any @xmath419 ( see lemma [ lem : n ] ) . the case @xmath421 is far more awkward since no `` extension - to - zero '' result , such as lemma [ lem - ext ] , is available . however , in the case @xmath79 , we can state an analogous for theorem [ contr_lemma ] . in order to prove it , it is though required a further investigation of the behavior of the integral function @xmath24 . on the other hand , one easily checks that @xmath450 , so that it is left to prove that @xmath451_{\dot{h}^{1/2}(0,t)}<\infty$ ] ( since @xmath452 for all @xmath402 , by @xcite ) . an easy computation shows that @xmath453_{\dot{h}^{1/2}(0,t)}^2 = 2\int_0^t\int_0^{x/2}\left|\frac{{\mathcal{n}}(x)-{\mathcal{n}}(y)}{x - y}\right|^2dy\,dx+2\int_0^t\int_{x/2}^x\left|\frac{{\mathcal{n}}(x)-{\mathcal{n}}(y)}{x - y}\right|^2dy\,dx.\ ] ] looking at the first integral and recalling that @xmath101 is increasing , we find @xmath454 hence , combining with and , @xmath455 concerning the second integral , jensen inequality yields @xmath456 & \leq \int_0^t\int_{x/2}^x\frac{1}{x - y}\int_y^x{\mathcal{i}}^2(s)ds\,dy\,dx . \end{aligned}\ ] ] furthermore , since @xmath24 is positive and convex by lemma [ lem : conv ] , it is @xmath457 for every @xmath458 $ ] , so that @xmath459 now , noting that @xmath460 for all @xmath461 and using again , @xmath462 whereas , on the other hand , @xmath463 thus @xmath464 and , summing up , it is @xmath451_{\dot{h}^{1/2}(0,t)}<\infty$ ] , which concludes the proof . since @xmath421 , does not hold . however , defining @xmath420 as in , @xmath468 and hence , arguing as in the proof of theorem [ contr_lemma ] ( with @xmath469 ) , one can check that @xmath470 then , it is left to estimate @xmath471_{\dot{h}^{1/2}(0,t)}$ ] . first , we note that for every @xmath472 @xmath473 hence , @xmath474_{\dot{h}^{1/2}(0,t)}^2\leq & \ , 4\int_0^t\int_0^x\left|\frac{1}{x - y}\int_y^x{\mathcal{i}}(s)g(x - s)ds\right|^2dy\,dx+\nonumber\\[.3 cm ] & \ , 4\int_0^t\int_0^x\left|\int_0^y{\mathcal{i}}(s)\frac{g(x - s)-g(y - s)}{t - s}ds\right|^2dy\,dx . \end{aligned}\ ] ] now , one can easily see that , since @xmath372 , @xmath475 & = 2\|g\|_{l^\infty(0,t)}^2[{\mathcal{n}}]_{\dot{h}^{1/2}(0,t)}^2\nonumber\\[.3 cm ] & \leq 2\|g\|_{l^\infty(0,t)}^2\|{\mathcal{n}}\|_{\dot{h}^{1/2}(0,t)}^2 \end{aligned}\ ] ] ( where @xmath476 is finite by lemma [ lem : n ] ) . on the other hand , by jensen inequality and monotonicity of @xmath101 , @xmath477 \hspace{4cm}\displaystyle \leq4{\mathcal{n}}(t)\int_0^t\int_0^x\int_0^y{\mathcal{i}}(s)\left|\frac{g(x - s)-g(y - s)}{x - y}\right|^2ds\,dy\,dx . \end{array}\ ] ] from the fubini theorem and a change of variables @xmath478 \hspace{4cm}\displaystyle = \int_0^t{\mathcal{i}}(s)\int_0^{t - s}\int_0^x\left|\frac{g(x - s)-g(y - s)}{x - y}\right|^2dy\,dx\,ds\\[.7 cm ] \hspace{4cm}\displaystyle \leq [ g]_{\dot{h}^{1/2}(0,t)}^2{\mathcal{n}}(t ) . \end{array}\ ] ] consequently , @xmath479 and plugging and into , @xmath480_{\dot{h}^{1/2}(0,t)}^2\leq c\cdot\max\{\|{\mathcal{n}}\|_{\dot{h}^{1/2}(0,t)}^2,{\mathcal{n}}^2(t)\}\left(\|g\|_{l^\infty(0,t)}^2+\|g\|_{\dot{h}^{1/2}(0,t)}^2\right).\ ] ] finally , combining with , follows . it is also worth stressing that theorem [ contr_lemma_2 ] holds as well for any positive and integrable kernel @xmath0 whose integral function @xmath482 ( such as , for instance , abel kernels ) . however , since this is a very specific assumption , we preferred to present it in the relevant case of the volterra kernel , where @xmath483 can be clearly shown , leaving to the reader further generalizations . finally , we show that a version of theorem [ contr_lemma ] holds also in @xmath484 . this result could seem disconnected from the framework of our paper , but nevertheless it further clarifies some specific features of @xmath3 and , then , we mention it for the sake of completeness . recalling and arguing as in the proof of theorem [ contr_lemma ] , one finds that @xmath486,\ ] ] where @xmath487}(s)ds,\ ] ] with @xmath4880,t[\,,\\[.2 cm ] 0 , & \text{if } x\in{\mathbb{r}}\backslash\;]0,t[\ , , \end{array } \right . \quad\text{and}\quad g_t(x):=\left\ { \begin{array}{ll } g(x ) , & \text{if } x\in\;]0,t[\,,\\[.2 cm ] 0 , & \text{if } x\in{\mathbb{r}}\backslash\;]0,t[\ , . \end{array } \right.\ ] ] now , recalling that @xmath433}(s)=h(s)-h(s - x)$ ] ( @xmath434 denoting the heaviside function ) and that @xmath489 , reads @xmath490 then , by well known properties of the convolution product @xmath491 on the other hand , by , @xmath492 and thus @xmath493 consequently , since @xmath494 , arguing as before one finds . the proof of the previous theorem stresses a relevant difference between the cases @xmath80 and @xmath395 , that arises in fact from the lack of further integrability ( of `` power type '' ) of @xmath0 . indeed , if @xmath0 belongs only to @xmath336 the additional assumption @xmath210 clearly can not be removed in @xmath395 , whereas it is not necessary in @xmath80 . r. adami , g. dellantonio , r. figari and a. teta , _ the cauchy problem for the schrdinger equation in dimension three with concentrated nonlinearity _ h. poincar anal . non linaire * 20 * ( 2003 ) , no . 3 , 477500 . r. adami , g. dellantonio , r. figari and a. teta , _ blow - up solutions for the schrdinger equation in dimension three with a concentrated nonlinearity _ poincar anal . non linaire * 21 * ( 2004 ) , no . 1 , 121137 . a. erdlyi , w. magnus , f. oberhettinger , and f. g. tricomi , _ higher transcendental functions . iii _ , robert e. krieger publishing co. , inc . , melbourne , fla . , 1981 , based on notes left by harry bateman , reprint of the 1955 original . s. g. samko , a. a. kilbas and o. i. marichev , _ fractional integrals and derivatives _ , gordon and breach science publishers , yverdon , 1993 , theory and applications , edited and with a foreword by s. m. nikolski , translated from the 1987 russian original , revised by the authors .

we establish the existence of small - amplitude uni- and bimodal steady periodic gravity waves with an affine vorticity distribution . the solutions describe waves with critical layers and an arbitrary number of crests and troughs in each minimal period . our bifurcation argument differs slightly from earlier theory , and under certain conditions we prove that the waves found are different from the ones in previous investigations . an important part of the analysis is a fairly complete description of the small - amplitude solutions . finally , we investigate the asymptotic behavior of solutions on the local bifurcation set .

1508.04664

up until fairly recently , most authors working with steady water waves have made the assumption that the vorticity @xmath0 of the velocity field @xmath1 vanishes identically . such waves are known as irrotational , as opposed to rotational waves where @xmath2 is allowed to be nonzero . rotational waves can exhibit more exotic behavior than irrotational ones , including interior stagnation points and critical layers of closed streamlines @xcite . stagnation points correspond to fluid particles that are stationary with respect to the wave , and for irrotational flows this can only occur at a sharp crest @xcite . irrotational waves are mathematically simpler to work with than rotational ones , due to the availability of the velocity potential . the velocity potential is the harmonic conjugate of the stream function , thus enabling the use of tools such as complex analysis , which are not available with nonzero vorticity . the survey @xcite treats the theory of stokes waves an important class of irrotational waves and the results on the so - called stokes conjecture for such waves . this conjecture was not fully settled until the appearance of the paper @xcite . the first , and still the only known , explicit nontrivial traveling wave solution to the euler equations was given in @xcite and is rotational ( which was only later pointed out by stokes ) . much later came the first existence result for small - amplitude waves with general vorticity distributions @xcite . it was not , however , before the pioneering article @xcite that large - amplitude waves were constructed using an extension of the global bifurcation theory of rabinowitz @xcite , leading to renewed interest in rotational waves . due to the methods used , neither the waves in @xcite nor those in @xcite exhibit stagnation . the first waves with a critical layer were constructed in @xcite , having constant vorticity . a different approach was used in @xcite , allowing for wave profiles with overhang ( for which existence is still an open question , with some numerical evidence in the affirmative @xcite ) . the method of proof for the existence of nontrivial rotational waves is typically bifurcation from parallel flows with a prescribed vorticity distribution , which are described in @xcite . other authors have looked at waves with density stratification @xcite , waves with compactly supported vorticity @xcite and waves with a general vorticity distribution and stagnation @xcite . of particular interest to us are @xcite , which cover small - amplitudes waves with an affine vorticity distribution . this is the natural step up from the constant vorticity considered in @xcite , and the resulting waves can have an arbitrary number of critical layers @xcite . in this paper , which builds upon @xcite , we consider the same setting as in @xcite . small - amplitude solutions with an affine vorticity distribution are found by bifurcating from trivial solutions that depend naturally on three parameters . by using other choices for the bifurcation parameters in our argument , we obtain waves that , in general , are different from those found in @xcite . in order to show that these are new waves , we are led to examine the asymptotic behavior of the bifurcation curves ; in particular for carefully chosen special cases . a complicating factor for our choice of bifurcation parameters is that they require an additional condition on the parameters . this condition can be interpreted as a nondegeneracy condition for the equation governing the dimension of the linearized problem . another novel aspect of this work is a fairly complete description of the nontrivial solutions near any trivial solution where the linear problem is one - dimensional , and for a class of trivial solutions with a two - dimensional linearized problem . we also show , by explicit construction , that the dimension of the linear problem can become arbitrary large for certain wavenumbers . this opens up the possibility for waves with arbitrarily many modes . the outline of the paper is as follows : in we formulate the problem and describe the setting in which we will work . contains the bifurcation result for a one - dimensional kernel and gives the properties of the resulting bifurcation curves , while the final section , , covers two - dimensional bifurcation . some useful derivatives are listed in . we consider pure gravity waves . the fluid motion is assumed to be incompressible and two - dimensional , with the coordinate system oriented so that the @xmath3- and @xmath4-axes are horizontal and vertical , respectively . the fluid domain is bounded below by a flat bottom , and above by a free surface . within this setting , our aim is to construct solutions of the steady water - wave problem ; that is , to find a surface profile @xmath5 and a velocity field @xmath1 , defined in the fluid domain @xmath6 where @xmath7 is the depth of the undisturbed fluid , satisfying the euler equations [ eq : euler ] @xmath8 in @xmath9 . the surface profile is assumed to satisfy @xmath10 , so that the bottom is not exposed to air . in and the function @xmath11 is the pressure , and @xmath12 is the velocity at which the wave travels . in addition to the equations in , we impose the boundary conditions [ eq : euler_bc ] @xmath13 the first two boundary conditions are known as kinematic boundary conditions , and state that there is no flux through the surface or bottom . the dynamic boundary condition , , ensures that there is no jump in pressure across the free surface . we will only be searching for periodic waves , and so we introduce the wave number @xmath14 , and stipulate that all functions be @xmath15-periodic in @xmath3 . we now reformulate the water wave problem in terms of a potential @xmath16 , called the _ relative stream function_. from incompressibility , together with @xmath9 being simply connected , we know ( see for instance @xcite ) that there exists a function @xmath17 satisfying @xmath18 this function is uniquely determined by @xmath1 , up to a constant . the kinematic boundary condition can be reformulated as @xmath19 at @xmath20 , and so @xmath16 is constant on the bottom . similarly , we can use to deduce that @xmath16 is constant also on the surface . next , and can be used to show that @xmath21 where @xmath22 is the poisson bracket defined by @xmath23 furthermore , by also using the boundary conditions in , one can infer that the surface bernoulli equation @xmath24 holds for some @xmath25 . in terms of the stream function , the vorticity is given by @xmath26 which follows directly from the definition of vorticity in . observe also that @xmath16 is @xmath15-periodic in the horizontal variable . to see this , note that @xmath27 is also a stream function , taking the same constant values as @xmath16 on the boundary . by uniqueness , they must be identical . the motivation for introducing the stream function is that , for a prescribed vorticity , the preceding equations are in fact equivalent to the steady water - wave problem . a precise statement , taken from @xcite , can be found in below . we will use the subscript `` per '' to denote @xmath15-periodicity in the horizontal variable , and identify @xmath28-periodic functions with functions on the circle @xmath29 . [ sfform ] for @xmath30 , @xmath31 and a prescribed vorticity @xmath32 , the steady water - wave problem is equivalent to the stream function formulation [ eq : euler_stream ] @xmath33 for @xmath34 and constants @xmath35 and @xmath36 . as long as the fluid velocity does not exceed the wave velocity , so that there is no stagnation , the vorticity at a point only depends on the value of the stream function at that point . this dependency is described by what is known as the vorticity distribution . [ gammaexists ] suppose @xmath37 . then there exists a function @xmath38 such that @xmath39 in @xmath9 . a notable consequence of the existence of a vorticity distribution is that is trivially satisfied , because @xmath40 by the chain rule . observe also that the condition in is sufficient , but not necessary . by _ assuming _ the existence of a vorticity distribution , we will still obtain solutions of the water - wave problem , even if @xmath41 is not satisfied . in fact , the solutions that we will find can exhibit stagnation and critical layers . the introduction of the vorticity distribution is standard for rotational waves , and was used already in @xcite . we shall consider the case where @xmath38 is affine . by making a shift of @xmath16 , it is sufficient to consider the case of linear @xmath38 . after scaling to unit depth and scaling away the gravitational acceleration , the stream function formulation reduces to [ problem ] @xmath42 where we have introduced the bottom @xmath43 and the surface @xmath44 . the parameter @xmath45 in controls the vorticity , and will be assumed to be _ negative_. for positive @xmath45 , one - dimensional but not higher - dimensional bifurcation is possible . more discussion on this can be found in @xcite . observe now that the system makes sense also for less regular function spaces than those specified in , and we will therefore allow for less regular ( but still classical ) solutions . more precisely , we will search for solutions @xmath46 where @xmath47 , and the subscript `` even '' signifies the subspace of functions which are even in the horizontal variable . the motivation for working in these hlder spaces is that then holds . [ rem : optimal_regularity ] due to and elliptic regularity ( * ? ? ? * theorem 9.26 ) for the differential operator @xmath48 , the stream function @xmath16 is smooth ( actually analytic ) in @xmath9 . in fact , we show in that our solutions will be of the best possibility regularity under the conditions and function spaces given here . the solutions of that we shall construct will be perturbations of laminar flows for which the velocity field is horizontal , but depth - dependent . these are the trivial solutions of , in the sense that @xmath49 and the stream function @xmath16 only depends on @xmath4 . by integrating , we arrive at trivial solutions of the form @xmath50{|\alpha|^{1/2 } ( y-1)+\lambda } , \qquad \lambda = ( \mu,\alpha,\lambda)\in { \mathbb{r } } ^3,\ ] ] with corresponding @xmath51 , @xmath52 and @xmath53 determined from as @xmath54 our goal is to find nontrivial solutions of for certain values of @xmath55 , corresponding to these particular values of @xmath36 , @xmath56 and @xmath57 . for technical reasons which we will elucidate later in , it is assumed that @xmath58 as in , we will often omit the dependence on @xmath55 in our notation . the main difficulty with the system is that it is a free - boundary problem , which entails that the domain is a priori unknown . there are several ways to remedy this . here , we will use the flattening transform @xmath59 giving a bijection from the sets @xmath60 , @xmath61 and @xmath62 onto @xmath63 \ } , \qquad \hat b \coloneqq \{(x , s ) : s=0 \ } , \qquad \hat s \coloneqq \{(x , s):s=1\},\ ] ] respectively , where we have introduced @xmath64 . using that @xmath65 , the map @xmath66 is a @xmath67-diffeomorphism , with inverse given by @xmath68{x,(1+\eta(x))s}.\ ] ] if we define @xmath69 on @xmath70 by @xmath71 , become @xmath72{\partial_x - \frac{s\eta_x}{1+\eta}\partial_s}\paren[\bigg]{\hat \psi_x - \frac{s \eta_x \hat \psi_s}{1+\eta } } + \frac{\hat{\psi}_{ss}}{(1+\eta)^2 } & = \alpha\hat{\psi } & \text{in $ \hat{\omega}$},\\ \frac{(1+\eta_x^2)\hat{\psi}_s}{2(1+\eta ) } + \eta & = q & \text{on $ \hat{s}$ } , \end{aligned}\ ] ] in the new flattened variables , for which we have the following : [ mcelem ] for functions @xmath73 and @xmath74 , the stream function formulation is equivalent to the transformed problem in for @xmath73 and @xmath75{\hat{\psi } \in c^{2,\beta}_{\mathrm{per , even}}(\overline{\hat{\omega } } , { \mathbb{r } } ) : \hat{\psi}|_{s=0 } = m_0,\ , \hat{\psi}|_{s=1}=m_1}.\ ] ] moreover , in this setting , a pair @xmath76 solves if and only if @xmath77 . with the trivial solutions found and the flattening transform introduced , we now elaborate on . any solution which is sufficiently close to a trivial solution is in fact analytic , as long as holds . the precise statement can be found in below . [ thm : optimal_regularity ] suppose that a solution @xmath78 of the problem in @xmath79 is such that the normal derivative @xmath80 of the stream function vanishes at no point on the surface . then we have the following : a. the surface profile @xmath5 is analytic . b. the stream function @xmath16 extends to an analytic function on an open set containing @xmath81 . the assumption on @xmath82 holds when @xmath55 satisfies and @xmath69 is sufficiently close to @xmath83 in @xmath84 . we start by showing that @xmath5 is analytic . for this , we will use the approach taken in @xcite , which is to apply ( * ? ? ? * theorem 3.2 ) . in @xcite this was done under the assumption of no stagnation , but it is sufficient to assume that stagnation does not occur on the surface . without loss of generality , we assume that @xmath85 . let @xmath86 be the component of @xmath87 that does not contain @xmath9 . proceed to define the function @xmath88 by @xmath89 and the differential operator @xmath90 by @xmath91 . observe that imply that @xmath92 on @xmath62 . all the assumptions of ( * ? ? ? * theorem 3.2 ) are now satisfied , with @xmath93 ( see the remark immediately after the theorem ) . we conclude that @xmath5 is analytic . note now that the differential operator @xmath90 is strongly elliptic in the sense of ( * ? ? ? * equation ( 1.7 ) ) . equipped with the fact that @xmath5 is analytic , we can use ( * ? ? ? * theorem a ) to conclude that @xmath16 extends to an analytic function on an open set containing @xmath81 . the final part of the theorem follows because @xmath94 where @xmath95 is bounded away from @xmath96 as long as @xmath69 is sufficiently close to @xmath97 in @xmath84 , due to the assumption that holds . is a local result at heart . it is clear from the proof that if @xmath98 , then @xmath5 is analytic in a neighborhood of @xmath99 . this , in turn , implies that @xmath16 extends analytically across the surface near the point @xmath100 . recall that the stream function @xmath16 is analytic on @xmath9 , regardless of whether the condition on @xmath82 on the surface in is satisfied . it is worth noting that this implies that the streamlines are analytic curves away from stagnation points by the analytic implicit function theorem ( * ? ? ? * theorem 4.5.4 ) . in order to linearize around a trivial solution @xmath97 , we write @xmath101 , and introduce the spaces @xmath102{\hat{\phi } \in c^{2,\beta}_{\text{per , even}}(\overline{\hat{\omega } } , { \mathbb{r } } ) : \hat{\phi}|_{s=0 } = \hat{\phi}|_{s=1}=0 } \end{aligned}\ ] ] and @xmath103 we will write @xmath104 for elements of @xmath105 . to capture our assumptions , it is convenient to define the sets @xmath106 and , to enforce that @xmath107 and hold , @xmath108 we now define the map @xmath109 by [ mcfdef ] @xmath110\paren[\bigg]{\partial_x - \frac{s\eta_x}{1+\eta}\partial_s}\paren[\bigg]{\hat{\phi_x } - \frac{s\eta_x(\psi_{0s}+\hat{\phi}_s)}{1+\eta}}\\ + \frac{\psi_{0ss}+\hat{\phi}_{ss}}{(1+\eta)^2 } - \alpha(\psi_{0ss}+\hat{\phi}),\end{multlined } \notag \end{aligned}\ ] ] where @xmath97 is as in and @xmath51 is given in . in , it is understood that the functions @xmath111 and @xmath112 are evaluated at @xmath113 . it is clear that @xmath114 is well defined as a map @xmath115 . we wish to solve the equation @xmath116 from @xcite , we have the following lemma . the first part is a corollary of and the second part is an immediate consequence of the definition of @xmath114 . the problem for @xmath117 is equivalent to , and @xmath118 for @xmath119 if and only if @xmath120 . moreover , @xmath121 . we obtain the linearized problem by taking the partial derivative of @xmath114 with respect to @xmath122 at the point @xmath123 . this yields @xmath124 where it again is understood that the functions are evaluated at @xmath113 in . by introducing an isomorphism , see , we can transform @xmath125 into a simpler operator . for this purpose , define @xmath126 { \phi \in c^{2,\beta}_{\mathrm{per , even}}(\overline{\hat{\omega } } , { \mathbb{r } } ): \phi|_{s=0 } = 0 } , \quad \tilde{x } \coloneqq \brac[\big]{(\eta,\phi ) \in x_1 \times \tilde{x}_2 } , \ ] ] where we have the inclusions @xmath127 and @xmath128 . we will denote elements of @xmath129 by @xmath130 . [ mctprop ] the bounded linear operator @xmath131 given by @xmath132 is an isomorphism of banach spaces . the operator @xmath133 satisfies @xmath134 that @xmath135 is well defined and an isomorphism is almost immediate . the expression for @xmath136 in follows by direct computation . introduce the complex parameter @xmath137 where @xmath138 will be some integer multiple of the wavenumber @xmath139 in general . this parameter will appear in functions of the form @xmath140 and @xmath141 , which are always real - valued . we record that @xmath142 \cos(|\theta_{k}|s ) , & \alpha + k^2 < 0 , \end{cases}\\ \frac{\sinh(\theta_{k } s)}{\theta_{k } } & = \begin{cases } \sinh(\theta_{k } s)/\theta_{k } , & \alpha+k^2 \ge 0 , \\[3pt ] \sin(|\theta_{k}| s)/|\theta_{k}| , & \alpha + k^2 < 0 . \end{cases } \end{aligned}\ ] ] when @xmath143 , we will interpret expressions with @xmath144 as extended by continuity . in particular , @xmath145 is interpreted as @xmath146 . we now describe the kernel of @xmath136 , which is related to the kernel of @xmath147 through @xmath148 , in terms of the above functions . the following lemma is stated , but not proved , in @xcite . we include its proof because it is instructive . [ kernellemma ] let @xmath149 . a basis for @xmath150 is then given by @xmath151 , where @xmath152 and @xmath153 is the finite set of all @xmath154 satisfying the _ kernel equation _ @xmath155 where @xmath156 suppose that @xmath157 , and expand it in a fourier series @xmath158 from @xmath159 , we deduce that the coefficients satisfy @xmath160 while @xmath161 and @xmath162 yield the boundary conditions @xmath163 for all @xmath164 . the general solution of with the boundary condition is @xmath165 for which the robin condition reduces to @xmath166 hence , if @xmath167 ( and thus @xmath168 ) is nonzero , then @xmath169 holds . observe that implies that @xmath170 ; otherwise we would have @xmath171 , and therefore @xmath172 . thus , by inserting the definition of @xmath97 into , we arrive at . this condition is also sufficient for @xmath173 to lie in the kernel . the set @xmath153 of @xmath174 such that holds is finite , because there are only finitely many @xmath164 such that @xmath175 , and the function @xmath176 is strictly increasing as soon as @xmath177 . [ remark : nontrivial_solutions ] in order to get nontrivial solutions , we shall assume that all solutions of the kernel equation lie in @xmath178 . the function @xmath179 , see , does not depend on @xmath3 . the bifurcation results in are valid under the assumption that @xmath150 is respectively one- and two - dimensional . below is a general result on the kernel equation from @xcite , which in particular shows that it is indeed possible to choose @xmath149 such that the dimension of the kernel is one or two . [ bif kernels ] let @xmath180 and @xmath149 . a. for every @xmath45 and any @xmath181 there are @xmath182 and @xmath183 such that the kernel equation holds for at least @xmath184 . b. the values of @xmath45 for which there exist @xmath183 and @xmath182 such that the kernel equation holds for more than one @xmath185 are isolated . c. for any @xmath183 with @xmath186 , and any pair @xmath187 with @xmath188 , there are @xmath189 and @xmath182 such that holds for @xmath190 and for no other @xmath191 . for later use , we now give some explicit examples of one- and two - dimensional kernels for @xmath136 , all corresponding to @xmath192 . to simplify the parameters involved , we choose specific values for @xmath139 . [ example : explicit_kernels ] let @xmath193 be the smallest positive solution of the equation @xmath194 . a. when @xmath195 , @xmath196 , @xmath197 and @xmath198 , the kernel is one - dimensional , being spanned by @xmath199 . b. let @xmath200 . when @xmath201 , @xmath202 and @xmath203 , the kernel is two - dimensional , with @xmath204 . c. let @xmath205 . when @xmath206 , @xmath207 and @xmath208 , the kernel is two - dimensional , with @xmath209 before we move on , we address a question that was raised in @xcite : does there exist @xmath149 such that @xmath210 is at least three - dimensional ? by also letting the wave number @xmath139 vary , this question was answered in the affirmative for dimension three in @xcite . in essence , their result says that many two - dimensional kernels can be modified in order to yield a three - dimensional kernel . here , we use a different approach to find kernels of arbitrary dimension for any @xmath139 in a set @xmath211 which is dense in @xmath212 . for any @xmath213 and @xmath183 such that @xmath214 modulo @xmath215 , we can obtain @xmath216 by choosing @xmath182 to satisfy @xmath217 which reduces the kernel equation to finding @xmath218 and @xmath154 such that @xmath219 is satisfied . this can be written on the form @xmath220 we first consider the case @xmath221 . [ lemma : arbitrary_kernel_pi ] for @xmath221 and any @xmath222 , there exists @xmath149 such that @xmath223 . when @xmath224 and @xmath225 for an odd number @xmath226 , becomes the diophantine equation @xmath227 the size of the kernel then corresponds to the number of representations of @xmath228 as the sum of two squares . as long as @xmath228 is not a square number , any representation has @xmath229 ( see ) . in order to conclude , we therefore need to find an odd non - square number @xmath228 such that @xmath228 has exactly @xmath230 representations as a sum of squares . by ( * ? ? ? * theorem 3 in chapter 2 ) , this is the case when @xmath231 for a prime @xmath232 . some examples of are listed below , using the smallest @xmath228 yielding @xmath230 representations in . these values of @xmath228 can easily be deduced by using the general formula given in @xcite for the number of representations as a sum of squares . a. since @xmath233 ( with no other representations ) , the choice @xmath234 yields a three - dimensional kernel , with @xmath235 . b. since @xmath236 , the choice @xmath237 yields a four - dimensional kernel , with @xmath238 . let @xmath239 denote the set of positive rational numbers with odd numerators when reduced to lowest terms . we can then generalize in the following way : [ theorem : arbitrary_kernel ] for @xmath240 and any @xmath222 , there exists @xmath149 such that @xmath223 . write @xmath241 , with @xmath242 and @xmath146 coprime . when @xmath243 , becomes @xmath244 choosing @xmath231 for a prime @xmath232 , we know that has exactly @xmath230 solutions @xmath245 in @xmath246 . the pairs @xmath247 defined by @xmath248 then solve . moreover , these are the only solutions : suppose that @xmath249 solves . then @xmath250 and @xmath251 , by coprimality of @xmath242 and @xmath146 . it follows that @xmath252 and @xmath253 , where @xmath254 and @xmath255 solve @xmath256 . since @xmath228 is odd , @xmath257 . uniqueness in now yields the result . although we provide kernels of arbitrary dimension in , the corresponding triples @xmath55 satisfy @xmath216 , unlike the three - dimensional kernels obtained in @xcite . in particular , this means that the two - dimensional bifurcation result in does not apply for the kernels from with @xmath258 . we remark that an obstacle for higher - dimensional bifurcation is that there are only four parameters to work with , namely @xmath55 and @xmath139 . this may be remedied by for instance including surface tension . an application of is one - dimensional bifurcation with several different wave numbers for fixed @xmath55 . if the set @xmath259 is such @xmath260 for all @xmath261 , we can make restrictions to each @xmath262 and then apply . this will yield @xmath230 different solution curves . two examples for which the condition on @xmath153 is fulfilled are @xmath263 and @xmath264 , corresponding to @xmath265 and @xmath266 , respectively . we will need the following technical result . [ mclisfred ] for each @xmath149 , the operator @xmath136 is fredholm , with index @xmath96 . the range of @xmath136 is the orthogonal complement of @xmath267 { ( \eta_{\phi},\phi ) : \phi \in \ker { \mathcal{l}}(\lambda ) } \subset \tilde{x } \subset y\ ] ] in @xmath268 with respect to the inner product @xmath269 let @xmath270 for @xmath271 , where @xmath272 and @xmath153 are as in . the projection @xmath273 onto @xmath274 along @xmath275 is then given by @xmath276 [ rem : fredholm ] that @xmath277 be fredholm is the main reason for assuming . when we have equality in , reduces to @xmath278 , whence @xmath279 . since @xmath280 is not closed in @xmath281 , the operator @xmath277 can not be fredholm . let @xmath282 be a triple @xmath283 such that @xmath284 has a nontrivial kernel , say with basis @xmath285 . then the pairs @xmath286 span the kernel of @xmath287 . since the kernel is finite - dimensional , there exists a closed subspace @xmath288 such that @xmath289 by , we can also decompose @xmath268 into the direct sum @xmath290 where @xmath291 , since these are orthogonal complements in the inner product on @xmath268 . applying the lyapunov schmidt reduction ( see e.g. kielhfer @xcite ) for these decompositions of @xmath105 and @xmath268 , we obtain the following lemma . [ lyasch ] there exist open neighborhoods @xmath292 of @xmath96 in @xmath293 , @xmath294 of @xmath96 in @xmath295 , and @xmath296 of @xmath282 in @xmath297 , and a uniquely determined function @xmath298 such that @xmath299 if and only if @xmath300 and @xmath301 solves the finite - dimensional problem @xmath302 where @xmath303 the function @xmath16 is smooth , and satisfies @xmath304 for all @xmath305 , and @xmath306 . we are now in a position to show that a curve of nontrivial solutions of bifurcates from each point @xmath307 where the kernel of @xmath308 is one - dimensional , given that @xmath282 satisfies an additional technical condition . this technical condition comes from below . [ ipcalclemmalambda ] suppose that @xmath309 , where @xmath310 is a basis function given by the formula . then , if @xmath311 is the corresponding basis function of @xmath274 , we have @xmath312 where @xmath313{\cot(\lambda ) + \frac{\mu |\alpha|^{3/2}}{2}}\ ] ] does not depend on @xmath314 . in particular , @xmath315 recalling , we find the derivatives @xmath316 using that @xmath317 , we get @xmath318 where @xmath319{\cot(\lambda ) + \frac{\mu^2 \abs{\alpha}^{3/2}}{2 } } , \end{aligned}\ ] ] by definition of @xmath97 and the kernel equation . since @xmath320 , with @xmath321 we now find @xmath322 which is with @xmath323 . we will refer to @xmath324 as _ the transversality condition _ , because it corresponds to transversality in the crandall rabinowitz theorem ( see @xcite or @xcite ) . if transversality fails at some @xmath149 , then it is possible to rechoose @xmath182 and @xmath183 , while keeping @xmath45 constant , in such a way that the transversality condition holds and the right hand side @xmath325 of the kernel equation remains unchanged . this ensures that the dimension of the kernel does not change . [ orthogonalitylemma ] suppose that @xmath326 is such that the transversality condition fails . then there are @xmath327 with @xmath328 such that @xmath329 the triple @xmath55 can be chosen arbitrarily close to @xmath330 . first observe that since @xmath331 we can use the implicit function theorem to deduce that there is an open rectangle @xmath332 containing @xmath333 , and a smooth function @xmath334 such that @xmath335 for all @xmath336 . by the definition of @xmath182 , we have @xmath337 because @xmath338{\cot(\lambda_0 ) + \frac{\mu_0 ^ 2 |\alpha_0|^{3/2}}{2}}=0,\ ] ] due to the assumption that transversality fails at @xmath330 . since implies that @xmath339{\cot(\lambda ) + \frac{\mu(\lambda)^2|\alpha_0|^{3/2}}{2}}'(\lambda_0 ) = - \frac{1}{\sin^2(\lambda_0 ) } \neq 0,\ ] ] the result follows . the one - dimensional bifurcation result is an application of the crandall rabinowitz bifurcation theorem . to clarify the proof of the two - dimensional bifurcation in the next section , we will nonetheless spell out the details of the proof . [ 1dbif ] suppose that @xmath340 is such that @xmath341 where @xmath342 , with @xmath343 for the unique @xmath185 that satisfies the kernel equation . if the transversality condition holds , then there exists a @xmath344-curve @xmath345 of nontrivial small - amplitude solutions to @xmath346 in @xmath347 , passing through @xmath348 , with @xmath349 in a neighborhood of @xmath350 in @xmath347 , these are all the nontrivial solutions of . using , we know that there exists a neighborhood of @xmath350 in @xmath347 for which the equation @xmath351 is equivalent to @xmath352 , where @xmath353 . from the same lemma we also have the identity @xmath354 , and hence we can write @xmath355{\phi(tz,\lambda)}{\ , \mathrm{d}z}=t\psi(t,\lambda),\ ] ] where @xmath356 is smooth . for nontrivial solutions ( @xmath357 ) , the equations @xmath358 and @xmath359 are equivalent , whence we need only concern ourselves with the latter equation . we want to apply the implicit function theorem to @xmath360 , which requires that @xmath361 and @xmath362 ( recall that @xmath274 is one - dimensional ) . now , from , we find @xmath363 so these are the derivatives of @xmath364 we need to compute . by the definition of @xmath364 , @xmath365 and so by evaluating in @xmath366 , and using the properties of @xmath16 listed in , we have @xmath367 which also yields @xmath368 we now obtain @xmath369 because @xmath370 by the last part of , and because @xmath371 projects along @xmath372 . similarly , @xmath373 note that @xmath374 , and hence @xmath375 which implies that can be written @xmath376 again using that @xmath371 projects along @xmath372 . we can now use to deduce that @xmath377 , due to the assumption of transversality . finally , since @xmath378 and @xmath379 , we can invoke the implicit function theorem to deduce that there exists an @xmath380 and a @xmath344-function @xmath381 with @xmath382 such that @xmath383 . moreover , the curve @xmath384 describes all solutions to @xmath385 in a neighborhood of @xmath350 . the corresponding solution curve to @xmath386 is @xmath387 , where @xmath388 . it follows that @xmath389 and we can conclude , once again using the properties of @xmath16 given in , that @xmath390 and @xmath391 . consequently , we obtain . if @xmath392 is sufficiently small , the waves obtained from are stokes waves . this can be seen from the asymptotic formula in . the one - dimensional bifurcation result in is analogous to ( * ? ? ? * theorem 4.6 ) , which uses @xmath182 instead of @xmath183 as the bifurcation parameter . other than the parameters , the main difference between the theorems is the addition of the transversality condition for bifurcation with respect to @xmath183 . we will show that the resulting solutions are different in general . this can be done by showing that @xmath393 and @xmath394 , in the notation of @xcite , are not constant . ( if they were constant , we would obtain the same solutions by uniqueness in . ) shows that we need to consider at least second - order properties of the bifurcation curve in order to achieve this . [ lambdadotzero ] under the hypothesis of , the function @xmath395 satisfies @xmath396 and so the bifurcation parameter is constant to the first order along the bifurcation curve . we adopt the notation used in the proof of . differentiation of the identity @xmath397 , and evaluation at @xmath366 , yields the equation @xmath398 for the derivative of @xmath395 at the origin . from and the discussion immediately after , we know that @xmath399 which means that uniquely determines @xmath400 . however , we still need to compute @xmath401 . from , we obtain @xmath402 and differentiation in leads to @xmath403\pi_z d_{w}^2 { \mathcal{f}}(tw^*+\psi(tw^*,\lambda),\lambda)(w^*+d_w\psi(tw^*,\lambda)w^*)^2 \\+ \pi_z d_w { \mathcal{f}}(tw^*+\psi(tw^*,\lambda),\lambda)d_w^2\psi(tw^*,\lambda)(w^*)^2.\end{multlined}\ ] ] hence , by using the properties of @xmath16 given in , and using that @xmath371 projects along the range of @xmath308 , we find @xmath404 using and the formula for @xmath371 given in , we now find @xmath405 and so it is sufficient to show that the numerator , @xmath406 vanishes . since @xmath407 with @xmath408 being a separable function of @xmath3 and @xmath146 , so is @xmath409 . moreover , we see from that their @xmath3-dependence is through @xmath410 . thus each term in @xmath411 and @xmath412 has an @xmath3-dependence of the form @xmath413 with @xmath414 ( see the derivatives listed in ) . it follows that we will be integrating terms whose @xmath3-dependence is @xmath413 with @xmath415 . but if @xmath416 are nonnegative integers , then @xmath417 whenever @xmath418 is odd . also holds if @xmath182 is substituted for @xmath183 . since the proof of only depends on the fact that the numerator in vanishes , we can conclude that one also has @xmath419 when using @xmath182 as the bifurcation parameter . to consider the question of second - order behavior of the solution curve , let us return to the expression for @xmath420 we found in the one - dimensional bifurcation result , namely @xmath421 where @xmath422 and @xmath423 are evaluated at @xmath424 . taking another derivative in yields @xmath425 this expression can be simplified significantly when @xmath366 . an expression for @xmath426 can also be found , akin to how was derived . the details are omitted , see for instance ( * ? ? ? * section i.6 ) . [ prop : second_derivatives ] under the hypothesis of , we have @xmath427 and @xmath428 for the solution curve @xmath429 . in view of , we have that @xmath304 for all @xmath55 in an open neighborhood of @xmath282 , and thus @xmath430 and @xmath431 are both zero . finally , the second term in vanishes due to . we now give a more transparent description of @xmath432 , which shows is required for computing both @xmath433 and @xmath426 . [ lem : psi_second_derivative ] write @xmath434 and let @xmath435 be such that @xmath436 where the coefficients @xmath437 and @xmath438 solve the boundary value problems @xmath439 for @xmath440 . then @xmath441 that @xmath442 can always be written as in can be deduced from the expressions for the derivatives of @xmath114 listed in . the function @xmath16 satisfies the identity @xmath443 for sufficiently small @xmath444 . if we take two derivatives of this equation and evaluate at @xmath366 we obtain the equation @xmath445 for @xmath432 . since we established in the proof of that @xmath446 lies in the range of @xmath308 , and since @xmath371 projects along @xmath447 , this implies that @xmath448 which uniquely determines @xmath432 . if we now introduce the function @xmath449 by @xmath450 then can be written @xmath451 utilizing , this proves the theorem . due to the form of @xmath452 and the expressions for the derivatives of @xmath114 in , we know that the coefficients @xmath437 and @xmath438 in are polynomials in @xmath146 , @xmath453 and @xmath454 for @xmath455 . however , computing them explicitly for general @xmath282 is a daunting task , especially if one wishes to write them in a tractable form . we will content ourselves with considering the first special case of , which was constructed specifically to make @xmath408 and @xmath97 as simple as possible . this , in turn , yields particularly simple @xmath437 and @xmath438 . [ thm : special_case ] when @xmath195 and @xmath456 , the functions @xmath437 and @xmath438 are given by @xmath457 respectively . this yields @xmath458 in particular , this implies that the solutions obtained in are , in general , different from the ones obtained in @xcite . the kernel of @xmath284 is spanned by @xmath459 , and moreover @xmath460 . one may check that for this special case , the coefficients in are given by @xmath461 it follows by direct verification that the functions @xmath437 and @xmath438 in solve . finally , a long ( but direct ) computation from the expression for @xmath462 in yields . and @xmath182 in the setting of . we use superscripts to separate the two different @xmath463 . ] the image to have in mind for bifurcation with respect to @xmath183 and @xmath182 from the same point is that of , where the parameter @xmath464 is held fixed . because @xmath465 for this special case , the resulting bifurcation curves lie in different planes . we mention that there is a particularly simple connection between @xmath426 and @xmath466 , which holds for general @xmath282 . from the expression for @xmath426 in , and the corresponding one for @xmath466 , it follows that @xmath467 where we have used and ( * ? ? ? * ( 4.9 ) ) . in particular , this implies that @xmath426 and @xmath466 must have either the same or opposite sign , depending on the sign of @xmath182 and which `` side '' of the transversality condition @xmath282 is on . this results in for the special case in . note that equality in corresponds to the extreme point on the graph in . from the proof of , together with the bifurcation curves in the setting of . the dashed lines are asymptotes . ] using , we can describe all solutions of in a neighborhood of any @xmath468 in @xmath469 for which @xmath470 is one - dimensional and the transversality condition is fulfilled . suppose that we have such a point , and that the kernel of @xmath284 is spanned by @xmath471 . for this fixed @xmath138 , define @xmath472 where @xmath242 was defined in . then @xmath473 , and @xmath474 due to transversality ( see ) . by the implicit function theorem there is a smooth function @xmath475 , defined on an open neighborhood of @xmath476 , such that @xmath477 and @xmath478 . let @xmath479 denote the graph of this function . observe that for any @xmath480 with @xmath481 , we must necessarily have @xmath482 and therefore @xmath483 since the kernel equation must have the same solutions as for @xmath282 . combined with the second part of ( note that this also holds if we include @xmath484 ) , and by possibly shrinking @xmath479 , we can assume that @xmath485 for all @xmath480 . we can also assume that the transversality condition holds on all of @xmath479 . this allows us to invoke at each point of @xmath479 , obtaining a family of solution curves . these are in fact all the nontrivial solutions near @xmath468 : can be found by using on the points of @xmath479 . ] [ thm : local_description_1d ] the family @xmath486 of solution curves bifurcating from points @xmath123 for @xmath480 , where @xmath479 is the surface defined below , constitutes all nontrivial solutions in a neighborhood of @xmath468 in @xmath469 . for each @xmath487 we have uniqueness in a set @xmath488 in the sense that all nontrivial solutions in @xmath489 are given by the solution curve obtained in . using the quantitative version of the implicit function theorem , and shrinking @xmath479 further , we can assume that @xmath490 is constant on @xmath479 . it is then clear that the family @xmath486 yields all nontrivial solutions in the open neighborhood @xmath491 of @xmath468 , see . for two - dimensional bifurcation we will use @xmath45 as the second bifurcation parameter . we therefore need the following analogue of for the parameter @xmath45 . [ ipcalclemmaalpha ] suppose that @xmath309 , where @xmath310 is a basis function given by the formula . then , if @xmath320 is the corresponding basis function of @xmath274 , we have @xmath492 where @xmath493{\frac{1}{\mu^2 \abs{\alpha}^2 \sin^2(\lambda ) } - \frac{\cot(\lambda)}{2\abs{\alpha}^{1/2}}}\ ] ] and @xmath494 is defined by @xmath495 suppose we have a two - dimensional kernel @xmath496 , where @xmath497 and @xmath282 satisfies the transversality condition ( which also appears for two - dimensional bifurcation ) . denote the subspace of @xmath105 consisting of functions that are @xmath498-periodic in the horizontal variable by @xmath499 . then @xmath500 . using the local description from , we obtain the set @xmath501 of all nontrivial solutions of in a neighborhood of @xmath468 in @xmath502 . similarly , under the condition that @xmath503 , the kernel of @xmath504 is spanned by @xmath505 , and we obtain the set @xmath506 of all nontrivial solutions in a neighborhood of @xmath468 in @xmath507 . the next result describes solutions near @xmath468 which are neither @xmath508- nor @xmath509-periodic . as mentioned before , we will use @xmath45 as the second bifurcation parameter . hence , we will look for solutions of the equation @xmath510 for @xmath511 and @xmath512 . let , for @xmath513 , the set @xmath514 consist of all @xmath515 such that @xmath516 . these are the nontrivial @xmath517-periodic solutions of near @xmath468 . [ 2dbif ] suppose that @xmath282 is such that the transversality condition holds , and that @xmath518 with @xmath497 and @xmath519 as in . furthermore , suppose that either @xmath520 or @xmath521 ( in which case @xmath522 ) . a. if @xmath503 , there exists a smooth family of nontrivial small - amplitude solutions @xmath523 of in @xmath524 , passing through @xmath525 when @xmath526 , with @xmath527 in a neighborhood of @xmath525 in @xmath524 , the set @xmath528 captures all nontrivial solutions of . b. let @xmath529 . if @xmath530 , there exists a smooth family of nontrivial small - amplitude solutions @xmath531 of in @xmath524 , passing through @xmath525 when @xmath532 , with @xmath533 in a neighborhood of @xmath525 in @xmath524 , the set @xmath534 contains all nontrivial solutions of such that @xmath535 in their projection @xmath536 on @xmath470 along @xmath295 . the lyapunov schmidt reduction in tells us that there is a neighborhood of @xmath468 in @xmath469 in which is equivalent to the equation @xmath537 , where @xmath538 recall that @xmath539 , where @xmath540 . if we let @xmath541 denote the projection onto the span of @xmath542 along the image of @xmath284 , then @xmath543 . defining @xmath544 for @xmath545 , the equation @xmath546 can then be rewritten as the system of equations @xmath547 let us first consider case _ ( i ) _ , where @xmath503 . we claim that @xmath548 for all @xmath549 , @xmath550 and @xmath55 . we will show the first identity ; the proof of the second being similar . using that @xmath551 , an application of the lyapunov schmidt reduction in @xmath552 yields a function @xmath553 with mapping into @xmath554 satisfying @xmath555 for all @xmath550 in a neighborhood of @xmath556 . due to the @xmath509-periodicity of @xmath557 , holds with @xmath558 replaced by @xmath371 , whence uniqueness of @xmath16 yields @xmath559 . the first identity in then follows by definition of @xmath560 . following the proof of , we now introduce the function @xmath561 by @xmath562 and similarly the function @xmath563 . then @xmath564 by , and we get different cases for depending on the values of @xmath549 and @xmath550 . if both @xmath549 and @xmath550 vanish , we get trivial solutions . when @xmath565 but @xmath566 , we know from that the system reduces to @xmath567 , the solutions of which correspond to @xmath506 , while when @xmath568 but @xmath569 we get solutions of which lie in @xmath501 . the remaining case is where both @xmath549 and @xmath550 are allowed to be nonzero , which amounts to investigating the solutions of @xmath570 in a neighborhood of @xmath571 . we will look at those solutions with @xmath572 . for this , we intend to use the implicit function theorem . note that from we get @xmath573 , and that by definition of @xmath364 we have @xmath574 where @xmath575 . using the properties of @xmath16 and , we can therefore conclude that @xmath576 . it remains to show that the derivative @xmath577 is invertible . using analogous computations to those after in the proof of , we have @xmath578 for @xmath545 and @xmath579 . it follows that @xmath577 is invertible if and only if the determinant @xmath580 \ip{d_{\alpha } { \mathcal{l}}(\lambda^ * ) \phi_1^*}{\tilde{w}_1^*}_y & \ip{d_{\alpha } { \mathcal{l}}(\lambda^ * ) \phi_2^*}{\tilde{w}_2^*}_y \end{vmatrix}\ ] ] is nonzero . the inner products appearing in this determinant have already been computed in and . using these , and elementary properties of determinants , we have @xmath581 b \paren * { \frac{\sinh(\theta_{k_1})}{\theta_{k_1 } } } ^2 + f(\theta_{k_1 } ) & b \paren * { \frac{\sinh(\theta_{k_2})}{\theta_{k_2 } } } ^2 + f(\theta_{k_2 } ) \end{vmatrix}\\ & = a \paren[\bigg ] { \paren * { \frac{\sinh(\theta_{k_1})}{\theta_{k_1 } } } ^2 f(\theta_{k_2 } ) - \paren * { \frac{\sinh(\theta_{k_2})}{\theta_{k_2 } } } ^2 f(\theta_{k_1 } ) } . \end{aligned}\ ] ] observe that of @xmath582 and @xmath583 , only @xmath583 can vanish . suppose for the moment that also @xmath584 . then @xmath585{\paren*{\frac{\theta_{k_j}}{\sinh(\theta_{k_j})}}^2 - \theta_{k_j}\frac{\cosh(\theta_{k_j})}{\sinh(\theta_{k_j})}}\\ & = \frac{\pi}{2\kappa\theta_{k_j}^2}\paren*{\frac{\sinh(\theta_{k_j})}{\theta_{k_j}}}^2\paren[\big]{r(\lambda^*)^2-\theta_{k_j}^2 -r(\lambda^ * ) } , \end{aligned}\ ] ] whence the determinant in can be written as @xmath586{\frac{1}{\theta_{k_2}^2}-\frac{1}{\theta_{k_1}^2}},\ ] ] where @xmath587 is nonzero due to the assumption of transversality . hence , we immediately see that @xmath588 is nonzero if and only if @xmath520 . a similar computation shows that @xmath589 when @xmath521 . this concludes the proof of part @xmath590 . next , we move on to case _ ( ii ) _ , where @xmath530 . we still find that @xmath591 for all @xmath550 and @xmath55 , and so we can introduce @xmath592 as before ; only now written using the polar coordinates @xmath593 ( identifying @xmath594 and @xmath595 ) . then @xmath596 . for @xmath597 , the corresponding identity in is no longer true in general , but we still have @xmath598 . we therefore introduce @xmath563 through @xmath599 which yields @xmath600 . like for case _ ( i ) _ , the solutions of @xmath601 near @xmath468 for @xmath569 correspond to solutions in @xmath602 . when @xmath566 , also @xmath603 , and so is equivalent to the problem @xmath604 which we will now consider again , we will use the implicit function theorem to find solutions with @xmath605 . due to similar computations as those for case _ ( i ) _ , we have @xmath606 and @xmath607 for all @xmath608 and @xmath579 . to find the derivatives of @xmath563 , note that @xmath609 and so @xmath610 where we have used that @xmath611 and @xmath612 are orthogonal in @xmath613 . from the preceding , we see that the derivative @xmath614 is invertible if and only if the determinant @xmath615 \ip{d_{\alpha } { \mathcal{l}}(\lambda^ * ) \phi_1^*}{\tilde{w}_1^*}_y & \ip{d_{\alpha } { \mathcal{l}}(\lambda^ * ) \phi_2^*}{\tilde{w}_2^*}_y\sin(v)\\ \end{vmatrix}\\ & = c\sin(v ) \end{aligned}\ ] ] is nonzero , where @xmath588 is the determinant introduced in . we know that @xmath616 under the assumptions of the theorem , so we can apply the implicit function theorem at each @xmath617 as long as @xmath618 . this can be done uniformly in @xmath608 as long as @xmath619 is bounded away from zero . in case @xmath590 , the surface profiles have multiple crests and troughs in each minimal period , at least for sufficiently small @xmath620 . this follows from the asymptotic formula in . observe that the second special case listed in has @xmath621 , while the third has @xmath622 . they therefore fall into different cases in . we will now present some properties of the sheets of solutions that were found in our two - dimensional bifurcation result , . the main purpose of this is to show that these solutions , found by bifurcating with respect to @xmath183 and @xmath45 , are not , in general , the same solutions as those that were found in ( * ? ? ? * theorem 4.8 ) . like for one - dimensional bifurcation , our result differs from the one in @xcite by the use of @xmath183 instead of @xmath182 , and the addition of the transversality condition . the first step towards showing that our waves are new is , the two - dimensional counterpart of [ gradzero ] for the solution sheets obtained in , we have the following : * in case ( i ) , the solutions satisfy @xmath623 * in case ( ii ) , we have @xmath624 as long as @xmath625 . the proof for case _ ( i ) _ is a simpler variant of that for case _ ( ii ) _ , so we focus on the latter . by definition of @xmath626 and @xmath627 , we have the identity @xmath628 for @xmath629 . through taking derivatives with respect to @xmath242 , this implies that @xmath630 where the derivatives of @xmath631 are evaluated at @xmath632 . this linear system of equations can be solved for @xmath633 and @xmath634 because the determinant @xmath635 in is nonzero . in order to show that the derivatives vanish , it is therefore sufficient ( and necessary ) to show that @xmath636 for @xmath545 . using the definitions of @xmath561 and @xmath563 , we find @xmath637 \frac{1}{2 } \cos^2(v ) \phi_{2t_1 t_1}(0,0,\lambda^ * ) + \frac{1}{2 } \sin^2(v ) \phi_{2t_2 t_2}(0,0,\lambda^ * ) \\[3pt ] + \sin(v ) \cos(v ) \phi_{2t_1 t_2}(0,0,\lambda^*),\end{multlined } \end{aligned}\ ] ] where @xmath638{d_w^2 { \mathcal{f}}(0,\lambda^*)(w_i^ * , w_j^*)}{\tilde{w}_l^*}_y}{\norm{\tilde{w}_l^*}_y^2}\tilde{w}_l^ * \end{aligned}\ ] ] for @xmath639 . using orthogonality in @xmath640 , like in the proof , one can show that the derivatives in are zero , except possibly when @xmath641 or @xmath642 . this is only the case when @xmath643 and either @xmath644 and @xmath645 or @xmath646 and @xmath647 . we now show that @xmath633 and @xmath634 can indeed be nonzero when @xmath648 , which is not covered by , by considering the second special case listed in . [ special ii ] [ thm : special_case_ii ] let @xmath193 be the smallest positive solution of @xmath649 . when @xmath200 and @xmath650 , we have @xmath651 and so by , and that @xmath652 for a nonsingular matrix @xmath653 not depending on @xmath608 . in particular , @xmath633 and @xmath634 are both nonzero , except possibly for isolated values of @xmath608 . shows that the solutions obtained in are , in general , not the same as those obtained in @xcite at least when @xmath654 . we finish by using to prove a version of describing all nontrivial solutions in a neighborhood of points falling into case _ ( i)_. let therefore @xmath340 be such that it fulfills the assumptions in case _ ( i ) _ of . proceeding as in , we introduce the function @xmath655 where we know that @xmath656 as long as @xmath481 . then @xmath657 and @xmath658 , and so by the implicit function theorem there is a smooth function @xmath659 , defined on an open neighborhood of @xmath660 , such that @xmath661 and @xmath662 . let @xmath479 denote the points on the curve @xmath663 . if @xmath480 , then @xmath482 and @xmath664 , and by possibly shrinking @xmath479 we may also assume that the transversality condition holds . we can thus apply to each point of @xmath479 , yielding a family of bifurcating sheets . these are all the nontrivial solutions near @xmath468 . we omit the proof , which is essentially the same as for . the family @xmath486 of solution sheets bifurcating from points @xmath123 for @xmath480 , where @xmath479 is the curve defined below , constitutes all nontrivial solutions in a neighborhood of @xmath468 in @xmath469 . the purpose of this appendix is simply to record the derivatives of @xmath114 with respect to @xmath122 at @xmath123 , up to the third order . these are used to obtain derivatives of the bifurcation curves from and the bifurcation sheets from . we have @xmath665 for @xmath666 , where we suppress the evaluation at @xmath113 , and the derivatives @xmath6676\psi_{0ss}\eta^2 + 2s\psi_{0s}\eta\eta_{xx}+(4s\psi_{0s}+2s^2\psi_{0ss})\eta_x^2 \\-4\eta\hat{\phi}_{ss } - 4s\eta_x \hat{\phi}_{xs } -2s\eta_{xx}\hat{\phi}_s\end{multlined}\\ d_w^3 { \mathcal{f}}_2(0,\lambda)w^3 & = \begin{multlined}[t]-24\psi_{0ss}\eta^3 -6s\psi_{0s}\eta^2\eta_{xx } -(24s\psi_{0s}+12s^2 \psi_{0ss})\eta\eta_x^2\\ + 18\eta^2 \hat{\phi}_{ss } + 12s\eta\eta_x\hat{\phi}_{xs } + 6s\eta\eta_{xx}\hat{\phi}_s + 12s\eta_x^2 \hat{\phi}_s \\+6s^2\eta_x^2\hat{\phi}_{ss}\end{multlined } \end{aligned}\ ] ] for @xmath668 .

accurate atomic mass measurements of neutron - deficient and neutron - rich nuclides around the doubly - magic and of neutron - rich cesium isotopes were performed with the penning trap mass spectrometer isoltrap at isolde / cern . the masses of , , , , , , , , , , , and were determined . the obtained relative mass uncertainty in the range of to is not only required for safe identification of isomeric states but also allows mapping the detailed structure of the mass surface . a mass adjustment procedure was carried out and the results included into the atomic mass evaluation . the resulting separation energies are discussed and the mass spectrometric and laser spectroscopic data are examined for possible correlations . , , , , , , , , , and atomic mass , binding energy , penning trap , radionuclide , isomer , cesium , thallium , lead , bismuth , radium , francium + 07.75.+h mass spectrometers , 21.10.dr binding energies and masses , 27.70.+q , 27.80.+w , 27.90.+b

0801.2068

the accurate knowledge of atomic masses is required for many areas of physics @xcite . in particular , the mass gives access to the nuclear binding energy which in turn reveals distinct features of nuclear structure , notably shell effects and their relative strength as a function of isospin . this paper reports on mass data in the vicinity of the doubly - magic @xmath0 nucleus . strong nuclear structure effects are observed in particular at the _ neutron - deficient _ side from @xmath17 to @xmath18 . they have been revealed by laser spectroscopic measurements of the optical isotope shift @xcite . such measurements allow one to determine the changes of the root - mean - square charge radii @xmath19 relative to a reference nucleus . characteristic irregularities were observed in the isotopic chains of iridium ( @xmath17 ) , platinum ( @xmath20 ) , gold ( @xmath21 ) , mercury ( @xmath22 ) , and thallium ( @xmath23 ) at approximately mid - shell ( @xmath24 ) . the most prominent example is the odd - even shape staggering effect within the mercury isotopes @xcite . these phenomena were first explained for mercury as two different minima in the potential energy surface at a weakly oblate and a stronger prolate deformation ( shape coexistence ) @xcite . + in addition , several rotational bands were observed in the nuclear excitation spectra of these nuclei . an overview on nuclear shape coexistence of even - mass nuclides is given in @xcite . in @xmath25 , even triple shape coexistence has been observed @xcite . here , the three lowest energy states having spin / parity @xmath26 correspond to spherical , oblate and prolate shapes , all within an energy spanning only @xmath27 . these phenomena are described by particle - hole excitations of proton pairs across the closed shell @xcite . the interaction of additional proton - hole pairs and valence neutrons will lead to a minimum in energy around mid - shell . if the energies of the ground state and the intruder state are close , tiny changes in the nuclear binding energy are decisive for the nuclear ground state shape . + such minute nuclear binding energy effects in this mass range ( @xmath28 ) are only accessible by mass determinations with mass uncertainties @xmath29 of @xmath30 . this has been demonstrated by isoltrap measurements on neutron - deficient mercury isotopes @xcite . in that work , isomeric states , a 54-kev level in @xmath31 and a 128-kev level in @xmath32 were even identified . this was the first example of nuclear spectroscopy by mass spectrometry . the present work complements the information on high - precision masses in this area of the chart of nuclei with data on bismuth , lead , and thallium isotopes . the interconnection of the nuclides studied in this work by @xmath33-decay chains results in an improvement of further mass values up to @xmath34 . + in case no direct experimental data existed , previous mass values were obtained by a combination of up to four @xmath35 links with results from mass spectrometry . for example , the @xmath33-decay of neutron - deficient bismuth was studied at the lisol facility for odd @xcite and even isotopes @xcite . all direct data being available in this region are either isoltrap measurements @xcite or mass determinations by the schottky technique at the experimental storage ring esr @xcite . a detailed comparison to the latter data sets is made in sec . [ esrcomp ] . mass values of neutron - rich cesium isotopes were previously determined by triplet measurements with a mattauch - herzog spectrometer @xcite and @xcite . + a particular challenge while addressing the neutron - deficient nuclides experimentally is caused by the existence of two or even three isomeric states with both short half - lives and low excitation energies of sometimes less than @xmath36 . this requires an extremely high mass resolution of up to @xmath37 . + masses on the _ neutron - rich _ side of the doubly magic @xmath0 nucleus are of great importance for predictions on the stability of superheavy elements , which are stabilized by shell corrections @xcite . a variation of this correction by about @xmath38 leads to changes in the calculated half - lives for spontaneous fission or @xmath33 decay by several orders of magnitude . since no experimental mass values are available in this region , model calculations of the shell correction can be verified at the adjacent shell closure of @xmath39 . + furthermore , accurate mass values are required in order to provide reliable reference masses for the calibration of mass data obtained in measurements at the experimental storage ring esr @xcite . in an esr mass determination on neutron - deficient nuclides @xcite 117 reference masses were used in order to calibrate the 466 mass values studied in that experiment . in the neutron - rich region the number of reference masses is particularly low . the penning trap mass spectrometer isoltrap @xcite is installed at the on - line isotope separator isolde @xcite at cern . here , mass - separated beams of short - lived nuclides are provided for different experiments at an energy of @xmath40 . they are produced by bombarding a thick target with high - energy ( @xmath41 ) proton pulses ( @xmath42 protons / pulse ) which induce spallation , fission , or fragmentation reactions . reaction products diffuse from the heated target into an ion source region . here , they are ionized either by surface ionization , by electron impact in a hot plasma , or a resonant photo ionization in a resonance ionization laser ion source ( rilis ) @xcite . + in this experiment nuclides around the shell closure of @xmath43 were produced by proton - induced ( 1.4-gev ) spallation and fragmentation reactions on a @xmath44 target ( @xmath45 ) . for lead and bismuth a resonant laser ionization was applied , whereas cesium , thallium , francium , and radium were surface - ionized . finally , the ions were mass - selected with a resolving power of @xmath46 in the high - resolution separator ( hrs ) . + figure [ a_kap52_is_setup ] shows a schematic of the isoltrap mass spectrometer . the radiofrequency quadrupole ( rfq ) cooler and buncher @xcite decelerates and accumulates the continuous 60-kev isolde beam and delivers a cooled , low - energy ion bunch for efficient injection into the preparation penning trap @xcite . here , a mass - selective buffer - gas cooling scheme @xcite is employed in order to separate isobars . a resolving power , @xmath47 , of up to @xmath48 can be achieved in this process . in the precision penning trap , the mass @xmath49 of a stored ion species with charge @xmath50 is determined via a measurement of its cyclotron frequency @xmath51 , where @xmath52 denotes the strength of the magnetic field . a cyclotron resonance spectrum is obtained by the time - of - flight detection method @xcite : the extra radial kinetic energy resulting from the resonant excitation with an azimuthal quadrupolar rf - field at @xmath53 is detected by a reduction in the time of flight of the ejected ions towards a detector ( mcp5 in fig . [ a_kap52_is_setup ] ) . the resolving power is approximately given by the product of the cyclotron frequency and the observation time : @xmath54 @xcite . a more detailed description of a typical measurement cycle at isoltrap can be found in a recent publication @xcite . + an illustrative example of a cyclotron resonance curve is shown in fig . [ a_kap54_zyklres_133 ] . in such a resonance curve , the mean time of flight of the ions is displayed as a function of the applied radiofrequency , and a resonant excitation can be recognized by a shorter flight time . the diagram comprises 16 frequency scans containing a total amount of 1820 detected ions . the linewidth of the central resonance , as given by the fourier limit @xmath55 , results in a resolving power of @xmath56 ( mass resolution @xmath57 ) with an excitation time of the quadrupolar rf - field @xmath58 . this is sufficiently high to resolve ground and isomeric states of @xmath59 . the theoretical lineshape @xcite was used to fit the data points . in order to calibrate the strength of the magnetic field @xmath52 , the cyclotron frequency of a reference nuclide with well - known mass value is determined in regular time intervals . this allows for the accurate mass determination of short - lived nuclides far from stability with relative mass uncertainties @xmath60 @xcite . examples of recent measurements on @xmath61 @xcite as well as on several isotopes of potassium @xcite , nickel , copper , gallium @xcite , and krypton @xcite cover a wide mass range from @xmath62 . the primary result of an isoltrap measurement is the cyclotron frequency ratio @xmath63 of a reference ion with well - known mass relative to the ion species to be determined . thus , these experimental ratios are independent from the uncertainty of the reference mass @xmath64 . from the ratios @xmath65 , the atomic mass of the nuclide under study can be deduced by @xmath66 here , the atomic mass of the reference ion is indicated by @xmath67 and @xmath68 is the mass of the electron . a correction for the binding energy of the missing electron in the order of ev can be neglected . + extensive studies using carbon cluster ions @xcite helped to improve the assessment of systematic effects . for example , systematic errors like mass - dependent frequency shifts or the effect of a variation in the magnetic field strength were quantified . in addition , these investigations result in a better estimate of the final uncertainty of the mass determination . the residual systematic error of the isoltrap mass measurements , estimated previously to be @xmath69 , was determined in @xcite to be @xmath70 the data analysis was performed as described in @xcite . some aspects that are essential in the analysis of the data presented here are summarized in the following . simultaneous storage of ions of different mass values may result in a shift of the line center of the cyclotron frequency resonance . such possible contaminations depend on the ion production process . in the case of measurements at isoltrap , isobars or nuclides in isomeric states might be present in addition to the nuclide under investigation . systematic experimental studies using 10 to 70 ions stored in the trap as well as simulations with two ions interacting by coulomb forces are discussed in @xcite and @xcite : if the two ions have different masses , the resulting frequency shifts depend on the relative distance of the line centers divided by the linewidth that is chosen during the measurement process . if both lines can not be resolved , only _ one _ cyclotron resonance is observed at the frequency of the center of gravity . if the cyclotron resonances of both ion species can be resolved , both line centers are shifted to lower frequencies . the size of the shift has been found to be proportional to the number of stored ions . + to be able to exclude the possible presence of ions of different mass in these measurements , the line center of the cyclotron resonance is analyzed with respect to the number of ions registered by mcp 5 ( see fig . [ a_kap52_is_setup ] ) in a so - called count - rate - class analysis . the number of ions counted by the microchannel plate detector after each irradiation of the stored ion cloud with radiofrequency and ejection from the trap is subdivided into at least three count - rate classes with comparable average ion counts @xmath71 . a possible shift induced by a _ contamination _ is hence corrected by extrapolating the cyclotron frequency to a vanishing number of detected ions . two examples of a count - rate - class analysis , both with the same average number of ions per count - rate class @xmath71 , are shown in fig . [ a_kap54_z - class ] . the upper plot shows a measurement done with stable @xmath72 ions from the off - line alkali ion source ( see fig . [ a_kap52_is_setup ] ) , which delivers only surface - ionized alkali elements . the resulting fit of the slope of @xmath73 shows no count - rate dependent effect . the lower one shows a measurement of the radionuclide @xmath74 as delivered by isolde . with an increasing number of detected ions per cycle a significant slope of @xmath75 of the cyclotron frequency towards lower values is observed . this might be due to the presence of ions in the isomeric state . see also the discussion on the nuclide @xmath74 in sec . [ evena ] . combining several measurements , the final value of a frequency ratio is obtained from the weighted mean of individual results . an ion species @xmath76 experiences a frequency shift which can originate for example from electric field imperfections or a possible misalignment of the magnetic and electric field axis versus each other @xcite . since these shifts are to first order _ mass - independent _ , this results in a systematic shift in the determination of the frequency ratios , depending on the mass difference @xmath77 between the mass of interest _ m _ and the reference mass @xmath78 . the upper limit of this effect has been previously estimated to @xmath79 @xcite . frequency ratios between carbon clusters whose values are exactly known were quantified by a linear dependance @xcite , @xmath80 where @xmath81 is the difference between the measured and the true frequency ratio . in the experiments reported here , covering the mass region between @xmath82 and @xmath83 , the reference mass of @xmath72 ( @xmath84 ) differs by about 70 mass units from the mass to be determined . after applying the correction given by eq . ( [ 54 - 8 ] ) an uncertainty of the same size , @xmath85 is added quadratically to the uncertainty . due to the large mass difference @xmath77 , this yields an additional error on the order of 1 kev to the deduced mass excess value . in the end , the residual systematic uncertainty of @xmath86 ( eq . ( [ 54 - 4 ] ) ) is added quadratically to obtain the final uncertainty . an unambiguous mass assignment to a specific state , ground state or isomer , is not possible if a non - resolved mixture of nuclei in their ground and isomeric states is present . if the production ratio @xmath87 ( isomer to ground state ) , and the isomeric excitation energy @xmath88 are known , the ground state mass _ m _ can be extrapolated from the mass @xmath89 , corresponding to the observed cyclotron resonance , by @xcite @xmath90 if the production ratio of two ion species in ground and isomeric state is not known , the presence of the so - called contamination can be observed in the data analysis by studying the behavior of the resulting cyclotron frequency as a function of the count rate class ( see fig . [ a_kap54_z - class ] ) . only in the rare case of very similar half - lives of the ion under investigation and contaminating ion , contaminations might remain undiscovered . + table [ a_kap5_nubase ] lists all relevant data of the nuclides that have been studied : half - life @xmath91 , spin state @xmath92 and the isomeric excitation energy @xmath88 . in case several isomeric states are known to exist , those having a half - life longer than @xmath93 are listed are not considered as contaminants in the studies reported here , since the typical storage time of @xmath94 exceeds by far @xmath91 . ] . the assignment of a measured mass value to an isomeric state is performed under the assumption that isomeric states with a given @xmath92 and excitation energy might appear as listed in tab . [ a_kap5_nubase ] . the chosen rf - excitation time results in a width of the cyclotron resonance @xmath95 , which corresponds to a width @xmath96 in energy units . this value is also called mass resolution and is listed in the last column , since it serves as an important criterion for the observation of a contamination and hence for the identification of the states . lrrlccc + nuclide & & & spin / parity & excitation & @xmath97 + & & & @xmath92 & energy @xmath88 [ kev ] & [ kev ] + + nuclide & & & spin / parity & excitation & @xmath97 + & & & @xmath92 & energy @xmath88 [ kev ] & [ kev ] + @xmath98 & & 582 & ms & @xmath99 & & 220 + @xmath100 & & 225 & ms & ( @xmath99 ) & & 660 + @xmath101 & & 3.4 & s & @xmath102 # & & 340 + @xmath103 & & 2.7 & ms & @xmath104 # & 850(30 ) & 340 + @xmath105 & & 6.9 & s & @xmath102 # & & 340 + @xmath106 & & 60 & ms & @xmath104 # & 625(17 ) & 340 + @xmath74 & * & & 40 # & s & ( @xmath107 ) & & 210 + @xmath3 & * & & 27.5 & s & ( @xmath108 ) & 40(190 ) & 210 + @xmath109 & & 2.9 & s & ( @xmath110 ) & 420(190 ) & 210 + @xmath4 & & @xmath11151 & s & ( @xmath102 ) & & 540 + @xmath112 & & 15.6 & s & ( @xmath104 ) & 332(4 ) & 540 + @xmath113 & & 1.84 & h & @xmath107 & & 120 + @xmath5 & & 1.41 & h & ( @xmath108 ) & 394.2(5 ) & 120 + @xmath6 & & & @xmath102 & & 430 + @xmath114 & & 8 & min & @xmath115 & & 120 + @xmath7 & & 43 & min & @xmath116 & 319.3(7 ) & 120 + @xmath0 & & & @xmath117 & & 270 + @xmath118 & * & 6.3 & s & ( @xmath119 ) & & 560 + @xmath120 & * & 6.2 & s & ( @xmath110 ) & 150(190 ) & 560 + @xmath121 & & @xmath122 500 & ns & @xmath108 # & 420(190 ) & 560 + @xmath123 & & 12.3 & s & ( @xmath104 ) & & 110 + @xmath124 & & 128 & ms & ( @xmath102 ) & 242(7 ) & 110 + @xmath125 & & 34.6 & s & ( @xmath119 ) & & 110 + @xmath126 & & 39.6 & s & ( @xmath110 ) & 210(50 ) # & 110 + @xmath127 & & 67 & s & ( @xmath104 ) & & 110 + @xmath128 & & 3.2 & s & ( @xmath102 ) & 308(7 ) & 110 + @xmath129 & * & 95 & s & ( @xmath119 ) & & 580 + @xmath130 & * & 125 & s & ( @xmath131,@xmath108 ) & 100(70 ) # & 580 + @xmath132 & * & 115 & s & ( @xmath110 ) & 230(310 ) & 580 + @xmath59 & & 183 & s & @xmath104 # & & 120 + @xmath133 & & 87 & s & @xmath102 # & 399(6 ) & 120 + @xmath134 & & 5.1 & min & ( @xmath119 ) & & 590 + @xmath135 & & 0.6 & s & ( @xmath108 ) & 167(3 ) & 590 + @xmath136 & & 4 & min & ( @xmath110 ) & 270(4 ) & 590 + @xmath137 & & 9.3 & min & ( @xmath104 ) & & 600 + @xmath138 & & 5.04 & min & ( @xmath102 ) & 350(160 ) & 600 + @xmath139 & & 19 & ey & @xmath104 & & 270 + @xmath140 & & 7.6 & min & ( @xmath104 ) & & 470 + @xmath141 & & 36.4 & min & ( @xmath142 ) & 1347.5(2.5 ) & 470 + @xmath143 & & 2.17 & min & @xmath144 # & & 480 + @xmath145 & & 550 & ms & @xmath104 # & & 420 + @xmath146 & & 3.85 & s & ( @xmath104 ) & & 260 + @xmath147 & & 50.2 & s & @xmath102 # & & 810 + @xmath148 & & 2.46 & s & @xmath117 & & 470 + @xmath149 & & 4 & min & @xmath150 & & 810 + @xmath151 & & 93 & min & @xmath117 & & 330 + in the measurements reported here possible contaminations are mostly isomeric states , which could not be separated in the preparation penning trap . exceptions are the heavy isobars @xmath147 and @xmath149 . the spallation process does not transfer large angular momentum and hence low - spin states are predominantly produced . fragmentation reactions lead to a production of higher spin states . nuclides in states with half - lives shorter than few 10 ms are suppressed due to the slow release time of the reaction products from the isolde target and the time needed for the isobaric separation in the preparation penning trap ( see fig . [ a_kap52_is_setup ] ) . + a further selection of an atom in a specific nuclear state can be obtained by resonance laser ionization ( ris ) with the isolde laser ion source and a proper choice of laser wavelengths and bandwidths @xcite . due to different hyperfine splittings and isotope shifts in the ground and isomeric state , ris can enhance the selectivity . + in the case of bismuth investigated in the present work , the lasers were used in the broad - band mode . the production ratios in this experiment can not be deduced from previous measurements with the conventional ion source at isolde . for lead , the applied broad - band mode ( laser bandwidth = 10 ghz ) does not result in a selection of isomeric states , since the isomeric shift and hyperfine splitting are smaller or of the same size as the laser bandwidth . table [ a_kap5_ratios ] lists all frequency ratios that were determined relative to the cyclotron frequency of @xmath72 ions . if possible , the assignment to a particular isomeric state is indicated and will be discussed in sec . [ assign ] . the last column lists all mass excess values deduced from the determined frequency ratios @xmath65 . llcr + nuclide & frequency ratio & @xmath152 & mass excess + & @xmath153 & & [ kev ] + + nuclide & frequency ratio & @xmath152 & mass excess + & @xmath153 & & [ kev ] + @xmath98 & 1.090516444(91 ) & @xmath154 & -60052(11 ) + @xmath100 & 1.10562975(48 ) & @xmath155 & -52011(60 ) + @xmath101 & 1.361768475(82 ) & @xmath156 & -12802(10 ) + @xmath105 & 1.376786217(81 ) & @xmath157 & -16592(10 ) + @xmath3 & 1.399332234(69 ) & @xmath158 & -19874.4(8.6 ) + @xmath159 & 1.40683800(18 ) & @xmath160 & -22154(23 ) + @xmath5 & 1.474515617(98 ) & @xmath161 & -27103(12 ) + @xmath6 & 1.542259739(82 ) & @xmath162 & -23818(10 ) + @xmath7 & 1.482061386(45 ) & @xmath163 & -24429.5(5.5 ) + @xmath0 & 1.564849038(42 ) & @xmath164 & -21742.2(5.2 ) + @xmath165 & 1.42950438(20 ) & @xmath166 & -10535(25 ) + @xmath123 & 1.437006669(64 ) & @xmath167 & -13244.9(8.0 ) + @xmath126 & 1.444529605(72 ) & @xmath158 & -13398.5(9.0 ) + @xmath127 & 1.452033781(84 ) & @xmath168 & -15875(10 ) + @xmath130 & 1.45955793(12 ) & @xmath169 & -15878(50 ) is increased in the ame2003 due to a possible contamination by the @xmath170 and @xmath171 states . ] + @xmath59 & 1.467064775(45 ) & @xmath172 & -18023.7(5.6 ) + @xmath135 & 1.47459021(12 ) & @xmath173 & -17868(50 ) @xmath174 + @xmath175 & 1.482099666(89 ) & @xmath176 & -19706(25 ) in the ame2003 for possible contamination from @xmath138 . ] + @xmath139 & 1.572401384(35 ) & @xmath177 & -18254.9(4.3 ) + @xmath140 & 1.61770721(12 ) & @xmath178 & 1648(15 ) + @xmath143 & 1.625265520(93 ) & @xmath179 & 5873(11 ) + @xmath145 & 1.52741073(13 ) & @xmath180 & 861(16 ) + @xmath146 & 1.542441561(73 ) & @xmath181 & -1308.6(9.0 ) + @xmath147 & 1.72332168(30 ) & @xmath182 & 35816(37 ) + @xmath148 & 1.61017053(17 ) & @xmath183 & 100(20 ) + @xmath149 & 1.72329550(16 ) & @xmath184 & 32575(19 ) + @xmath151 & 1.73083537(10 ) & @xmath168 & 34518(12 ) + the atomic mass evaluation ame is an adjustment procedure giving mass data for all those nuclides where relevant experimental information , like for example mass values or half - lives , are known . such an evaluation was performed in december 2002 before these isoltrap data were included . this atomic mass evaluation is referred to in the following as intermediate atomic mass evaluation 2002 ( iame2002 ) . it is an up - to - date adjustment after the published issue of ame1995 @xcite and the nubase compilation @xcite . the assignment of the observed cyclotron resonances to nuclear states was not straightforward at all since this area of the chart of nuclei is rich in low - lying isomers . subsequently , the isoltrap results , reported in this publication , were included into the atomic mass evaluation 2003 . a detailed description of the adjustment procedure is given in the particular issues of the ame @xcite . here , only some general features that are relevant for this work are described . as a result of the ame , adjusted mass values are obtained from different types of experimental input data . even though a possible existence of more than 6000 different combinations of protons and neutrons is predicted , data from only approximately 2830 nuclides are available @xcite . if direct experimental data are not existing for nuclides near to the region of known masses , mass values are extrapolated from systematic , experimental trends . all experimental results can be represented as a relation between two ( and more ) nuclides . they span a net of interconnections between nuclides across the whole nuclear chart as it can be seen in the diagram for @xmath185 to @xmath186 ( fig . [ connect ] ) . for the present isoltrap data giving a frequency ratio to a nuclide with precisely known atomic mass value , these connections are linked to the mass of @xmath72 , which has been determined with a relative uncertainty of only @xmath187 @xcite . + all input data are transformed into a set of linear equations . for the case of experimental frequency ratios this is explained in @xcite . each datum @xmath188 with a standard deviation @xmath189 represents a link between two , three , or four unknown masses @xmath190 . the over - determined set of _ x _ equations for _ y _ mass values @xmath191 is represented as a set of linear equations with _ y _ parameters and is finally solved by the least - squares method as described in @xcite . most of the isoltrap frequency ratios that are published in the present work were introduced already into the most recently published atomic mass evaluation , _ ame2003 @xcite . the assignment of a measured mass value to an isomeric state is performed under the assumption that isomeric states with a given @xmath92 and excitation energy might appear as listed in tab . [ a_kap5_nubase ] . in this section those odd - mass nuclides having a long - lived isomeric state are discussed . according to tab . [ a_kap5_nubase ] the ground states of @xmath193 and @xmath114 have always a low spin value , whereas in the bismuth isotopes @xmath194 the ground state has a high spin . + the comparison of the results of tab . [ a_kap5_ratios ] with those of the intermediate adjustment ( iame2002 ) @xcite , helps to indicate which spin states have most probably been produced during the experiment . shown in fig . [ odd ] is such a comparison of the isoltrap masses to the iame2002 masses with and without the added isomeric energy . the zero line depicts the experimental isoltrap data with their uncertainties . the differences to the literature values are indicated by open symbols , whereas the position of excited nuclear states is taken from the current nubase data ( tab . [ a_kap5_nubase ] ) @xcite . the uncertainty of the excited state is derived by adding the uncertainties of the ground state mass value and the excitation energy quadratically . + * @xmath101 * and * @xmath105 * : here , the half - lives of the excited isomeric states ( @xmath195 ) of @xmath196 for @xmath101 and @xmath197 for @xmath105 , are much shorter than the duration of a measurement cycle ( approximately @xmath198 ) . therefore , the measured cyclotron frequency ratios are clearly assigned to the respective ground states . the measurements of isoltrap agree with the values from the intermediate adjustment iame2002 ( compare fig . [ odd ] ) , but the uncertainties got significantly improved . + * @xmath4 * : the measurement of the nuclide @xmath4 has been performed with a mass resolution of @xmath199 , which is insufficient to resolve both ground and isomeric states , differing by @xmath200 . hence , an eventually present admixture of either of the states can not be excluded . the determined isoltrap value suggests a predominant production of the isomer . + * @xmath7 * : the mass values of both ground and isomeric states in @xmath114 are listed in iame2002 with uncertainties of only @xmath201 . the value resulting from the isoltrap measurement with a final uncertainty of @xmath202 agrees within the error bars to the one of the excited @xmath203 state ( @xmath204 ) and hence is assigned to it . + * @xmath194 , @xmath140 * : a series of bismuth isotopes with odd mass numbers @xmath205 and @xmath206 has been studied . with the exception of the case of @xmath137 the mass resolution @xmath207 ( see tab . [ a_kap5_nubase ] ) is at least a factor of two smaller than the mass difference between both states . therefore , a contamination of the respective second state is definitely excluded in the count - rate - class analysis . the remaining question which of the two states has been measured can not be solved easily . yield measurements for lead and bismuth isotopes at isolde reveal that states having higher spin values are predominantly produced @xcite . these are for @xmath208 the @xmath209 ground states of bismuth . indeed , the comparison of the isoltrap results shows a good agreement with the values of the intermediate evaluation iame2002 for the ground states of @xmath194 . the measurements of @xmath137 were performed with a resolution of @xmath210 which is not sufficient to resolve ground and isomeric state , differing by @xmath211 . measurements of the neighboring odd-@xmath192 isotopes @xmath127 and @xmath59 with approximately @xmath212 resolution did not show any hint of a contamination by an excited state . a systematic comparison of individual measurements with low ( @xmath213 ) and high ( @xmath212 ) resolution was used in order to quantify a correction by a possible contamination by the excited @xmath138 state . this correction of @xmath214 increases the uncertainty of the isoltrap value in the atomic mass table ame2003 from @xmath215 to @xmath216 . + considering the isomeric states of all odd - mass nuclides studied here a consistent picture is observed . except for cases where the presence of a state can be excluded right away due to its short half - life , higher spin states are predominantly produced . in @xmath4 and @xmath114 these are the _ excited _ states with spins of @xmath217 and @xmath218 . in bismuth ( @xmath219 ) the ground states have a higher spin value of @xmath217 . for the nuclide @xmath140 with an excited state of @xmath220 energy and a spin @xmath221 , the @xmath217 ground state mass value has been determined . in order to facilitate a spin - to - state assignment , the present isoltrap data are plotted in fig . [ even ] and compared relative to ground state mass values and excitation energies from the intermediate evaluation iame2002 . these nuclides have sometimes _ two _ excited , long - lived states ( see also tab . [ a_kap5_nubase ] ) . excitation energies of states with half - lives longer than @xmath222 are shown . + apart from the @xmath222-state of @xmath118 all nuclides have half - lives that are sufficiently long ( @xmath223 ) in order to be studied with the penning trap mass spectrometer isoltrap . due to the lack of any quantitative information on production ratios , a conclusion can be only based on a systematic comparison within the current data . in analogy to the nuclides of odd mass number , it is expected that lower spin states will be produced with lower yields . + * @xmath3 * : an unambiguous assignment of the isoltrap mass value to one of the long - lived states in @xmath74 is not straightforward . this is also reflected by the different spin assignments given in nubase ( tab . [ a_kap5_nubase ] ) and in the evaluated nuclear structure data files ( ensdf ) . mass values of both excited states are given with uncertainties of @xmath224 . the energy difference between @xmath3 and @xmath109 is @xmath225 @xcite . the attempt to employ a very long rf - excitation of @xmath226 , corresponding to a mass resolution of @xmath227 , failed . for excitation times beyond @xmath228 , a decay of the stored ions has been observed ( see half - lives in tab . [ a_kap5_nubase ] ) . + figure [ tl186 ] shows the results of the four cyclotron frequency measurements . with a mass resolution of @xmath229 ( measurement # 2 - 4 ) a mixture between @xmath3 and @xmath109 is resolved and the presence of either of the states should have been detected . measurement # 2 and # 4 yield count - rate - class corrected frequencies . the slope @xmath230 in [ mhz / ion ] of the linear extrapolation of the cyclotron frequency of @xmath231 is indicated for each measurement . the third data point ( # 3 ) , which has been obtained after an additional waiting time of @xmath232 in the upper precision trap , does not show a significant count rate effect , since this measurement allowed for a decay of the @xmath233-state . to avoid any undetected contamination from the low - resolution measurement # 1 , it was not considered in the final weighted average of the frequency ratio . after a comparison to the spin systematics , as given in tab . [ a_kap5_nubase ] , and the agreement of the isoltrap value with the intermediate evaluation iame2002 this result is assigned to the @xmath3 state . it is not excluded that a small contamination by the ground state ( @xmath234 ) might be present . + * @xmath5 * : the mass resolution , being @xmath235 during this measurement , was sufficiently high to separate both states of @xmath113 . the resulting isoltrap value agrees with the mass value of the excited , medium spin state @xmath5 ( @xmath236 , @xmath237 ) though the uncertainties of the iame2002 values are rather large . + * @xmath238 * : these nuclides have three long - living states similar to almost all even-@xmath192 bismuth isotopes up to @xmath239 . the excitation energies of about @xmath240 are all known with uncertainties of around @xmath241 , except for @xmath134 ( fig . [ even ] ) . the resolving power employed during those cyclotron frequency measurements of approximately @xmath242 corresponds to a mass resolution between @xmath243 and @xmath244 . in order to obtain a mass resolution of @xmath224 an excitation time of @xmath245 is required . + the mass value of isoltrap for @xmath134 agrees well to the one of the @xmath246-state ( @xmath135 ) . however , a contamination can not be excluded . accounting for a possible admixture of either the @xmath170 or the @xmath171 state the original uncertainty of @xmath247 had been increased to @xmath224 in the ame2003 . + due to the good agreement for @xmath134 , it is concluded that the medium spin state has been predominantly produced for @xmath129 as well . since the possible presence of both isomeric states , @xmath170 and @xmath171 , whereas each of them is about @xmath36 different in energy , can not be excluded with a resolution of @xmath213 , the original error of @xmath247 is increased in the atomic mass evaluation 2003 to @xmath224 . + in @xmath118 the medium spin state exhibits a half - life in the nanoseconds regime . + the measured isoltrap value with @xmath216 uncertainty agrees within @xmath248 to the value of the @xmath249-state ( @xmath118 ) . due to the minor resolution during the measurement , a mixture of both states @xmath118 and @xmath120 differing by @xmath250 might be present . + _ * @xmath125 * _ : in contrast to other even-@xmath192 bismuth isotopes from @xmath118 to @xmath251 , no medium nuclear spin of , for example , @xmath252 or @xmath253 has been observed . however , the general trend in this experiment points to a favored production of medium spin states , @xmath254 or 7 . the question on the production ratio between the @xmath170 ground state and the excited @xmath171 state remains unsolved . from the general trend throughout all these measurements it is observed , that it is the higher spin state that is produced with a higher weight and measured in the trap . consequently the mass value has been tentatively assigned to @xmath126 in the atomic mass evaluation 2003 . in the mass measurement of this nuclide two observations are made : first , the counting class analysis of two individual measurements with a resolution of @xmath255 indicate the presence of a contamination and give a shifted , extrapolated value . secondly , the characteristic pattern for an in - trap decay with short half - life is observed in the time - of - flight spectrum although the half - lives of both known states are more than ten times longer than the excitation time of @xmath256 . this might indicate a third , so far unknown , isomeric state with shorter half - life or any other contamination . the nuclides @xmath147 and @xmath149 were stored simultaneously in the precision trap since the applied mass resolving power @xmath257 in the preparation penning trap was not sufficient to separate the @xmath186-isobars . figure [ fr229_ra229 ] shows the cyclotron resonance curves of the isobars , which differ in mass by @xmath258 . the center frequencies including the first sidebands are clearly separated with a resolving power of @xmath259 , using an excitation time of @xmath260 . a superposition of two individual cyclotron resonance curves , weighed by their relative abundance , was fitted to the data points . assuming a detection efficiency of the microchannel plate of 30 % , the average number of detected ions of 2.5 corresponds to about eight ions that were simultaneously stored . therefore , the possibility of a frequency shift dependent on the number of ions ( see sec . [ counti ] ) was investigated in a detailed count - rate - class analysis . + the result of such an analysis is plotted in fig . [ z - class_229 ] for the cyclotron frequencies of @xmath147 ( left ) and @xmath149 ( right ) . here , a total number of 2340 registered ions was subdivided into three ( upper diagrams ) as well as into four ( lower diagrams ) count - rate classes . _ different _ subdivisions were made in order to exclude any possible count - rate dependance produced within the statistics . in both cases no significant count - rate dependance is observed . figure [ rest ] shows the mass excess values of other nuclides studied in this work , including the stable isotopes @xmath6 , @xmath0 , and @xmath139 . the masses of these stable isotopes , which are known with high accuracy , agree within one standard deviation with iame2002 . the largest discrepancies are observed for @xmath98 ( @xmath261 ) , @xmath100 ( @xmath262 ) , and @xmath149 ( @xmath263 ) . some of the mass values which are reported in this work have also been determined by schottky mass spectrometry in the experimental storage ring esr at gsi @xcite . figure [ trap - esr ] compares the results of both experiments and the values of the iame2002 . a larger set of data is obtained by including also the indirectly determined results at esr using @xmath264 values from literature @xcite . in most of the cases the data from both measurement techniques agree within their uncertainties . an exception is the nuclide @xmath101 , where both values differ by more than @xmath265 . since the esr value has been indirectly determined , this discrepancy could be attributed as well to the @xmath266 value . table [ a_kap5_ma8_input ] compiles the mass excesses of those radionuclides discussed here and compares the data with values given in nubase 1997 @xcite and @xcite . in addition , the results of the atomic mass evaluation published in 2003 ( ame2003 ) are given , which include the isoltrap results . due to the manifold interconnections , the present isoltrap data improve not only the directly measured nuclides , but also a large region from @xmath267 to @xmath268 interconnected via @xmath33 chains in particular . [ linked nuclides ] shows the respective section of the nuclear chart from tantalum to neptunium isotopes , indicating all related nuclides in this region . + for primary nuclides , which directly contribute to the adjustment procedure , the total influence of each input value in the ame is given in tab . [ a_kap5_ma8_input ] by the _ significance _ in per cent . it is the sum of all influences on individual mass values @xmath190 . the quantity v / s denotes the difference between both values , divided by the uncertainty of the isoltrap value . the last column marks those data which were not used in the adjustment procedure . this is the case , if , for example , a contamination of the isoltrap measurement by a further isomeric state could not be excluded , or if , for the stable nuclides , another high - accuracy mass determination already exists , which is at least three times more precise than the isoltrap value . + in addition , the mass values of around 100 other nuclides changed after the atomic mass evaluation with the isoltrap data reported here . the largest modifications in the adjusted mass values is observed for @xmath269 ( @xmath270 ) . + lrrrcrc + nuclide & & sign . & v / s & remark + & isoltrap & nubase1997 and @xcite & ame2003 & % & & + + nuclide & & sign . & v / s & remark + & isoltrap & nubase1997 and @xcite & ame2003 & % & & + @xmath98 & -60052(11 ) & -60180(50 ) & -60057(11 ) & 94 & -0.4 & + @xmath100 & -52011(60 ) & -52290(150 ) & -52020(50 ) & 79 & -0.1 & + @xmath101 & -12802(10 ) & -12490#(300 # ) & -12801(9 ) & 92 & -0.2 & + @xmath105 & -16592(10 ) & -16290#(190 # ) & -16587(10 ) & 91 & 0.4 & + @xmath3 & -19874.4(8.6 ) & -19980(50 ) & -19874(9 ) & 100 & & + @xmath112 & -22154(23 ) & -22085(22 ) & -22109(8 ) & & 2 & f + @xmath271 & -27103(12 ) & -27470#(100 # ) & -27497(12 ) & 100 & & e + @xmath6 & -23818(10 ) & -23820.8(1.4 ) & -23820.6(1.3 ) & & -0.3 & u + @xmath7 & -24429.5(5.5 ) & -24423(15 ) & -24429(6 ) & 100 & & + @xmath0 & -21742.2(5.2 ) & -21749.6(1.3 ) & -21748.5(1.2 ) & & -1.2 & u + @xmath120 & -10535(25 ) & -10590(50 ) & -10483(10 ) & & 2.1 & f + @xmath123 & -13244.9(8.0 ) & -13211(23 ) & -13240(7 ) & 86 & 0.6 & + @xmath126 & -13398.5(9.0 ) & -13220#(120 # ) & -13399(9 ) & 100 & & + @xmath127 & -15875(10 ) & -15870(80 ) & -15873(10 ) & 100 & 0.2 & + @xmath130 & -15878(50 ) & -15970#(440 # ) & -15880(50 ) & 100 & & + @xmath59 & -18023.7(5.6 ) & -18039(21 ) & -18024(6 ) & 100 & & + @xmath135 & -17868(50 ) & -17856(30 ) & -17842(25 ) & 100 & 0.5 & + @xmath137 & -19706(25 ) & -19689(15 ) & -19688(8 ) & 100 & 0.8 & + @xmath139 & -18254.9(4.3 ) & -18259.1(1.5 ) & -18258.5(1.4 ) & & -0.8 & u + @xmath140 & 1648(15 ) & 1710(100 ) & 1649(15 ) & 100 & & + @xmath143 & 5873(11 ) & 5780#(100 # ) & 5874(11 ) & 100 & & + @xmath145 & 861(16 ) & 870(70 ) & 861(16 ) & 100 & & + @xmath146 & -1308.6(9.0 ) & -1310(50 ) & -1310(8 ) & 100 & -0.1 & + @xmath147 & 35816(37 ) & 35790#(360 # ) & 35820(40 ) & 100 & & + @xmath148 & 100(20 ) & 100(10 ) & 101(9 ) & 100 & 0 & + @xmath149 & 32575(19 ) & 32440(60 ) & 32563(19 ) & 91 & -0.6 & + @xmath151 & 34518(12 ) & 34550(30 ) & 34518(12 ) & 100 & & + the mass values and the corresponding nuclear binding energies represent the underlying forces acting in a nucleus . in analogy to the ionization potential of atoms , plots of separation energies for the removal of two protons , @xmath272 or two neutrons , @xmath273 reveal the details of nuclear shell structure . in this representation , odd - even effects caused by the usually much larger binding energy for even proton or neutron numbers are removed . + figures [ bbb ] and [ ddd ] show the behavior of separation energies for protons and neutrons versus the respective nucleon number . the results from ame1995 and ame2003 are compared . a global trend , a continuous decrease in the separation energies , is observed . discontinuities from this general trend appear at shell closures or in regions of strong deformations . the @xmath274 shell closure is clearly visible in fig . the much less pronounced strength of the proton shell closure @xmath275 is larger at the magic neutron number @xmath276 than at the half - filled shell @xmath24 ( fig . [ ddd ] ) . the main differences between the appearance of the mass surface in 1995 and 2003 are substantially reduced uncertainties , due to improved as well as numerous new experimental input data . these are isoltrap measurements in the isotopic chains of mercury , platinum , lead , polonium , radon , and radium @xcite as well as a large number of new results obtained at the esr by the schottky technique @xcite . in particular the small uncertainty ( @xmath277 ) of data from isoltrap along the extended isotopic chain from @xmath278 to @xmath279 enables the observation of finer nuclear structure effects and to study , for example , nuclear shape coexistence and the influence of low - excitation intruder states @xcite . note the pronounced irregularities in the thallium isotopic chain at neutron numbers @xmath280 and 114 . + the two - proton separation energies are much more regular and smooth in this region of the nuclear chart than those for two neutrons . however , a peculiar irregularity in the two - proton separation energies appears for the isotopes of thallium ( @xmath281 ) with even neutron numbers @xmath282 from 98 to 108 : some separation energies are nearly equal or even higher than those of the isotopes with one additional neutron . such an effect is very uncommon in the chart of nuclei . in order to investigate this anomalous behavior further , the two - proton separation energies are plotted in fig . [ stagg ] against the neutron number and the isotopes of an element are connected by lines . this presentation shows very clearly in the region of the lead isotopes at @xmath43 the steadily increasing shell gap from mid - shell @xmath24 towards the magic neutron number @xmath274 . also the above mentioned irregularities in the thallium and ( less pronounced ) in the mercury isotopes are revealed distinctly as an odd - even effect . the reason for this effect is unclear . three possible explanations are discussed in the following : + 1 . the masses of @xmath101 and @xmath105 , determined with isoltrap cyclotron resonances and included in ame2003 , were wrongly assigned to the ground states of these isotopes . however , an assignment to the isomeric states ( excitation energy @xmath283 or @xmath284 ) would increase the staggering even further . also the mass of @xmath4 is determined to more than 60 % by isoltrap , but in this case indirectly via the @xmath264 value of the @xmath33 decay and the directly measured mass of @xmath123 . also here , a possible problem in the assignment can be excluded due to the sufficiently high resolving power for separating ground and isomeric states and because of the short half - life of the isomer . some mass measurements might be wrong . this explanation is contradictory to the fact that the mass values in question stem from different sources and mass spectrometric techniques ( isoltrap , esr , @xmath264 values , etc . ) . furthermore , the odd - even staggering is not an isolated phenomenon but appears also ( less pronounced ) in the neighboring isotopic chain of mercury . this is not easily explicable since the mercury and the thallium isotopes are not interconnected by @xmath266 values . the odd - even staggering of the two - proton separation energies in the thallium isotopic chain is factual and represents a nuclear structure effect . this would imply a stronger pairing effect for neutron numbers around mid - shell ( @xmath24 ) than for nearly double - magic thallium nuclei with @xmath23 and @xmath285 . similar to the drastic odd - even staggering observed in the charge radii of the mercury isotopes ( caused by shape transitions ) @xcite these phenomena depend very critically on the neutron and proton configurations and very elaborated calculations are required to reproduce these effects . nuclear masses , as determined in this work , and nuclear charge radii , as obtained by optical spectroscopy from isotope shift measurements , illustrate global and collective properties of nuclear matter . the mass is a measure of the binding energy and the charge radii reflect static and dynamic deformations . the nuclear system , as a function of protons and neutrons , maximizes the binding energy according to the nuclear shape . the pairing energy , which is about @xmath38 , plays an important role in this context and becomes visible as odd - even effect in the masses as well as in the charge radii . + due to the complex interplay of single - particle and collective degrees of freedom , one observes in the region of the nuclear chart investigated in this work both a spherical and deformed shape in the same nucleus at low excitation energies . this phenomenon , first observed at isolde / cern in the seventies by optical spectroscopy @xcite is known as `` shape coexistence or shape isomerism '' and nowadays an extensive bulk of data has been collected throughout the chart of nuclei @xcite . shape coexistence is visible in the vicinities of the shell closures @xcite . most clearly it is observed in the slope of charge radii in the mercury isotopic chain : it manifests itself in a huge odd - even shape staggering near the @xmath286 mid - shell region @xcite and a large difference in charge radii and therefore nuclear deformation between the ground and isomeric state in @xmath287 @xcite ( fig . [ figrad3 ] ) . since the nuclear states of very different configuration are ( almost ) degenerate in energy , one can not expect to observe any major trend in the slope of the masses or binding energies as a function of neutron number . in fact , only now the high mass resolving power @xmath288 of up to @xmath37 of the penning trap mass spectrometer isoltrap enables one to resolve ground and isomeric states in this region of the nuclear chart where shape coexistence occurs . the implementation of many new accurate mass data from isoltrap ( @xcite and this work ) to the data base of the atomic mass evaluation and a large data set of charge radii @xcite stimulated an attempt to reveil a correlation between masses and charge radii @xcite . + in order to uncover finer details in the binding energies regarding odd - even staggering and general trends @xcite , the absolute values of the deduced pairing gap parameters for neutrons are plotted instead of the experimental masses . + the three - point indicator for the shell gap parameter @xmath289 is given by @xmath290,\ ] ] which is often interpreted as a measure of the empirical pairing gap @xcite . another commonly used relation is the four - point indicator , which averages the @xmath291 values and is given by @xmath292 . \vspace*{-0.045cm}\ ] ] figures [ figrad1 ] - [ figrad5 ] compare the results from mass measurements with those from isotope shift determinations for the isotopic chains of platinum , gold , mercury , thallium , and lead . for these elements , extensive data are available . in order to allow an easy comparison of pairing gap parameters with charge radii for the different elements , the scales of the x- and y - axis are made identical in figs . [ figrad1 ] - [ figrad5 ] . the pairing gap parameters of the isotopic chains of the even-@xmath293 elements lead , mercury and platinum are nearly identical in the regions of @xmath294 and @xmath295 , respectively . the pairing gap parameters for the isotopes of the odd-@xmath293 elements thallium and gold show a similar behavior but the pairing gap is smaller in magnitude . around mid - shell neutron number @xmath24 a minimum develops which is barely visible in the case of lead . it becomes sharper and is shifted to @xmath296 for thallium , distinct and narrow for mercury but very much pronounced and broader for gold and platinum . + the slopes of the isotopic chains of lead , thallium and mercury are almost identical in the region where data are available for the same neutron number . only the odd - even staggering is much more pronounced for the thallium isotopes as compared to those of lead and mercury . + down to neutron number @xmath297 , the slope of the charge radii for the isotopes of gold and platinum is a little bit smaller than for the other isotopic chains which indicates increasing deformation . for @xmath296 in the case of mercury and @xmath298 for gold a sudden change of the charge radii is observed which is due to a sharp change in deformation . + comparing the slopes of pairing parameters with those of the charge radii , one observes that the extremely sharp discontinuity at the shell closure at @xmath274 is visible in the charge radii only as a small kink in their slope , that the pronounced increase of deformation at mid - shell @xmath24 is reflected as a local minimum in the pairing parameters , and that the size of the odd - even staggering in the pairing parameter @xmath291 is not reflected in the size of the one of the charge radii . there is an interesting observation : for thallium an unusual odd - even staggering between the isomers below mass number @xmath299 is observed . + the new data obtained in this region of the nuclear chart near @xmath300 will hopefully stimulate theoretical work to develop nuclear models further which are able to describe simultaneously experimental mean square charge radii as well as experimental masses of spherical and deformed nuclei in a satisfactory way . up to now , only little attention has been paid to such investigations which concern model - independent gross properties of nuclear matter . for example , the charge radii are only poorly described by macroscopic - microscopic mass models which are , however , quite successful in describing nuclear masses @xcite . even more demanding is a description of the odd - even staggering of masses and of the charge radii as attempted in ref . @xcite for spherical nuclei . the fine structure of the mass surface is important for the elucidation of the dramatic shape changes that appear so suddenly in this region of the nuclear chart . for most of the nuclides that were investigated in this work a clear mass - to - state assignment was possible . few exceptions due to an insufficient resolving power during the measurement are the nuclides @xmath4 , @xmath118 , @xmath301 , and @xmath137 . spectroscopic data giving production ratios could assist these identifications using eq . ( [ 54 - 10 ] ) . the excitation energies of bismuth isotopes ( odd - odd nuclei ) suffer from particulary high uncertainties . hence , these cases should be addressed in a future mass measurement requiring the following conditions : ( i ) if the half - lives of the states are sufficiently long in order to allow an extremely high - resolution measurement : a mass resolution of , for example , @xmath302 can be obtained with an excitation time of @xmath232 . in these cases the count - rate - class analysis can give a meaningful interpretation , _ i.e. _ a possible contamination is detected . ( ii ) similar to the measurements of both states of @xmath303 @xcite , a narrow - band laser ionization with a bandwidth of @xmath304 can be used to selectively ionize one of the states . as a prerequisite , the element of interest has to be accessible for laser ionization and the hyperfine structure splitting of the isomeric components with different nuclear spin has to be known . ( iii ) in cases where the half - life is not sufficiently long to perform a measurement with extended excitation time , the mass determination should be supported by nuclear spectroscopy in order to enable a clear identification of states . + * acknowledgements * + we gratefully acknowledge the isolde technical group and the rilis group for assistance during the experiments . this work was supported by the german ministry for education and research ( bmbf contracts 06mz962i , 06mz21s , and 06lm968 ) , by the european commission ( contracts hpri - ct-1998 - 00018 ( lsf ) and hpri - ct-2001 - 50034 ( nipnet ) ) , and by the association of helm- holtz research centers ( contract vh - ng-037 ) . 77 d. lunney , j.m . pearson , and c. thibault , rev . phys . * 75 * ( 2003 ) 1021 . k. blaum , phys . rep . * 425 * ( 2006 ) 1 . otten , treatise on heavy ion physics * 8 * ( 1989 ) 517 , and references therein . kluge and w. nrtershuser , spectrochim . acta b * 58 * ( 2003 ) 1031 , and references therein . j. bonn _ et al . _ , phys . b * 38 * ( 1972 ) 308 . p. dabkiewicz _ et al . _ , phys . lett . b * 82 * ( 1979 ) 199 . s. frauendorf and v.v . pashkevich , phys . * 55b * ( 1975 ) 365 . wood _ et al . . rep . * 215 * ( 1992 ) 101 . andreyev _ et al . _ , nature * 405 * ( 2000 ) 430 . k. heyde _ et al . . rep . * 102 * ( 1983 ) 291 . k. heyde , nucl . a. * 507 * ( 1990 ) 149c . s. schwarz , ph.d . thesis , university of mainz ( 1998 ) , isbn 3 - 8288 - 0735 - 6 . s. schwarz _ et al . _ , nucl . phys . a * 693 * ( 2001 ) 533 . r. fossion _ et al . _ , nucl . phys . a * 697 * ( 2002 ) 703 . e. coenen _ et al . _ , phys . lett . * 54 * ( 1985 ) 1783 . van duppen _ et al . _ , nucl . phys . a * 529 * ( 1991 ) 268 . g. bollen _ et al . _ , j. mod . opt . * 39 * ( 1992 ) 257 . f. herfurth _ et al . _ , eur . j. a * 25 * ( 2005 ) 17 . t. radon _ et al . _ , nucl . phys . a * 677 * ( 2000 ) 75 . litvinov , ph.d . thesis , university of giessen ( 2003 ) . litvinov _ et al . _ , nucl . phys . a * 756 * ( 2005 ) 3 . m. epherre _ et al . _ , c * 19 * ( 1979 ) 1504 . et al . _ , phys . a * 378 * ( 1982 ) 443 . z. patyk _ et al . _ , nucl . phys . a * 491 * ( 1989 ) 267 . g. bollen _ et al . _ , nucl . instr . meth . a * 368 * ( 1996 ) 675 . f. herfurth _ et al . _ , nucl . instr . meth . a * 469 * ( 2001 ) 254 . m. mukherjee _ et al . _ , submitted to eur . j. a. e. kugler , hyp . int . * 129 * ( 2000 ) 23 . mishin _ et al . . instr . meth . b * 73 * ( 1993 ) 550 . u. kster _ et al . _ , spectrochim . acta * 58 * ( 2003 ) 1047 . h. raimbault - hartmann _ et al . _ , nucl . b * 126 * ( 1997 ) 378 . g. savard _ et al . _ , phys . lett . a * 158 * ( 1991 ) 247 . g. grff , h. kalinowski , and j. traut , z. phys . a * 297 * ( 1980 ) 35 . m. knig _ et al . _ , int . j. mass . spectrom . ion process . * 142 * ( 1995 ) 95 . g. bollen , nucl . phys . a * 693 * ( 2001 ) 3 . g. sikler _ et al . _ , nucl . phys . a * 763 * ( 2005 ) 45 . a. kellerbauer _ et al . _ , j. d * 22 * ( 2003 ) 53 . k. blaum _ et al . _ , j. phys . b * 36 * ( 2003 ) 921 . s. george _ et al . _ , phys . lett . * 98 * ( 2007 ) 162501 . c. yazidjian _ et al . _ , c * 76 * ( 2007 ) 024308 . c. gunaut _ et al . _ , phys . rev . c * 75 * ( 2007 ) 044303 . p. delahaye _ et al . _ , phys . c * 74 * ( 2006 ) 034331 . k. blaum _ et al . _ , eur . j. a * 15 * ( 2002 ) 245 . m. knig , diploma thesis , university of mainz ( 1991 ) . g. bollen _ et al . _ , phys . c * 46 * ( 1992 ) r2140 . d. beck _ et al . _ , nucl . meth . b * 126 * ( 1997 ) 374 . g. audi _ et al . _ , nucl . a * 624 * ( 1997 ) 1 . g. audi , compilation of an intermediate atomic mass evaluation in 2002 . this evaluation did not contain the isoltrap data presented in this work but all other mass data published up to this date . iame is used for comparison with our mass values , since those are already included in the ame 2003 @xcite . bradley _ et al . _ , phys . lett . * 83 * ( 1999 ) 4510 . g. audi and a.h . wapstra , nucl . phys . a * 595 * ( 1995 ) 409 . g. audi and a.h . wapstra , nucl . a * 565 * ( 1993 ) 1 . g. audi , a.h . wapstra , and c. thibault , nucl . phys . a * 729 * ( 2003 ) 337 . g. audi , hyp . 132 * ( 2001 ) 7 . wapstra , g. audi and c. thibault , nucl . phys . a * 729 * ( 2003 ) 129 . d. beck _ et al . _ , eur . phys . j. a * 8 * ( 2000 ) 307 . a. kohl , ph.d . thesis , university of heidelberg ( 1999 ) . r. julin _ et al . _ , j. phys . g * 27 * ( 2001 ) r109 . g. ulm _ et al . _ , z. phys . a * 325 * ( 1986 ) 247 . j. bonn _ et al . _ , phys . lett . b * 36 * ( 1971 ) 41 . h. de witte _ et al . _ , phys . * 98 * ( 2007 ) 112502 . c. weber _ et al . j. a * 25 * supplem . 1 ( 2005 ) 201 . w. satula _ et al . _ , * 81 * ( 1998 ) 3599 . j. dobaczewski _ et al . _ , phys . rev . * 63 * ( 2001 ) 024308 . f. buchinger _ et al . _ , c * 72 * ( 2005 ) 057305 . fayans and d. zawischa , int . j. modern phys . b * 15 * ( 2001 ) 1684 . c. weber _ et al . _ , lett . a * 347 * ( 2005 ) 81 .

the observed surface dynamics of jupiter and saturn are dominated by a banded system of fierce zonal winds . the depth of these winds remains unclear but they are thought to be confined to the very outer envelopes where hydrogen remains molecular and the electrical conductivity is small . the dynamo maintaining the dipole - dominated magnetic fields of both gas giants , on the other hand , likely operates in the deeper interior where hydrogen assumes a metallic state . here , we present numerical simulations that attempt to model both the zonal winds and the interior dynamo action in an integrated approach . using the anelastic version of the mhd code magic , we explore the effects of density stratification and radial electrical conductivity variations . the electrical conductivity is mostly assumed to remain constant in the thicker inner metallic region and it decays exponentially towards the outer boundary throughout the molecular envelope . our results show that the combination of a stronger density stratification and a weaker conducting outer layer is essential for reconciling dipole dominated dynamo action and a fierce equatorial zonal jet . previous simulations with homogeneous electrical conductivity show that both are mutually exclusive , with solutions either having strong zonal winds and multipolar magnetic fields or weak zonal winds and dipole - dominated magnetic fields . all jets tend to be geostrophic and therefore reach right through the convective shell in our simulations . the particular setup explored here allows a strong equatorial jet to remain confined to the weaker conducting outer region where it does not interfere with the deeper seated dynamo action . the flanking mid to high latitude jets , on the other hand , have to remain faint to yield a strongly dipolar magnetic field . the fiercer jets on jupiter and saturn only seem compatible with the observed dipolar fields when they remain confined to a weaker conducting outer layer . atmospheres , dynamics , jupiter , interior , variable electrical conductivity , numerical dynamos

1210.3245

the gas giants , jupiter and saturn , mainly consist of a hydrogen - helium mixture . due to the large pressures and temperatures reached inside these planets , hydrogen acquires metallic properties @xcite . the transition happens at 8590% of jupiter s and @xmath0% of saturn s radii . a classical view is that the lower metallic layer likely hosts the dynamo of these planets , while the upper molecular envelope accommodates the observed fierce zonal jets . higher densities , lorentz forces and ohmic diffusion would lead to a more sluggish dynamics in the metallic layer and confine the zonal winds to the upper region . traditional dynamical models therefore treat the two layers separately with dynamo simulations modelling only the metallic layer and jet simulations concentrating on the molecular envelope . the zonal jets have been investigated since the 70s by tracking cloud features ( see , for example , @xcite , @xcite for jupiter and @xcite , @xcite for saturn ) . their driving forces and depth are still debated . some authors argue that they are a shallow weather phenomenon @xcite while others promote deeper - rooted jets that extend through the whole molecular envelope @xcite . both gas giants emit roughly twice as much energy as they receive from the sun which implies vigorous interior convection . in the rotationally - dominated dynamics ruling planetary atmospheres , interior convection naturally drives zonal winds via reynolds stresses ( i.e. a statistical correlation between the convective flow components ; * ? ? ? * ; * ? ? ? these winds follow a geostrophic structure , minimizing variations in the direction of the rotation axis , and therefore reach through the whole fluid atmosphere . @xcite show that even when the forcing is restricted to a shallow weather layer the jets may reach much deeper into the planet . @xcite , on the other hand , present an anelastic deep convection model where the equatorial zonal flow is geostrophic and the higher latitude jets are confined to the outer few percent in radius . saturn s magnetic field is very axisymmetric and strongly concentrated at higher latitudes @xcite which is incompatible with the results of a classical earth - like dynamo model . a stably stratified layer at the top of the dynamo region @xcite or a completely different dynamo driven by differential rotation @xcite are two proposed alternatives for the special situation encountered at saturn . here we concentrate on jupiter whose field is very similar to the geomagnetic field so that the well - explored geodynamo models also seem to apply at first sight . these models typically adopt the boussinesq approximation where the mild @xmath1% density stratification of earth s core is simply ignored . in jupiter , however , the density increases by more than a factor of 5000 below the @xmath2bar level . while the stratification is mostly concentrated in the outer molecular envelope , the density still rises by about one order of magnitude across the metallic layer ( fig . 1 of * ? ? ? some newer numerical models therefore use the anelastic approximation which allows to incorporate the effects of the background density stratification while filtering out fast sound waves @xcite . in an extensive parameter study , @xcite ( hereafter referred to as gdw12 ) show that dipole - dominated dynamos are rather rare when stronger stratifications are assumed . gdw12 quantify the stratification in their anelastic models in terms of the number of density scale heights @xmath3 , where @xmath4 and @xmath5 are the densities at the inner and outer boundaries of the simulated shell , respectively . for the larger density stratifications @xmath6 , a value that corresponds to an increase by a factor @xmath7 , no dipole - dominated solutions were found . this is attributed to the fact that the focus of convective action moves progressively outward in cylindrical radius when the stratification is intensified . once the convective columns are mainly confined to a relatively thin outer shell , a non - axisymmetric dynamo mode is preferred that has previously only been observed in mean field dynamo simulations @xcite . we will refer to this as the thin - shell dynamo model in the following . for the smaller to intermediate stratifications @xmath8 , gdw12 find dipole dominated magnetic fields when the local rossby number remains smaller than a critical value of @xmath9 . this is consistent with the findings of @xcite who introduced @xmath10 as a measure for the relative importance of inertia in their boussinesq models ( see eq . [ eq : rol ] ) . multipolar solutions with weaker magnetic fields on the other hand exist for all @xmath10 values which means that both types of solutions coexist below @xmath11 for identical model parameters , forming two distinct branches . this so - called bistability can be attributed to the fact that free - slip boundary conditions were employed @xcite . these conditions allow strong zonal winds to develop that compete with large scale magnetic fields . on the dipolar branch , zonal winds are weak , on the multipolar branch they are stronger . when no - slip conditions are used zonal flows generally remain weaker and only the dipolar branch is found for @xmath12 @xcite . _ ab initio _ calculations suggest that there is actually no clear phase transition between the regions of molecular and metallic hydrogen states @xcite . in the dynamo context , the electrical conductivity profile is of particular importance . due to the increasing degree of hydrogen ionization , the conductivity rises super - exponentially with depth and matches the conductivity of the metallic region at the transition radius without any pronounced jump . the classical separation of the dynamics for the two envelopes thus becomes questionable . @xcite argue that this has important consequences for the depth of the zonal winds which should remain confined to a shallow outer layer where the conductivity remains negligible . the strong shear associated with the zonal winds would otherwise create strong azimuthal magnetic field and lead to ohmic heating incompatible with the observed luminosity ( see however * ? ? ? * ) . @xcite present an anelastic simulation of a relatively thin shell with exponentially decaying electrical conductivity to model the very outer part of the shell . the model uses extreme parameters ( i.e. low ekman and prandtl number and high rayleigh number ) and a dipole - dominated magnetic field develops in the presence of strong geostrophic zonal winds . however , since a detailed discussion and a systematic parameter study are missing , it remains impossible to disentangle the effects of density stratification , varying conductivity , and the particular parameter choice . @xcite and @xcite also include a radial conductivity profile in their deep shell boussinesq models , with a constant conductivity in the deeper interior and an exponential decay in the outer part . these models also demonstrate that well - pronounced deep - rooted zonal winds can be compatible with dipole - dominated dynamo action . the present paper extends the work of gdw12 by adding an electrical conductivity profile loosely based on the _ ab initio _ calculations by @xcite . following @xcite and @xcite , the electrical conductivity profile assumes a constant value in the metallic region and an exponential decay in the molecular region . the aim is to systematically explore under which circumstances dipole - dominated dynamo action and strong zonal surface winds can coexist in anelastic dynamo models . we describe our model in section [ model ] with special attention to the anelastic formulation and the electrical conductivity profile . the numerical results are presented in section [ results ] , first concentrating on the question of dipole - dominance and then on the dynamo mechanism . section [ conclusions ] summarizes our main results and discusses their implications for the gas giants . the fluid and convective interior of the planet is modelled by solving the mhd equations in a rapidly - rotating spherical shell . previous models typically used the boussinesq approximation , which neglects the background density and temperature variations . this is questionable in gas planets and , following @xcite , @xcite and @xcite , we therefore adopt the anelastic approximation . this allows to include background variations while ruling out sound waves by neglecting fast local density variations . we solve the equations in a dimensionless form ( e.g. * ? ? ? * ) , using the shell thickness @xmath13 as a length scale and the viscous diffusion time @xmath14 as a timescale . here , @xmath15 and @xmath16 are the outer and inner radii , respectively , and @xmath17 is the kinematic viscosity . temperature and density are both non - dimensionalized by their values at the outer boundary , @xmath18 and @xmath5 . we employ constant entropy boundary conditions and use the imposed contrast @xmath19 across the shell as the entropy scale . there are no internal heat sources and all the heating coming into the shell via the inner boundary leaves it through the outer . while this is not the most realistic heating mode for gas giants , it has been chosen to ease the comparisons with more classical boussinesq simulations . the magnetic field is scaled by @xmath20 , where @xmath21 is the rotation rate of the shell and @xmath22 is the inner boundary reference value of the magnetic diffusivity @xmath23 . here , @xmath24 is the magnetic permeability and @xmath25 is the prescribed electrical conductivity profile . below we will also use the normalized magnetic diffusivity and electrical conductivity profiles related via : @xmath26 . the medium is assumed to be an electrically conducting ideal polytropic gas . generally , dynamo simulations solve for small variations around an adiabatic hydrostatic background state that we mark with a tilde in the following . the background temperature profile is then defined by the background temperature gradient @xmath27 and the density profile by @xmath28 , where @xmath29 is the polytropic index . for simplicity , we adopt a gravity profile proportional to radius which implicitly assumes a homogeneous density . the other extreme is to assume that all the mass is concentrated in the centre , which leads to a gravity profile proportional to @xmath30 @xcite . gdw12 show that both gravity profiles lead to very similar results . the true profile of the gas giants lies somewhere in - between . the temperature reference state is then given by @xmath31 where @xmath32 @xmath33 is the number of density scale heights between the inner and the outer boundaries of the shell and @xmath34 is the ratio between the corresponding radii ( see * ? ? ? * ; * ? ? ? * for the full derivation of the reference state ) . the dimensionless form of the anelastic equations is @xmath35 @xmath36 @xmath37 @xmath38 @xmath39 the traceless rate - of - strain tensor @xmath40 for the homogeneous kinematic viscosity assumed here is given by @xmath41 \;\;\;\textrm { and } \;\;\ ; \textsf{e}_{ij } = \frac{1}{2}\bigg(\frac{\partial u_i}{\partial x_j}+\frac{\partial u_j}{\partial x_i}\bigg ) \textrm { , } \label{eq : straintensor}\ ] ] where @xmath42 is the identity matrix . the viscous and ohmic heating contributions are @xmath43 \label{eq : vischeat}\ ] ] and @xmath44 the system of eqs . ( [ eq : navierstokeseq][eq : divfieldeq ] ) is governed by the dimensionless ekman number @xmath45 , rayleigh number @xmath46 , prandtl number @xmath47 and magnetic prandtl number at the inner boundary @xmath48 : @xmath49 @xmath50 @xmath51 @xmath52 the specific heat @xmath53 , the thermal diffusivity @xmath54 , magnetic diffusivity @xmath55 and kinematic viscosity @xmath17 are all assumed to be homogeneous . to quantify gravity we use the reference value @xmath56 at the outer boundary . to simulate the variable electrical conductivity of hydrogen in the interior of jupiter , we employ a profile that corresponds to a constant conductivity in the metallic hydrogen layer and an exponential decay in the outer molecular envelope . both branches are matched via a polynomial that also ensures that the first radial derivative is continuous : @xmath57 & r\ge r_m \end{array } \right . \textrm{. } \label{varcondeq}\ ] ] the exponential decay with a rate @xmath58 starts at a radius @xmath59 where the normalized conductivity has already decreased from @xmath60 to @xmath61 . for convenience we also define the relative transition radius in percentage : @xmath62 . this profile has first been used by @xcite and it seems a fair first approximation to the results from _ ab initio _ calculations by @xcite . the super - exponential increase of electrical conductivity over the molecular layer is not feasible to model numerically ( see fig . [ fig : vconds ] ) . we thus mainly use a rate of @xmath63 for our simulations , but we also tested @xmath64 in a few cases ( see tab . [ tab2 ] ) and @xmath65 for one case with a different @xmath66 ( grey profile in fig . [ fig : vconds ] , discussed in section [ diporflow ] ) . in all the other cases , @xmath61 was fixed to @xmath67 and @xmath68 was varied assuming values of @xmath69 , @xmath70 , @xmath71 and @xmath72% . corresponding simulations for homogeneous conductivity with @xmath73 can be found in gdw12 . for the numerical simulation of the model described above , we use the anelastic version of the magic code @xcite . this is a pseudo - spectral code that solves eqs . ( [ eq : navierstokeseq][eq : divfieldeq ] ) in a spherical shell using a poloidal / toroidal decomposition of the vector fields @xmath74 and @xmath75 : @xmath76 for the spectral representation of the dependence on latitude @xmath77 and longitude @xmath78 , the poloidal potentials @xmath79 and @xmath80 , the toroidal potentials @xmath81 and @xmath58 , the entropy @xmath82 and the pressure @xmath83 are expanded in spherical harmonic functions up to degree and order @xmath84 . chebyshev polynomials up to degree @xmath85 are used in the radial direction . for the parameter studies presented here , we use different resolutions , because the higher gradient in density and electrical conductivity demand an increase of both radial and horizontal resolutions . for the chebyshev polynomial truncations between @xmath86 and @xmath87 are used while @xmath84 ranges between @xmath88 and @xmath89 . each simulation is run for at least one magnetic diffusion time with the exception of some cases at the lower ekman number @xmath90 ( see tab . [ tab2 ] ) . regarding the velocity boundary conditions , we apply a no - slip condition at the inner core boundary and a free - slip condition at the outer boundary in most of our simulations which seems appropriate for a gas planet with a rocky core . a few test cases with no - slip conditions at both boundaries allow to explore the impact of the boundary condition on the dynamics . the existence of an inner core in jupiter and its possible size is still unclear . here , we assume a small and electrically conducting solid inner core with @xmath91 . gdw12 explore @xmath92 and @xmath93 in very similar models with homogeneous electrical conductivity and find generally very similar results . furthermore , we use constant entropy boundary conditions and match the magnetic field to a diffusive solution at the inner boundary and to a potential field at the outer boundary . the ekman number is either @xmath94 or @xmath90 . the larger value allows a more extensive scan of the other system parameters like rayleigh number , density stratification @xmath95 and electrical conductivity transition radius @xmath68 . at @xmath90 , we could only afford to run eleven cases in a more restricted parameter regime . we assume a prandtl number of @xmath96 and an inner boundary magnetic prandtl of typically @xmath97 for @xmath94 and @xmath98 for @xmath90 . nine additional cases with @xmath99 at @xmath94 and @xmath100 at @xmath90 have also been computed . [ cols="^,^,^,^",options="header " , ] [ tab1 ] all together , we ran 74 cases with rayleigh numbers between @xmath101 and @xmath102 times supercritical . @xcite examine anelastic convection for an aspect ratio of @xmath103 and show that the critical rayleigh number increases with increasing stratification @xmath95 . table [ tab1 ] demonstrates that we observe a similar trend for the smaller aspect ratio @xmath104 employed here . at a certain stratification , the critical wave number jumps from lower to high values . this is the point where the centre of the flow convection moves from close to the inner to close the outer boundary . the respective transition happens at larger stratifications @xmath105 when the ekman number is decreased ( see also * ? ? ? * ) . in our anelastic simulations , we consider a polytropic index of @xmath106 and we explore density scale heights ranging from the boussinesq case @xmath107 to @xmath108 , where the latter corresponds to a density jump of @xmath109 . while _ ab initio _ simulations suggest a jovian stratification of @xmath110 from the bottom of the molecular hydrogen layer to the @xmath2bar level @xcite . however , since the density gradient rapidly steepens with radius in the planets outer shell our largest stratifications already cover the inner 99% of jupiter s radius . the parameters of all numerical experiments discussed here are listed in tab . [ tab2 ] along with several diagnostic quantities that characterize the solution and are defined in the following . the amplitude of the zonal flow contribution is measured in terms of the rossby number @xmath111 : @xmath112 where @xmath113 is the rms volume - averaged flow velocity and the triangular brackets denote the angular average @xmath114 @xmath115 is the axisymmetric azimuthal flow component , and @xmath116 is the volume of the spherical shell . overbars correspond to azimuthal averages . we use the relative kinetic energy @xmath117 to quantify the relative importance of zonal flows . the magnetic reynolds number @xmath118 estimates the ratio of magnetic field production and diffusion and we use a modified form here to account for the radial - dependent magnetic diffusivity : @xmath119 the local rossby number has been introduced by @xcite to quantify the relative importance of the advection term in the navier - stokes equation ( eq . [ eq : navierstokeseq ] ) and is defined as @xmath120 here , @xmath121 is a typical flow length scale given by @xmath122 where @xmath123 is the flow contribution of spherical harmonic degree @xmath124 . we use a modified form of @xmath10 based exclusively on the inner conducting region ( @xmath125 ) : @xmath126 the magnetic field strength is quantified by the elsasser number which measures the ratio of lorentz to coriolis forces using the modified form @xmath127 the geometry of the surface field is characterized by the dipolarity @xmath128 which measures the relative energy in the axial dipole contribution at the outer boundary @xmath15 . following @xcite , we restricted the magnetic field to spherical harmonic degrees and orders below 12 in eq . ( [ eq : dip ] ) . tab . [ tab2 ] lists time averages of the properties defined above for all our models and we always refer to the time - averaged properties for characterizing our solutions in the following . the time variability of the dipolarity , also listed in tab . [ tab2 ] , is quantified by its standard deviation @xmath129 . in the complex models explored here , the magnetic field geometry not only depends on the local rossby number @xcite but also on the density stratification , on the thickness of the weaker conducting layer , on the ekman number , and on the magnetic prandtl number . fig . [ fig : dipvsrol ] shows the dependence of the dipolarity @xmath130 on the local rossby number for all our cases , excluding runs with @xmath131 , with the exception of case 67 from tab . [ tab2 ] which is similar to one of @xcite s runs . to illustrate the relation between the field geometry and the zonal flows , we plot @xmath130 versus the relative kinetic energy of axisymmetric azimuthal flows in fig . [ fig : dipvsekin ] . in both figures , the symbol type refers to the different stratifications while the symbol colour identifies the four transitional radii @xmath68 explored here . we start by analysing the different dynamo regimes based on the results for @xmath94 and @xmath132 and come back to the solutions for larger magnetic prandtl numbers and for @xmath90 further below . when the weakly conducting layer is relatively thin ( @xmath133 and @xmath134 ) and the stratification is mild to intermediate @xmath135 , we find two distinct branches . a dipolar branch , characterized by @xmath136 and weak zonal flows , is restricted to cases with local rossby numbers below the critical value of @xmath137 . this is significantly lower than the values of @xmath138 suggested for homogeneous electrical conductivity by gdw12 . the dipole - dominated solutions forming this branch are located in the upper left corner of fig . [ fig : dipvsrol ] and in the left portion of the yellow high - dipolarity regime in fig . [ fig : dipvsekin ] . a second branch with multipolar magnetic fields at @xmath139 but intermediate zonal flows exists for all @xmath10 values . these solutions can be found in the lower part of fig . [ fig : dipvsrol ] and the cyan low - dipolarity regime in fig . [ fig : dipvsekin ] . for local rossby numbers below @xmath140 , we thus find both types of solutions while only multipolar solutions remain stable beyond @xmath11 . figures [ fig : dipvsrol ] and [ fig : dipvsekin ] contain seven examples ( dashed lines ) where a solution on each branch is found for identical model parameters , clearly demonstrating the bistability for @xmath12 . which branch a specific numerical simulation will chose depends on the initial magnetic field configuration . note that the multipolar attractor always has the more intense zonal flows ( see fig . [ fig : dipvsekin ] ) . comparing magnetic reynolds numbers and local rossby numbers for bistable cases shows that the relative difference is smaller in the latter than in the former measure . this indicates that the weaker flow amplitude caused by the larger lorentz forces in the dipole - dominated cases is accompanied by a growth in the flow length scale . increasing the stratification to values beyond @xmath141 while keeping @xmath68 large always leads to solutions of the multipolar thin - shell type discussed by gdw12 . altogether , the behaviour for a thin weakly conducting layer is similar to that for a homogeneous electrical conductivity with the exception of the lower critical rossby number @xmath11 . for a thicker weakly conducting layer with @xmath142% , the influence of the stratification on the dipolarity is reversed . clearly , dipolar solutions with @xmath136 now exclusively exist for stratifications of @xmath143 or @xmath108 . since the relative zonal flow amplitude reaches intermediate values , these cases can be found in the middle section of the yellow regime in fig . [ fig : dipvsekin ] . a second branch of solutions is characterized by low to intermediate dipolarity that increases with @xmath105 and by large relative zonal flow amplitudes . these cases populate the pink region in fig . [ fig : dipvsekin ] . for stratification of @xmath144 , the solutions on this secondary branch become strongly time - dependent as indicated by the large error bars in figs . [ fig : dipvsrol ] and [ fig : dipvsekin ] . demonstrates that the time dependence reflects an oscillation between dipolar and multipolar field configurations without ever establishing a solution on the dipole - dominated branch . polarity reversals or excursions become possible when the dipolarity is relatively low . once more , both branches coexist for not too large local rossby numbers and we could identify two bistable cases for @xmath142% , @xmath143 and @xmath145 , @xmath146 . when increasing the rayleigh number to @xmath147 , however , only the multipolar solution remains which suggests a critical local rossby number of about @xmath148 ( see tab . [ tab2 ] ) . for @xmath149% , the thickest weakly conducting outer shell explored here , even stronger stratification seems required to establish a dipole dominated magnetic field than at @xmath142% . for the magnetic prandtl number @xmath132 , generally used at @xmath94 , only the highly time - dependent solutions with intermediate dipolarity and strong zonal winds ( on average ) were found , even at @xmath143 . however , the mean dipolarity increases with @xmath105 , just as in the @xmath142% cases and stratifications of @xmath150 may finally establish a dipole - dominated solution . highlights the role of the density stratification at @xmath151% , @xmath70% , @xmath71% and @xmath72% for models at @xmath94 and @xmath132 with similar local rossby numbers . when the weakly conducting layer is relatively thin ( @xmath133 or @xmath134 ) , dipole - dominated solutions can only be found for @xmath152 while multipolar solutions exist for all stratifications . for a thicker layer , however , the stratification has a reversed effect . the mean dipolarity increases with stratification and highly dipolar solutions are only found for stronger stratifications @xmath153 . note that at @xmath154% and @xmath107 , we could only find multipolar solutions , even at low rayleigh numbers where @xmath10 is small . the reason for this is not yet understood . what finally helped to establish a dipolar solution here was increasing the magnetic prandtl number from @xmath155 to @xmath156 . we also tested the effect of larger magnetic prandtl numbers for several other parameter combinations and this often promoted dipole - dominated solutions . for example , at @xmath142% and @xmath107 a multipolar case became dipolar when increasing @xmath157 from @xmath155 to @xmath158 . likewise , the highly time - dependent case at @xmath142% , @xmath159 , @xmath160 and @xmath132 developed into a stable dipole - dominated solution when doubling @xmath157 . the same behaviour was found at @xmath149% , @xmath159 and @xmath161 . this indicates a certain trade - off between larger stratifications and higher electrical conductivities . at @xmath142% or @xmath72% , @xmath162 and @xmath163 , however , an increase from @xmath132 to @xmath164 was not sufficient to establish a dipole - dominated solution . even higher magnetic prandtl numbers may be required here . illustrates the different types of solutions with snapshots of the radial magnetic field at the outer boundary for @xmath94 . the top row shows @xmath133 cases at two different stratifications : a dipole - dominated boussinesq case and a multipolar solution at @xmath165 . the latter shows a large scale wave number ( @xmath166 ) structure similar to that reported for multipolar dynamos with homogeneous electrical conductivity and free - slip boundaries ( * ? ? ? * gdw12 ) . the bottom row of fig . [ fig : brs ] depicts the two branches found for @xmath167 . the left panel shows a snapshot of a boussinesq multipolar case and the right panel illustrates the dipolar configuration found at strong stratifications ( @xmath168 ) . the results by @xcite prompted us to also conduct simulations at the lower ekman number of @xmath169 used in their study . [ tab2 ] lists the respective models with different rayleigh numbers and stratifications . the thicker weakly conducting layer of @xmath142% was generally chosen except for one model with @xmath151% . the boussinesq case 67 is identical to one of the models presented by @xcite and has a larger aspect ratio of @xmath170 instead of @xmath92 . the outer grey line of the symbols in figs . [ fig : dipvsrol ] and [ fig : dipvsekin ] corresponds to the twelve runs with @xmath90 . while for @xmath94 and @xmath142% we had to increase the stratification to @xmath153 to find strongly dipolar solutions , this is not the case any more at @xmath171 . even the boussinesq models now clearly have dipole - dominated magnetic fields . the only multipolar case , which seems to be of the highly time - dependent type , is found at @xmath172 and it has a local rossby number of @xmath173 . another model at @xmath172 but @xmath174 is strongly dipolar so that the critical local rossby number can be estimated to @xmath175 . note however , that @xmath11 may depend on stratification . the coexistence of dipolar and multipolar branches indicates a competition between zonal winds and dipolar magnetic fields already discussed by gdw12 . the stronger lorentz forces associated to the larger dipolar fields effectively compete with the reynolds stresses responsible for driving the zonal winds . the zonal wind amplitude and the relative zonal wind energy thus remain typically small . fierce zonal winds , on the other hand , seem to promote multipolar fields . this is at least the situation for @xmath94 and @xmath151% or @xmath70% . but why are stronger stratifications and/or larger magnetic prandtl numbers required to yield dipole - dominated dynamo action for thicker weakly conducting outer layers ? + + illustrates the zonal flow structure and the poloidal magnetic field lines for different stratifications at @xmath151% ( middle row ) and @xmath142% ( bottom row ) for @xmath94 . the top row shows non - magnetic cases and demonstrates that the inner retrograde jet decreases in amplitude when the stratification intensifies . this reflects the progressive outward concentration of the convective motions and thus of the reynolds stresses driving the zonal flows ( gdw12 ) . the dominance of coriolis forces at this relatively low ekman number enforces the taylor - proudman theorem and the intense zonal jets remain strongly geostrophic , i.e. variations in the direction of the rotation axis are much smaller than variations perpendicular to it . for the thinner weakly conducting layer ( middle row in fig . [ fig : zonalflow ] ) , the lorentz forces associated with the stronger dipolar field at mild stratifications effectively suppress the zonal flows in the whole shell . for @xmath6 , the weaker multipolar fields created by the thin - shell dynamo allow the outer prograde jet to survive , albeit with a significantly reduced amplitude and a restricted width than in the non - magnetic simulations . the thickness of the weakly conducting layer now determines the width of the outer jet , confirming previous work by @xcite . at @xmath142% ( lower row in fig . [ fig : zonalflow ] ) , the zonal flows generally remain more energetic than for @xmath151% . since these flows are largely geostrophic , the force balances on geostrophic cylinders ( i.e. on cylinders aligned with the rotation axis ) should be considered . the lorentz forces now have a harder time to brake the zonal flows since they act in a significantly reduced volume . dipole - dominated dynamo action only becomes possible when the retrograde inner zonal jet is already relatively weak in the non - magnetic simulations , which happens at stronger stratifications . the thin - shell dynamo mechanism generating the multipolar field for @xmath176% does not apply here , since it would have to operate , at least partly , in the weakly conducting layer where the magnetic reynolds number is now too low to support dynamo action . instead , a strongly dipolar magnetic field is generated in the deeper interior where it does not interfere with the remaining prograde outer zonal jet . the change in the depth of the poloidal dynamo action is further illustrated by the radial profiles of magnetic energy shown in fig . [ fig : emagvsrad ] . up to a stratification of @xmath159 , poloidal and toroidal magnetic energies have similar profiles and peak in the outer part of the conducting region around @xmath177 . for @xmath143 , however , the profiles are different with a pronounced focus on deeper parts of the shell around @xmath178 . for @xmath179 , the toroidal magnetic field is larger than the poloidal , suggesting that the induction mechanism is different from the dipole - dominated case at @xmath143 where the poloidal field is stronger . at @xmath149% , the volume over which lorentz forces can efficiently brake zonal winds is even further reduced . dipolar solutions can only be found at even larger stratification than at @xmath142% where the driving of the inner zonal jet is yet weaker . the role of lorentz forces in defeating zonal winds and thereby enabling dipole - dominated magnetic fields also offers an explanation why larger magnetic prandtl numbers help . the reason likely is that larger @xmath157 values lead to stronger magnetic fields and thus stronger lorentz forces . we can also now interpret the highly time - dependent solutions with intermediate mean dipolarities . here , the balance seems to be undecided ( fig . [ fig : diptilt ] ) . stronger lorentz forces successfully suppress the zonal winds at times but never enough to establish the solution on the highly dipolar more stable branch . at other times , reynolds stresses succeed in driving stronger zonal flows that mostly create a weaker multipolar magnetic field . to further test the theory that the zonal flows are decisive for the field geometry we ran a few @xmath94 cases with a no - slip outer boundary condition that largely prevents zonal flows from developing . the results are mixed and not entirely conclusive , which may have to do with the fact that other flow components are also affected by this change in boundary conditions . at @xmath151% , @xmath107 and @xmath180 , the no - slip boundary conditions indeed promote a dipole - dominated solution with weak zonal flows where we only find multipolar solutions with strong zonal flows for a free - slip outer boundary condition ( compare cases 3 and @xmath181 ) . the same positive effect was found for @xmath154% , @xmath107 and @xmath182 ( cases 7 and @xmath183 ) . at @xmath154% , @xmath172 and @xmath184 , however , we find bistable cases for both type of boundary conditions ( cases 22d / m and @xmath185d / m ) . in the no - slip case , both the dipole - dominated and the multipolar solution have weak zonal flows . free - slip outer boundary condition promotes dipolarity , but it is not a necessary condition to find this feature . note that such a bistable case for no - slip conditions has already been reported by @xcite . at @xmath142% , @xmath107 and @xmath180 , the suppression of the zonal flows by the no - slip condition is not sufficient to yield a dipole - dominated solution and the same is true at @xmath142% , @xmath162 and @xmath186 or @xmath163 ( cases 17 , 32 and 36 ) . in the latter two examples , the particular thin shell dynamo described by gdw12 , rather than the stronger zonal flows , may be the reason for the multipolarity which could explain why the no - slip condition has no effect . we also varied the electrical conductivity profile in a few cases . increasing the exponential decay rate from @xmath187 to @xmath64 for two simulations at @xmath142% required a finer radial numerical grid and thus more expensive numerical simulations . the zonal flows in the weakly conducting layer were intensified in both cases , likely because of the further decreased weaker lorentz forces there . the type of solution , however , remained unchanged ( see cases 33/34 and 51/52 ) . we also tested a more realistic profile that models the approximately linear decrease of electrical conductivity in the metallic layer ( see fig . [ fig : vconds ] ) and a steeper decrease at larger radii . at @xmath142% , @xmath143 and @xmath188 the solution is bistable for our standard conductivity profile . for the more realistic profile we so far only found a clearly dipole - dominated at the same supercriticality but we can not exclude that the multipolar case also exists . compares the radial profiles of the convective magnetic reynolds number @xmath189 for both profiles ( yellow lines and grey line ) . being based on rms flows velocities that exclude zonal winds , @xmath189 is appropriate for characterizing poloidal magnetic field production . numerical simulations suggest that a magnetic reynolds number larger than @xmath190 is required to support dynamo action @xcite . for our standard electrical conductivity profiles , @xmath189 typically falls below this value for radii beyond @xmath191 or @xmath192 . the linear decrease in the metallic layer , however , further reduces the convective reynolds number which is already very low at depth . @xmath189 values larger than @xmath190 are now restricted to the inner region of @xmath193 . a multipolar dynamo where the outer parts of the shell play a sizeable role thus becomes unlikely . in jupiter , @xmath189 is generally significantly higher in the metallic region and only decreases below the critical value for dynamo action in the molecular envelope . we therefore refrained from further exploring this profile since the decrease in magnetic reynolds number artificially limits the dynamo region . shows zonal flows and axisymmetric poloidal field lines for the dipole - dominated solutions at @xmath94 with the modified electrical conductivity profile ( first panel from the left ) and the standard profile ( second panel ) . the poloidal fields are very similar and produced at greater depth in both cases . this explains why the low convective magnetic reynolds number in the outer part of the shell has little impact on the dynamo mechanism for dipolar dominated solutions . once more , the weaker lorentz force in the outer layer allows for more vigorous zonal winds for the more realistic conductivity profile . ( cases 50 in grey , 51 in red , 60 in blue and 66 in green , from tab . [ tab2 ] ) . the two additional yellow cases ( 53d / m in tab . [ tab2 ] ) yield a higher @xmath46 and bistability , at @xmath143 and @xmath194 . [ fig : rms],title="fig:",height=307 ] how do the results at @xmath90 fit into the picture we outlined above ? compares two @xmath90 cases at @xmath195 ( third panel , case 60 ) and @xmath196 ( fourth panel , case 66 ) with the dipole - dominated solutions for the more realistic profile ( first panel , case 50 ) and for our standard profile at @xmath168 and @xmath94 ( second panel , case 51 , see also fig . [ fig : zonalflow ] ) . the magnetic field and zonal flow structures are very similar in all cases . at both ekman numbers , the relative amplitude of the retrograde jets decreases with increasing @xmath105 ( see also tab . [ tab2 ] ) . the absolute zonal flow amplitude , however , is significantly smaller in all lower ekman number models . for example , the zonal flow rossby number is @xmath197 in the @xmath90 / @xmath196 case ( 66 ) depicted in fig . [ fig : vppoln1 ] , but @xmath198 in the @xmath94 / @xmath168 simulation ( 51 ) . in non - magnetic free - slip simulations , the flow amplitude roughly scales with the modified rayleigh number @xmath199 , as it has been shown by @xcite for boussinesq and @xcite for anelastic models . for example , @xcite suggest the dependence @xmath200 . this scaling describes an asymptotic behaviour for larger rayleigh numbers where zonal flows clearly dominate so that @xmath201 . for the smaller rayleigh numbers typically examined here , it may only serve as a rough estimate for the zonal flow amplitude . for case 51 with @xmath94 , we have @xmath202 and the scaling predicts @xmath203 . for case 66 with @xmath90 and @xmath204 , it predicts @xmath205 . both values are not too far from the numerical results @xmath206 and @xmath207 , respectively , which suggests that the difference in @xmath208 is indeed the main reason for the much weaker zonal flows at the lower ekman number . because of the quadratic ekman number dependence of @xmath208 , @xmath46 has to be increased by two orders of magnitude to reach the same zonal flow amplitudes in the @xmath90 as in the @xmath94 cases . this leads to larger @xmath10 values and thus possibly multipolar fields @xcite . the rayleigh number increase from @xmath209 ( case 61 ) to @xmath210 ( case 62 ) at @xmath172 already leads to a multipolar field while only doubling the zonal flow amplitude . the similar elsasser numbers in the dipole - dominated cases at both ekman numbers indicate that the lorentz forces also have comparable amplitudes . these forces have a much easier job to brake the systematically weak zonal flows at @xmath90 , allowing a dipole - dominated field to develop even at mild stratifications . the more extensive parameter study at @xmath94 suggests that stronger stratifications should allow for more vigorous outer jets while retaining dipole - dominated dynamo action . gdw12 reported that the multipolar solutions with stronger zonal flows are dynamos of an @xmath211 or an @xmath212 type . dynamos of the @xmath213 type , on the other hand , are known to produce dipole - dominated magnetic fields @xcite . the @xmath214 stands for poloidal and toroidal field production by local helical structures , while @xmath21 stands for the production of toroidal field by global zonal wind shear . following @xcite , the @xmath21-effect is given by @xmath215 and describes the production of the axisymmetric azimuthal magnetic field @xmath216 which is purely toroidal . the mean ohmic diffusion of @xmath216 is given by @xmath217 compares @xmath216 , @xmath21 and md for two @xmath142% cases , a multipolar solution at @xmath162 ( left ) and dipole dominated solution at @xmath143 ( right ) . both cases were depicted previously in figs . [ fig : zonalflow ] and [ fig : emagvsrad ] . for the multipolar solution , the @xmath21-effect plays an important role , as demonstrated by the high degree of correlation with the azimuthal field over the shell . for the dipole - dominated solution , however , the @xmath21-effect is only strong in the weakly conducting region where it is effectively balanced by the large ohmic diffusion . the change in field geometry is thus once more coupled to a switch from an @xmath211 or @xmath212 mechanism at weaker stratifications to an @xmath213 mechanism at stronger density stratifications . we ran a suite of dynamo simulations with an electrical conductivity profile geared to combine the dynamics of the metallic and the molecular hydrogen layers of the gas giants in one integrated model . in most models , the conductivity is assumed to remain constant over the inner part of the shell representing the metallic hydrogen region . beyond a relative radius of @xmath68 , it decays exponentially with radius , to model the molecular envelope . the use of the anelastic mhd code allowed us to also study the effects of density stratification . free - slip outer boundary and no - slip inner boundary conditions seem appropriate for the gas giants and were used in the majority of our simulations . in gdw12 , we had explored the dynamo action for homogeneous electrical conductivity in an otherwise identical setup . dipole - dominated solutions were only found for mild stratifications and local rossby numbers below @xmath218 . the respective solution branch is characterized by weak zonal winds and coexists with a second branch with weaker multipolar magnetic fields but stronger zonal winds at identical parameters . this indicates a competition between zonal winds and dipolar magnetic fields @xcite . translated to jupiter and saturn , these simulations would predict multipolar magnetic fields , should the observed zonal winds reach into the dynamo region . the strong stratification within the gas giants should also promote multipolar solutions . we largely recover these results when the outer weakly conducting layer occupies only @xmath156 or @xmath158 percent in radius ( @xmath133 and @xmath134 ) . however , the critical local rossby number , below which dipole - dominated solutions are possible , decreases to @xmath219 . @xcite already showed that even very thin weakly conducting outer layers promote multipolar magnetic field configurations . they speculate that the separation of the ekman and hartmann boundary layers may play a role in their models with no - slip boundaries , although this explanation is difficult to apply for the free - slip models predominantly explored here . more research is required to clarify this point in the future , specially concerning no - slip boundaries . for a thicker weakly conducting outer layer covering the outer 20 or 30% in radius ( @xmath142% or @xmath72% ) , the volume over which the lorentz forces can act to balance reynolds stresses is more significantly reduced . the competition between zonal wind and stronger dipolar fields thus becomes even more of an issue . at an ekman number of @xmath94 , the mean zonal winds tend to be relatively strong even at low rayleigh numbers . dipole - dominated dynamo action is nevertheless possible in the deeper interior of strongly stratified models , where the zonal flows remain relatively weak even in the non - magnetic case . alternatively , dipole - dominated solutions are found for larger magnetic prandtl numbers which help to keep zonal flows at bay by increasing lorentz forces . in the dipole - dominated solutions , the zonal winds are then mainly restricted to a fierce prograde jet that resides within the weakly conducting outer envelope . at the lower ekman number of @xmath90 , dipole - dominated magnetic fields can even be maintained at weak stratifications because the zonal flow amplitudes are lower than at @xmath94 . the peak velocity of jupiter s equatorial jet is around @xmath220 @xcite and about @xmath221 for saturn @xcite . for example , the @xmath194 case illustrated in fig . [ fig : vppoln1 ] ( second panel from the left ) reaches @xmath222 which is somewhat too high for jupiter . the @xmath90 simulation depicted in the same figure has @xmath223 at the lower stratification of @xmath196 . the amplitude of the equatorial jet decreases with ekman number and increases with density stratification . we speculate that the higher stratifications within the gas giants may allow to reach appropriate zonal jet amplitudes at the much lower realistic ekman number , around @xmath224 @xcite , while retaining dipole - dominated dynamo action . the number of zonal jets is much smaller in our simulations than for the gas giants . also , the strong decrease in the zonal flow amplitude from the equatorial to the flanking jets , that is necessary to retain dipole - dominated dynamo action in our models , is not compatible with the observations for jupiter . a dipolar configuration nevertheless seems possible should the higher latitude jets remain too shallow to interfere with the deeper dynamo process . the equatorial jet does not pose a problem in this respect because it can reside completely within the lower conductivity envelope . an argument against deep reaching winds is that the associated strong @xmath21-effect and ohmic dissipation may not be compatible with jupiter s observed luminosity @xcite . a first analysis of our results confirms that the @xmath21-effect and associated ohmic dissipation can be significant . @xcite argues that the magnetic field may assume a configuration where the poloidal field lines are aligned with the rotation axis in regions of strong zonal flow shear . since the shear is perpendicular to the rotation axis , this would minimize the @xmath21-effect and related ohmic dissipation . illustrates that the field lines indeed approach such an alignment in the very outer part of the shell where the electrical conductivity is still important . the ohmic dissipation nevertheless remains significant in all our simulations with strong zonal flows . further investigation is necessary to quantify this effect and extrapolate it to the planetary situation . any problems related to ohmic dissipation and dipolar dynamo action would not be an issue when stronger zonal winds remain confined to a thin outer envelope with ( @xmath225 ) , where the electrical conductivity remains small enough @xcite . in our simulations , however , all the stronger jets obey the taylor - proudman theorem and reach through the planet . shallow jets have been found by @xcite , who use a different anelastic approximation and a different internal heating mode . further investigations are required to clarify which specific model features influence the depth on the zonal jets . all the computations have been carried out in the gwdg computer facilities in gttingen , in the norddeutscher verbund fr hoch- und hchstleistungsrechnen ( hlrn ) in hannover and in the max - planck - institut fr sonnensystemforschung . the authors would like to specially thank the reviewers and editor for the much helpful comments and suggestions and also hlrn for providing the possibility of carrying out demanding `` last minute '' simulations that allowed improvement of the work . this work was supported by the special priority program 1488 ( planetmag , http://www.planetmag.de ) of the german science foundation . , u. r. , aubert , j. , jul . 2006 . scaling properties of convection - driven dynamos in rotating spherical shells and application to planetary magnetic fields . geophysical journal international 166 , 97114 . , p. a. , glatzmaier , g. a. , feb . compressible convection in a rotating spherical shell . i - anelastic equations . ii - a linear anelastic model . iii - analytic model for compressible vorticity waves . 45 , 335388 . , a. p. , beebe , r. f. , collins , s. a. , mitchell , j. l. , terrile , r. j. , hunt , g. e. , muller , p. , smith , b. a. , aug . 1979 . zonal velocity and texture in the jovian atmosphere inferred from voyager images . 280 , 773775 . , c. a. , kuzanyan , k. m. , mitchell , r. h. , aug . 2009 . linear theory of compressible convection in rapidly rotating spherical shells , using the anelastic approximation . journal of fluid mechanics 634 , 291 . & & & + 01d & 95 & 0.0 & 11.5 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath238 & @xmath239 & @xmath240 & 0.04 & 130 & @xmath241 & @xmath242 & @xmath243 + 01 m & 95 & 0.0 & 11.5 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath244 & @xmath245 & @xmath246 & 0.36 & 157 & @xmath247 & @xmath248 & @xmath249 + 02 & 95 & 0.0 & 17.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath250 & @xmath251 & @xmath252 & 0.37 & 236 & @xmath253 & @xmath254 & @xmath255 + 03 & 95 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath256 & @xmath257 & @xmath258 & 0.35 & 290 & @xmath259 & @xmath260 & @xmath261 + 04@xmath262 & 95 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath263 & @xmath264 & @xmath265 & 0.04 & 239 & @xmath266 & @xmath267 & @xmath268 + 05 & 90 & 0.0 & 9.2 & @xmath235 & @xmath269 & 61@xmath23764 & @xmath270 & @xmath271 & @xmath272 & 0.07 & 242 & @xmath273 & @xmath274 & @xmath275 + 06 & 90 & 0.0 & 10.3 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath276 & @xmath277 & @xmath278 & 0.57 & 130 & @xmath279 & @xmath280 & @xmath281 + 07 & 90 & 0.0 & 11.5 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath282 & @xmath283 & @xmath284 & 0.50 & 153 & @xmath285 & @xmath286 & @xmath287 + 08@xmath262 & 90 & 0.0 & 11.5 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath288 & @xmath289 & @xmath290 & 0.04 & 114 & @xmath291 & @xmath292 & @xmath287 + 09 & 90 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath293 & @xmath294 & @xmath295 & 0.38 & 270 & @xmath296 & @xmath297 & @xmath298 + 10@xmath262 & 90 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath299 & @xmath300 & @xmath301 & 0.10 & 241 & @xmath250 & @xmath302 & @xmath303 + 11 & 80 & 0.0 & 11.5 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath304 & @xmath305 & @xmath306 & 0.65 & 140 & @xmath307 & @xmath308 & @xmath309 + 12 & 80 & 0.0 & 11.5 & @xmath235 & @xmath310 & 73@xmath237106 & @xmath311 & @xmath312 & @xmath313 & 0.09 & 403 & @xmath256 & @xmath314 & @xmath268 + 13@xmath262 & 80 & 0.0 & 11.5 & @xmath235 & @xmath310 & 73@xmath237106 & @xmath315 & @xmath316 & @xmath290 & 0.03 & 395 & @xmath317 & @xmath318 & @xmath319 + 14 & 80 & 0.0 & 17.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath320 & @xmath321 & @xmath322 & 0.64 & 206 & @xmath323 & @xmath324 & @xmath261 + 15@xmath262 & 80 & 0.0 & 20.7 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath325 & @xmath326 & @xmath327 & 0.11 & 162 & @xmath328 & @xmath329 & @xmath298 + 16 & 80 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath330 & @xmath331 & @xmath279 & 0.62 & 276 & @xmath332 & @xmath333 & @xmath334 + 17@xmath262 & 80 & 0.0 & 23.0 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath335 & @xmath336 & @xmath337 & 0.09 & 177 & @xmath338 & @xmath339 & @xmath281 + 18 & 80 & 0.0 & 45.9 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath340 & @xmath341 & @xmath342 & 0.62 & 454 & @xmath343 & @xmath344 & @xmath345 + 19d & 95 & 1.0 & 4.1 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath346 & @xmath347 & @xmath348 & 0.05 & 90 & @xmath349 & @xmath350 & @xmath351 + 19 m & 95 & 1.0 & 4.1 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath352 & @xmath353 & @xmath354 & 0.21 & 103 & @xmath355 & @xmath356 & @xmath351 + 20d & 95 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath357 & @xmath358 & @xmath359 & 0.04 & 121 & @xmath360 & @xmath361 & @xmath362 + 20 m & 95 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath363 & @xmath364 & @xmath365 & 0.18 & 145 & @xmath366 & @xmath367 & @xmath368 + 21 & 95 & 1.0 & 7.8 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath369 & @xmath370 & @xmath371 & 0.21 & 234 & @xmath325 & @xmath372 & @xmath298 + 22d & 90 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath373 & @xmath374 & @xmath375 & 0.06 & 103 & @xmath376 & @xmath377 & @xmath255 + 22 m & 90 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath378 & @xmath379 & @xmath380 & 0.26 & 126 & @xmath381 & @xmath382 & @xmath261 + 23d@xmath383 & 90 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath357 & @xmath384 & @xmath385 & 0.04 & 110 & @xmath341 & @xmath386 & @xmath236 + 23m@xmath383 & 90 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath387 & @xmath388 & @xmath389 & 0.05 & 131 & @xmath390 & @xmath391 & @xmath392 + 24 & 90 & 1.0 & 7.8 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath393 & @xmath394 & @xmath395 & 0.30 & 204 & @xmath396 & @xmath397 & @xmath255 + 25 & 80 & 1.0 & 5.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath264 & @xmath398 & @xmath399 & 0.58 & 95 & @xmath400 & @xmath401 & @xmath402 + 26 & 80 & 1.0 & 10.3 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath403 & @xmath404 & @xmath405 & 0.65 & 224 & @xmath406 & @xmath407 & @xmath236 + 27 & 95 & 2.0 & 2.9 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath408 & @xmath360 & @xmath409 & 0.28 & 107 & @xmath410 & @xmath411 & @xmath368 + 28 & 95 & 3.0 & 3.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath412 & @xmath413 & @xmath414 & 0.29 & 158 & @xmath415 & @xmath416 & @xmath417 + 29 & 90 & 3.0 & 3.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath418 & @xmath419 & @xmath420 & 0.45 & 124 & @xmath421 & @xmath422 & @xmath368 + 30 & 80 & 3.0 & 3.2 & @xmath235 & @xmath236 & 73@xmath23785 & @xmath423 & @xmath424 & @xmath322 & 0.72 & 76 & @xmath425 & @xmath426 & @xmath427 + 31 & 80 & 3.0 & 3.2 & @xmath235 & @xmath428 & 121@xmath237106 & @xmath429 & @xmath430 & @xmath431 & 0.52 & 215 & @xmath432 & @xmath433 & @xmath345 + 32@xmath262 & 80 & 3.0 & 3.2 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath379 & @xmath434 & @xmath435 & 0.23 & 90 & @xmath436 & @xmath437 & @xmath281 + 33 & 80 & 3.0 & 4.3 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath438 & @xmath439 & @xmath440 & 0.67 & 134 & @xmath332 & @xmath441 & @xmath442 + 34@xmath443 & 80 & 3.0 & 4.3 & @xmath235 & @xmath236 & 121@xmath237106 & @xmath444 & @xmath445 & @xmath446 & 0.71 & 128 & @xmath447 & @xmath448 & @xmath362 + 35 & 80 & 3.0 & 4.3 & @xmath235 & @xmath428 & 145@xmath237106 & @xmath449 & @xmath317 & @xmath450 & 0.45 & 341 & @xmath451 & @xmath452 & @xmath453 + 36@xmath262 & 80 & 3.0 & 4.3 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath454 & @xmath455 & @xmath395 & 0.19 & 137 & @xmath331 & @xmath456 & @xmath268 + 37 & 80 & 3.0 & 8.6 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath457 & @xmath458 & @xmath381 & 0.55 & 258 & @xmath459 & @xmath460 & @xmath351 + 38 & 70 & 3.0 & 4.3 & @xmath235 & @xmath236 & 73@xmath237106 & @xmath461 & @xmath462 & @xmath317 & 0.77 & 90 & @xmath463 & @xmath464 & @xmath268 + 39 & 70 & 3.0 & 4.3 & @xmath235 & @xmath428 & 129@xmath237106 & @xmath465 & @xmath466 & @xmath400 & 0.71 & 241 & @xmath317 & @xmath467 & @xmath468 + 40 & 95 & 4.0 & 5.5 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath469 & @xmath470 & @xmath471 & 0.27 & 168 & @xmath472 & @xmath473 & @xmath319 + 41 & 80 & 4.0 & 4.4 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath474 & @xmath475 & @xmath476 & 0.71 & 69 & @xmath273 & @xmath477 & @xmath478 + 42 & 80 & 4.0 & 5.5 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath479 & @xmath480 & @xmath481 & 0.66 & 102 & @xmath482 & @xmath483 & @xmath484 + 43 & 80 & 4.0 & 5.5 & @xmath235 & @xmath485 & 97@xmath237170 & @xmath486 & @xmath487 & @xmath488 & 0.43 & 156 & @xmath447 & @xmath489 & @xmath298 + 44 & 80 & 4.0 & 8.8 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath490 & @xmath491 & @xmath432 & 0.61 & 184 & @xmath492 & @xmath493 & @xmath351 + 45 & 70 & 4.0 & 6.6 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath494 & @xmath495 & @xmath496 & 0.78 & 93 & @xmath497 & @xmath498 & @xmath281 + 46 & 70 & 4.0 & 6.6 & @xmath235 & @xmath485 & 81@xmath237170 & @xmath499 & @xmath500 & @xmath501 & 0.61 & 118 & @xmath502 & @xmath503 & @xmath298 + 47 & 70 & 4.0 & 8.8 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath504 & @xmath505 & @xmath506 & 0.79 & 129 & @xmath432 & @xmath507 & @xmath508 + 48 & 90 & 5.0 & 7.4 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath509 & @xmath510 & @xmath511 & 0.31 & 140 & @xmath512 & @xmath513 & @xmath298 + 49 & 90 & 5.0 & 9.3 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath514 & @xmath515 & @xmath516 & 0.28 & 181 & @xmath293 & @xmath517 & @xmath453 + 50@xmath518 & 90 & 5.0 & 9.3 & @xmath235 & @xmath519 & 81@xmath237192 & @xmath520 & @xmath521 & @xmath425 & 0.57 & 50 & @xmath338 & @xmath522 & @xmath298 + 51 & 80 & 5.0 & 7.4 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath523 & @xmath524 & @xmath525 & 0.38 & 69 & @xmath526 & @xmath527 & @xmath287 + 52d@xmath443 & 80 & 5.0 & 7.4 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath528 & @xmath529 & @xmath530 & 0.60 & 58 & @xmath531 & @xmath532 & @xmath345 + 52m@xmath443 & 80 & 5.0 & 7.4 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath533 & @xmath534 & @xmath291 & 0.72 & 100 & @xmath535 & @xmath536 & @xmath268 + 53d & 80 & 5.0 & 9.3 & @xmath235 & @xmath236 & 81@xmath237170 & @xmath537 & @xmath538 & @xmath539 & 0.37 & 94 & @xmath540 & @xmath541 & @xmath319 + 53 m & 80 & 5.0 & 9.3 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath542 & @xmath543 & @xmath544 & 0.56 & 111 & @xmath545 & @xmath546 & @xmath268 + 54 & 80 & 5.0 & 11.2 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath547 & @xmath548 & @xmath549 & 0.49 & 128 & @xmath550 & @xmath551 & @xmath453 + 55 & 70 & 5.0 & 11.2 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath552 & @xmath553 & @xmath369 & 0.72 & 95 & @xmath554 & @xmath555 & @xmath468 + 56 & 80 & 5.5 & 9.7 & @xmath235 & @xmath236 & 97@xmath237170 & @xmath556 & @xmath557 & @xmath558 & 0.30 & 76 & @xmath559 & @xmath560 & @xmath261 + 57 & 95 & 0.0 & 10.0 & @xmath169 & @xmath345 & 81@xmath237133 & @xmath561 & @xmath562 & @xmath563 & 0.06 & 98 & @xmath564 & @xmath565 & @xmath298 + 58 & 80 & 0.0 & 12.5 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath566 & @xmath567 & @xmath568 & 0.11 & 90 & @xmath569 & @xmath570 & @xmath345 + 59 & 80 & 0.0 & 20.8 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath571 & @xmath572 & @xmath573 & 0.17 & 138 & @xmath525 & @xmath574 & @xmath192 + 60 & 80 & 1.0 & 6.7 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath575 & @xmath576 & @xmath577 & 0.40 & 101 & @xmath578 & @xmath579 & @xmath298 + 61 & 80 & 1.0 & 10.0 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath580 & @xmath581 & @xmath582 & 0.41 & 158 & @xmath583 & @xmath584 & @xmath298 + 62 & 80 & 1.0 & 16.7 & @xmath169 & @xmath345 & 97@xmath237170 & @xmath585 & @xmath586 & @xmath587 & 0.60 & 329 & @xmath436 & @xmath588 & @xmath589 + 63 & 80 & 2.0 & 3.6 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath288 & @xmath590 & @xmath591 & 0.54 & 84 & @xmath592 & @xmath593 & @xmath468 + 64 & 80 & 2.0 & 5.4 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath594 & @xmath595 & @xmath596 & 0.73 & 151 & @xmath597 & @xmath598 & @xmath281 + 65 & 80 & 3.0 & 3.4 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath599 & @xmath600 & @xmath601 & 0.81 & 102 & @xmath602 & @xmath603 & @xmath281 + 66 & 80 & 3.0 & 4.5 & @xmath169 & @xmath345 & 81@xmath237170 & @xmath604 & @xmath247 & @xmath605 & 0.61 & 141 & @xmath549 & @xmath606 & @xmath345 + & & & + 67@xmath607 & 80 & 0.0 & 7.2 & @xmath169 & @xmath478 & 121@xmath237170 & @xmath608 & @xmath609 & @xmath610 & 0.50 & 306 & @xmath611 & @xmath612 & @xmath613 +

we present a new derivation of the casimir force between two parallel plane mirrors at zero temperature . the two mirrors and the cavity they enclose are treated as quantum optical networks . they are in general lossy and characterized by frequency dependent reflection amplitudes . the additional fluctuations accompanying losses are deduced from expressions of the optical theorem . a general proof is given for the theorem relating the spectral density inside the cavity to the reflection amplitudes seen by the inner fields . this density determines the vacuum radiation pressure and , therefore , the casimir force . the force is obtained as an integral over the real frequencies , including the contribution of evanescent waves besides that of ordinary waves , and , then , as an integral over imaginary frequencies . the demonstration relies only on general properties obeyed by real mirrors which also enforce general constraints for the variation of the casimir force .

quant-ph0210174

an important prediction of quantum theory is the existence of irreducible fluctuations of electromagnetic fields in vacuum . besides their numerous observable consequences in microscopic physics , vacuum fluctuations also have observable effects in macroscopic physics , for example the casimir force they exert on mirrors @xcite . casimir calculated this force in a geometrical configuration where two plane mirrors are placed a distance @xmath0 apart and parallel to each other , the area @xmath1 of the mirrors being much larger than the squared distance @xmath2 . he considered the ideal case of perfectly reflecting mirrors and obtained an expression which , remarkably , depends only on the geometrical quantities @xmath1 and @xmath0 and on the fundamental constants @xmath3 and @xmath4 @xmath5 this attractive force has been observed in a number of ` historical ' experiments @xcite which confirmed its existence and main properties @xcite . several recent experiments reached an accuracy in the % range by measuring the force between a plane and a sphere @xcite or two cylinders @xcite . similar experiments were also performed with mems @xcite ( see also @xcite ) . an experiment studied the plane - plane configuration considered by casimir @xcite but , as a consequence of the difficulties associated with this geometry , reached only a 15% accuracy ( see reviews of recent experiments in @xcite ) . the casimir force is the most accessible experimental consequence of vacuum fluctuations in the macroscopic world while vacuum energy is known to raise a serious problem with respect to gravity and cosmology ( see references in @xcite ) . this is a reason for testing the predictions of quantum field theory concerning the casimir effect with the greatest care and accuracy . the theory of the casimir force is also a key point for the experiments searching for the new weak forces predicted by theoretical unification models to arise at distances between nanometer and millimeter @xcite . the casimir force is indeed the dominant effect between two neutral objects at @xmath6 m or sub@xmath6 m distances so that an accurate knowledge of its theoretical expectation is as crucial as the precision of measurements in such experiments @xcite . in this context , it is essential to account for the differences between the ideal case considered by casimir and the real experimental situation . recent experiments use metallic mirrors which show perfect reflection only at frequencies below their plasma frequency . they are performed at room temperature , with the effect of thermal fluctuations superimposed to that of vacuum fluctuations . in the most accurate experiments , the force is measured between a plane and a sphere , and not between two parallel planes . the surface state of the plates , in particular their roughness , should also affect the force . a large number of works have been devoted to the study of these effects and we refer the reader to @xcite for a bibliography . the evaluation of the casimir force between imperfect lossy mirrors at non zero temperature has given rise to a burst of controversial results @xcite which constitutes a part of the motivations for the present work . for the sake of comparing experimental measurements and theoretical expectations , it is necessary to have at one s disposal a reliable expression of the casimir force in the experimental situation . in the present paper , we focus our attention on the effect of imperfect reflection of the mirrors . other effects , in particular the effect of temperature , will be addressed in follow - on papers . we consider the original casimir geometry with two perfectly plane and parallel mirrors . except for these assumptions , we consider arbitrary frequency dependences for the mirrors which , in particular , may be lossy . we evaluate the casimir force as the effect of vacuum radiation pressure on the fabry - perot cavity formed by the two mirrors . the net force results from the balance between the repulsive and attractive contributions associated respectively with resonant or antiresonant frequencies . it is obtained as an integral over the axis of real frequencies , including the contribution of evanescent waves besides that of ordinary waves . it is then transformed into an integral over imaginary frequencies by using physical properties fulfilled by all real mirrors . the formula obtained here for the casimir force turns out to be identical to the expression already published in @xcite but the new derivation has a wider scope of validity than the previous one since it remains valid for lossy mirrors . the fact that the formula keeps the same form despite the widening of the assumptions is intimately related to a theorem which relates the spectral density of the fields inside the cavity to the reflection amplitudes seen by the same fields . this theorem was demonstrated in @xcite and @xcite in specific cases and we prove it in the present paper without any restriction . to this aim , we introduce a systematic treatment of lossy mirrors and cavities as dissipative networks @xcite . we define scattering and transfer matrices for elementary networks like the interface between two media or the propagation over a given length in a medium . we then deduce the matrices associated with composed networks , like the optical slab or the multilayer mirror . the results obtained in this manner are therefore applicable to a large variety of mirrors , still with the assumption of perfect plane geometry . in the particular case of a slab with a large width , the lifshitz expression @xcite is recovered . at the limit of perfectly reflectors , the ideal casimir formula ( [ eqcasimir ] ) is obtained . more generally , the expression gives the casimir force as an integral written in terms of the reflection amplitudes characterizing the two mirrors . this integral is finite as soon as the amplitudes obey the general properties of scattering theory already alluded to . in other words , the difficulties usually associated with the infiniteness of vacuum energy are solved by using the properties of real mirrors themselves rather than through an additional formal regularization technique . we finally show that the same physical properties constrain the variation of the casimir force . in particular , they invalidate proposals which has been done for ` tayloring ' the force at will by using mirrors with specially designed scattering amplitudes @xcite . in these proposals , the balance between attractive and repulsive contributions to the force is change , leading to the hope that the casimir force could reach large or have its sign changed from an attractive force to a repulsive one @xcite . using the simple model of a one - dimensional space , it has already been shown @xcite that these hopes can not be met for arbitrary mirrors built up with dielectric layers . here , the argument is generalized to the casimir geometry in three - dimensional space with the following conclusions : the casimir force can not exceed the value obtained for perfect mirrors , it remains attractive for any cavity length and its value is a decreasing function of the cavity length . this is true for any mirror obtained by piling up layers of media described by dielectric functions . this definition of multilayer dielectric mirrors includes the case of metallic layers , provided that magnetic effects play a negligible role in the optical response . as explained in the introduction , we consider in this paper the original casimir geometry with perfectly plane and parallel mirrors aligned along the directions @xmath7 and @xmath8 . this configuration obeys a symmetry with respect to time translation as well as transverse space translations along these directions . we use bold letters for two - dimensional vectors along these directions and denote @xmath9 the transverse position . as a consequence of this symmetry , the frequency @xmath10 , the transverse vector @xmath11 and the polarization @xmath12 are preserved throughout the scattering processes on a mirror or a cavity . the scattering couples only the free vacuum modes which have the same values for the preserved quantum numbers and differ by the sign of the longitudinal component @xmath13 of the wavevector . in the present section , we introduce notations for the vacuum field modes , first in empty space and then in a dielectric medium . these notations are chosen to be well - adapted to the symmetry of the problem . in empty space , the components of the wavevector are given for each field mode by the frequency @xmath14 , the incidence angle @xmath15 and the azimuthal angle @xmath16 @xmath17 @xmath18 is the modulus of the transverse wavevector and the longitudinal component @xmath13 may be expressed in terms of the preserved quantities @xmath14 and @xmath19 @xmath20 @xmath21 is defined as the sign of @xmath22 and represents the direction of propagation with @xmath23 and @xmath24 corresponding respectively to rightward and leftward propagation . the two polarizations @xmath12 are defined by the transversality with the incidence plane of electric and magnetic fields respectively . they are given by the unit electric vectors @xmath25 @xmath26 or , equivalently , the unit magnetic vectors @xmath27 and @xmath28 . for each mode , the wavevector and polarization vectors form an orthogonal spatial basis . we have chosen linear polarizations described by real components ; hence the unit vectors @xmath29 and @xmath30 are not affected by the complex conjugation appearing below in the relation between positive and negative frequencies . the two modes corresponding to the same values of @xmath14 , @xmath19 and @xmath31 but opposite values of @xmath21 are coupled by scattering on a mirror . for this reason , we introduce a label @xmath32 gathering the values of @xmath10 , @xmath19 and @xmath31 . a mode freely propagating in vacuum is thus labeled by @xmath33 and @xmath34 and the summation over modes is described by the symbols @xmath35 note that @xmath34 appears implicitly as the sign of @xmath13 in the first form whereas it appears explicitly in the second one . the free vacuum fields are then written as linear superpositions of modes @xmath36 the vacuum impedance @xmath37 describes the electromagnetic constants in vacuum . in the following , the symbol @xmath38 will be reserved to the relative permittivity with the value 1 in vacuum . the quantum field amplitudes @xmath39 and @xmath40 correspond to positive and negative frequency components . they fit the definition of annihilation and creation operators of quantum field theory and obey the canonical commutation relations @xcite @xmath41 & = & \left ( 2\pi \right ) ^{3}\delta ^{(2)}\left ( \mathbf{k - k}^{\prime } \right ) \delta \left ( k_{z}-k_{z}^\prime \right ) \delta _ { pp^{\prime } } \delta _ { \phi \phi ^\prime } \nonumber \\ & \equiv & \delta _ { mm^\prime } \delta _ { \phi \phi ^{\prime } } \nonumber \\ \left [ e_{m^\prime } ^{\phi ^\prime } , e_{m}^{\phi}\right ] & = & \left [ e_{m^\prime } ^{\phi ^\prime } { } ^{\dagger},e_{m}^{\phi}{}^\dagger \right ] = 0 \label{commut}\end{aligned}\ ] ] in the vacuum state , the anticommutators of quantum amplitudes are derived from the corresponding commutators @xmath42 = \frac{1}{2 } \delta _ { mm^\prime } \delta _ { \phi \phi ^{\prime } } \nonumber \\ \left\langle e_{m^\prime } ^{\phi ^\prime } \cdot e_{m}^{\phi } \right\rangle _ { \mathrm{vac } } & = & \frac{1}{2 } \left [ e_{m^\prime } ^{\phi ^\prime } , e_{m}^{\phi } { } \right ] = 0 \label{anticommvacuum}\end{aligned}\ ] ] the dot symbol represents a symmetrized product . the energy density per unit volume @xmath43 is a quadratic form of the fields @xmath44 and @xmath45 @xmath46 when subsituting the expression of free fields , @xmath43 is obtained as a bilinear form of the field amplitudes . here , we study the averaged radiation pressure in the vacuum state which leads to a contraction @xmath47 in the sums over modes . using the vacuum property ( [ anticommvacuum ] ) , we find the averaged energy density in vacuum equal to the sum over the modes of @xmath48 @xmath49 as it is well - known , this energy density is infinite . the radiation pressure on plane mirrors oriented along @xmath50 directions is determined by the component @xmath51 of the maxwell stress tensor @xmath52 here , the dot symbol represents a symmetrized product of the quantum amplitudes and , simultaneously , a scalar product of the vectors ; the overline symbol describes the mathematical reflexion of a vector with respect to the plane @xmath50 @xmath53 as for @xmath43 , averaging @xmath51 in vacuum state leads to a contraction over the modes with the result @xmath54 this expression is similar to the expression ( [ t00vac ] ) of the energy density with an extra factor @xmath55 well - known in studies of radiation pressure . the sum over modes is still infinite but this infiniteness problem will be solved in the forthcoming calculation of the casimir force . in the following , we consider mirrors built up as dielectric multilayers . each dielectric medium is characterized by a relative permittivity @xmath56 $ ] or , equivalently , an index of refraction @xmath57 = \sqrt{\varepsilon \left [ \omega \right]}$ ] depending on frequency . the magnetic permeability is kept equal to its vacuum value since this corresponds to all experimental situations studied so far . we stress again that this definition of dielectric mirrors includes the case of metals as long as the magnetic response plays a negligible role . we consider layers thick enough so that the dielectric response is local , i.e. described by a wavevector - independent permittivity @xmath56 $ ] . we will sometimes take the plasma model as a first description of metallic optical response @xmath58 = 1 - \frac{\omega_\mathrm{p}^2}{\omega^2 } \qquad \omega_\mathrm{p } = \frac{2\pi c}{\lambda_\mathrm{p } } \label{plasmamodel}\ ] ] where @xmath59 and @xmath60 represent respectively the plasma frequency and the plasma wavelength . this simple model is not sufficient for an accurate evaluation of the casimir force between real mirrors @xcite . to this aim , it is necessary to describe the optical response of metals with a dissipative part associated with electronic relaxation processes . as a consequence of causality , the real and imaginary parts of @xmath61 are related to each other through the kramers - kronig dispersion relations @xcite . for any function of frequency more generally , causality is unambiguously characterized in terms of analyticity properties : @xmath57 $ ] or @xmath56 $ ] are analytical functions of @xmath10 in the ` physical domain ' of the complex frequency plane , that is the domain of frequencies @xmath14 with a positive imaginary part @xmath62 . this property is obeyed by other response functions to be encountered below and it will play an important role in the derivation of the casimir force . we will introduce an equivalent notation @xmath63 for complex frequencies with the physical domain now defined by a positive real part for @xmath63 @xmath64 the dispersion relation ( [ dispersionrelation ] ) is changed inside a refractive medium to @xmath65 \frac{\omega } { c } \cos \theta \label{dispersionrelationindex}\end{aligned}\ ] ] the preservation of @xmath14 and @xmath19 at the traversal of an interface is equivalent to the snell - descartes law of refraction . the sign has to be carefully chosen when extracting the square root to express @xmath13 in terms of the conserved quantities @xmath10 and @xmath19 . as soon as the refractive index contains an imaginary part , this is also the case for @xmath13 and the dephasing @xmath66 associated with propagation includes an extinction factor . in order to ensure that this factor is effectively a decreasing exponential , we have to choose a specific root defined differently for the two propagation directions @xmath67 of the field @xmath68 \frac{\xi ^{2}}{c^{2}}+\mathbf{k}^{2 } } \qquad \re\kappa > 0 \label{dispersionkappa}\end{aligned}\ ] ] the argument has been presented for freely propagating modes but it holds as well for evanescent waves confined to the vicinity of an interface between two media . in this case , the sign of @xmath13 is also chosen so that it corresponds to an extinction when the distance to the interface increases and this choice is still described by equation ( [ dispersionkappa ] ) . in the following , we will use systematically the notations @xmath63 and @xmath69 , keeping in mind that the causality relations have to be written for each value of the conserved quantity @xmath19 . besides the dispersion relation ( [ dispersionrelationindex ] ) , the dielectric medium also changes the impedance , that is the ratio between magnetic and electric field amplitudes . precisely , the impedance is changed from the value @xmath70 in empty space to the value @xmath71 in a dielectric medium of index @xmath61 , resulting in reflection at the interface . we now introduce the description of mirrors as optical networks . we present the scattering and transfer representations and the relations between them . the transfer approach is well adapted to the composition of networks which are piled up . we first consider elementary networks such as an interface or propagation inside a refractive medium . we then use the composition law to study composed networks such as the slab and multilayer . in the present section , we only consider classical fields or , equivalently , mean quantum fields . the next section will be devoted to the full quantum treatment including the addition of noise associated with the losses inside the mirror . we first introduce the scattering and transfer representations for an arbitrary network represented with two ports and four fields . these fields are identified as lefthand / righthand ( symbols ` l ' and ` r ' ) , rightward / leftward ( arrows @xmath72 and @xmath73 ) or input / output fields ( labels ` in ' and ` out ' ) , as shown on figure [ fignetwork ] . let us emphasize that the arrows are a symbolic representation of the two modes coupled by the network which correspond to the same label @xmath33 and to the two opposite signs @xmath74 . the geometrical directions of propagation are given by the wavevectors of equation ( [ dispersionrelationindex ] ) . the coupling between the fields is described by reflection and transmission amplitudes represented below by scattering or transfer matrices . in the scattering point of view , we gather the input and output fields in twofold columns related by a @xmath75matrix @xmath76 @xmath77 and @xmath78 are the reflection amplitudes while @xmath79 and @xmath80 are the transmission amplitudes . we will also use an equivalent convention where the output ket is defined with the upper and lower components exchanged @xmath81 this convention simplifies some algebraic manipulations while being completely equivalent to the former convention . for comparison with previous works , note that the former notation ( [ smatrix ] ) was used in @xcite whereas the latter one ( [ smatrixtilde ] ) was used in @xcite . in the transfer point of view , the network is described by lefthand and righthand columns related by a @xmath82matrix @xmath83 the matrix @xmath84 introduced in ( [ smatrixtilde ] ) exchanges the two directions of propagation . we also use in the following the matrices @xmath85 which project onto each direction @xmath86 these matrices obey simple rules which define an algebraic calculus in the space @xmath87 of @xmath88 matrices with complex coefficients @xmath89 the identification of figure ( [ fignetwork ] ) is written as @xmath90 it relates the transfer and scattering amplitudes . we decompose the scattering equations ( [ smatrix ] ) on the two components and use ( [ identify ] ) to rewrite them as @xmath91 this linear system may be put under a matrix form @xmath92 it is equivalent to the transfer equation ( [ tmatrix ] ) with the @xmath82matrix obtained as @xmath93 the converse transformation is obtained by performing the same manipulations in the reverse order . starting from the transfer equation ( [ tmatrix ] ) and using ( [ identify ] ) , one obtains a linear system which is equivalent to the scattering equation ( [ smatrixtilde ] ) with @xmath94 the relations ( [ stot ] ) and ( [ ttos ] ) have the same form . they represent an idempotent homographic transformation in the space @xmath95 , care being taken for the non - commutativity of multiplications in this space . when inverting algebraically the homographic relations ( [ stot ] ) and ( [ ttos ] ) , one obtains equivalent expressions @xmath96 other equivalent expressions are obtained from the equalities @xmath97 all these expressions may be written in terms of the scattering and transfer amplitudes @xmath98 the more formal homographic transformations written above are nevertheless useful , as it will become clear in forthcoming calculations . the @xmath82matrices are perfectly adapted to the composition of optical networks corresponding to a piling up process ( see figure [ figcomposition ] ) . on each network , the transfer equations are written as @xmath99 the brackets @xmath100 specify the network for which the @xmath82matrix or field column is written . identifying the fields according to figure ( [ figcomposition ] ) @xmath101 we deduce that the piling up process is equivalent to the product of @xmath102-matrices @xmath103 we have assumed the two networks to be in the immediate vicinity of each other but without any electronic exchange between them , which again corresponds to the assumption of thick enough layers . we now study two elementary networks , that is the traversal of an interface and the propagation over a given length inside a dielectric medium . for the scattering at the plane interface between two media with indices @xmath104 and @xmath105 , we write the reflection and transmission amplitudes as the fresnel scattering amplitudes @xcite . reflection amplitudes @xmath106 are obtained from characteristic impedances @xmath107 defined for plane waves with polarization @xmath31 in each medium and from the continuity equations at the interface @xmath108 then the transmission amplitudes are obtained as @xmath109 we deduce the expression of the transfer matrix @xmath110 we now consider the process of field propagation over a propagation length @xmath111 inside a dielectric medium characterized by a permittivity @xmath112 . for this elementary network , the @xmath82matrix has the simple form @xmath113 the optical depth @xmath114 does not depend on the polarization . note that the composition is commutative within the class of interfaces or that of propagations : it corresponds to the multiplication of the @xmath115parameters for interfaces and to the addition of @xmath116parameters for propagations . but the composition is no longer commutative when interfaces and propagations are piled up . we now prove a reciprocity theorem obeyed by arbitrary dielectric multilayers , _ i.e. _ networks obtained by piling up interfaces and propagations . to this aim , we first remark that the ratio of the two transmission amplitudes is related to the determinant of the @xmath82matrix @xmath117 this follows from the relations ( [ sandtamplitudes ] ) between @xmath75 and @xmath82amplitudes for an arbitrary network . then , it is clear from ( [ compot ] ) that the determinant of @xmath102 is simply multiplied under composition @xmath118 for the two kinds of elementary networks studied previously ( see eqs [ tint]-[tprop ] ) , the determinant of @xmath102 is the ratio of the values of @xmath69 at the right and left sides of the network @xmath119 it follows that this relation is valid for any optical network composed by piling up interfaces and propagations . in the particular case where the network has its two ports corresponding to vacuum , which is the case for a mirror , the values of @xmath69 are equal on its two sides and the @xmath82matrix has a unit determinant @xmath120 note that reciprocity corresponds to a symmetrical @xmath75matrix and has to be distinguished from the spatial symmetry of the network with respect to its mediane plane which entails @xmath121 . this theorem is the specific form , when the symmetry of plane mirrors is assumed , of the general reciprocity theorem demonstrated by casimir @xcite as an extension to electromagnetism of onsager s microreversibility theorem @xcite . we have disregarded any static magnetic field which could affect these reciprocity relations . we now consider the dielectric slabs and multilayers as composed networks and we deduce their transfer and scattering amplitudes from the preceding results . the slab is obtained by piling up a vacuum / matter interface with indices @xmath122 and @xmath123 at its left and righthand sides , propagation over a length @xmath124 inside matter , and a matter / vacuum interface with now @xmath123 and @xmath122 at its left and righthand sides . we denote @xmath125 the @xmath82matrix associated with the first interface and obtain the @xmath82matrix associated with the second interface as the inverse of @xmath125 . as a consequence of the composition law ( [ compot ] ) , the @xmath82matrix associated with the slab is obtained as @xmath126 using the expressions ( [ tint],[tprop ] ) of @xmath127 and @xmath128 , we evaluate @xmath129 as @xmath130 we deduce the form of the @xmath75matrix which is simultaneously reciprocal ( @xmath131 ) and symmetrical in the exchange of its two ports ( @xmath121 ) @xmath132 in the limiting case of a small thickness @xmath133 , we find @xmath134 and @xmath135 , which means that the slab tends to become transparent . in this case indeed , the propagation can be forgotten and the two inverse interfaces have their effects cancelled by each other . the opposite limiting case of a large thickness is often considered since it fits the usual experimental situations . more precisely , experiments are performed with metallic mirrors having a thickness much larger than the plasma wavelength . this is why the limit of a total extinction of the field through the medium is assumed in most calculations . this corresponds to the so - called ` bulk limit ' with @xmath136 and @xmath137 in eq.([slab ] ) : the reflection amplitude is determined entirely by the first interface . let us emphasize however that the bulk limit raises several delicate problems . first , the transmission amplitude @xmath138 vanishes in this limit so that the @xmath82matrix is not defined , with the drawback of invalidating the general method used in the present paper . then , the bulk limit can not be met in the case of non absorbing media where @xmath139 remains a complex number with unit modulus for any value of @xmath124 . even in the presence of absorption , a large value of the width @xmath124 does not necessarily imply a large value of the optical thickness @xmath114 since @xmath69 may go to zero at normal incidence and zero frequency , leading to a transparent slab in contrast with the results of the bulk limit . therefore a reliable calculation must consider the experimental situation of mirrors with a large but finite thickness . in the present paper , we consider the general case of arbitrary mirrors and test the reliability of the bulk limit in the end of the calculations . we can deal with the case of dielectric multilayers similarly . if we consider as an example the multilayer obtained by piling up a vacuum / matter interface with indices @xmath122 and @xmath123 at its left and righthand sides , propagation over a length @xmath140 inside the medium 1 , an interface between media 1 and 2 , propagation over a length @xmath141 inside the medium 2 , and an interface between medium 2 and vacuum , its @xmath82matrix is obtained as the product @xmath142 alternatively , the same multilayer may be obtained by piling up two slabs each corresponding to one of the layers @xmath143 in the last two equations , the indices specify the different interfaces , propagations or slabs using an obvious convention . since any multilayer mirror is obtained by piling up slabs connecting two vacuum ports and thus obeying the reciprocity relation @xmath131 , we can use a simple form of the composition law written in terms of scattering amplitudes @xcite @xmath144 for readibility , we have specified the networks by using subscripts rather than brackets . we will proceed similarly in forthcoming specific computations . iterating this composition law , we can compute the scattering amplitudes for any dielectric multilayer . this systematic technique is quite similar to the classical computation techniques used for studying multilayers @xcite . it is generalized to the full quantum treatment in the next section . it also leads in the following to general results constraining the variation of the casimir force for arbitrary dielectric mirrors . it reproduces the known results for the multilayer systems which have already been studied @xcite . up to now , we have performed a classical analysis which is not sufficient for the purpose of describing the scattering of vacuum fluctuations . real mirrors consist of absorbing media which scatter incident fields to spontaneous emission modes and reciprocally scatter fluctuations from noise modes to the modes of interest . the @xmath75matrix calculated previously can not be unitary for a lossy mirror but it should be the restriction to the modes of interest of a larger @xmath75matrix which includes the noise modes and obeys unitarity . in the present section , we characterize the additional fluctuations for a lossy mirror by using the corresponding ` optical theorem ' , that is also the unitarity of the larger @xmath75matrix ( see @xcite and references therein ) . we assume that the scattering restricted to the modes of interest still fulfills the symmetry of plane mirrors considered in the previous classical calculations . this amounts to neglect multiple scattering processes which could couple different modes through their coupling with noise modes . except for this assumption , we consider arbitrary dissipative media and discuss the optical theorem in the scattering and transfer points of view . we use the latter one to deal with composition of additional fluctuations when lossy mirrors are piled up . should we use the previous classical equations for the quantum amplitudes , we would find that the output fields can not obey the canonical commutators , except in the particular case of lossless mirrors . this implies that the input / output transformation for quantum field must include additional fluctuations superimposed to the classical equations @xmath145 @xmath146 and @xmath147 are defined as in ( [ smatrix ] ) with the quantum amplitudes @xmath148 in place of the classical fields @xmath149 , @xmath150 is the same matrix as previously and @xmath151 is a twofold column matrix describing the additional fluctuations . all these quantities depend on the quantum number @xmath33 which is common to all fields coupled in the scattering process . the additional fluctuations are linear superpositions of all modes coupled to the main modes @xmath152 by the microscopic couplings which cause absorption . as an example , the atoms constituting a dielectric medium couple the main modes to all electromagnetic modes through spontaneous emission processes , represented symbolically by the wavy arrows on figure [ figaddfluct ] . the stationarity assumption implies that only modes having the same frequencies are coupled . in particular , it forbids parametric couplings which could couple modes with different frequencies and ` squeeze ' the vacuum fluctuations @xcite . the whole scattering matrix which takes into account all coupled field modes is unitary and this basic property makes the canonical commutation relations compatible for input and output fields . in contrast , the reduced scattering matrix containing only the classical scattering amplitudes coupling the main modes @xmath152 is not unitary , except in the particular case of lossless mirrors . in order to write the unitarity property of the whole scattering matrix , it is convenient to represent the additional fluctuations @xmath151 by introducing auxiliary noise modes @xmath153 and auxiliary noise amplitudes gathered in a noise matrix @xmath154 @xmath155 the components of the twofold column @xmath156 are defined to have the same canonical commutators as the input fields in the main modes . in fact , they are linear superpositions of the input vacuum modes responsible for the fluctuation process . they are defined up to an ambiguity : any canonical transformation of the noise modes leads to an equivalent representation of the additional fluctuations , which corresponds to a different form for the noise amplitudes while leading to the same physical results at the end of the computations . for any of these equivalent representations , the norm matrix @xmath157 has the same expression determined by the optical theorem , that is the unitarity condition for the whole scattering process , @xmath158 where @xmath159 is the @xmath88 unity matrix . this is easily proven by a direct inspection of the explicit expressions of the commutators of the output fields . the same inspection shows that noise modes corresponding to different values of @xmath33 are not correlated to each other . condition ( [ unitarity ] ) is made more explicit when @xmath160 and @xmath161 are developed in terms of scattering amplitudes @xmath162 more detailed discussions are presented for the case of the slab in appendix [ appslab ] . the description of noise may as well be represented with the alternative representation ( [ smatrixtilde ] ) of the scattering process @xmath163 the additional fluctuations are then represented in terms of the same noise modes and of a modified noise matrix @xmath164 we now present the description of additional fluctuations in the transfer approach . performing the same manipulations as in the previous section , we transform equation ( [ smatrixtildenoise ] ) into @xmath165 we thus get transfer equations with additional fluctuations described by a twofold column @xmath166 @xmath167 the @xmath82 matrix has the same expression ( [ stot ] ) as previously and the additional fluctuations @xmath168 are a linear expression of the fluctuations @xmath169 defined in the scattering approach . this linear relation may be written under alternative forms by using the relations ( [ sandtsym ] ) @xmath170 in the scattering approach , the norm of additional fluctuations is described by matrices @xmath161 and @xmath171 which are themselves determined by the optical theorem ( [ unitarity ] ) or ( [ unitaritytilde ] ) . in order to translate these properties to the transfer approach , we rewrite ( [ ftog ] ) in terms of the canonical noise modes @xmath156 and of noise amplitudes gathered in a matrix @xmath172 @xmath173 the associated norm matrix is @xmath174 using equations ( [ unitaritytilde ] ) and ( [ ttos ] ) , we rewrite it as @xmath175 @xmath176 is a diagonal matrix with two eigenvalues representing the directions of propagation @xmath67 of the field . using these tools , we now write composition laws for the fluctuations and their norms . we start from transfer equations written for each network a and b @xmath177 using the identifications ( [ identifycompo ] ) associated with the composition law , we deduce for the composed network @xmath178 the fluctuations @xmath179 are a linear superposition of fluctuations @xmath180 and @xmath181 added in a and b. in order to obtain the composition law for the norm matrices , we develop the additional fluctuations @xmath179 on the canonical noise modes associated with the two elements @xmath182 since the noise modes associated with different elements are uncorrelated , @xmath179 may be rewritten in terms of new canonical noise modes and new noise amplitudes such that @xmath183 using expression ( [ unitaryt ] ) of the optical theorem for both networks a and b , we deduce that the composed network ab obeys the same relation @xmath184 equivalently , the @xmath75matrix of the composed network ab obeys the optical theorem ( [ unitarity ] ) as soon as the two networks a and b do . we have studied the scattering or , equivalently , the lefthand / righthand transfer of fields by a composed network ab . we want now to characterize the properties of the fields inside the cavity formed between a and b. this problem will play a key role in the evaluation of the casimir force ( see next section ) . the situation is illustrated by figure [ figcavity ] which , in contrast to figure [ figcomposition ] , keeps the trace of the intracavity fields . in algebraic terms , the cavity fields are defined by rewriting the identifications ( [ identifycompo ] ) as @xmath185 from now on , we drop the label @xmath186 for the composed network and use subscripts for the networks a and b. in order to express the cavity fields in terms of the input modes and additional fluctuations , we first write the cavity fields @xmath187 in terms of the righthand ones @xmath188 @xmath189 we then identify the two components of @xmath188 as @xmath190 using the expression of @xmath191 in terms of @xmath166 and the composition law ( [ compog ] ) for @xmath166 , we deduce @xmath192 as already explained , the unitarity of scattering entails that the output fields have the same commutators as the input ones . but this is not the case for the cavity fields which have their commutators determined by the matrix @xmath193 expanding this quadratic form and using the composition law ( [ componoiset ] ) , we rewrite @xmath194 as @xmath195 using relation ( [ unitaryt ] ) for the three networks a , b and ab , we obtain a simpler expression after a few rearrangements @xmath196 we now proceed to explicit calculations of these matrices . we note that @xmath197 where @xmath79 is the transmission amplitude of the network ab and deduce @xmath198 @xmath79 is simply the inverse of the transfer amplitude @xmath199 associated with the network ab ( see eq.[sandtamplitudes ] ) and the latter is deduced from the composition law ( [ compot ] ) @xmath200 then , the transfer amplitudes of the networks a and b may be substituted by the associated scattering amplitudes , leading to @xmath201 collecting these results and proceeding to slight rearrangements , we finally get @xmath202 in the following we will use the diagonal terms of the matrix @xmath194 to evaluate the casimir force . in order to prepare the evaluation of the casimir force , we generalize the preceding expression to the case of the fabry - perot cavity containing a zone of field propagation between the two mirrors m1 and m2 ( see figure [ figfabryperot ] ) . the distance between the two mirrors is denoted @xmath0 and the cavity fields are defined at an arbitrary position inside the cavity , say at distances @xmath203 from m1 and @xmath204 from m2 with @xmath205 . in these conditions , the study of the fabry - perot cavity is reduced to the problem studied in the preceding subsection through the following identifications : the network a contains the mirror m1 and the propagation l1 with @xmath206 while the network b contains the propagation l2 and the mirror m2 with @xmath207 . the transfer amplitudes for the networks a and b are derived from those corresponding to m1 and m2 and from phase factors corresponding to the propagations l1 and l2 @xmath208 we have labeled the amplitudes for the mirrors m1 and m2 with mere indices 1 and 2 ; @xmath209 is defined in vacuum . these results entail that the reflection amplitudes @xmath210 and @xmath211 are seen from a point inside the cavity as the product of phase factors by the reflection amplitudes @xmath212 and @xmath213 seen from a point in the immediate vicinity of m1 and m2 . we then deduce the scattering amplitudes for the whole cavity @xmath214 and the expression of @xmath194 @xmath215 the diagonal terms in the matrix @xmath194 coincide with the airy function @xmath216 this result will play the central role in the derivation of the casimir force in the next section . it means that the commutators of the intracavity fields are not the same as those of the input or output fields . they correspond to a spectral density modified through a multiplication by the airy function @xmath217 . this is the basic property used in cavity quantum electrodynamics @xcite . it is clear from the present derivation that this result has a quite general status : it is obtained for any inner field in any composed network , assuming the symmetry of plane mirrors . this property was already known for non absorbing mirrors @xcite and for lossy mirrors symmetrical with respect to their mediane plane @xcite . the present derivation proves that it is also valid for arbitrary dielectric multilayers with dissipation . the final result only depends on the reflection amplitudes @xmath218 and @xmath219 of the mirrors as they are seen from the inner side of the cavity . the reflection amplitudes seen from the outer side and the transmission amplitudes do not appear in expressions ( [ gmatrix],[gfunction ] ) . this can be interpreted as resulting from the unitarity of the whole scattering processes . we may now deal with the radiation pressure of vacuum fields on the mirrors of a fabry - perot cavity . we show that the resulting casimir force is a regular integral which can be written over real or imaginary frequencies . we then derive general constraints obeyed by the casimir force for arbitrary dielectric mirrors . if we first consider a mirror isolated in vacuum , the radiation pressure is obtained by adding the contributions of the 4 fields coupled in the scattering process @xmath220 the identification of these fields is given by figure ( [ fignetwork ] ) . we have developed the sum over @xmath34 and kept the symbol @xmath33 to represent the quantum numbers @xmath221 . we assume that the whole system is in vacuum , that is at zero temperature , so that the anticommutators of input fields are given by relation ( [ anticommvacuum ] ) . since the commutators are the same for the output and input fields , the vacuum radiation pressure vanishes in the case of an isolated mirror . in other words , the two sides of the mirror play equivalent roles so that no mean force can appear . when we consider two mirrors forming a fabry - perot cavity , the two sides of a given mirror are no longer equivalent since one is an inner side and the other an outer side . it follows that the compensation observed for an isolated mirror does no longer hold , resulting in the appearance of the casimir force . in order to evaluate the force , we write the mean radiation pressures @xmath222 and @xmath223 on mirrors m1 and m2 ( see figure [ figfabryperot ] ) @xmath224 for the same reasons as previously , the field anticommutators are given by ( [ anticommvacuum ] ) for input and output fields . for intracavity fields , they are multiplied by the airy function ( [ gfunction ] ) like the commutators @xmath225 \nonumber \\ & = & \frac{1}{2}g_{m } \delta _ { mm^\prime } \delta _ { \phi \phi ^\prime } \label{anticommcav}\end{aligned}\ ] ] as shown in the previous section , these expressions do not depend on the position inside the cavity where the cavity fields are defined . we finally deduce the mean radiation pressures on mirrors m1 and m2 @xmath226 at this point , it is worth emphasizing that we have assumed equilibrium at zero temperature for the whole system : not only the input fields but also any fluctuations associated with loss mechanisms inside the mirrors correspond to zero - point fluctuations , whatever their microscopic origin may be . otherwise , the expression of the force discussed in the following would be affected . the pressures have opposite values on the two mirrors m1 and m2 . this entails that the global force exerted by vacuum upon the cavity vanishes , in consistency with the translational invariance of vacuum . in the following , we denote @xmath227 the casimir force calculated for m1 when considering the limit of a large area @xmath228 @xmath229 the sign conventions used here are such that the positive value obtained below for @xmath227 corresponds to an attraction of the two mirrors to each other . we now perform a change of variable to rewrite the summation symbol as specified in ( [ mphi ] ) @xmath230 \right ) \label{casimirairy}\end{aligned}\ ] ] we will now specify the domain of integration for @xmath10 . up to now , we have discussed the scattering for ordinary waves which freely propagate in vacuum and correspond to frequencies @xmath14 larger than the bound @xmath231 fixed by the norm of the transverse wavevector . but we must also take into account the contribution of evanescent waves which correspond to frequencies @xmath14 smaller than @xmath231 . these waves are fed by the additional fluctuations coming from the noise lines into the dielectric medium and propagating with an incidence angle larger than the limit angle . they are thus transformed at the interface into evanescent waves decreasing exponentially when the distance from the interface increases . as is well known @xcite , the properties of these evanescent waves are conveniently described through an analytical continuation of those of ordinary waves . this analytical continuation can only be dealt with in terms of functions having a well defined analyticity behaviour . this is not the case for the airy function @xmath232 $ ] but we know that this function is the sum ( [ gfunction ] ) of parts having well defined analyticity properties @xmath233 & = & 1+f_{\mathbf{k}}^{p}\left [ \omega \right ] + f_{\mathbf{k}}^{p}\left [ \omega \right ] ^{\ast } = \frac{1-\left| \rho _ { \mathbf{k}}^{p}\left [ \omega \right ] \right| ^{2 } } { \left| 1-\rho _ { \mathbf{k}}^{p}\left[\omega \right ] \right| ^{2 } } \nonumber \\ f_{\mathbf{k}}^{p}\left [ \omega \right ] & = & \frac{\rho _ { \mathbf{k}}^{p}\left [ \omega \right ] } { 1-\rho _ { \mathbf{k}}^{p}\left [ \omega \right ] } \nonumber \\ \rho _ { \mathbf{k}}^{p}\left [ \omega \right ] & = & r_{\mathbf{k},1}^{p}\left [ \omega \right ] r_{\mathbf{k},2}^{p}\left [ \omega \right ] e^{-2\kappa _ { 0}l } \label{loopfunctions}\end{aligned}\ ] ] @xmath234 $ ] is the ` open loop function ' corresponding to one round trip of the field inside the cavity and defined as the product of the reflection amplitudes @xmath235 $ ] and @xmath236 $ ] of the two mirrors and of the propagation phaseshift @xmath237 ; it is an analytical function in the physical domain of complex frequencies @xmath238 with the branch of the square root chosen so that @xmath239 . since the transverse wavevector is spectator throughout the whole scattering process , analyticity is defined with @xmath19 fixed . then , @xmath240 $ ] is the ` closed loop function ' built up on the open loop function @xmath234 $ ] . it is also an analytical function , thanks to analyticity of the open loop and to a stability property which has a natural interpretation : the system formed by the fabry - perot cavity and the vacuum fluctuations is stable because neither the mirrors nor the vacuum would have the ability to sustain an oscillation . in some cases , the stability can be derived from a more stringent passivity property @xcite which may essentially be written @xmath241 \right| < 1 $ ] . however , the passivity property is sometimes too stringent to be obeyed by real mirrors ( see more detailed discussions in appendix [ appevan ] ) . in any case , the stability property , _ i.e. _ the absence of self sustained oscillations , is sufficient for the present derivation of the casimir force . we are now able to give more precise specifications of the domain of integration in ( [ casimirairy ] ) . using the decomposition ( [ loopfunctions ] ) , we write the contribution of ordinary waves to this integral as the sum of two conjugated expressions @xmath242 \end{aligned}\ ] ] the integral @xmath243 is built on the retarded function @xmath240 $ ] which may be extended through an analytical continuation from the sector of ordinary waves to that of evanescent waves . the contribution of evanescent waves to the force is thus obtained as @xmath244 \end{aligned}\ ] ] the final expression of the casimir force is the sum of the contributions of ordinary and evanescent waves that is also the integral over the whole axis of real frequencies @xmath245 \label{forcereal}\end{aligned}\ ] ] as far as ordinary waves are concerned , this corresponds to the intuitive picture where the casimir force results from the radiation pressure of vacuum fluctuations filtered by the cavity @xcite . the contribution of evanescent waves is but the extension of the domain of integration to the whole real axis with the cavity response function @xmath240 $ ] extended through an analytical continuation . in the evanescent sector , the cavity function @xmath240 $ ] is written in terms of reflection amplitudes calculated for evanescent waves and exponential factors corresponding to evanescent propagation through the cavity . this means that it describes the ` frustration ' of total reflection on one mirror due to the presence of the other . this explains why the radiation pressure of evanescent waves is not identical on the two sides of a given mirror and , therefore , how evanescent waves have a non null contribution to the casimir force . using the cauchy theorem , we now rewrite the casimir force ( [ forcereal ] ) as an integral over the axis of imaginary frequencies . since @xmath246 $ ] is analytical in the domain @xmath247 , its integral over a closed contour lying in this domain has to vanish . we choose the contour drawn on figure [ figcontour ] which consists of the positive part of the real axis including ordinary ( c@xmath248 ) and evanescent ( c@xmath249 ) waves , a quarter of circle c@xmath250 with a very large radius and , finally , the imaginary axis c@xmath251 run from infinity to zero . now the function @xmath246 $ ] goes to zero for large values of the frequency , as a consequence of transparency at high frequency , a property certainly valid for any realistic model of optical mirror . thanks to this property , the contribution to the integral of c@xmath250 vanishes . we then deduce that the integrals over the real axis @xmath252 and over the imaginary axis @xmath253 are equal . we thus get a new expression of the force @xmath227 as an integral over imaginary frequencies @xmath14 , that is also as an integral over real values of @xmath63 , @xmath254 \nonumber \\ & & \kappa _ { 0}=\sqrt{\mathbf{k}^{2}+ \frac{\xi ^{2}}{c^2 } } \label{force}\end{aligned}\ ] ] we have used the fact that @xmath255 is real , so that @xmath256 is simply equal to @xmath255 . this property is less obvious , but also true , with @xmath255 written as an integral over real frequencies . we wish to emphasize more generally that expression ( [ force ] ) is mathematically equivalent to ( [ forcereal ] ) . the former expression is closer to the physical intuition whereas the latter is better adapted to explicit computations of the force . expression ( [ force ] ) gives the casimir force between real mirrors described by arbitrary frequency dependent reflection amplitudes . it is a regular integral as soon as these amplitudes obey the physical assumptions used in the derivation : causality , unitarity and high frequency transparency for each mirror , stability of the system formed by the two mirrors and the scattered vacuum fields . the demonstration holds for dissipative mirrors and not only for lossless ones . the limit of perfect mirrors is obtained in expression ( [ force ] ) by letting the reflection amplitudes go to unity , which leads to the casimir formula ( [ eqcasimir ] ) . this can be considered as an alternative demonstration of the casimir formula without any reference to a renormalization or regularization technique . basically , the properties of real mirrors , in particular their high frequency transparency , are sufficient to provide a regular expression of the force , as it was guessed a long time ago by casimir @xcite . as a simple model of the mirrors used in the experiments , let us consider a metallic slab with a large width , that is a width @xmath111 larger than a few plasma wavelengthes . we use expression ( [ force ] ) of the force written as an integral over imaginary values of the frequencies ( @xmath257 , @xmath63 real ) . hence , the phase factor corresponding to one round trip inside the slab is a decreasing exponential with a real exponent @xmath258 . for the plasma model ( [ plasmamodel ] ) , @xmath259 is given by @xmath260 and it is larger than @xmath261 for all values of @xmath63 and @xmath19 . when relaxation is taken into account , this is still the case except in a very narrow domain with values of @xmath63 and @xmath19 both close to zero . this domain has a negligible contribution to the integral ( [ force ] ) and it follows that the reflection amplitude of the slab may be replaced by the limiting expression obtained for the bulk . one thus recovers the lifshitz expression for the casimir force @xcite which is widely used for comparing experimental results with theoretical expectations @xcite . we now deduce general constraints which invalidate proposals made for tayloring the casimir force at will by using specially designed mirrors @xcite . this generalizes to 3d space the results obtained for 1d space in @xcite to which the reader is referred for further discussions . expression ( [ force ] ) is an integral over the axis of imaginary frequencies essentially determined by the reflection amplitudes @xmath262 $ ] and @xmath263 $ ] for @xmath63 real . these amplitudes always have a modulus smaller than unity , for arbitrary dielectric multilayers ( see appendix [ apppassiv ] ) . they are negative for arbitrary dielectric slabs ( see appendix [ appslab ] ) and we deduce from the composition law ( [ composlab ] ) that this is still the case for arbitrary dielectric multilayers . it follows that the product of the reflection amplitudes of the two mirrors is always positive with a modulus smaller than unity @xmath264 r_{2}\left [ i\xi \right ] < 1\ ] ] from this , we deduce first that the casimir force has an absolute value smaller than the value ( [ eqcasimir ] ) reached for perfect mirrors and that it remains attractive @xmath265 we also derive that the casimir force decreases as a function of the length @xmath266 this means that the properties obeyed by real mirrors strongly constrain the possible variation of the casimir force , contrarily to what might have been expected at first sight @xcite . note that we have considered mirrors used in the experiments which have electric permittivity but no magnetic permeability . different results would be obtained with magnetic mirrors , precisely with one of the mirrors dominated by electric response and the other one by magnetic response . the product of the two reflection amplitudes would indeed be negative in this case and the casimir force repulsive @xcite . we have presented a derivation of the casimir force between lossy mirrors characterized by arbitrary frequency dependent reflection amplitudes , in the casimir geometry where the cavity is made with two parallel plane mirrors . we have shown how mirrors and cavities may be dealt with by using a quantum theory of optical networks . we have deduced the additional fluctuations accompanying dissipation from expressions of the optical theorem adapted to quantum network theory . the optical theorem is equivalent to the unitarity of the whole scattering process which couples the modes of interest and the noise modes and it ensures that the quantum commutators of the output fields are the same as those of the input fields . the situation is different for the cavity fields which do not freely propagate . we have given a general proof of a theorem previously demonstrated in particular cases @xcite which states that the modification of the commutators is determined by the usual airy function , that is the spectral density associated with the fabry - perot cavity . for arbitrary lossy mirrors , the spectral density is determined by the reflection amplitudes as they are seen by the intracavity fields . it determines the radiation pressure exerted by vacuum fluctuations upon the mirrors with repulsive and attractive contributions associated respectively with resonant or antiresonant frequencies . the casimir force is then obtained as an integral over the whole axis of real frequencies , including the contribution of evanescent waves besides that of ordinary waves . it is equivalently expressed as an integral over imaginary frequencies . the derivation only uses a few general assumptions certainly valid for real optical mirrors , namely causality , unitarity , high - frequency transparency for each mirror and stability of the compound cavity - vacuum system . it leads to a finite result without any further reference to a regularization technique @xcite . the formula obtained in the present paper for the casimir force was already known @xcite but its scope of validity is widened by the present demonstration . it has been used to discuss the effect of imperfect reflection for the metallic mirrors used in the experiments . different descriptions of the optical response of metals have been used , from the crude application of the plasma model ( [ plasmamodel ] ) to a more complete characterization of the dielectric constant derived from tabulated optical data and dispersion relations . this kind of calculations , discussed in great detail for the mirrors corresponding to the recent experiments ( see for example @xcite ) , has not been reproduced here . instead , we have presented general results valid for any real mirrors obeying the physical properties already evoked and shown that they strongly constrain the variation of the casimir force . in the present paper , we have restricted our attention on the limit of zero temperature although our work was partly motivated by a recent polemical discussion of the effect of temperature on the casimir force between real mirrors @xcite . since contradictory results may have raised doubts about the validity and consistency of various derivations of the casimir force , we have considered it was important to come back to the first principles in this derivation . this has been done in the present paper for the case of zero temperature . a follow - on publication will show how to include the effect of thermal fluctuations in the treatment in order to obtain an expression free from ambiguities for the casimir force between arbitrary lossy mirrors at non zero temperatures . in this appendix , we discuss in more detail the specific case of the dielectric slab . we consider lossy as well as lossless slabs . for a lossless dielectric medium , the permittivity @xmath38 is real at real frequencies . for ordinary waves , @xmath267 and @xmath268 are purely imaginary , so that the impedance ratios are real for both polarizations . hence @xmath269 is real ( @xmath270 ) and @xmath114 purely imaginary ( @xmath271 ) so that the scattering amplitudes ( [ slab ] ) are read as @xmath272 the sum of the squared amplitudes is unity @xmath273 while the reflexion and transmission amplitudes are in quadrature to each other @xmath274 , which means that @xmath150 is a unitary @xmath275 matrix , as it was expected for a lossless mirror . this implies that the reflection amplitude has a modulus smaller than unity @xmath276 . this property also holds for lossy mirrors thanks to positivity of dissipation ( see appendix [ apppassiv ] ) . unitarity is defined without ambiguity only in the case of ordinary waves . for a lossless slab and evanescent waves , @xmath209 is real - it is just the inverse of the penetration length of evanescent wave in vacuum - whereas @xmath277 remains purely imaginary . hence @xmath269 as well as @xmath114 are purely imaginary and it is no longer possible to obtain general bounds for the scattering amplitudes @xmath278 in particular , @xmath279 does not remain always smaller than 1 ( see more explicit discussions in appendix [ appevan ] with different results for the te and tm polarizations ) . for imaginary frequencies finally , @xmath280 and @xmath281 are positive real numbers , for lossy as well as lossless slabs . in this case , general bounds are easily obtained for the amplitudes @xmath282 the fact that @xmath77 is negative with a modulus smaller than unity plays an important role in the derivation of constraints on the casimir force . interesting results are also obtained for the eigenvalues of the @xmath75matrix , which have a simple form @xmath283 since the slab is symmetrical in the exchange of its two ports . in the sector of ordinary waves , unitarity ( [ unitarity ] ) has a simple form in terms of @xmath283 and of the similar quantities @xmath284 defined on the noise matrix @xmath285 @xmath286 for the lossless slab , @xmath287 have a unit modulus and @xmath288 vanish . for a lossy slab , we have @xmath289 this can be considered as a consequence of ( [ unitarityssprime ] ) with @xmath290 . equivalently , it can be considered that unitarity ( [ unitarityssprime ] ) fixes the modulus of @xmath288 when the modulus of @xmath287 is known . condition ( [ bounds ] ) will be found in appendix [ apppassiv ] to express a passivity property for the slab , here for ordinary waves . this property still holds in the sector of imaginary frequencies , as a consequence of ( [ boundrtimaginary ] ) and of the following inequalities obeyed for all positive real numbers @xmath114 and @xmath269 @xmath291 using the terms of appendix [ apppassiv ] , this means that the domain of passivity always includes the sectors of ordinary waves and imaginary frequencies , in the case of a dielectric slab . however , it does not necessarily include the sector of evanescent waves ( see appendix [ appevan ] ) . ordinary waves correspond to frequencies @xmath292 and real wavevectors @xmath13 whereas evanescent waves correspond to frequencies @xmath293 and imaginary values of @xmath13 . causal scattering amplitudes can be extended from ordinary to evanescent waves , by an analytical continuation through the physical domain of complex frequencies @xmath257 with @xmath238 and @xmath239 . the ` energy conditions ' which bear on quadratic forms are not necessarily preserved in this process . in order to illustrate the idea , let us consider the reflection amplitude ( [ r01 ] ) at the interface between vacuum ( @xmath294 ) and a lossless dielectric medium ( @xmath295 real for @xmath14 real ) . in the sector of evanescent waves , @xmath277 is imaginary and @xmath296 real , so that @xmath77 and @xmath78 are complex numbers with a unit modulus , that is also pure dephasings corresponding to the phenomenon of total reflection . meanwhile , the transmission amplitudes differ from zero , which describes how evanescent waves in vacuum are fed by the fields coming from the dielectric medium with an incidence angle larger than the limit angle . in these conditions , it is clear that the condition @xmath297 fails . for the te polarization , it turns out that @xmath298 in the evanescent sector at the interface between vacuum and any dielectric medium . this property is always true in the sectors of ordinary waves and imaginary frequencies for an arbitrary mirror ( see the appendices [ appslab ] and [ apppassiv ] ) . using high frequency transparency , it follows from the phragmn - lindelf theorem @xcite that inequality ( [ rteeva ] ) holds in the whole physical domain in the complex plane . this ensures that the closed loop function @xmath299 is analytic and , in particular , has no pole in the domain @xmath247 . in other words , since the open loop gain is smaller than unity , the closed loop can not reach the oscillation threshold , leading to the stability property used in the derivation of the casimir force . although it seems quite natural , this argument is not valid in the general case . for metallic mirrors for example , the condition @xmath300 is violated in the evanescent sector for tm modes . the reflection amplitude is even known to reach large resonant values at the plasmon resonances @xcite . of course , this does not prevent the stability property to be fulfilled : the fabry - perot cavity is in this case a stable closed loop built on an open loop exceeding the unit modulus but with a phase such that the oscillation threshold is not reached . we stress again that the stability property is necessary in the derivation of the casimir force since it entails that the closed loop function is properly defined in the evanescent sector . when the more stringent property @xmath300 is also obeyed , it follows from expression ( [ loopfunctions ] ) that the airy function , which has been defined with the significance of a positive spectral density on ordinary waves , remains positive in the evanescent sector . when the property @xmath300 fails , the airy function can no longer be thought of as a spectral density in the whole physical domain , but this does not invalidate the derivation of the casimir force . in this appendix , we discuss the related but not identical properties corresponding to positivity of dissipation and passivity . we consider an arbitrary mirror , that is a reciprocal network connecting two vacuum ports . for ordinary waves , we define the power dissipated by the mirror @xmath301 where we have introduced row vectors conjugated to the column vectors @xmath302 this power is positive as a consequence of unitarity @xmath303 which corresponds to the positivity of the matrix @xmath304 @xmath305 where @xmath306 represents arbitrary input fields . positivity can also be expressed in terms of the eigenvalues @xmath111 of @xmath307 @xmath308 these eigenvalues are always real and positivity of dissipation is equivalent to the fact that they are smaller than unity @xmath309 passivity is a property directly related to positivity of dissipation but defined more generally for complex frequencies in the physical domain . in order to discuss it , we extend the matrix @xmath150 from the sector of ordinary waves through the analytical continuation already discussed . we extend @xmath310 similarly , with the complex conjugation cautiously defined since it involves complex frequencies : conjugation corresponds to @xmath311 and @xmath312 and it preserves the physical domain @xmath313 ; the derivations performed for an amplitude in the domain @xmath314 are thus translated to similar derivations for the conjugated amplitude in the quarter plane @xmath315 . then , the domain of passivity of @xmath150 is defined by the domain of @xmath63 for which @xmath316 is a positive matrix ( eq.[defpassiv ] ) that is also for which the eigenvalues @xmath124 of @xmath307 are smaller than unity ( eq.[def2passiv ] ) . an important feature of this property is that it is stable under composition : when two networks a and b are piled up as in figure [ figcomposition ] , the quadratic forms appearing in ( [ defpassiv ] ) simply add up so that passivity of the network ab follows from passivity of the two networks a and b. this is a special case of a general theorem @xcite which states that networks built up with passive elements are passive . passivity means that the eigenvalues @xmath317 of the matrix @xmath316 are both positive , which is equivalent to the following inequalities @xmath318 it may be written in terms of the scattering amplitudes @xmath319 passivity implies that the scattering amplitudes have a modulus smaller than unity @xmath320 conversely , the latter conditions are necessary but not sufficient for passivity . for a mirror symmetrical in the exchange of its two ports , a slab for example , the passivity conditions take the simple form @xmath321 . the results of appendix [ appslab ] thus entail that the domain of passivity always includes the sectors of ordinary waves and imaginary frequencies , for arbitrary slabs ( eq.[bounds ] ) . using the stability of passivity under composition , we deduce that this is also the case for arbitrary multilayers . it follows that the reflection amplitudes always have a modulus smaller than unity for imaginary frequencies ( @xmath63 real ) @xmath322 \right| \leq 1 \label{passivimaginary}\ ] ]

let and be convex bodies in , such that contains the origin , and define the process , , as follows : let be a uniform random point in , and set . clearly , is a nested sequence of convex bodies which converge to a non - empty limit object , again a convex body in . we study this process for being a regular simplex , a cube , or a regular convex polygon with an odd number of vertices . we also derive some new results in one dimension for non - uniform distributions .

1406.6590

the following problem was formulated by blint tth some 20 years ago with @xmath9 being the unit disc of the plane . let @xmath0 and @xmath1 be convex bodies in @xmath2 , such that @xmath0 contains the origin , and define the process @xmath3 , @xmath4 , as follows : let @xmath5 be a uniform random point in @xmath6 , and set @xmath7 . clearly , @xmath8 is a nested sequence of convex bodies which converge to a non - empty limit object , again a convex body in @xmath2 . what can we say about the distribution of this limit body ? what can we say about the speed of the process ? in figure [ fig:7gon ] one can see the evolution of the process up to @xmath10 on the right , and @xmath11 on the left , when @xmath12 is a regular heptagon . the evolution of the process for @xmath9 being a regular heptagon ] in @xcite ambrus , kevei and vgh investigated the process in @xmath13 dimension , when @xmath14 $ ] . in this case the limit object is a random unit interval , whose center has the arcsine distribution ( see theorem @xmath13 in @xcite ) . so even in the simplest case the process has very interesting features . moreover , in theorem @xmath15 in @xcite it is shown that if @xmath16 is the radius of the interval @xmath6 , then @xmath17 converges in distribution to a standard exponential random variable . the idea of the proof is to observe that @xmath18 behaves as the minimum of iid random variables , and thus obtain the limit theorem via extreme value theory . we also would like to point out the formal relationship between the diminishing process and the so called rnyi s parking problem from 1958 @xcite . rnyi studied the following random process : consider an interval @xmath19 of length @xmath20 , and sequentially and randomly pack ( non - overlapping ) unit intervals into @xmath19 . in each step we choose the center of the next unit interval uniformly from the possible space . the process stops when there is no space for placing a new unit interval . ( intuitively @xmath19 is the parking lot and the unit intervals are the cars . ) the first possible question is to determine the expectation @xmath21 of the covered space . many other variants of this problem has been studied in the last more than 50 years , for an up - to - date state of the art we refer to clay and simnyi @xcite . the connection between the diminishing process and rnyi s parking problem can be seen easily as follows : if we choose in the definition of the diminishing process @xmath22 , and we drop the conditions we put on @xmath0 , and define @xmath0 as the complement of the closed interval of length @xmath15 centered at @xmath23 , then we get exactly rnyi s parking problem . in the present paper we analyze the diminishing process in more general cases . in section [ sect:1dim ] we consider the case , when instead of choosing @xmath5 uniformly in the interval , we choose it according to a translated and scaled version of a fixed distribution @xmath24 . again , the limit object is a random unit interval . in theorem [ thm:1dim - rapid ] we determine the asymptotic behavior of the speed , while in theorem [ thm:1dim - center ] we show that for appropriate choice of @xmath24 the distribution of the center has the beta law . in sections [ sect : cube ] and [ sect : simplex ] we consider the case when @xmath12 is a cube and a regular @xmath25-dimensional simplex , respectively . the cube process can be represented as @xmath25 independent interval processes , thus the results in section [ sect : cube ] follow from the corresponding results in @xcite . in the case of the simplex process , the limit object is also a random regular simplex . the main result of this part is that the center of the limit simplex in barycentric coordinates has multidimensional dirichlet law , which is a natural generalization of the beta laws to any dimension . the rate of the process is also determined . the processes considered this far are ` self - similar ' in the sense that at each step the process is a scaled and translated version of the original one . in sections [ sect:5 ] , [ sect : pentagon ] and [ sect : sok ] we consider diminishing processes in the plane . in case of the pentagon process even the shape of the limiting object is random . we prove that it is a pentagon with equal angles , however it is not regular a.s . this process is not ` self - similar ' , and its behavior is more complicated . we determine the rate of the convergence of the maximal height , but as the area of the limit object is random , limit theorem with deterministic normalization is not possible . also the behavior of the center of mass is intractable with our methods . finally , in section [ sect : sok ] we consider regular polygons with odd number of vertices , i.e. @xmath26 is a regular polygon . using the theory of stochastic orderings for random vectors we prove that the rate of the speed is @xmath27 . we conjecture that in the case , when the number of vertices is even the speed of the process is @xmath28 . this is established in case of square , but in general it is open . in this section we consider the process in the interval @xmath29 $ ] , and the random point is chosen according to a not necessarily uniform distribution . fix a distribution on @xmath30 $ ] with distribution function @xmath24 , and in each step we choose the random point according to this distribution . that is , if the center and radius is @xmath31 the random point @xmath5 is given by @xmath32 , where @xmath33 is independent from @xmath34 , and has distribution function @xmath24 . the initial condition is @xmath35 , i.e. we start from the interval @xmath29 $ ] . let @xmath36 iid random variables with distribution function @xmath24 . it is easy to see that for @xmath4 @xmath37 to simplify the recursions above we have to pose some assumptions on @xmath24 . the following lemmas contain these assumptions . to determine the rapidness of the process we only need part ( i ) , while for the limit distribution of the center we need both parts . in fact , in both cases we only need the ` if ' part . in the following , for a random variable @xmath38 and an event @xmath39 the notation @xmath40 stands for the conditional distribution of @xmath38 given @xmath39 . [ lemma : konnyu ] let @xmath38 be a random variable with distribution function @xmath24 , such that @xmath41 \ } = 1 $ ] . * for all @xmath42 $ ] , for which @xmath43 , the distributional equality @xmath44 holds , if and only if either @xmath38 is a degenerate random variable at 0 , or @xmath45 , @xmath46 $ ] , for some @xmath47 . * the random variables @xmath48 and @xmath49 are independent if and only if @xmath50 , or @xmath51 , or @xmath52 for all @xmath53 $ ] . the simple proof of lemma [ lemma : konnyu ] is given in the appendix . as an immediate consequence we obtain the following . [ lemma : corollary ] let @xmath54 be a random variable in @xmath30 $ ] with continuous distribution function @xmath24 . then for any @xmath55 the distributional equality @xmath56 holds , and @xmath57 and @xmath58 are independent if and only if @xmath59 , \\ 1 - ( 1-c ) 2^\delta ( 1-x)^\delta , & x \in [ 1/2 , 1 ] , \end{array } \right.\ ] ] for some @xmath60 $ ] and @xmath47 . during the analysis of diminishing processes we frequently end up with a recursion of the following type . let @xmath61 be a sequence of iid random variables with distribution function @xmath62 , @xmath46 $ ] , for some @xmath47 , and let @xmath63 be a sequence of bounded nonnegative random variables , such that @xmath64 , a.s . , where @xmath65 is deterministic . assume that @xmath66 , and for @xmath4 , for some @xmath67 @xmath68 where @xmath69 $ ] and the abbreviation w.p . stands for ` with probability ' . to be precise this means here and later on the following . on our probability space @xmath70 there is a filtration @xmath71 . the filtration is usually generated by the random points @xmath72 , i.e. @xmath73 . the random variables @xmath74 and @xmath75 are @xmath76 measurable , and almost surely @xmath77 . conditionally on @xmath74 and @xmath75 let @xmath78 be a bernoulli@xmath79 random variable and independently @xmath80 is a random variable with distribution function @xmath81 , @xmath46 $ ] . then @xmath82 whenever @xmath83 , and @xmath84 otherwise . ( here and in the following section @xmath74 is simply a function of @xmath75 . however , when dealing with the polygon process @xmath74 is the area of @xmath6 , and it does depend on the chosen points , and not only on @xmath75 . this is the reason of the complication . ) in the next lemma we determine the asymptotic behavior of such sequence @xmath75 . the idea of the proof is to show that @xmath75 behaves like the minimum of @xmath85 iid random variables , as in the proof of theorem 1 in @xcite . the proof is deferred to the appendix . for @xmath47 , the weibull@xmath86 distribution function is given by @xmath87 , for @xmath88 , and 0 otherwise . [ lemma : recursion ] assume that @xmath75 is defined by ( [ eq : ell - recursion ] ) . then @xmath89 moreover , for any @xmath90 @xmath91 where @xmath92 is the usual gamma - function . with the help of these lemmas we can analyze the speed of the process . [ thm:1dim - rapid ] assume that for the distribution of @xmath38 we have @xmath93,\ ] ] for some @xmath94 . then as @xmath95 @xmath96 i.e. for any @xmath88 @xmath97 moreover , for any @xmath98 @xmath99 using the assumption and lemma [ lemma : konnyu ] ( i ) we see that ( [ eq : gen - rho - recursion ] ) can be rewritten as @xmath100 with @xmath101 , and @xmath61 are iid , @xmath62 , @xmath46 $ ] . now the theorem follows from lemma [ lemma : recursion ] , with @xmath102 and @xmath103 . to determine the limit distribution of the center consider the thinned process @xmath104 , which is obtained from the original process @xmath31 by dropping those steps when nothing changes , i.e. when @xmath105 . clearly , the limit of the center is not affected . after some calculation we obtain the recursion @xmath106 where @xmath33 has the distribution of @xmath38 conditioned on @xmath107 . note that in ( [ eq : df - form ] ) in lemma [ lemma : corollary ] for @xmath108 the distribution is concentrated on @xmath109 $ ] , in which case the center always moves towards @xmath110 , so the limit distribution of the center is degenerate at @xmath110 . similarly , for @xmath111 the limit is deterministic @xmath112 . in the following theorem we exclude these cases . for @xmath113 the random variable @xmath38 has beta@xmath114 law , if its density is @xmath115 , @xmath116 , where @xmath117 is the usual beta function . [ thm:1dim - center ] let us assume that for some @xmath118 and @xmath47 ( [ eq : df - form ] ) holds . then the distribution of @xmath119 is the translated beta@xmath120 law , i.e. its density function is @xmath121 by lemma [ lemma : konnyu ] and ( [ gen - recursion ] ) we obtain the recursion @xmath122 where @xmath123 , and @xmath124 are iid random variables , such that @xmath125 , and independently from @xmath126 , the sequence @xmath127 are iid @xmath128 random variables , i.e. with distribution function @xmath129 . the initial value is @xmath130 . formula ( [ eq : spec - recursion ] ) implies the infinite series representation of the limit @xmath131 and thus the distributional equation perpetuity @xmath132 where on the right - hand side @xmath133 are independent . corollary 1.2 in hitczenko and letac @xcite ( or the proof of theorem 3.4 in sethuraman @xcite ) implies that @xmath134 has @xmath135 distribution . note that once we have the infinite series representation ( [ eq : infinite - repr ] ) the proof can be finished using the properties of @xmath86 ( or poisson dirichlet ) law ; see hirth @xcite , or bertoin @xcite section 2.2.5 . distributional equations of type @xmath136 where @xmath137 are random vectors , and @xmath138 is a random variable , are called _ perpetuities_. equation ( [ eq : perpet ] ) is an example . necessary and sufficient conditions for the existence of a unique solution of one - dimensional perpetuities is given by goldie and maller @xcite . however , in special cases ( for example for @xmath139 $ ] ) the existence of a unique solution in any dimension was known earlier , see lemma 3.3 by sethuraman @xcite . therefore , in ( [ eq : perpet ] ) above , or in @xmath25 dimension in ( [ eq : d - perpet ] ) below , the assertion that certain distribution @xmath140 satisfies the perpetuity equation is equivalent to saying that the perpetuity equation has a unique solution @xmath140 . the perpetuities ( [ eq : perpet ] ) and ( [ eq : d - perpet ] ) are interesting in their own right , because there are relatively few perpetuities when the exact solution is known . the results of sethuraman @xcite ( proof of theorem 3.4 ; see also theorem 1.1 in @xcite ) cover those equations which appear in our investigations . for more general perpetuity equations with exact solutions we refer to the recent paper by hitczenko and letac @xcite in the cube process @xmath14^d$ ] . now the limiting convex body is a cube of unit edgelength . denote @xmath141 the edgelengths of the rectangular box @xmath6 , and @xmath142 the center of @xmath6 . properties of the uniform distribution imply that the processes @xmath143,@xmath144 , @xmath145 are @xmath25 independent copies of the segment process . therefore the following theorem is a consequence of theorem 1 and 2 in @xcite . [ th : rectangle ] for the speed of the cube process we have @xmath146 where @xmath147 are independent exponential random variables with parameter 1 . for the maximum of the edgelengths @xmath148 we have @xmath149 where @xmath150 , @xmath151 . for the limit distribution of the center @xmath152 where @xmath153 are independent translated arcsine random variables , that is with density function @xmath154 _ remark . _ similarly , the results obtained in section [ sect:1dim ] can be generalized for a ` non - uniform cube process ' . the details are left to the interested reader . now we turn to the simplex process in any dimension . let @xmath0 be a regular @xmath25-dimensional simplex with centroid @xmath155 and vertices @xmath156 , such that @xmath157 let denote @xmath158 the radius of the inscribed sphere of @xmath0 . let the initial simplex be @xmath159 ( for reasons explained below ) , and for @xmath6 given , choose a random point @xmath5 uniformly in @xmath6 and let @xmath7 . let @xmath160 denote the height of @xmath6 . then @xmath6 is a nested sequence of regular simplicies and the limit object is a regular simplex with height @xmath161 . it turns out that this process can be investigated by the same methods as for @xmath162 , in case of the segment process , in @xcite . the idea is that for the simplex in any dimension the process is ` self - similar ' , i.e. after each step the process is a translated and scaled version of the original one . if in the @xmath163st step the point @xmath5 falls close to the center , then nothing happens , i.e. @xmath164 . the ` change regions ' are @xmath165 congruent , regular simplicies of height @xmath166 , each of them sits at a vertex of @xmath6 . note that since the height of @xmath6 is @xmath167 then these simplicies are disjoint , so the process is simpler . this is the reason we assume @xmath168 , since its height @xmath169 . although , if we would start with a larger @xmath1 , as @xmath170 a.s . , in a random number of steps the height of @xmath6 would be @xmath167 , thus the assumption @xmath168 has no effect on the rapidness of the process . for the height process @xmath160 @xmath171 moreover , for any @xmath90 @xmath172^{\alpha / d}}{\rho_d^\alpha } ( m_n - \rho_d)^\alpha = \frac{\alpha}{d } \gamma \left ( \frac{\alpha}{d } \right).\ ] ] with disjoint change regions for the height process we have @xmath173 where @xmath174 are iid random variables , with distribution function @xmath175,\ ] ] which is the distribution of the distance from the base of a uniformly distributed random point in a regular simplex with height 1 , see figure [ fig:3szog ] . ( -1.67,-0.3 ) rectangle ( 6.97,6.26 ) ; ( 0,0) ( 6,0 ) ; ( 6,0) ( 3,5.2 ) ; ( 3,5.2) ( 0,0 ) ; ( 1.75,3.03) ( 4.25,3.03 ) ; ( -0.11,4.2 ) node @xmath176 ; ( 4.34,3.03 ) ( 4.34,3.79 ) ; ( 4.34,3.79 ) ( 4.34,3.03 ) ; ( 0.77,5.2 ) ( 0.77,3.03 ) ; ( 0.77,3.03 ) ( 0.77,5.2 ) ; ( -0.69,3.03 ) ( -0.69,0 ) ; ( -0.69,0 ) ( -0.69,3.03 ) ; ( -1.18,1.55 ) node @xmath177 ; ( 5.8,3.6 ) node@xmath178 ; ( 2.45,1.3 ) node @xmath6 ; ( 3.14,3.79 ) circle ( 1.5pt ) ; ( 3.1,4.1 ) node @xmath5 ; putting @xmath179 , we have @xmath180 a.s . , and @xmath181 the theorem follows from lemma [ lemma : recursion ] with @xmath182 and @xmath183 . let @xmath184 denote the center of the regular simplex @xmath6 . in this subsection we determine the limit distribution of @xmath184 . as we emphasized previously the limit distribution of @xmath185 is not affected if we start from any smaller regular simplex , in particular which has height @xmath186 . however , this is not true for the limit distribution of the center @xmath184 . to handle the process we have to assume that the change regions are disjoint , and so in each step the center can only move towards one of the vertices , or stay . in order to investigate the limit distribution of the centroid , we can consider the thinned ( centroid , height ) process @xmath187 , skipping the steps when nothing happens . put @xmath188 . since the disjoint change regions have the same volume , in each step the center moves towards any of the vertices with the same probability @xmath189 , according to the change region in which the chosen point falls . the size of the shift is @xmath190 , where @xmath191 is the distance of the chosen point from the base of the change region . see figure [ fig:3szog ] . thus @xmath192 where @xmath193 are iid random variables with distribution function ( [ h_d ] ) , @xmath124 are independent , uniformly distributed random variables on the set @xmath194 , and the initial condition is @xmath195 , @xmath196 . to obtain a more symmetric description of the center process we introduce the barycentric coordinates . the center of the limiting simplex falls in @xmath197 , i.e. in a regular simplex with height @xmath161 . put @xmath198 , that is @xmath199 are the vertices of @xmath200 . to parametrize the center we may use barycentric coordinates in terms of @xmath200 . that is , for @xmath184 the center of @xmath6 , we have @xmath201 , with @xmath202 , @xmath203 , @xmath204 . it is well - known that this parametrization is unique . put @xmath205 . we can rewrite ( [ mc1-simp ] ) in terms of the barycentric coordinates of @xmath206 . after some calculation we have @xmath207 where @xmath208 is the constant @xmath209 vector , except its @xmath210th coordinate being @xmath25 . the initial values are @xmath211 , @xmath196 . before stating the theorem , we define the multidimensional dirichlet distribution . let @xmath212 be positive numbers . the random vector @xmath213 has dirichlet@xmath214 distribution , if its components are nonnegative , @xmath215 , and @xmath216 has density function @xmath217 on the set @xmath218 . [ thm : simlex - kp ] the barycentric coordinates of the center of the limit simplex have @xmath219 distribution . let @xmath220 be the barycentric coordinates of the center of the limit . from ( [ mc1-simp-2 ] ) we obtain that @xmath221 rearranging we get @xmath222 . \end{split}\ ] ] notice that the infinite sum in brackets is equal in distribution with @xmath220 and it is independent of @xmath223 and @xmath224 . since @xmath225 , where @xmath226 are the usual unit vectors in @xmath227 , we obtain the distributional equality @xmath228 where on the right - hand side @xmath229 are independent . applying now theorem 1.1 in @xcite ( or the results in the proof of theorem 3.4 in @xcite ) with @xmath230 , and @xmath231 , we obtain obtain the theorem . let @xmath232 be an odd positive integer , and assume @xmath233 . let @xmath0 be a regular @xmath232-gon with circumradius @xmath13 , centroid @xmath234 , such that @xmath235 is a vertex and the side @xmath236 is parallel to the @xmath237-axis . we denote the vectors pointing from the origin to the vertices of @xmath0 in the counterclockwise order by @xmath238 . ( to avoid confusion , we distinguish between points and vectors . ) put @xmath239 , and consider the process as before . for simplicity we usually omit @xmath232 from our notation , and assume that @xmath232 is fixed , odd , and clear from the circumstances . obviously , @xmath6 is a polygon for each @xmath85 , and since it is the intersection of translated copies of @xmath0 , its sides are parallel to the sides of @xmath0 . however , note that @xmath6 is not necessarily a @xmath232-gon . for convenience , we are still going to consider @xmath6 as a ( possibly degenerated ) @xmath232-gon with the following definitions . let @xmath240 and @xmath241 be two parallel support lines of @xmath6 with equations @xmath242 and @xmath243 , where @xmath244 . now , we denote @xmath245 by @xmath246 and we consider it as the @xmath247th vertex of @xmath6 . similarly , @xmath248 is denoted by @xmath249 and we call it the @xmath247th side of @xmath6 . note that with these notation some vertices might coincide and correspondingly some sides might degenerate into a point . we also introduce the @xmath247th height of @xmath6 as @xmath250 . we put @xmath251 , and @xmath252 . the radius of the inscribed circle of @xmath0 is denoted by @xmath253 . we also introduce the notion of change region here : @xmath254 intuitively , the @xmath247th side moves , if we choose the next random point in @xmath255 . ( note that , this is not entirely true , since a degenerated side can move in other ways . ) obviously , if @xmath256 , then @xmath257 . we define @xmath258 the so called _ limit object_. [ lemma : limit - obj ] the limit object @xmath259 is a possibly degenerated , closed @xmath232-gon whose sides are parallel to the sides of @xmath0 . furthermore , the maximal height of @xmath259 is exactly @xmath260 almost surely . since @xmath259 is the intersection of closed half - planes with possible outer normals @xmath261 , it follows , that @xmath259 is a closed , possibly degenerated @xmath232-gon with sides parallel to the sides of @xmath0 . first we show that no height of @xmath259 is larger than @xmath260 . suppose that @xmath262 , in this case @xmath263 is of positive area . observe that no point was selected from @xmath263 by definition , which is a contradiction . next we prove that the maximal height of @xmath259 is at least @xmath260 . clearly , it is enough to see that @xmath264 for every @xmath85 . this follows from the observation that if @xmath256 , then @xmath257 . in the following lemma we show that @xmath6 always contains a small circle of radius @xmath265 . in particular this implies that the area of @xmath6 ( and thus the area of @xmath259 as well ) is uniformly bounded from below by @xmath266 . to ease the notation we put @xmath267 . [ termin ] let @xmath233 , and assume that @xmath268 where @xmath269 for all @xmath210 . then @xmath6 contains a circle of radius @xmath265 . denote @xmath270 the unit circle centered at the origin , which is the circumcircle of @xmath0 by definition . also by definition @xmath271 is the incircle of @xmath0 . we consider @xmath272 and we observe that @xmath273 holds for all @xmath85 . we claim that for all @xmath274 , we have @xmath275 . by definition @xmath276 . suppose that @xmath277 , then there exist an index @xmath278 with @xmath279 such that @xmath280 , and thus @xmath281 . but by definition @xmath282 , a contradiction . we obtained that @xmath283 is the intersection of the unit circles @xmath284 such that all centers @xmath285 are contained in @xmath283 . this readily implies that the minimal width of @xmath283 is at least one . then blaschke s theorem ( see @xcite , p. 18 , th . 2 - 5 . ) implies that there exists @xmath237 such that @xmath286 . obviously for all @xmath287 we have that @xmath288 , and thus @xmath289 implies that for all @xmath287 we have @xmath290 , which proves the statement . there exists a @xmath291 such that if every height of @xmath6 is smaller than @xmath292 , then the change regions @xmath255 are pairwise disjoint . we show that for every @xmath293 @xmath255 and @xmath294 are disjoint . first we show that the statement is true for adjacent regions . suppose that @xmath295 ( see figure [ fig : consecutivedr ] ) . ( 5.38,-3.74 ) rectangle ( 16.98,4.7 ) ; ( 10.38,-3.74 ) ( 10.38,4.7 ) ; plot(,(-3.811.52*)/3.38 ) ; plot(,(27.74 - 1.52*)/3.38 ) ; ( 7.86,2.4) ( 10.38,3.54 ) ; ( 10.38,3.54) ( 15.22,1.36 ) ; ( 7.48,2.23) ( 9.82,-2.98 ) ; ( 9.82,-2.98) ( 12.35,2.65 ) ; plot(,(-7.051.52*)/3.38 ) ; plot(,(32.65 - 2.18*)/4.84 ) ; ( 9.82,2.33) ( 9.82,-2.98 ) ; ( 10.38,3.54 ) circle ( 1.5pt ) ; ( 11.4,3.51 ) node @xmath296 ; ( 7.86,2.4 ) circle ( 1.5pt ) ; ( 7.34,2.71 ) node @xmath297 ; ( 15.22,1.36 ) circle ( 1.5pt ) ; ( 16.06,1.47 ) node @xmath298 ; ( 8.8,3.2 ) node @xmath299 ; ( 13.24,2.65 ) node @xmath300 ; ( 9.82,-2.98 ) circle ( 1.5pt ) ; ( 9.78,-3.26 ) node @xmath38 ; ( 9.14,-0.02 ) node @xmath301 ; ( 11.98,-0.04 ) node @xmath301 ; ( 9.82,2.33 ) circle ( 1.5pt ) ; ( 9.8,2.66 ) node @xmath138 ; ( 9.1,1.08 ) node @xmath302 ; according to figure [ fig : consecutivedr ] we draw two lines parallel to @xmath303 and @xmath304 respectively that are at distance exactly @xmath260 from the point @xmath38 , these two lines meet in the point @xmath138 . obviously , there exists a @xmath305 ( depending only on @xmath232 ) , such that @xmath306 . readily follows that @xmath307 , a contradiction . next we prove that if @xmath308 , and @xmath309 , then @xmath310 . this obviously implies the statement of the lemma . we proceed by induction on @xmath311 . for @xmath312 we are done . now we assume that the statement is true till @xmath313 , and we prove it for @xmath311 . pick @xmath314 . we may assume that @xmath315 for any @xmath316 , otherwise we would be done by applying the hypothesis twice . we may also assume that we changed the coordinate system such that the slope of @xmath317 is positive , the slope of @xmath303 is negative , and the bisectors of the line @xmath303 and @xmath317 are vertical and horizontal , see figure [ fig : nonconsecutivedr ] . ( -2.42,-6.84 ) rectangle ( 22.7,5.32 ) ; ( 10.1,-6.84 ) ( 10.1,5.32 ) ; ( 0.2,-1.52) ( 2.08,1.76 ) ; ( 2.08,1.76) ( 5.32,3.48 ) ; ( 18.7,0.76) ( 19.85,-1.26 ) ; ( 14.28,3.1) ( 18.7,0.76 ) ; plot[domain=1.21:1.85,variable=](1 * 14.25*cos(r)+0 * 14.25*sin(r),0 * 14.25*cos(r)+1 * 14.25*sin(r ) ) ; ( 9.78,-4.14 ) circle ( 4.56 cm ) ; ( 1.44,0.64) ( 9.78,-4.14 ) ; ( 9.78,-4.14) ( 5.5,3.91 ) ; ( 9.78,-4.14) ( 14.06,3.91 ) ; ( 9.78,-4.14) ( 18.6,0.92 ) ; plot(,(-5.393.28*)/1.88 ) ; plot(,(3.221.72*)/3.24 ) ; plot(,(-36.861.72*)/-3.24 ) ; plot(,(36.04 - 2.01*)/1.15 ) ; ( 9.78,-4.14) ( 18.82,-5.34 ) ; ( 0.2,-1.52 ) circle ( 1.5pt ) ; ( -0.65,-1.8 ) node @xmath318 ; ( 2.08,1.76 ) circle ( 1.5pt ) ; ( 4.15,1.2 ) node @xmath319 ; ( 5.32,3.48 ) circle ( 1.5pt ) ; ( 7.36,3.06 ) node @xmath320 ; ( 0.6,0.28 ) node @xmath321 ; ( 4.54,2.1 ) node @xmath322 ; ( 14.28,3.1 ) circle ( 1.5pt ) ; ( 12.4,3.06 ) node @xmath323 ; ( 18.7,0.76 ) circle ( 1.5pt ) ; ( 20.62,1 ) node @xmath324 ; ( 19.85,-1.26 ) circle ( 1.5pt ) ; ( 20.78,-1.88 ) node @xmath298 ; ( 18.76,-0.48 ) node @xmath300 ; ( 15.98,1.64 ) node @xmath299 ; ( 9.78,-4.14 ) circle ( 1.5pt ) ; ( 9.7,-4.55 ) node @xmath38 ; ( 18.82,-5.34 ) circle ( 1.5pt ) ; ( 14.46,-5.06 ) node @xmath260 ; ( 3.18,2.68 ) circle ( 2.5pt ) ; ( 1.68,3.4 ) node @xmath325 ; ( 16.38,2.68 ) circle ( 2.5pt ) ; ( 18.24,2.94 ) node @xmath326 ; draw the translated copy @xmath327 of @xmath0 whose center is @xmath38 , the incircle of @xmath327 is of radius @xmath260 and of center @xmath38 . consider the vertices @xmath324 and @xmath319 of @xmath6 , and the vertices @xmath326 and @xmath325 of @xmath327 . from the assumptions it clearly follows that the ` horizontal distance ' ( the difference of the @xmath237 coordinates ) of @xmath324 and @xmath319 is larger than the horizontal distance of @xmath326 and @xmath325 . but this is a contradiction , since the sides @xmath328 form a fixed angle with the @xmath237-axis , and each of them is at most as long as the side length of @xmath0 , and thus the horizontal distance of @xmath326 and @xmath325 is maximal . a configuration is called _ reduced _ if the change regions are disjoint . in a reduced state it is possible to follow the process . that gives the importance of the following simple corollary which readily follows from the fact that @xmath329 is componentwise monotone decreasing and @xmath330 . [ cor : reduced ] the process a.s . reaches a reduced state in a random number of steps . after reaching a reduced state , the process always stays in a reduced state . ( -3.5,5 ) rectangle ( 8.89,14.18 ) ; ( -0.14,5.84) ( 5.52,5.84 ) ; ( 5.52,5.84) ( 6.91,10.09 ) ; ( 6.91,10.09) ( 2.31,13.43 ) ; ( 2.31,13.43) ( -1.67,10.54 ) ; ( -1.67,10.54) ( -0.14,5.84 ) ; ( 1.33,12.72) ( 3.29,12.72 ) ; ( 0.28,11.96) ( -0.93,8.25 ) ; ( 5.34,11.23) ( 6.31,8.25 ) ; ( 6.13,7.7) ( 3.56,5.84 ) ; ( 1.06,5.84) ( -0.52,6.98 ) ; ( 3.56,5.84) ( 5.52,5.84 ) ; ( 5.52,5.84) ( 6.13,7.7 ) ; ( 6.13,7.7) ( 3.56,5.84 ) ; ( 6.31,8.25) ( 6.91,10.09 ) ; ( 6.91,10.09) ( 5.34,11.23 ) ; ( 5.34,11.23) ( 6.31,8.25 ) ; ( 3.29,12.72) ( 2.31,13.43 ) ; ( 2.31,13.43) ( 1.33,12.72 ) ; ( 1.33,12.72) ( 3.29,12.72 ) ; ( 0.28,11.96) ( -1.67,10.54 ) ; ( -1.67,10.54) ( -0.93,8.25 ) ; ( -0.93,8.25) ( 0.28,11.96 ) ; ( -0.14,5.84) ( 1.06,5.84 ) ; ( 1.06,5.84) ( -0.52,6.98 ) ; ( -0.52,6.98) ( -0.14,5.84 ) ; ( -0.66,5.56 ) node @xmath331 ; ( 0.3,6.9 ) node @xmath332 ; ( 6.11,5.56 ) node @xmath333 ; ( 5.0,6.23 ) node @xmath334 ; ( 7.37,10.32 ) node @xmath335 ; ( 6.3,10.0 ) node @xmath336 ; ( 2.81,13.73 ) node @xmath337 ; ( 2.2,12.4 ) node @xmath338 ; ( -2.1,10.8 ) node @xmath339 ; ( -1.0,10.34 ) node @xmath340 ; ( -0.14,5.84 ) circle ( 1.5pt ) ; ( -1.67,10.54 ) circle ( 1.5pt ) ; ( 5.52,5.84 ) circle ( 1.5pt ) ; ( 6.91,10.09 ) circle ( 1.5pt ) ; ( 2.31,13.43 ) circle ( 1.5pt ) ; in this section we consider the pentagon process . this is the simplest case when not only the position , but also the shape of the limit object is random . we show that exactly one height of the limit object is @xmath341 , which allows us to determine the speed of the process . first we prove that the process can not degenerate in the following sense . @xmath6 is always a pentagon with equal inner angles . the key observation is that the directions of the sides of @xmath6 are prescribed , thus the only thing we have to show that a side can not disappear . suppose the opposite , and seek a contradiction . let @xmath6 be the first non - pentagonal state , and first assume that it is a quadrilateral and the side @xmath342 disappears . it easy to calculate the inner angles of @xmath6 , three of them equals the inner angle of a regular pentagon , @xmath343 ( at vertices @xmath344 , @xmath345 and @xmath346 ) , while the fourth one is @xmath347 ( at the vertex @xmath348 ) . also note , that the side lengths of @xmath6 can not exceed the side length of @xmath0 . thus @xmath6 is contained in a deltoid , see figure [ deltoid ] , where @xmath349 is the side length of @xmath0 . this implies that the heights @xmath350 and @xmath351 of @xmath6 are at most @xmath352 . a simple argument shows that we may assume that @xmath346 was a vertex of @xmath353 , but @xmath348 and @xmath344 were not . this implies that the side @xmath354 comes from @xmath0 ( more precisely , @xmath355 ) , and so @xmath356 . but this is not possible , since @xmath357 , a contradiction . similar argument settles the case when @xmath6 is a triangle . ( 3.8,-5 ) rectangle ( 16.04,3.98 ) ; ( 9.61,3.69) ( 5.13,0.43 ) ; ( 5.13,0.43) ( 6.84,-4.84 ) ; ( 6.84,-4.84) ( 12.38,-4.84 ) ; ( 12.38,-4.84) ( 14.09,0.43 ) ; ( 14.09,0.43) ( 9.61,3.69 ) ; ( 7.9,-1.58) ( 13.44,-1.58 ) ; ( 12.38,1.67) ( 7.9,-1.58 ) ; ( 12.38,1.67) ( 14.09,0.43 ) ; ( 14.09,0.43) ( 13.44,-1.58 ) ; ( 11.07,0.72) ( 11.07,-1.58 ) ; ( 11.07,0.72) ( 12.54,-0.35 ) ; ( 12.54,-0.35) ( 12.14,-1.58 ) ; ( 7.9,-1.58 ) circle ( 1.5pt ) ; ( 7.6,-1.84 ) node @xmath348 ; ( 13.44,-1.58 ) circle ( 1.5pt ) ; ( 12.38,1.67 ) circle ( 1.5pt ) ; ( 10.04,-1.88 ) node @xmath349 ; ( 9.6,0.42 ) node @xmath349 ; ( 12.14,-1.58 ) circle ( 1.5pt ) ; ( 12.34,-2.04 ) node @xmath344 ; ( 11.07,0.72 ) circle ( 1.5pt ) ; ( 10.96,1.28 ) node @xmath346 ; ( 12.54,-0.35 ) circle ( 1.5pt ) ; ( 12.96,-0.18 ) node @xmath345 ; ( 11.07,-1.58 ) circle ( 1.5pt ) ; ( 11.38,-0.56 ) node @xmath351 ; by corollary [ cor : reduced ] in a random number of steps we reach a reduced state , and so as in the simplex case we may and do assume that the process starts from a reduced state . it also follows that in a reduced state the change regions are always triangles . note that if the random point falls in @xmath358 then beside @xmath359 , the opposite heights @xmath360 and @xmath351 also decrease . some calculation shows that if @xmath359 decreases by @xmath237 then @xmath360 and @xmath351 both decreas by @xmath361 , with @xmath362 being the ratio of the golden section . we say that @xmath363 and @xmath364 are _ competing _ heights , if @xmath365 , @xmath366 , and they are not adjacent . to describe the dynamics of the process we define the following vectors : @xmath367 , @xmath368 , @xmath369 , @xmath370 , and @xmath371 . with this notation , if in a reduced state in the @xmath163th step the random point falls in @xmath372 , then @xmath373 where @xmath174 are iid with common distribution function @xmath374 in ( [ h_d ] ) , i.e. @xmath375 is the distribution of the distance from the base of a uniformly chosen point in a triangle with height 1 . that is , @xmath376 is the distance of @xmath5 and the side of @xmath372 which is opposite to @xmath377 . the probability of this event is @xmath378 , where @xmath379 is the area . [ nincsnemszomszedos ] the limit pentagon can not have non - adjacent heights equal to @xmath341 . emphasizing that the process can be at any reduced state we omit the index @xmath85 . assume that there is a state with at least 2 competing heights @xmath380 . let , say , @xmath359 be the maximum height , which has a competing pair , say @xmath360 . if the maximum height has no competing pair @xmath380 than its change has no affect on the two competing heights . thus @xmath359 will change eventually . so we may and do assume that @xmath359 is the largest height . * case 1 * : @xmath381 , with @xmath382 defined in ( [ c ] ) . then the probability that in the next change step the uniform random point falls in @xmath358 is @xmath383 , and given this the probability that @xmath359 decrease at least with @xmath384 equals @xmath385 . in this case @xmath360 decreases below @xmath341 , and so the probability of this @xmath386 . * case 2 * : @xmath387 . the probability that in the next change step the random point falls in @xmath388 is @xmath389 we show that with positive probability we end up in a state corresponding to case 1 . in the next step @xmath390 we want an @xmath391 , such that @xmath392 . some calculation shows that this happens if and only if @xmath393 where the right side is @xmath394 the probability of this event is at least @xmath395 so we are almost in case 1 , but it can happen that @xmath396 is not maximal . notice that @xmath397 which implies that the probability of choosing in @xmath358 in the next change step is @xmath398 . so we showed that starting from any state with at least two competing heights @xmath380 , the probability that in two change steps one of them decreases below @xmath341 is @xmath399 this proves that the process can not have this configuration for infinite number of steps . [ nincsketszomszedos ] there is no non - regular pentagon with equal angles , in which the two largest heights are consecutive . as a first step we prove a somewhat surprising result that provides a linear relationship between any four heights of the pentagon . we assume that @xmath400 and @xmath351 are given , and we express @xmath350 as a linear combination of the previous three . to simplify the calculations , we place the pentagon into a new coordinate system such that @xmath348 is the origin and @xmath401 agrees with the @xmath237-axis , and the whole pentagon lies in the upper half plane . recall that @xmath402 , @xmath403 are the outer normals of the sides , as we defined earlier . from the setup the equations of @xmath404 and @xmath405 readily follow : @xmath406 and @xmath407 . using the definition of @xmath359 we obtain @xmath408 . and again by the definition of @xmath360 and @xmath351 , @xmath346 is on the line of equation @xmath409 and @xmath345 is on @xmath410 . we can express @xmath345 and @xmath346 by solving the system of equations : @xmath411 now , we can find the equation of @xmath304 and @xmath412 . after suitable simplifications , introducing the golden ratio @xmath413 , we obtain @xmath414 thus @xmath415 , and to obtain @xmath350 we need to calculate the distance between @xmath344 and @xmath416 : @xmath417 from ( [ eqe2 ] ) and ( [ eqe5 ] ) it readily follows that @xmath418 and @xmath419 , hence @xmath420 now , suppose that @xmath359 and @xmath350 are the two largest heights . if @xmath421 , then we have a contradiction by ( [ aranymetszes ] ) . if @xmath422 , then since @xmath359 and @xmath350 are the two largest , it follows that @xmath423 , and hence the pentagon is regular . as a consequence of the previous lemmas we obtain [ thm : limit - pentagon ] the limit pentagon has exactly one height equal to @xmath341 a.s . _ remark . _ with a rather tedious case analysis one can prove that for any height of the limit pentagon @xmath424 , which is sharp . in the previous section we proved that the limit pentagon has exactly one height equal to @xmath341 a.s . , i.e. after finite number of steps @xmath6 has only one height greater than @xmath341 . this observation allows us to prove some asymptotic results for the speed , however , as the area of the limit is now _ random _ , we can not prove limit theorem , only upper and lower bounds . let denote @xmath425 the maximum and @xmath426 the minimum of the area of the possible limit pentagons . note that @xmath427 by lemma [ termin ] . then we have the following . [ th : pentagon - rapid ] for any @xmath88 @xmath428 moreover , @xmath429 where @xmath430 denotes the area of the limit pentagon . put @xmath431 . once there is only one height @xmath380 the limit pentagon is determined and so is its area @xmath432 . the area of the only non - empty change region @xmath433 . this means that the height process @xmath434 behaves as @xmath435 where @xmath436 are iid , @xmath437 , for @xmath46 $ ] . since @xmath438 a.s . , by lemma [ lemma : recursion ] with @xmath439 , @xmath440 , @xmath441 we obtain that given @xmath430 we have for any @xmath88 @xmath442 or @xmath443 the convergence of the moments also hold ( as in lemma [ lemma : recursion ] ) , in particular @xmath444 \right ] \\ & \to { \mathbf{e}}\int_0^\infty { \mathrm{e}}^{-\frac{x^2}{t } } { \mathrm{d}}x = \frac{{\mathbf{e}}\sqrt{t } } { 4 \sqrt{\pi } } , \end{split}\ ] ] and the theorem is proved . in general the polygon process is too complicated to say anything more about the limit object than lemma [ lemma : limit - obj ] . according to this lemma the maximal height of the limit object is @xmath260 . using stochastic majorization and minorization we are able to determine the order of the convergence . [ mainodd ] for any @xmath88 we have @xmath445 the evolution of @xmath446 for 1800 iterations of the heptagon process ] let @xmath447 be the height vector , @xmath160 its maximum , and @xmath448 the area of the change regions . by corollary [ cor : reduced ] we may and do assume that the change regions are already disjoint . the probability of no change is the probability that the random point does not fall in @xmath449 , is @xmath450 . the probability of change is @xmath451 , in particular @xmath452 is the probability that we choose the point in @xmath372 . in this case @xmath453 , and all the other heights decrease at most with @xmath454 , where @xmath455 is the distance from the base of a uniformly chosen point in @xmath372 , and so @xmath456 is the distance from the base of a uniformly chosen point in @xmath457 , i.e. we scale the change region to have height 1 . so we have that in case of change @xmath458 , where @xmath459 stands for the constant 1 vector , and so @xmath460 . we want to construct simple processes , serving as lower and upper bound for @xmath160 . in order to do so we recall same basic properties of stochastic ordering . for random variables @xmath38 and @xmath54 we say that @xmath38 is _ stochastically larger _ than @xmath54 ( @xmath461 ) if @xmath462 for any @xmath463 . this is equivalent to the condition @xmath464 for any increasing function @xmath465 . for random vectors the definition is somewhat trickier . in @xmath466 a set @xmath467 is an _ upper set _ if for @xmath468 , @xmath469 imply @xmath470 . for @xmath232-dimensional random vectors @xmath471 and @xmath472 we have @xmath473 if @xmath474 for any upper set @xmath467 . this is equivalent to the condition @xmath475 for any componentwise increasing function @xmath476 . we refer to shaked and shanthikumar @xcite chapter 1.a and chapters 6.a and 6.b . the first step is to obtain a stochastic majorant and minorant for @xmath477 for any type of scaled change regions . let us fix such a region , and let @xmath478 be the area of those points in the region , which are farther than @xmath479 from the base . if @xmath375 is the distance of the random point from the base then @xmath480 . the angle of the upper vertex is @xmath481 , and the corresponding angle bisector is orthogonal to the base , so for all @xmath46 $ ] @xmath482 by lemma [ termin ] a disc of radius @xmath265 is contained in @xmath6 , which together with convexity imply that the angle of the upper vertex is @xmath483 . therefore @xmath484 summarizing , we have @xmath485 where @xmath486 . note that @xmath487 in the lower bound does not depend on @xmath232 . for @xmath151 put @xmath488 for the corresponding distribution functions of @xmath489 . the previous reasoning also shows that @xmath490 and so @xmath491 by the trivial bound and by lemma [ termin ] we have the following upper and lower bounds for the area : @xmath492 * the lower bound . * using ( [ eq : a_n - bound ] ) and ( [ eq : ter - est ] ) the change probability can be estimated as @xmath493 let us define the process @xmath494 where @xmath495 iid , and @xmath489 has distribution function @xmath496 in ( [ h - major ] ) . we claim that @xmath497 indeed , @xmath498 decreases with higher probability , and if it decreases , decreases larger . putting @xmath499 @xmath500 we can write @xmath501 , with @xmath502 independent from @xmath503 and having distribution function @xmath504 , \\ \textrm{anything } , & x > { { \mathbf{\ell}}}_n ' . \end{array } \right.\ ] ] if @xmath505 has distribution function @xmath506 then @xmath507 for any @xmath85 and @xmath508 . for @xmath127 iid with distribution function @xmath509 , put @xmath510 . we obtained that for all @xmath85 @xmath511 combining this with ( [ 1-step - dom ] ) we deduce @xmath512 we claim that these inequalities imply the unconditional inequality . the latter process can be written as ( we assume that the process starts from a sufficiently small state ) @xmath513 where @xmath514 are iid with distribution function @xmath515 , @xmath46 $ ] . short calculation gives that @xmath516 which is decreasing in @xmath517 for any fix @xmath237 . let us assume that @xmath518 , and it is sufficiently small . the law of total probability and ( [ 1-step - dom - v ] ) imply @xmath519 . assume that for any @xmath88 , @xmath520 for some @xmath521 . then @xmath522 where we used the law of total probability , ( [ 1-step - dom - v ] ) , the induction hypothesis , the monotonicity of the conditional probabilities , and that for two distribution functions @xmath523 , such that @xmath524 , and for a monotone decreasing function @xmath465 we have @xmath525 ( @xcite chapter 1.a ) . so we proved that @xmath526 for every @xmath85 . for the asymptotic behaviour of @xmath527 we have @xmath528 and since @xmath526 , we obtain @xmath529 in particular we have @xmath530 & = \int_0^\infty { \mathbf{p}}\left\ { \sqrt{c_1 c_2 } \sqrt{n } ( m_n - \rho_k ) > x \right\}{\mathrm{d}}x \\ & \geq \int_0^\infty { \mathbf{p}}\left\ { \sqrt{c_1 c_2 } \sqrt{n } \underline v_n > x \right\ } { \mathrm{d}}x \\ & \to \int_0^\infty { \mathrm{e}}^{-x^2 } { \mathrm{d}}x,\end{aligned}\ ] ] where at the last convergence we used the uniform integrability of @xmath531 . * upper bound . * now we turn to the construction of the upper bound process . if the random point falls in the change region @xmath372 then we have @xmath532 , and the other heights may change or may not . in any case @xmath533 , where @xmath534 is the @xmath247th standard , @xmath232 dimensional unitvector . the probability of this event is @xmath378 for which by ( [ eq : a_n - bound ] ) and ( [ eq : ter - est ] ) @xmath535 instead of @xmath536 we put the stochastically smaller @xmath375 , for which @xmath537 has distribution function @xmath538 defined in ( [ h - minor ] ) . note that for this @xmath375 we have @xmath539 . we define the @xmath232-dimensional process @xmath540 as follows . let @xmath541 such that @xmath542 . then define @xmath543 where @xmath174 are iid and @xmath489 has distribution function @xmath538 in ( [ h - minor ] ) , that is in each step at most one component decreases , and component @xmath247 can decrease only in steps @xmath544 , @xmath545 . from the construction it is clear that for each @xmath546 , and for each upper set @xmath467 @xmath547 now we show that @xmath548 is a monotone increasing function of @xmath549 for any fixed upper set @xmath467 . to do so , let @xmath542 , and define @xmath550 . we may assume that @xmath551 , @xmath552 and @xmath553 , otherwise the statement is obvious . recall that in one step only coordinate @xmath247 can change , and so by ( [ hat ] ) we have @xmath554 by the properties of the upper set we have that @xmath555 and so the conditional probability is monotone increasing . as in the case of the lower estimation this allows us to prove the majorization @xmath556 as follows : if @xmath557 in distribution , then we have the majorization for @xmath558 , and if it is true for some @xmath521 , then for any upper set @xmath467 @xmath559 where we used the law of total probability , ( [ 1-step - maj ] ) , the induction hypothesis , the monotonicity of the conditional probabilities , and that for two distribution functions @xmath523 , such that @xmath560 ( understood componentwise ) , and for a monotone increasing function @xmath465 we have @xmath525 ( @xcite , chapter 6.b ) . putting @xmath561 as before we see that @xmath562 where @xmath563 are iid random variables with distribution function @xmath564 . we have @xmath565^{k } \\ & = \left [ 1 - \left ( 1- \frac{x^2}{n } \right)^{\lfloor n / k \rfloor } \right]^{k } \\ & \to \left ( 1 - { \mathrm{e}}^{- \frac{x^2}{k } } \right)^{k}. \end{split}\ ] ] this , together with the stochastic majorization @xmath556 implies that @xmath566 in particular we have @xmath567 & = \int_0^\infty { \mathbf{p}}\left\ { \sqrt{c_3 n } ( m_n - \rho_{k } ) > y \right\}{\mathrm{d}}y \\ & \leq \int_0^\infty { \mathbf{p}}\left\ { \sqrt{c_3 n } ( \widehat m_n - \rho_{k } > y \right\ } { \mathrm{d}}y \\ & \to \int_0^\infty \left [ 1 - \left ( 1 - { \mathrm{e}}^{- \frac{y^2}{k } } \right)^{k } \right ] { \mathrm{d}}y.\end{aligned}\ ] ] the major difference between regular polygons with odd and even number of vertices hides in the fact that while in the odd case the change regions are always triangles , in the even case change regions might be trapezoids or ( in the degenerated case ) triangles , hence their area might be of different order ( see figure [ figure : caps ] ) . we conjecture that in the latter case the ` typical ' change regions are trapezoids , which would imply that the speed of the process is @xmath568 . ( compare with theorem [ mainodd ] , where we obtained @xmath569 for the speed in the odd case . ) this conjecture is well supported by numerical experiments . we conclude the paper with the results of some computer simulations , see figure [ fig:7 ] and figure [ fig:8 ] . it is transparent that @xmath570 and @xmath85 are the right normalizations , respectively . ( -2.92,0.44 ) rectangle ( 23.72,4.52 ) ; ( 3.08,3 ) ( 0.09,1.68 ) ( 6.07,1.68 ) cycle ; ( 13.72,3 ) ( 10.73,1.68 ) ( 20.89,1.68 ) ( 17.9,3 ) cycle ; plot(,(44.46 - 0*)/14.81 ) ; ( 3.08,3) ( -1.76,0.86 ) ; ( 3.08,3) ( 7.92,0.86 ) ; ( 13.72,3) ( 8.88,0.86 ) ; ( 17.9,3) ( 22.74,0.86 ) ; plot(,(24.9 - 0*)/14.82 ) ; ( 13.72,3) ( 17.9,3 ) ; ( -2.22,1.68 ) ( -2.22,3 ) ; ( -2.22,3 ) ( -2.22,1.68 ) ; ( 3.08,3) ( 0.09,1.68 ) ; ( 0.09,1.68) ( 6.07,1.68 ) ; ( 6.07,1.68) ( 3.08,3 ) ; ( 13.72,3) ( 10.73,1.68 ) ; ( 10.73,1.68) ( 20.89,1.68 ) ; ( 20.89,1.68) ( 17.9,3 ) ; ( 17.9,3) ( 13.72,3 ) ; ( 3.08,3 ) circle ( 1.5pt ) ; ( 17.9,3 ) circle ( 1.5pt ) ; ( 2.97,2.32 ) node @xmath571 ; ( 13.72,3 ) circle ( 1.5pt ) ; ( 15.64,2.32 ) node @xmath572 ; ( -1.68,2.32 ) node @xmath237 ; @xmath573 for 200 outcomes of the heptagon process ( the minimal value is @xmath574 , the maximal value is @xmath575 ) ] @xmath576 for 200 outcomes of the octagon process ( the minimal value is @xmath577 , the maximal value is @xmath578 ) ] lemma [ lemma : konnyu ] to prove part ( i ) note that the distributional equality means @xmath579 for all @xmath580 . the monotonicity of @xmath24 easily implies that the solution has the stated form for some @xmath47 . the ` if ' part follows by simple calculation . lemma [ lemma : recursion ] after some calculation one obtains that given @xmath75 and @xmath74 , @xmath583 , where @xmath54 is a nonnegative random variable , such that @xmath584 is uniformly distributed on @xmath585 $ ] . to prove the reverse inequality , let us fix @xmath596 , @xmath597 . given that @xmath598 we have @xmath599 , and thus given that @xmath600 we have @xmath601 by the law of total probability @xmath602 by the assumption @xmath77 a.s . , so the second term goes to 0 , while from extreme value theory ( see e.g. @xcite , p.192 ) we have @xmath603 that is , by the stochastic dominance @xmath604 since @xmath596 and @xmath597 are as small as we want , we obtain @xmath605 and the convergence in distribution is proved . once we have the distributional convergence , to prove the moment convergence it is enough to show that @xmath606 is uniformly integrable ( see e.g. @xcite theorem 25.12 ) . since @xmath74 is bounded , for some @xmath607 we have @xmath608 a.s . for all @xmath521 , and thus @xmath609 . therefore @xmath610 and the uniform integrability follows .

the absolute muon flux between 20 and 3000 is measured with the l3 magnetic muon spectrometer for zenith angles ranging from to . due to the large exposure of about 150 msrd , and the excellent momentum resolution of the l3 muon chambers , a precision of 2.3% at 150 in the vertical direction is achieved . the ratio of positive to negative muons is studied between 20 and 500 , and the average vertical muon charge ratio is found to be 1.285 0.003 ( stat . ) 0.019 ( syst . ) . _ the l3+c group dedicates this publication to the late bianca monteleoni . _

hep-ex0408114

atmospheric muons are among the final products of cosmic ray induced air - showers . the absolute muon flux and its momentum dependence are mainly determined by the flux of nucleons entering the atmosphere and the inclusive meson production cross sections in high - energy hadronic interactions . the ratio of the fluxes of positive to negative muons , denoted as charge ratio in the following , reflects the proton to neutron ratio at the top of the atmosphere , folded with the production and decay spectra of charged pions and kaons . while the knowledge of the primary cosmic ray spectrum below a few 100 has improved considerably in the recent past @xcite , large uncertainties still exist in the primary energy range between 0.1 and 500 responsible for the production of secondaries with momenta in the range under study here . moreover , the details of high energy hadronic interactions still lack theoretical understanding and there is little experimental data in the relevant energy and phase space regions @xcite . therefore the ground - level muon flux and charge ratio are widely used to tune or verify the parameters of atmospheric cascade calculations @xcite . currently these calculations are of great interest , as they predict the absolute atmospheric neutrino fluxes @xcite which are needed to interpret the observed muon neutrino flux deficit @xcite and to evaluate the backgrounds for neutrino astronomy . the muon flux and charge ratio have been extensively studied with different experimental techniques @xcite . however , results show discrepancies of about 10@xmath420% with respect to each other , which exceed the uncertainties assigned to the individual measurements and thus indicate the presence of systematic effects not accounted for . here a new measurement of the atmospheric muon flux is presented using the precise muon spectrometer of the l3 detector located at the lep collider at cern , near geneva , switzerland . special attention is given to the precise determination of all relevant detector and environmental parameters needed to convert the raw - data distributions into an absolute surface level flux . the large statistics available permits extensive studies of the residual systematic uncertainties . the momentum distribution of atmospheric muons is measured with the upgraded l3 setup of the l3 detector @xcite known as l3+c @xcite . the parts of the detector used in this analysis are sketched in figure [ fig : l3c ] . after passing through the stratified rock overburden , called `` molasse '' , the arrival time @xmath5 of a muon is measured with a resolution of 1.7 ns by a 202 m@xmath2 scintillator array placed on top of the l3 detector . the array is composed of 34 modules , each read out by two photomultipliers in coincidence to reduce noise . inside a volume of about 1000 m@xmath6 with a magnetic field of 0.5 t , the coordinates and slopes of a muon track are measured in up to six drift chambers in the bending plane and up to eight times in the non - bending plane . these chambers are arranged concentrically around the lep beam line in two groups of eight octants , each containing three layers of drift cells . by subtracting the @xmath5 time from the arrival times of the drift electrons at the sense wires , a track position in each chamber can be reconstructed with a precision of about 60 @xmath7 m in the bending plane and 1 mm in the non - bending plane . only three points are needed to determine the radius of the track in the magnetic field , therefore the momentum of a muon traversing two octants can be measured twice . this redundancy is used to evaluate the detector efficiencies and the resolution of the apparatus . the best resolution is obtained when fitting the six points together over the full track length of 11 meters . the multiple scattering and energy loss inside the l3 inner detectors , as well as the effect of the inhomogeneous magnetic field are taken into account using the procedure proposed in reference @xcite . equipped with a trigger and data acquisition system independent of the normal l3 data - taking , l3+c recorded @xmath8 atmospheric muon triggers during its operation in the years 1999 and 2000 . the l3+c experiment was located 450 m above sea level at a longitude of 6.02@xmath9 e and a latitude of 46.25@xmath9 n. for vertically incident muons , the mean energy loss in the molasse ( @xmath10= 6854 g@xmath11 ) and the magnet ( @xmath10=1227 g@xmath11 ) is 19 at low momenta and reaches 57 at 1 . the geometrical acceptance of the l3+c detector and the stochastic energy loss in the molasse overburden are evaluated using the following simulation procedure : monte carlo events are generated on the surface using a parameterization of the zenith angle and momentum dependence of the muon spectrum as obtained with the ` corsika ` @xcite program . these simulated muons are then tracked through a ` geant ` @xcite model of the l3+c environment which includes the molasse , access shafts and the concrete structures around the cavern which hosts the apparatus . finally , the detector response is simulated with a detailed ` geant ` description of the l3 detector . the generated detector signals are reconstructed with the same program used for the data . in total @xmath12 reconstructed monte carlo events are used in this analysis . the data analysis is restricted to events with three position measurements in at least one octant , a scintillator hit and good running conditions during data - taking . a total of @xmath13 reconstructed muon tracks are retained . the shielding of the 30 m of molasse overburden absorbs most of the charged air - shower particles other than muons . the number of muons produced in @xmath14 collisions by lep is negligible compared to the flux of atmospheric muons . therefore no background rejection is needed . the data selection focuses on two topics . firstly , fiducial volume cuts are defined to assure a good description of the data by the simulation . secondly , selection cuts are imposed on the track quality to enhance the momentum and angular resolution . these selection criteria are : * the muon track positions must be measured in six layers in the bending plane . * the momentum resolution , calculated from the quality of the track position measurements , should not exceed its nominal value by more than 50% . * at least four position measurements ( two in each octant ) should be present in the non - bending plane . * the @xmath15 of a fit of the tracks to a circle within an octant must satisfy @xmath16 . * the difference between the two photomultiplier time measurements from the same scintillator module must be below 8 ns . after these cuts , @xmath17 data events remain for the muon spectrum analysis . + the selection efficiencies depend on the charge , momentum and direction of the muon . for muons with momenta above 80 , the average efficiency is 7.6% for the fiducial volume cut and 33.3% for the quality selection . + 4 % of the raw events are multi - muon events . each individual muon is counted as an input to the spectrum data . the single - octant resolution is inferred directly from the data by comparing the two independent curvature measurements of muons traversing two octants . an example of the curvature difference is shown in figure [ fig : resoexa ] . these measurements are used to tune the detector simulation , from which the resolution of the full fit is determined . the relative momentum resolution @xmath18 as a function of momentum at the detector - level is shown in figure [ fig : reso](a ) . and @xmath19 . the fraction of events with width @xmath20 is denoted by @xmath21 . ] the maximum detectable momentum of the spectrometer , defined as the momentum at which @xmath18 reaches unity , is 0.78 for muons measured in only one octant and about 5 for muons measured in two octants . the efficiency of each subdetector is studied by exploiting redundancies in the measurement process . for about 50% of the tracks , the muon arrival time is also deduced from the muon chambers . these tracks are used to determine the scintillator efficiencies as a function of time and position on the array . a mean efficiency of 95.6% is found at the start of data - taking decreasing continuously to 94.5% towards the end of 2000 . the possibility of reconstructing a muon within a single octant is used to scan the drift - layer performance of the facing octant . on average , a fraction of 10.5% of the drift cells are found to have an efficiency lower than 80% . these regions are excluded in both the data and monte carlo reconstruction . under these conditions the trigger efficiency is determined from redundant trigger classes to be 99.85% on average . during data - taking , the total effective running time was continuously measured with a 10 mhz live - time counter , which is disabled whenever the trigger system is not ready to accept new data . in addition , each second an external trigger signal was sent to the l3+c trigger system . the number of these external triggers on tape compared to the total number of running seconds gives another estimate of the effective running time and agrees with the value from the live - time counter within 0.02% . using the possibility to measure a muon independently in two detector parts , the selection efficiencies are determined in the following way : the detector is subdivided into two hemispheres , @xmath22 and @xmath23 , and the conditional hemisphere selection probabilities @xmath24 are measured for data and monte carlo separately as a function of the muon charge @xmath25 , momentum @xmath26 and zenith angle @xmath27 . in the absence of correlated inefficiencies , the total selection efficiency for accepting a track in the two hemispheres is given by the product @xmath28 . the ratio @xmath29 is used to correct the differences between data and monte carlo . depending on the zenith angle range and the data - taking year , @xmath21 varies from 0.84 to 0.90 . a large fraction of this correction factor originates from a defect in the tdcs used to read out the muon chambers , giving rise to an 8% inefficiency for the full track selection . the relation between the momentum distribution measured in l3+c and the muon surface spectrum is given by @xmath30 here @xmath31 is the vector of events @xmath32 with measured momenta between @xmath33 $ ] . the effective live - time is given by @xmath34 and @xmath35 denotes the migration matrix , _ i.e. _ the conditional probability of measuring a momentum @xmath36 given a surface momentum @xmath37 . @xmath38 is the diagonal matrix of geometrical acceptances as a function of the surface momentum and @xmath39 is the diagonal matrix of detector efficiencies as a function of momentum at the detector - level . the vector @xmath40 contains the true surface spectrum integrated over a surface momentum bin . the complete detector matrix , @xmath41 , is evaluated from the measured detector efficiencies and the detector simulation as follows : @xmath42 where @xmath43 is the surface area used in the monte carlo generator , @xmath44 the solid angle of the zenith bin under study , @xmath24 includes the scintillator and trigger efficiencies and @xmath45 is the selection efficiency correction discussed above . @xmath46 denotes the number of selected monte carlo events found within a detector - level momentum bin @xmath22 , which were generated within the momentum bin @xmath23 at the surface , and @xmath47 is the total number of monte carlo events generated within this surface momentum bin . the effective acceptance of this analysis is calculated by summing over the columns of the detector matrix , which yields the geometrical factor for a muon being registered in the detector and fulfilling the selection cuts . the product of the effective acceptance and the live - time gives the total exposure , shown in figure [ fig : reso](b ) for positive and negative muons as a function of surface momentum . it rapidly decreases at low energies due to the momentum cut - off caused by the molasse overburden . below 200 , positive and negative muons have different acceptances , because the magnetic field bends their tracks in opposite directions and correspondingly into different detector regions . at large momenta the acceptance decreases with the performance of the full detector fit and a more difficult reconstruction caused by the increasing production of delta rays . the measurement equation ( [ eq : spec ] ) is solved using the least squares method , by minimizing @xmath48 where @xmath49 contains the statistical errors of the data and the detector matrix : @xmath50\ , m_j^2}\;. \label{eq : newerr}\ ] ] in the first step , the statistical monte carlo variances @xmath51 $ ] are set to zero , such that equation ( [ eq : lsqchi2 ] ) becomes linear with respect to the surface spectrum @xmath52 and its solution is @xmath53 with covariance matrix @xmath54=\left(\mathbf{d}^t \mathbf{w } \mathbf{d}\right)^{-1}\ , . \label{eq : vm}\ ] ] here @xmath55 denotes the diagonal weight matrix containing the statistical errors of the data and the monte carlo , @xmath56 . the minimization of equation ( [ eq : lsqchi2 ] ) is then repeated using the solution @xmath57 of the previous iteration for the calculation of the errors in equation ( [ eq : newerr ] ) . this process is repeated until the maximum relative difference to the result of the previous iteration is below @xmath58 . typically four iterations are needed . uncertainties on the live - time and the trigger and scintillator efficiencies give rise to a normalization uncertainty of 0.7% . the uncertainty of the detector acceptance is assessed in three studies : first , the results obtained for statistically independent data subsamples , as for instance the data collected in 1999 and 2000 or in different detector parts , are compared . second , the muon flux and charge ratio are measured as a function of the azimuthal angle . at large momenta , geomagnetic effects and the variation of the molasse overburden are not important , and therefore a flat distribution is expected . finally , the stability of the measured flux and charge ratio with respect to a variation of the selection criteria is investigated . from these studies , additional normalization uncertainties in the absolute muon flux are derived . these range from 1.7% to 3.7% depending on the zenith angle . for the charge ratio normalization uncertainties between 1.0% and 2.3% are estimated . figure [ fit : rvst ] shows an example of the stability of the measured muon flux within the two data - taking years for zenith angles between @xmath59 and @xmath60 degrees . since the muon production is known to change with atmospheric conditions @xcite , the observed variation of the muon flux is compared to an air - shower simulation with the ` target ` @xcite program using atmospheric density profiles measured in balloon flights close to the experiment @xcite . good overall agreement between data and monte carlo is observed . however , the full comparison to the rates in 26 weeks and 14 momentum bins yields a @xmath61 of 526/364 . the assumption that this large value is caused by detector inefficiencies not accounted for , leads to an additional normalization uncertainty of 0.3% , which is well within the above estimated uncertainties . + comparison of data and monte carlo is also shown . ] due to the steepness of the muon spectrum , even small uncertainties in the absolute momentum scale can introduce a considerable bias in the muon flux measurement . the uncertainty on the l3 magnetic field strength introduces a momentum scale bias of less than 0.4% @xcite . furthermore , the momentum measurement is subject to uncertainties of the detector alignment . a systematic shift of the chamber positions may introduce a constant offset @xmath62 . the measurement of the curvature , @xmath63 , and the alignment related momentum scale uncertainty , @xmath64 is given by @xmath65 and depends on the muon charge . within one octant , the alignment is measured by an optical alignment system @xcite with a precision corresponding to 0.19 @xmath66 @xcite . the relative alignment of the muon chamber octants , relevant for this analysis , is determined from the data itself with a precision between 0.075 and 0.152 @xmath66 @xcite , depending on the zenith angle . the uncertainty due the molasse overburden affects the conversion of the measured flux at the detector to the surface . the results of two survey drillings at different locations close to l3+c provide an absolute measurement of the l3+c matter overburden . the influence of molasse inhomogeneities and of surface installations not included in the l3+c simulation is estimated by studying the variance of the muon flux as a function of the azimuthal angle near the momentum threshold . this leads to an uncertainty of the average rock density of 2% , which is equivalent to an energy loss uncertainty of 0.4 in the vertical direction . good agreement between the muon energy - loss calculation used here @xcite and other approaches @xcite is found . the residual differences correspond to a momentum scale uncertainty below or less than 0.3% in the vertical direction . the relative momentum scale uncertainties for vertically incident muons are displayed in figure [ fig : scsys](a ) . at low energies the molasse uncertainty contributes the most , whereas above 100 the alignment uncertainties dominate . [ fig : crats ] the limited monte carlo statistics affects the precision of the detector matrix @xmath67 . below 200 , it dominates the total statistical uncertainty in the denominator of equation ( [ eq : lsqchi2 ] ) , contributing about 0.5% to the total uncertainty per zenith angle bin . in order to estimate the influence of the uncertainty of the momentum resolution on the measured muon flux , the minimization of equation ( [ eq : lsqchi2 ] ) is repeated with different detector matrices , for which the momentum resolution is altered by @xmath38% . this corresponds to the estimated uncertainty of its monte carlo prediction . as expected , no differences are found at low momenta . above 200 , the observed relative flux change @xmath68 is well described by @xmath69 with @xmath70 . the observed difference between the high - energy muon flux measured in different detector regions leads to a somewhat larger value of @xmath71 . the total uncertainties of the muon flux and charge ratio are obtained by adding the individual contributions in quadrature . the different sources of the vertical uncertainties are shown in figures [ fig : scsys](b ) and ( c ) . the muon flux uncertainty is dominated by the uncertainty of the molasse overburden at low momenta and by the alignment and resolution uncertainty at high momenta . the minimal uncertainty is 2.3% at 150 in the vertical direction . the vertical charge ratio uncertainty is below 2% up to momenta of 100 . above this momentum , it rises rapidly with the alignment uncertainties . these uncertainties are fully correlated between different momenta for a given zenith angle bin . as approximately the same detector parts are used to measure the muons in neighboring zenith angles , the systematic uncertainties are also correlated with respect to the zenith angle . the estimated correlation coefficients are listed in table [ tab : zsyscorr ] . the understanding of the detector is validated by analyzing the muons produced at lep via the process @xmath72 recorded during the lep calibration runs at a mean centre - of - mass energy of 91.27 . the selection criteria include the requirement of a muon track close to the collision point and an event - time in coincidence with the lep beam crossing time . the number of selected muons with a momentum above 60% of the beam energy is converted to an absolute cross section resulting in @xmath73 here the quoted systematic uncertainty includes only sources which are not relevant to the muon spectrum measurement , such as the luminosity . + using the lep precision measurements @xcite , the standard model expectation of @xmath74 is calculated @xcite to be @xmath75 which is in excellent agreement to the value measured here . + thus this study verifies the l3+c acceptance calculation and a normalization uncertainty of @xmath765.2% at 68% c.l . can be stated . although this number is larger than the estimated systematic uncertainty of the muon flux normalization , it provides an absolute systematic cross - check qualitatively different from the relative studies described above . + the momentum distribution of the selected events , displayed in figure [ fig : z0e ] , shows good agreement between the data and the simulation . from the peak position of the data , an absolute momentum scale uncertainty of @xmath76370 and a single octant alignment uncertainty of @xmath760.1 @xmath66 is derived . + as can be seen in figure figure [ fig : reso](a ) , the momentum resolution derived from the @xmath77 muon sample agrees well with the one measured with atmospheric muons . ) . the arrow indicates the low momentum cut . ] the muon fluxes , @xmath78 , conventionally multiplied by the third power of the momentum , and the charge ratios , @xmath79 , are listed for each zenith angle bin in tables [ tab : flux10]@xmath4[tab : flux03 ] with their statistical and systematic uncertainties . the statistical correlation coefficients , @xmath80 , between neighboring momentum bins , as derived from equation ( [ eq : vm ] ) , are also given . due to the limited detector resolution these correlations are inevitable . however , the momentum binning is chosen such that only neighboring bins have a significant correlation . the average momenta , @xmath81 , within a momentum range @xmath82 $ ] are calculated @xcite by fitting the phenomenological muon flux function from reference @xcite to our data and solving @xmath83 it should be noted that the fluxes are neither corrected for the altitude of l3+c nor for the atmospheric profile ( bottom ) to 58@xmath84 ( top ) . the inner bars denote the statistical uncertainty , the full bars show the total uncertainty . for better visibility , an offset of 0.05 @xmath2@xmath11s@xmath66sr@xmath66 was added consecutively and lines are shown to guide the eye . ] to avoid additional theoretical uncertainties . instead , we quote the average atmospheric mass overburden @xmath10 above l3+c , which was continuously measured with balloon flights from close to the experiment to altitudes of over 30 km @xcite . the parameterization of reference @xcite is used to describe the mass profile @xmath10 in g@xmath11 as a function of the altitude @xmath85 in km above sea level : @xmath86 a fit to the live - time weighted balloon data yields @xmath87 , @xmath88 , @xmath89 , @xmath90 and @xmath91 . the measured muon fluxes at the l3+c altitude are shown in figure [ fig : spectrum ] for each zenith angle bin . as no previous continuous zenith angle measurements exist in the large energy range examined here , only the vertical flux can be compared to other experiments , as shown in figure [ fig : vertflux ] . only measurements providing an absolute normalization @xcite are taken into account . the data are extrapolated to sea level using the muon flux predictions of the ` target ` @xcite program . the comparison to low energy experiments @xcite gives a good overall agreement with this analysis above about 40 . at lower momenta , a systematic slope difference seems to be present , which corresponds to about three standard deviations of the systematic molasse uncertainty estimated above . only three previous experiments measured a normalized spectrum at high energies . the shape of the kiel measurements @xcite agree with this result over the full momentum range , but a lower flux normalization is determined by l3+c . + the data obtained with the mars apparatus @xcite significantly disagree with this result , both in shape and normalization . + above momenta of 50 , the recent muon flux measurement from bess - tev @xcite is in very good agreement with this result . the measured charge ratios at the l3+c altitude are shown in figure [ fig : crat ] for each zenith angle bin up to momenta of 500 . in the considered momentum range , the charge ratio is independent of the momentum within the experimental uncertainties . the mean value in the vertical direction is found to be 1.285 @xmath3 0.003 ( stat . ) @xmath3 0.019 ( sys . ) with a @xmath929.5/11 . this is in good agreement with the average of all previous measurements , 1.270 @xmath3 0.003 ( stat . ) @xmath3 0.015 ( sys . ) it is worth noting , that the precision of the data of a single l3+c zenith angle bin is comparable to the combined uncertainty of all data collected in the past . ( bottom ) to 58@xmath84 ( top ) . the inner bars denote the statistical uncertainty , the full bars show the total uncertainty . for better visibility , an offset of 0.5 was added consecutively and lines are shown to guide the eye . ] we wish to acknowledge the contribution of all the engineers and technicians who have participated in the construction and the maintenance of the l3 and l3+c experiments . those of us who are not from member states thank cern for its hospitality and help . furthermore , we would like to thank the bartol group for providing the ` target ` program . + t. k. gaisser , m. honda , p. lipari and t. stanev , in proceedings of the 27th icrc ( hamburg , 7 - 15 aug , 2001 ) 1643 , and references therein . r. engel , t. k. gaisser and t. stanev , phys . b * 472 * ( 2000 ) 113 . m. honda , t. kajita , k. kasahara and s. midorikawa , phys . d * 52 * ( 1995 ) 4985 . v. agrawal , t. k. gaisser , p. lipari and t. stanev , phys . d * 53 * ( 1996 ) 1314 . e. v. bugaev _ et al . _ , phys . d * 58 * ( 1998 ) 054001 . g. fiorentini , v. a. naumov and f. l. villante , phys . b * 510 * ( 2001 ) 173 . t. k. gaisser and m. honda , ann . nucl . part . * 52 * ( 2002 ) 153 , and references therein . super - kamiokande collaboration , y. fukuda _ et al . _ , phys . * 81 * ( 1998 ) 1562 . soudan 2 collaboration , w. w. m. allison _ et al . _ , phys . b * 391 * ( 1997 ) 491 . macro collaboration , m. ambrosio _ et al . _ , phys . b * 434 * ( 1998 ) 451 . t. hebbeker and c. timmermans , astropart . phys . * 18 * ( 2002 ) 107 . l3 collaboration , b. adeva _ et al . _ , nucl . instrum . a * 289 * ( 1990 ) 35 . l3+c collaboration , o. adriani _ et al . _ , nucl . instrum . a * 488 * ( 2002 ) 209 . v. innocente and e. nagy , nucl . instrum . a * 324 * ( 1993 ) 297 . d. heck _ et al . _ , `` corsika : a monte carlo code to simulate extensive air showers , '' forschungszentrum karlsruhe , fzka 6019 ( 1998 ) . r. brun _ et al . _ , `` geant 3 , '' cern report dd / ee/84 - 1 ( 1987 ) . s. bottai and l. perrone , nucl . instrum . meth . a * 459 * ( 2001 ) 319 . p. h. barrett _ et al . _ , rev . * 24 * ( 1952 ) 133 . service de climatologie , geneva . c. brouwer _ _ , nucl . instrum . a * 313 * ( 1992 ) 50 . p. duinker _ et al . _ , nucl . instrum . meth . a * 273 * ( 1988 ) 814 . m. fabre , `` the dimuon mass resolution of the l3 experiment at lep , '' ph.d . thesis , eth zrich , 1992 . m. unger , `` measurement of the momentum spectrum of atmospheric muons with the l3 detector , '' ph.d . thesis , humboldt universitt berlin , 2004 , desy - thesis-2004 - 008 . w. lohmann , r. kopp and r. voss , `` energy loss of muons in the energy range 110000 gev , '' cern report 85 - 03 ( 1985 ) . d. e. groom , n. v. mokhov and s. i. striganov , atom . data nucl . data tabl . * 78 * ( 2001 ) 183 . d. chirkin and w. rhode , `` muon monte carlo : a new high precision tool for muon propagation through matter , '' desy - proc-2002 - 01 ( 2002 ) . particle data group collaboration , k. hagiwara _ et al . _ , phys . d * 66 * , ( 2002 ) 010001 , and references therein . d. y. bardin _ et al . _ , comput . phys . commun . * 133 * ( 2001 ) 229 . s. jadach , b. f. l. ward and z. was , comput . * 79 * ( 1994 ) 503 . t. sjstrand , comput . * 82 * ( 1994 ) 74 . k. maeda , fortsch . * 21 * ( 1973 ) 113 . o. c. allkofer , k. carstensen and d. w. dau , phys . b * 36 * ( 1971 ) 425 . c. a. ayre _ et al . _ , j. phys . g * 1 * ( 1975 ) 584 . b. j. bateman _ et al . _ , phys . b * 36 * ( 1971 ) 144 . j. green _ et al . _ , d * 20 * ( 1979 ) 1598 . de pascale _ et al . _ , j. geophys . * 98 * ( 1993 ) 3501 . caprice collaboration , j. kremer _ et al . _ , phys . * 83 * ( 1999 ) 4241 . caprice collaboration , m. boezio _ et al . _ , phys . d * 67 * ( 2003 ) 072003 . bess collaboration , t. sanuki _ et al . _ , phys . b * 541 * ( 2002 ) 234 [ erratum - ibid . b * 581 * ( 2004 ) 272 ] . bess collaboration , s. haino _ et al . _ , `` measurements of primary and atmospheric cosmic - ray spectra with the bess - tev spectrometer '' [ arxiv : astro - ph/0403704 ] . p.achard o.adriani m.aguilar-benitezm.van den akker j.alcaraz g.alemannij.allabya.aloisio m.g.alviggih.anderhub v.p.andreevf.anselmoa.arefiev t.azemoon t.aziz p.bagnaiaa.bajo g.baksayl.baksayj.bhrs.v.baldew s.banerjee sw.banerjee a.barczyk r.barillre p.bartalini m.basilen.batalovar.battistona.bay f.becattiniu.beckerf.behnerl.bellucci r.berbeco j.berdugo p.berges b.bertuccib.l.betevm.biasinim.bigliettia.biland j.j.blaising s.c.blyth g.j.bobbink a.bhml.boldizsarb.borgia s.bottaid.bourilkovm.bourquins.braccinij.g.bransonf.brochu j.d.burgerw.j.burgerx.d.cai m.capellg.cara romeog.carlinoa.cartacci j.casausf.cavallarin.cavallo c.cecchi m.cerradam.chamizot.chiarusiy.h.chang m.chemarina.chen g.chen g.m.chen h.f.chen h.s.cheng.chiefari l.cifarellif.cindoloi.clarer.clare g.coignet n.colino s.costantini b.de la cruzs.cucciarellij.a.van dalen r.de asmundisp.dglon j.debreczenia.degr k.dehmeltk.deiters d.della volpe e.delmeire p.denes f.denotaristefania.de salvo m.diemoz m.dierckxsens l.k.dingc.dionisi m.dittmara.doriam.t.dovad.duchesneau m.dudai.duranb.echenarda.elinea.el hageh.el mamounia.engler f.j.eppling p.extermanng.faber m.a.falagans.falcianoa.favaraj.fay o.fedinm.felcinit.ferguson h.fesefeldt e.fiandrinij.h.field f.filthautw.fisheri.fiskg.forconi k.freudenreichc.furettayu.galaktionovs.n.ganguli p.garcia-abiam.gataullins.gentiles.giaguz.f.gongh.j.graboschg.grenier o.grimmh.groenstege m.w.gruenewald m.guida y.n.guos.guptav.k.gupta a.gurtul.j.gutayd.haasch.hallerd.hatzifotiadouy.hayashiz.x.het.hebbekera.herv j.hirschfelderh.hoferh.hofer,jun . m.hohlmanng.holzner s.r.houa.x.huoy.hun.ito b.n.jin c.l.jing l.w.jonesp.de jongi.josa-mutuberrav.kantserov m.kaurs.kawakamim.n.kienzle-focaccij.k.kimj.kirkbyw.kittela.klimentov a.c.knige.koka.kornm.kopalv.koutsenko m.krberh.h.kuang r.w.kraemera.krgerj.kuijpers a.kunin p.ladron de guevarai.laktinehg.landim.lebeaua.lebedevp.lebrunp.lecomte p.lecoq p.le coultre j.m.le goffy.leih.leichr.leiste m.levtchenkop.levtchenkoc.li l.liz.c.lis.likhoded c.h.linw.t.linf.l.lindel.listaz.a.liuw.lohmanne.longo y.s.lu c.luci l.luminariw.lustermannw.g.ma x.h.may.q.mal.malgeria.malinin c.maaj.mans j.p.martin f.marzano k.mazumdarr.r.mcneil s.melex.w.mengl.merola m.meschini w.j.metzgera.mihula.van milh.milcentg.mirabelli j.mnichg.b.mohanty b.monteleonig.s.muanzaa.j.m.muijsb.musicar m.musy s.nagyr.nahnhauerv.a.naumov s.natalem.napolitanof.nessi-tedaldih.newman a.nisatit.novakh.nowak r.ofierzynski g.organtinii.palc.palomaresp.paoluccir.paramatti j .- t.paulm.pauluzzic.pausf.paussm.pedaces.pensottid.perret-gallix b.petersend.piccolo f.pierella m.pierim.pioppip.a.pirou e.pistolesiv.plyaskin m.pohl v.pojidaevj.pothierd.prokofiev j.quartieric.r.qingg.rahal-callotm.a.rahaman p.raics n.rajar.ramelli p.g.rancoitar.ranieri a.raspereza k.c.ravindranp.razisd.ren m.rescignos.reucroftp.rewiersmas.riemannk.rilesb.p.roea.rojkovl.romero a.rosca c.rosemannc.rosenblecks.rosier-leess.rothj.a.rubio g.ruggiero h.rykaczewski r.saidia.sakharovs.saremi s.sarkarj.salicio e.sanchezc.schferv.schegelskyv.schmittb.schoeneichh.schopperd.j.schotanusc.sciaccal.servolic.q.shens.shevchenkon.shivarovv.shoutko e.shumilov a.shvorobd.sonc.sougap.spillantini m.steuerd.p.stickland b.stoyanova.straessnerk.sudhakarh.sulankeg.sultanovl.z.suns.sushkovh.suter j.d.swainz.szillasix.w.tangp.tarjanl.tauscherl.taylorb.tellili d.teyssier c.timmermanssamuel c.c.ting s.m.ting s.c.tonwar j.tth g.trowitzschc.tullyk.l.tungj.ulbricht m.ungere.valente h.verkooijenr.t.van de waller.vasquezv.veszpremig.vesztergombii.vetlitsky d.vicinanza g.viertel s.villam.vivargent s.vlachosi.vodopianov h.vogelh.vogt i.vorobiev a.a.vorobyov m.wadhwar.g.wangq.wangx.l.wangx.w.wang z.m.wangm.weberr.van wijkt.a.m.wijnenh.wilkenss.wynhoff l.xia y.p.xuj.s.xuz.z.xu j.yamamoto b.z.yang c.g.yang h.j.yangm.yangx.f.yangz.g.yaos.c.yeh z.q.yu an.zaliteyu.zalitec.zhangf.zhangj.zhangs.zhangz.p.zhang j.zhaos.j.zhoug.y.zhur.y.zhuh.l.zhuangq.q.zhua.zichichib.zimmermann m.zllera.n.m.zwart . 20.0&27.0 & 23.18 & 0.217 & 0.4&[1.5ex ] -0.24 & 4.4 & 1.274 & 0.7&[1.5ex ] -0.24 & 1.5 + 27.0&34.5 & 30.47 & 0.231 & 0.3&[1.5ex ] -0.22 & 3.7 & 1.284 & 0.6&[1.5ex ] -0.22 & 1.5 + 34.5&42.0 & 38.02 & 0.244 & 0.4&[1.5ex ] -0.24 & 3.3 & 1.295 & 0.8&[1.5ex ] -0.24 & 1.5 + 42.0&50.0 & 45.78 & 0.252 & 0.5&[1.5ex ] -0.26 & 3.0 & 1.269 & 0.9&[1.5ex ] -0.26 & 1.5 + 50.0&58.5 & 54.04 & 0.257 & 0.5&[1.5ex ] -0.31 & 2.8 & 1.286 & 1.0&[1.5ex ] -0.31 & 1.5 + 58.5&68.5 & 63.25 & 0.261 & 0.5&[1.5ex ] -0.31 & 2.6 & 1.298 & 1.0&[1.5ex ] -0.31 & 1.5 + 68.5&81.5 & 74.63 & 0.266 & 0.5&[1.5ex ] -0.30 & 2.5 & 1.273 & 1.1&[1.5ex ] -0.29 & 1.5 + 81.5&100 & 90.13 & 0.262 & 0.6&[1.5ex ] -0.26 & 2.4 & 1.293 & 1.2&[1.5ex ] -0.26 & 1.5 + 100&132 & 114.5 & 0.264 & 0.5&[1.5ex ] -0.22 & 2.3 & 1.286 & 1.0&[1.5ex ] -0.21 & 1.6 + 132&200 & 161.3 & 0.250 & 0.5&[1.5ex ] -0.22 & 2.2 & 1.287 & 1.1&[1.5ex ] -0.22 & 2.1 + 200&300 & 243.0 & 0.233 & 0.8&[1.5ex ] -0.27 & 2.2 & 1.327 & 1.7&[1.5ex ] -0.27 & 3.7 + 300&500 & 381.9 & 0.203 & 1.2&[1.5ex ] -0.28 & 2.4 & 1.276 & 2.6 & & 7.2 + 500&1000 & 687.2 & 0.151 & 2.3&[1.5ex ] -0.30 & 3.5 + 1000&3000 & 1599 & 0.087 & 6.0 & & 8.4 + 27.0&34.5 & 30.47 & 0.237 & 0.3&[1.5ex ] -0.23 & 4.5 & 1.265 & 0.6&[1.5ex ] -0.22 & 2.3 + 34.5&42.0 & 38.02 & 0.254 & 0.4&[1.5ex ] -0.24 & 4.1 & 1.261 & 0.8&[1.5ex ] -0.24 & 2.3 + 42.0&50.0 & 45.78 & 0.259 & 0.5&[1.5ex ] -0.26 & 3.8 & 1.265 & 1.0&[1.5ex ] -0.26 & 2.3 + 50.0&58.5 & 54.04 & 0.264 & 0.5&[1.5ex ] -0.30 & 3.7 & 1.297 & 1.1&[1.5ex ] -0.30 & 2.3 + 58.5&68.5 & 63.25 & 0.266 & 0.6&[1.5ex ] -0.31 & 3.5 & 1.250 & 1.1&[1.5ex ] -0.31 & 2.3 + 68.5&81.5 & 74.63 & 0.270 & 0.6&[1.5ex ] -0.29 & 3.4 & 1.319 & 1.2&[1.5ex ] -0.29 & 2.3 + 81.5&100 & 90.13 & 0.271 & 0.6&[1.5ex ] -0.25 & 3.3 & 1.259 & 1.3&[1.5ex ] -0.25 & 2.3 + 100&132 & 114.5 & 0.270 & 0.5&[1.5ex ] -0.21 & 3.2 & 1.296 & 1.1&[1.5ex ] -0.21 & 2.5 + 132&200 & 161.3 & 0.262 & 0.6&[1.5ex ] -0.21 & 3.1 & 1.281 & 1.2&[1.5ex ] -0.21 & 3.0 + 200&300 & 243.0 & 0.241 & 0.9&[1.5ex ] -0.27 & 3.1 & 1.273 & 1.8&[1.5ex ] -0.27 & 4.8 + 300&500 & 382.1 & 0.219 & 1.3&[1.5ex ] -0.27 & 3.2 & 1.389 & 2.7 & & 8.3 + 500&1000 & 687.8 & 0.165 & 2.4&[1.5ex ] -0.30 & 4.2 + 1000&3000 & 1604 & 0.093 & 6.5 & & 9.2 + 27.0&34.5 & 30.47 & 0.245 & 0.4&[1.5ex ] -0.24 & 4.3 & 1.265 & 0.7&[1.5ex ] -0.24 & 2.3 + 34.5&42.0 & 38.02 & 0.257 & 0.4&[1.5ex ] -0.26 & 3.9 & 1.296 & 0.9&[1.5ex ] -0.26 & 2.3 + 42.0&50.0 & 45.78 & 0.265 & 0.5&[1.5ex ] -0.28 & 3.6 & 1.302 & 1.1&[1.5ex ] -0.28 & 2.3 + 50.0&58.5 & 54.04 & 0.273 & 0.5&[1.5ex ] -0.31 & 3.4 & 1.267 & 1.1&[1.5ex ] -0.31 & 2.3 + 58.5&68.5 & 63.25 & 0.279 & 0.5&[1.5ex ] -0.31 & 3.2 & 1.287 & 1.1&[1.5ex ] -0.30 & 2.3 + 68.5&81.5 & 74.64 & 0.282 & 0.5&[1.5ex ] -0.27 & 3.1 & 1.272 & 1.1&[1.5ex ] -0.27 & 2.3 + 81.5&100 & 90.14 & 0.283 & 0.5&[1.5ex ] -0.23 & 3.0 & 1.288 & 1.1&[1.5ex ] -0.23 & 2.4 + 100&132 & 114.5 & 0.284 & 0.5&[1.5ex ] -0.19 & 2.9 & 1.303 & 0.9&[1.5ex ] -0.19 & 2.6 + 132&200 & 161.3 & 0.274 & 0.5&[1.5ex ] -0.19 & 2.8 & 1.325 & 1.0&[1.5ex ] -0.19 & 3.8 + 200&300 & 243.0 & 0.255 & 0.7&[1.5ex ] -0.23 & 2.8 & 1.300 & 1.4&[1.5ex ] -0.23 & 6.8 + 300&500 & 382.0 & 0.220 & 1.1&[1.5ex ] -0.25 & 2.9 & 1.437 & 2.2 & & 12.0 + 500&1000 & 687.3 & 0.172 & 1.9&[1.5ex ] -0.29 & 4.2 + 1000&3000 & 1599 & 0.091 & 5.7 & & 11.3 + 34.5&42.0 & 38.02 & 0.263 & 0.4&[1.5ex ] -0.28 & 3.9 & 1.248 & 0.8&[1.5ex ] -0.28 & 1.3 + 42.0&50.0 & 45.78 & 0.269 & 0.5&[1.5ex ] -0.29 & 3.5 & 1.288 & 1.1&[1.5ex ] -0.29 & 1.3 + 50.0&58.5 & 54.04 & 0.274 & 0.5&[1.5ex ] -0.32 & 3.3 & 1.280 & 1.1&[1.5ex ] -0.32 & 1.3 + 58.5&68.5 & 63.25 & 0.284 & 0.6&[1.5ex ] -0.31 & 3.1 & 1.285 & 1.2&[1.5ex ] -0.31 & 1.3 + 68.5&81.5 & 74.64 & 0.280 & 0.6&[1.5ex ] -0.28 & 2.9 & 1.274 & 1.2&[1.5ex ] -0.28 & 1.3 + 81.5&100 & 90.14 & 0.288 & 0.6&[1.5ex ] -0.24 & 2.8 & 1.302 & 1.3&[1.5ex ] -0.24 & 1.3 + 100&132 & 114.5 & 0.290 & 0.5&[1.5ex ] -0.20 & 2.6 & 1.280 & 1.1&[1.5ex ] -0.20 & 1.4 + 132&200 & 161.3 & 0.279 & 0.6&[1.5ex ] -0.20 & 2.6 & 1.313 & 1.1&[1.5ex ] -0.20 & 2.0 + 200&300 & 243.1 & 0.264 & 0.8&[1.5ex ] -0.27 & 2.5 & 1.301 & 1.7&[1.5ex ] -0.27 & 4.7 + 300&500 & 382.1 & 0.232 & 1.3&[1.5ex ] -0.30 & 2.6 & 1.318 & 2.6 & & 11.3 + 500&1000 & 686.8 & 0.167 & 2.5&[1.5ex ] -0.37 & 4.2 + 1000&3000 & 1587 & 0.087 & 7.9 & & 10.4 + 34.5&42.0 & 38.02 & 0.255 & 0.5&[1.5ex ] -0.32 & 4.6 & 1.279 & 1.0&[1.5ex ] -0.32 & 1.0 + 42.0&50.0 & 45.78 & 0.265 & 0.5&[1.5ex ] -0.32 & 4.2 & 1.284 & 1.1&[1.5ex ] -0.32 & 1.0 + 50.0&58.5 & 54.04 & 0.275 & 0.6&[1.5ex ] -0.35 & 3.9 & 1.263 & 1.1&[1.5ex ] -0.35 & 1.0 + 58.5&68.5 & 63.25 & 0.283 & 0.6&[1.5ex ] -0.33 & 3.7 & 1.261 & 1.2&[1.5ex ] -0.33 & 1.0 + 68.5&81.5 & 74.64 & 0.283 & 0.6&[1.5ex ] -0.29 & 3.6 & 1.271 & 1.2&[1.5ex ] -0.29 & 1.0 + 81.5&100 & 90.14 & 0.290 & 0.6&[1.5ex ] -0.26 & 3.4 & 1.281 & 1.2&[1.5ex ] -0.26 & 1.0 + 100&132 & 114.6 & 0.290 & 0.5&[1.5ex ] -0.20 & 3.3 & 1.256 & 1.0&[1.5ex ] -0.20 & 1.1 + 132&200 & 161.3 & 0.292 & 0.5&[1.5ex ] -0.22 & 3.2 & 1.280 & 1.1&[1.5ex ] -0.22 & 1.4 + 200&300 & 243.1 & 0.271 & 0.8&[1.5ex ] -0.28 & 3.2 & 1.297 & 1.6&[1.5ex ] -0.28 & 3.1 + 300&500 & 382.2 & 0.244 & 1.2&[1.5ex ] -0.31 & 3.3 & 1.428 & 2.5 & & 7.4 + 500&1000 & 687.5 & 0.191 & 2.2&[1.5ex ] -0.36 & 4.0 + 1000&3000 & 1594 & 0.087 & 7.6 & & 8.4 + 34.5&42.0 & 38.02 & 0.253 & 0.6&[1.5ex ] -0.38 & 4.6 & 1.273 & 1.2&[1.5ex ] -0.38 & 1.8 + 42.0&50.0 & 45.78 & 0.265 & 0.6&[1.5ex ] -0.35 & 4.1 & 1.292 & 1.1&[1.5ex ] -0.35 & 1.8 + 50.0&58.5 & 54.04 & 0.275 & 0.5&[1.5ex ] -0.38 & 3.7 & 1.263 & 1.1&[1.5ex ] -0.38 & 1.8 + 58.5&68.5 & 63.25 & 0.281 & 0.6&[1.5ex ] -0.34 & 3.4 & 1.273 & 1.2&[1.5ex ] -0.34 & 1.8 + 68.5&81.5 & 74.64 & 0.291 & 0.6&[1.5ex ] -0.31 & 3.1 & 1.263 & 1.2&[1.5ex ] -0.31 & 1.8 + 81.5&100 & 90.15 & 0.293 & 0.6&[1.5ex ] -0.27 & 2.9 & 1.260 & 1.2&[1.5ex ] -0.27 & 1.8 + 100&132 & 114.6 & 0.298 & 0.5&[1.5ex ] -0.22 & 2.7 & 1.266 & 1.0&[1.5ex ] -0.22 & 1.8 + 132&200 & 161.4 & 0.298 & 0.5&[1.5ex ] -0.23 & 2.5 & 1.281 & 1.0&[1.5ex ] -0.23 & 1.9 + 200&300 & 243.2 & 0.285 & 0.7&[1.5ex ] -0.29 & 2.4 & 1.267 & 1.6&[1.5ex ] -0.29 & 2.8 + 300&500 & 382.3 & 0.256 & 1.1&[1.5ex ] -0.31 & 2.6 & 1.394 & 2.5 & & 5.7 + 500&1000 & 688.2 & 0.203 & 2.0&[1.5ex ] -0.35 & 3.5 + 1000&3000 & 1601 & 0.120 & 5.3 & & 7.8 + 42.0&50.0 & 45.79 & 0.256 & 0.6&[1.5ex ] -0.41 & 4.7 & 1.262 & 1.3&[1.5ex ] -0.41 & 1.4 + 50.0&58.5 & 54.05 & 0.269 & 0.6&[1.5ex ] -0.42 & 4.3 & 1.284 & 1.2&[1.5ex ] -0.42 & 1.4 + 58.5&68.5 & 63.26 & 0.282 & 0.6&[1.5ex ] -0.36 & 4.0 & 1.261 & 1.3&[1.5ex ] -0.36 & 1.4 + 68.5&81.5 & 74.65 & 0.290 & 0.6&[1.5ex ] -0.31 & 3.7 & 1.277 & 1.3&[1.5ex ] -0.31 & 1.4 + 81.5&100 & 90.16 & 0.299 & 0.6&[1.5ex ] -0.26 & 3.5 & 1.267 & 1.2&[1.5ex ] -0.26 & 1.4 + 100&132 & 114.6 & 0.307 & 0.5&[1.5ex ] -0.22 & 3.3 & 1.287 & 1.0&[1.5ex ] -0.22 & 1.4 + 132&200 & 161.4 & 0.312 & 0.5&[1.5ex ] -0.24 & 3.1 & 1.274 & 1.0&[1.5ex ] -0.23 & 1.4 + 200&300 & 243.2 & 0.300 & 0.8&[1.5ex ] -0.32 & 3.0 & 1.278 & 1.6&[1.5ex ] -0.31 & 2.1 + 300&500 & 382.5 & 0.277 & 1.2&[1.5ex ] -0.35 & 3.1 & 1.317 & 2.4 & & 4.7 + 500&1000 & 688.5 & 0.231 & 2.0&[1.5ex ] -0.38 & 3.8 + 1000&3000 & 1602 & 0.117 & 6.3 & & 8.7 + 42.0&50.0 & 45.79 & 0.256 & 1.4&[1.5ex ] -0.49 & 5.6 & 1.281 & 2.8&[1.5ex ] -0.50 & 1.4 + 50.0&58.5 & 54.05 & 0.255 & 0.9&[1.5ex ] -0.45 & 5.3 & 1.269 & 1.7&[1.5ex ] -0.45 & 1.4 + 58.5&68.5 & 63.26 & 0.273 & 0.8&[1.5ex ] -0.38 & 5.1 & 1.290 & 1.6&[1.5ex ] -0.37 & 1.4 + 68.5&81.5 & 74.65 & 0.284 & 0.8&[1.5ex ] -0.32 & 4.9 & 1.280 & 1.5&[1.5ex ] -0.32 & 1.4 + 81.5&100 & 90.16 & 0.294 & 0.7&[1.5ex ] -0.25 & 4.7 & 1.278 & 1.4&[1.5ex ] -0.25 & 1.4 + 100&132 & 114.6 & 0.305 & 0.5&[1.5ex ] -0.22 & 4.6 & 1.258 & 1.1&[1.5ex ] -0.21 & 1.4 + 132&200 & 161.5 & 0.316 & 0.6&[1.5ex ] -0.22 & 4.5 & 1.254 & 1.2&[1.5ex ] -0.22 & 1.6 + 200&300 & 243.4 & 0.315 & 0.8&[1.5ex ] -0.30 & 4.4 & 1.285 & 1.7&[1.5ex ] -0.30 & 2.6 + 300&500 & 382.7 & 0.297 & 1.3&[1.5ex ] -0.33 & 4.4 & 1.332 & 2.6 & & 5.6 + 500&1000 & 689.3 & 0.243 & 2.3&[1.5ex ] -0.36&4.8 + 1000&3000 & 1604 & 0.153 & 5.9 & & 9.4 +

recently , the author and collaborators have developed a systematic program for proving the existence of homoclinic orbits in partial differential equations . two typical forms of homoclinic orbits thus obtained are : ( 1 ) . transversal homoclinic orbits , ( 2 ) . silnikov homoclinic orbits . around the transversal homoclinic orbits in infinite dimensional autonomous systems , the author was able to prove the existence of chaos through a shadowing lemma . around the silnikov homoclinic orbits , the author was able to prove the existence of chaos through a horseshoe construction . very recently , there has been a breakthrough by the author in finding lax pairs for euler equations of incompressible inviscid fluids . further results have been obtained by the author and collaborators . [ multiblock footnote omitted ]

math0302200

unlike chaos in finite dimensional systems , the area of chaos in partial differential equations has been a long standing open field . during the past ten years , there have been some remarkable developments in the area of chaos in partial differential equations by the author and collaborators . our main results can be briefly described as follows : a systematic program is developed for proving the existence of homoclinic orbits for perturbed soliton equations . the program involves machineries from integrable theory , dynamical systems , and partial differential equations . the types of homoclinic orbits thus obtained can be either transversal or of silnikov . with regard to transversal homoclinic orbits , shadowing lemmas are utilized or developed to prove the existence of chaos . with regard to silnikov homoclinic orbits , smale horseshoes are constructed to prove the existence of chaos . the main machineries from integrable theory are the isospectral theory and darboux transformations . the main machineries from dynamical systems are persistence of invariant manifolds , fenichel fibers , melnikov measurement , and other measurements . the main machineries from partial differential equations are local well - posedness and regularity . another exciting development happened to euler equations of incompressible inviscid fluids . lax pairs for both 2d and 3d euler equations was found by the author and steve childress . a darboux transformation for the lax pair of 2d euler equation was also found by artyom yurov and the author . the philosophical significance of the existence of lax pairs for euler equations is fundamentally important . if one defines integrability of an equation by the existence of a lax pair , then both 2d and 3d euler equations are integrable . more importantly , both 2d and 3d navier - stokes equations at high reynolds numbers are near integrable systems . such a point of view changes our old ideology on euler and navier - stokes equations . the most important application of the theory on chaos in partial differential equations in theoretical physics will be on the study of turbulence . our approach of studying turbulence is different from many other studies in which one starts with stokes equation to prove results on navier - stokes equations for small reynolds number . in our studies , we start with euler equations and view navier - stokes equations at large reynolds number as singular perturbations of euler equations . for this goal , we chose the 2d navier - stokes equations under periodic boundary conditions to begin a dynamical system study on 2d turbulence . we started with a simple fixed point of 2d euler equation , and studied the linearized 2d euler equation at the fixed point . a complete spectral theorem is obtained . in particular , we found unstable eigenvalues . then , naturally we are interested in the corresponding unstable manifold . such an unstable manifold forms the prototype of the attractor of navier - stokes equations at large reynolds number . and long term turbulence lives around the attractor . finding the unstable manifold is not successful yet . partial success is accomplished in finding the corresponding unstable manifold for some galerkin truncation . here the unstable manifold is the 2d surface of an ellipsoid . stable and unstable manifolds together form a lip - shape configuration . whether or not the stable and unstable manifolds for 2d euler equation have the same topology as above is not known . in fact , the existence of stable and unstable manifolds for 2d euler equation has not been proved . in terms of proving the existence of a homoclinic orbit , the most common tool is the so - called melnikov integral method @xcite @xcite . this method was subsequently developed by holmes and marsden @xcite , and most recently by wiggins @xcite . for partial differential equations , this method was mainly developed by li et al . @xcite @xcite @xcite @xcite . there are two derivations for the melnikov integrals . one is the so - called geometric argument @xcite @xcite @xcite @xcite @xcite . the other is the so - called liapunov - schmitt argument @xcite @xcite . the liapunov - schmitt argument is a fixed - point type argument which directly leads to the existence of a homoclinic orbit . the condition for the existence of a fixed point is the melnikov integral . the geometric argument is a signed distance argument which applies to more general situations than the liapunov - schmitt argument . it turns out that the geometric argument is a much more powerful machinary than the liapunov - schmitt argument . in particular , the geometric argument can handle geometric singular perturbation problems . i shall also mention an interesting derivation in @xcite . in establishing the existence of homoclinic orbits in high dimensions , one often needs other tools besides the melnikov analysis . for example , when studying orbits homoclinic to fixed points created through resonances in ( @xmath0)-dimensional near - integrable systems , one often needs tools like fenichel fibers , as presented in previous chapter , to set up geometric measurements for locating such homoclinic orbits . such homoclinic orbits often have a geometric singular perturbation nature . in such cases , the liapunov - schmitt argument can not be applied . for such works on finite dimensional systems , see for example @xcite @xcite @xcite . for such works on infinite dimensional systems , see for example @xcite @xcite . consider the regularly perturbed nonlinear schrdinger ( nls ) equation @xcite , @xmath1 q + i { \varepsilon}[\hat{{\partial}}^2_xq - { \alpha}q + { \beta } ] \ , \label{rpnls}\ ] ] where @xmath2 is a complex - valued function of the two real variables @xmath3 and @xmath4 , @xmath3 represents time , and @xmath4 represents space . @xmath5 is subject to periodic boundary condition of period @xmath6 , and even constraint , i.e. , @xmath7 @xmath8 is a positive constant , @xmath9 and @xmath10 are constants , @xmath11 is a bounded fourier multiplier , @xmath12 @xmath13 when @xmath14 , @xmath15 when @xmath16 , for some fixed large @xmath17 , and @xmath18 is the perturbation parameter . the following theorem was proved in @xcite . there exists a @xmath19 , such that for any @xmath20 , there exists a codimension 1 surface in the external parameter space @xmath21 where @xmath22 , and @xmath23 . for any @xmath24 on the codimension 1 surface , the regularly perturbed nonlinear schrdinger equation ( [ rpnls ] ) possesses a symmetric pair of silnikov homoclinic orbits asymptotic to a saddle @xmath25 . the codimension 1 surface has the approximate representation given by @xmath26 , where @xmath27 is plotted in figure [ kappa ] . [ rhorbit ] to prove the theorem , one starts from the invariant plane @xmath28 on @xmath29 , there is a saddle @xmath30 to which the symmetric pair of silnikov homoclinic orbits will be asymptotic to , where @xmath31 its eigenvalues are @xmath32\pm 2 \sqrt{(\frac{n^2}{2 } + { \omega}^2-i)(3i -{\omega}^2 -\frac{n^2}{2 } ) } \ , \label{qev}\ ] ] where @xmath33 , @xmath34 , @xmath35 when @xmath36 , @xmath37 when @xmath38 , for some fixed large @xmath17 , and @xmath39 is given in ( [ qec ] ) . the crucial points to notice are : ( 1 ) . only @xmath40 and @xmath41 have positive real parts , @xmath42 ; ( 2 ) . all the other eigenvalues have negative real parts among which the absolute value of @xmath43 is the smallest ; ( 3 ) . @xmath44 . actually , items ( 2 ) and ( 3 ) are the main characteristics of silnikov homoclinic orbits . the unstable manifold @xmath45 of @xmath46 has a fiber representation @xcite . the melnikov measurement measures the signed distance between @xmath45 and a center - stable manifold @xmath47 @xcite . by virtue of a fenichel fiber theorem @xcite , one can show that , to the leading order in @xmath48 , the signed distance is given by the melnikov integral @xmath49dxdt,\end{aligned}\ ] ] where @xmath50 is a homoclinic orbit for the integrable nls , which can be generated through darboux transformations @xcite ; and @xmath51 and @xmath52 are melnikov vectors which can be generated through floquet discriminant in the isospectral theory of the integrable nls @xcite . the zero of the signed distance implies the existence of an orbit in @xmath53 . the stable manifold @xmath54 of @xmath46 is a codimension 1 submanifold in @xmath47 . to locate a homoclinic orbit , one needs to set up a second measurement measuring the signed distance between the orbit in @xmath55 and @xmath54 inside @xmath47 . to set up this signed distance , first one can rather easily track the ( perturbed ) orbit by an unperturbed orbit to an @xmath56 neighborhood of @xmath29 , then one needs to prove the size of @xmath54 to be @xmath57 ( @xmath58 ) with normal form transform . to the leading order in @xmath48 , the zero of the second signed distance is given by @xmath59 where @xmath60 and @xmath61 is a phase shift of @xmath50 . to the leading order in @xmath48 , the common zero of the two second signed distances satisfies @xmath26 , where @xmath27 is plotted in figure [ kappa ] . then the claim of the theorem is proved by virtue of the implicit function theorem . for rigorous details , see @xcite . consider the singularly perturbed nonlinear schrdinger equation @xcite , @xmath62 q + i { \varepsilon}[q_{xx } - { \alpha}q + { \beta } ] \ , \label{spnls}\ ] ] where @xmath2 is a complex - valued function of the two real variables @xmath3 and @xmath4 , @xmath3 represents time , and @xmath4 represents space . @xmath5 is subject to periodic boundary condition of period @xmath6 , and even constraint , i.e. , @xmath7 @xmath63 is a positive constant , @xmath9 and @xmath10 are constants , and @xmath18 is the perturbation parameter . the following theorem was proved in @xcite . there exists a @xmath19 , such that for any @xmath20 , there exists a codimension 1 surface in the external parameter space @xmath64 where @xmath65 , @xmath66 is a finite subset , and @xmath23 . for any @xmath24 on the codimension 1 surface , the singularly perturbed nonlinear schrdinger equation ( [ spnls ] ) possesses a symmetric pair of silnikov homoclinic orbits asymptotic to a saddle @xmath25 . the codimension 1 surface has the approximate representation given by @xmath26 , where @xmath27 is plotted in figure [ kappa ] . [ shorbit ] the proof of the theorem is also completed through two measurements : the melnikov measurement and the second measurement . but more powerful machineries are needed @xcite . in recent years , novel results have been obtained on the solutions of the vector nonlinear schrdinger equations @xcite @xcite @xcite . abundant ordinary integrable results have been carried through @xcite @xcite , including linear stability calculations @xcite . specifically , the vector nonlinear schrdinger equations can be written as @xmath67 where @xmath68 and @xmath69 are complex valued functions of the two real variables @xmath3 and @xmath4 , and @xmath70 is a positive constant . these equations describe the evolution of two orthogonal pulse envelopes in birefringent optical fibers @xcite @xcite , with industrial applications in fiber communication systems @xcite and all - optical switching devices @xcite . for linearly birefringent fibers @xcite , @xmath71 . for elliptically birefringent fibers , @xmath70 can take other positive values @xcite . when @xmath72 , these equations are first shown to be integrable by s. manakov @xcite , and thus called manakov equations . when @xmath70 is not 1 or 0 , these equations are non - integrable . propelled by the industrial applications , extensive mathematical studies on the vector nonlinear schrdinger equations have been conducted . like the scalar nonlinear schrdinger equation , the vector nonlinear schrdinger equations also possess figure eight structures in their phase space . consider the singularly perturbed vector nonlinear schrdinger equations , @xmath73 p = i { \varepsilon } [ p_{xx } -{\alpha}p - { \beta}]\ , \label{pnls1}\\ & & iq_t + q_{xx } + \frac{1}{2 } [ ( |p|^2 + |q|^2)-{\omega}^2 ] q = i { \varepsilon } [ q_{xx } -{\alpha}q - { \beta}]\ , \label{pnls2}\end{aligned}\ ] ] where @xmath74 and @xmath5 are subject to periodic boundary condition of period @xmath75 , and are even in @xmath4 , i.e. @xmath76 @xmath77 @xmath78 , @xmath79 and @xmath80 are real constants , and @xmath18 is the perturbation parameter . we have there exists a @xmath19 , such that for any @xmath20 , there exists a codimension 1 surface in the space of @xmath81 where @xmath82 , @xmath66 is a finite subset , and @xmath83 . for any @xmath24 on the codimension 1 surface , the singularly perturbed vector nonlinear schrdinger equations ( [ pnls1])-([pnls2 ] ) possesses a homoclinic orbit asymptotic to a saddle @xmath25 . this orbit is also the homoclinic orbit for the singularly perturbed scalar nonlinear schrdinger equation studied in last section , and is the only one asymptotic to the saddle @xmath25 for the singularly perturbed vector nonlinear schrdinger equations ( [ pnls1])-([pnls2 ] ) . the codimension 1 surface has the approximate representation given by @xmath26 , where @xmath27 is plotted in figure [ kappa ] . consider the following perturbed discrete cubic nonlinear schrdinger equations @xcite , @xmath84+|q_n|^2(q_{n+1}+ q_{n-1})-2{\omega}^2 q_n \nonumber \\ & & + i{\varepsilon}\bigg[-{\alpha}q_n + { 1 \over h^2}(q_{n+1}-2q_n+q_{n-1 } ) + { \beta}\bigg ] , \label{pdnls}\end{aligned}\ ] ] where @xmath85 , @xmath86 s are complex variables , @xmath87 @xmath88 , and @xmath89 this is a @xmath90 dimensional system , where @xmath91 this system is a finite - difference discretization of the perturbed nls ( [ spnls ] ) . the following theorem was proved in @xcite . denote by @xmath92 the external parameter space , @xmath93 for any @xmath17 ( @xmath94 ) , there exists a positive number @xmath95 , such that for any @xmath96 , there exists a codimension @xmath97 surface @xmath98 in @xmath99 ; for any external parameters ( @xmath100 ) on @xmath98 , there exists a homoclinic orbit asymptotic to a saddle @xmath46 . the codimension @xmath97 surface @xmath98 has the approximate expression @xmath101 , where @xmath102 is shown in fig.[nkappa ] . [ dhorbit ] in the cases ( @xmath103 ) , @xmath104 is always negative . for @xmath105 , @xmath104 can be positive as shown in fig.[nkappa ] . when @xmath17 is even and @xmath106 , there is in fact a symmetric pair of homoclinic orbits asymptotic to a fixed point @xmath46 at the same values of the external parameters ; since for even @xmath17 , we have the symmetry : if @xmath107 solves ( [ pdnls ] ) , then @xmath108 also solves ( [ pdnls ] ) . when @xmath17 is odd and @xmath106 , the study can not guarantee that two homoclinic orbits exist at the same value of the external parameters . consider the perturbed davey - stewartson ii equations @xcite , @xmath109q + i\epsilon f \ , \\ & & \ \ \delta u = -4\partial_y |q|^2 \ , \end{aligned}\ ] ] where @xmath69 is a complex - valued function of the three variables ( @xmath110 ) , @xmath111 is a real - valued function of the three variables ( @xmath110 ) , @xmath112 , @xmath113 , @xmath114 is a constant . we also consider periodic boundary conditions . the perturbation @xmath115 can be for example a singular perturbation @xmath116 where @xmath117 , @xmath118 are constants , or a regular perturbation by mollifying @xmath119 into a bounded fourier multiplier @xmath120 in the case of periods ( @xmath121 ) , @xmath122 for some large @xmath17 , @xmath123 . under both regular and singular perturbations , the rigorous melnikov measurement can be established @xcite . it turns out that only local well - posedness is necessary for rigorously setting up the melnikov measurement , thanks to the fact that the unperturbed homoclinic orbits are classical solutions . the obstacle toward proving the existence of homoclinic orbits comes from a technical difficulty in solving a linear system to get the normal form for proving the size estimate of the stable manifold of a saddle . for details , see @xcite . transversal homoclinic orbits in continuous systems often appear in two types of systems : ( 1 ) . periodic systems where the poincar period map has a transversal homoclinic orbit , ( 2 ) . autonomous systems where the homoclinic orbit is asymptotic to a hyperbolic limit cycle . consider the periodically perturbed sine - gordon ( sg ) equation , @xmath124 , \label{psg}\ ] ] where @xmath125 for @xmath126 , @xmath127 is a smooth bump function , under odd periodic boundary condition , @xmath128 @xmath129 , @xmath130 , @xmath131 is a small perturbation parameter . there exists an interval @xmath132 such that for any @xmath133 , there exists a transversal homoclinic orbit @xmath134 asymptotic to @xmath135 in @xmath136 . consider the derivative nonlinear schrdinger equation , latexmath:[\[i q_t = q_{xx } + 2 |q|^2 q + i { \varepsilon}\bigg [ ( \frac{9}{16}-|q|^2 ) q + \mu @xmath69 is a complex - valued function of two real variables @xmath3 and @xmath4 , @xmath18 is the perturbation parameter , @xmath138 is a real constant , and @xmath139 is a bounded fourier multiplier , @xmath140 and some fixed @xmath141 . periodic boundary condition and even constraint are imposed , @xmath142 there exists a @xmath19 , such that for any @xmath20 , and @xmath143 , there exist two transversal homoclinic orbits asymptotic to the limit cycle @xmath144\}$ ] . the importance of homoclinic orbits with respect to chaotic dynamics was first realized by poincar @xcite . in 1961 , smale constructed the well - known horseshoe in the neighborhood of a transversal homoclinic orbit @xcite @xcite @xcite . in particular , smale s theorem implies birkhoff s theorem on the existence of a sequence of structurely stable periodic orbits in the neighborhood of a transversal homoclinic orbit @xcite . in 1984 and 1988 @xcite @xcite , palmer gave a beautiful proof of smale s theorem using a shadowing lemma . later , this proof was generalized to infinite dimensions by steinlein and walther @xcite @xcite and henry @xcite . in 1967 , silnikov proved smale s theorem for autonomous systems in finite dimensions using a fixed point argument @xcite . in 1996 , palmer proved smale s theorem for autonomous systems in finite dimensions using shadowing lemma @xcite @xcite . in 2002 , li proved smale s theorem for autonomous systems in infinite dimensions using shadowing lemma @xcite . for nontransversal homoclinic orbits , the most well - known type which leads to the existence of smale horseshoes is the so - called silnikov homoclinic orbit @xcite @xcite @xcite @xcite @xcite . existence of silnikov homoclinic orbits and new constructions of smale horseshoes for concrete nonlinear wave systems have been established in finite dimensions @xcite @xcite and in infinite dimensions @xcite @xcite @xcite . let @xmath145 be a set which consists of elements of the doubly infinite sequence form : @xmath146 where @xmath147 , m labels , and @xmath148 . we introduce a topology in @xmath145 by taking as neighborhood basis of @xmath149 the set @xmath150 for @xmath151 . this makes @xmath145 a topological space . the shift automorphism @xmath70 is defined on @xmath145 by @xmath152 the shift automorphism @xmath70 exhibits _ sensitive dependence on initial conditions _ , which is a hallmark of _ we continue from section [ horrnls ] and consider the regularly perturbed nonlinear schrdinger ( nls ) equation ( [ rpnls ] ) . starting from the homoclinic orbit theorem [ rhorbit ] , we have the following theorem on the existence of chaos . under certain generic assumptions for the perturbed nonlinear schrdinger system ( [ rpnls ] ) , there exists a compact cantor subset @xmath153 of @xmath154 ( the sobolev space ) , @xmath153 consists of points , and is invariant under a poincar map @xmath155 . @xmath155 restricted to @xmath153 , is topologically conjugate to the shift automorphism @xmath70 on four symbols @xmath156 . that is , there exists a homeomorphism @xmath157 such that the following diagram commutes : @xmath158 [ horserhthm ] although the symmetric pair of silnikov homoclinic orbits is not structurally stable , the smale horseshoes are structurally stable . thus , the cantor sets and the conjugacy to the shift automorphism are structurally stable . we continue from section [ horsnls ] and consider the singularly perturbed nonlinear schrdinger ( nls ) equation ( [ spnls ] ) . starting from the homoclinic orbit theorem [ shorbit ] , we have under certain generic assumptions for the perturbed nonlinear schrdinger system ( [ spnls ] ) , there exists a compact cantor subset @xmath153 of @xmath154 , @xmath153 consists of points , and is invariant under a poincar map @xmath155 . @xmath155 restricted to @xmath153 , is topologically conjugate to the shift automorphism @xmath70 on four symbols @xmath156 . that is , there exists a homeomorphism @xmath157 such that the following diagram commutes : @xmath158 [ horseshthm ] we continue from section [ hordnls ] . starting from the homoclinic orbit theorem [ dhorbit ] , we have the theorem on the existence of chaos for the perturbed discrete nonlinear schrdinger equation ( [ pdnls ] ) . under certain generic assumptions for the perturbed discrete nonlinear schrdinger system ( [ pdnls ] ) , there exists a compact cantor subset @xmath153 of @xmath159 , @xmath153 consists of points , and is invariant under a poincar map @xmath155 . @xmath155 restricted to @xmath153 , is topologically conjugate to the shift automorphism @xmath70 on two symbols @xmath160 . that is , there exists a homeomorphism @xmath157 such that the following diagram commutes : @xmath158 [ horsedhthm ] the finite - difference discretization of both the regularly and the singularly perturbed nonlinear schrdinger equations ( [ rpnls ] ) and ( [ spnls ] ) leads to the same discrete perturbed nonlinear schrdinger equation ( [ pdnls ] ) . in the chaotic regime , typical numerical output is shown in figure [ numo ] . notice that there are two typical profiles at a fixed time : ( 1 ) . a breather type profile with its hump located at the center of the spatial period interval , ( 2 ) . a breather type profile with its hump located at the boundary ( wing ) of the spatial period interval . these two types of profiles are half spatial period translate of each other . if we label the profiles , with their humps at the center of the spatial period interval , by `` c '' ; those profiles , with their humps at the wing of the spatial period interval , by `` w '' , then @xmath161 where @xmath162 is the symmetry group element representing half spatial period translate . the time series of the output in figure [ numo ] is a chaotic jumping between `` c '' and `` w '' , which we call `` chaotic center - wing jumping '' . we interpret the chaotic center - wing jumping as the numerical realization of the shift automorphism @xmath70 on symbols . we can make this more precise in terms of the phase space geometry . the figure - eight structure of the integrable nls projected onto the plane of the fourier component @xmath163 is illustrated in figure [ pfi ] , and labeled by @xmath164 and @xmath165 . from the symmetry , we know that @xmath166 @xmath164 has the spatial - temporal profile realization as in figure [ figlc ] with the hump located at the center . @xmath165 corresponds to the half spatial period translate of the spatial - temporal profile realization as in figure [ figlc ] , with the hump located at the boundary ( wing ) . an orbit inside @xmath164 , @xmath167 has a spatial - temporal profile realization as in figure [ figlcin ] . the half period translate of @xmath167 , @xmath168 is inside @xmath165 . an orbit outside @xmath164 and @xmath165 , @xmath169 has the spatial - temporal profile realization as in figure [ figlout ] . @xmath170 and @xmath171 are two phase blocks . from these figures , we can see clearly that the chaotic center - wing jumping ( figure [ numo ] ) is the realization of the shift automorphism on symbols in @xmath172 . since its invention @xcite , shadowing lemma has been a useful tool for solving many dynamical system problems . here we only focus upon its use in proving the existence of chaos in a neighborhood of a transversal homoclinic orbit . according to the type of the system , the level of difficulty in proving the existence of chaos with a shadowing lemma is different . a finite - dimensional periodic system can be written in the general form @xmath173 where @xmath174 , and @xmath175 is periodic in @xmath3 . let @xmath115 be the poincar period map . a doubly infinite sequence @xmath176 in @xmath177 is a @xmath178 pseudo - orbit of a @xmath179 diffeomorphism @xmath180 if for all integers @xmath181 @xmath182 an orbit @xmath183 is said to @xmath48-shadow the @xmath178 pseudo - orbit @xmath176 if for all integers @xmath181 @xmath184 a compact invariant set @xmath66 is hyperbolic if there are positive constants @xmath141 , @xmath185 and a projection matrix valued function @xmath186 , @xmath187 , of constant rank such that for all @xmath4 in @xmath66 @xmath188 @xmath189 @xmath190 let @xmath66 be a compact hyperbolic set for the @xmath179 diffeomorphism @xmath180 . then given @xmath191 sufficiently small there exists @xmath192 such that every @xmath178 pseudo - orbit in @xmath66 has a unique @xmath48-shadowing orbit . [ shal ] the proof of this theorem by palmer @xcite is overall a fixed point argument with the help of green functions for linear maps . let @xmath193 be a transversal homoclinic point asymptotic to a saddle @xmath194 of a @xmath179 diffeomorphism @xmath180 . then the set @xmath195 is hyperbolic . denote by @xmath196 and @xmath197 the two orbit segments of length @xmath198 @xmath199 let @xmath200 where @xmath201 , be any doubly infinite binary sequence . let @xmath202 be the doubly infinite sequence of points in @xmath66 , associated with @xmath203 @xmath204 when @xmath205 is sufficiently large , @xmath202 is a @xmath178 pseudo - orbit in @xmath66 . by the shadowing lemma ( theorem [ shal ] ) , there is a unique @xmath48-shadowing orbit that shadows @xmath202 . in this manner , palmer @xcite gave a beautiful proof of smale s horseshoe theorem . denote by @xmath206 the set of doubly infinite binary sequences @xmath200 where @xmath201 . we give the set @xmath207 the discrete topology and @xmath206 the product topology . the bernoulli shift @xmath70 is defined by @xmath208_j = a_{j+1}\ .\ ] ] let @xmath193 be a transversal homoclinic point asymptotic to a saddle @xmath194 of a @xmath179 diffeomorphism @xmath180 . then there is a homeomorphism @xmath209 of @xmath206 onto a compact subset of @xmath177 which is invariant under @xmath115 and such that when @xmath205 is sufficiently large @xmath210 that is , the action of @xmath211 on @xmath212 is topologically conjugate to the action of @xmath70 on @xmath206 . here one can define @xmath213 to be the point on the shadowing orbit that shadows the midpoint of the orbit segment @xmath214 , which is either @xmath194 or @xmath193 . the topological conjugacy can be easily verified . for details , see @xcite . other references can be found in @xcite @xcite . an infinite - dimensional periodic system defined in a banach space @xmath215 can be written in the general form @xmath173 where @xmath216 , and @xmath175 is periodic in @xmath3 . let @xmath115 be the poincar period map . when @xmath115 is a @xmath179 map which needs not to be invertible , shadowing lemma and symbolic dynamics around a transversal homoclinic orbit can both be established @xcite @xcite @xcite . other references can be found in @xcite @xcite @xcite @xcite . there exists also a work on horseshoe construction without shadowing lemma for sinusoidally forced vibrations of buckled beam @xcite . we continue from section [ ppsge ] , and use the notations in section [ palmer ] . for the periodically perturbed sine - gordon equation ( [ psg ] ) , the poincar period map is a @xmath179 diffeomorphism in @xmath154 . as a corollary of the result in last section , we have the theorem on the existence of chaos . there is an integer @xmath205 and a homeomorphism @xmath209 of @xmath217 onto a compact cantor subset @xmath218 of @xmath136 . @xmath219 is invariant under the poincar period-@xmath75 map @xmath155 of the periodically perturbed sine - gordon equation ( [ psg ] ) . the action of @xmath220 on @xmath219 is topologically conjugate to the action of @xmath70 on @xmath221 . that is , the following diagram commutes : @xmath222 a finite - dimensional autonomous system can be written in the general form @xmath223 where @xmath174 . in this case , a transversal homoclinic orbit can be an orbit asymptotic to a normally hyperbolic limit cycle . that is , it is an orbit in the intersection of the stable and unstable manifolds of a normally hyperbolic limit cycle . instead of the poincar period map as for periodic system , one may want to introduce the so - called poincar return map which is a map induced by the flow on a codimension 1 section which is transversal to the limit cycle . unfortunately , such a map is not even well - defined in the neighborhood of the homoclinic orbit . this poses a challenging difficulty in extending the arguments as in the case of a poincar period map . in 1996 , palmer @xcite completed a proof of a shadowing lemma and existence of chaos using newton s method . it will be difficult to extend this method to infinite dimensions , since it used heavily differentiations in time . other references can be found in @xcite @xcite @xcite @xcite . an infinite - dimensional autonomous system defined in a banach space @xmath215 can be written in the general form @xmath223 where @xmath216 . in 2002 , the author @xcite completed a proof of a shadowing lemma and existence of chaos using fenichel s persistence of normally hyperbolic invariant manifold idea . the setup is as follows , * * assumption ( a1 ) : * there exist a hyperbolic limit cycle @xmath66 and a transversal homoclinic orbit @xmath224 asymptotic to @xmath66 . as curves , @xmath66 and @xmath224 are @xmath225 . * * assumption ( a2 ) : * the fenichel fiber theorem is valid at @xmath66 . that is , there exist a family of unstable fenichel fibers @xmath226 and a family of stable fenichel fibers @xmath227 . for each fixed @xmath228 , @xmath229 and @xmath230 are @xmath225 submanifolds . @xmath229 and @xmath230 are @xmath231 in @xmath232 . the unions @xmath233 and @xmath234 are the unstable and stable manifolds of @xmath66 . both families are invariant , i.e. @xmath235 @xmath236 where @xmath237 is the evolution operator . there are positive constants @xmath104 and @xmath238 such that @xmath239 , @xmath240 and @xmath241 , @xmath242 @xmath243 * * assumption ( a3 ) : * @xmath244 is @xmath245 in @xmath3 , for @xmath246 , @xmath247 . for any fixed @xmath248 , @xmath244 is a @xmath231 diffeomorphism on @xmath215 . notice that we do not assume that as functions of time , @xmath66 and @xmath224 are @xmath249 , and we only assume that as curves , @xmath66 and @xmath224 are @xmath249 . under the above setup , a shadowing lemma and existence of chaos can be proved @xcite . another crucial element in the argument is the establishment of a @xmath250-lemma ( also called inclination lemma ) @xcite . we continue from section [ adns ] , and consider the derivative nonlinear schrdinger equation ( [ dernls ] ) . the transversal homoclinic orbit given in theorem [ thmdns ] is a classical solution . thus , assumption ( a1 ) is valid . assumption ( a2 ) follows from the standard arguments in @xcite @xcite @xcite . since the perturbation in ( [ dernls ] ) is bounded , assumption ( a3 ) follows from standard arguments . thus there exists chaos in the derivative nonlinear schrdinger equation ( [ dernls ] ) @xcite . the governing equations for the incompressible viscous fluid flow are the navier - stokes equations . turbulence occurs in the regime of high reynolds number . by formally setting the reynolds number equal to infinity , the navier - stokes equations reduce to the euler equations of incompressible inviscid fluid flow . one may view the navier - stokes equations with large reynolds number as a singular perturbation of the euler equations . results of t. kato show that 2d navier - stokes equations are globally well - posed in @xmath251 , and for any @xmath252 , the mild solutions of the 2d navier - stokes equations approach those of the 2d euler equations in @xmath253 ; h^s(r^2))$ ] @xcite . 3d navier - stokes equations are locally well - posed in @xmath254 ; h^s(r^3 ) ) , \ s>5/2 $ ] , and the mild solutions of the 3d navier - stokes equations approach those of the 3d euler equations in @xmath254 ; h^s(r^3))$ ] , where @xmath255 depends on the norms of the initial data and the external force @xcite @xcite . extensive studies on the inviscid limit have been carried by j. wu et al . @xcite @xcite @xcite @xcite . there is no doubt that mathematical study on navier - stokes ( euler ) equations is one of the most important mathematical problems . in fact , clay mathematics institute has posted the global well - posedness of 3d navier - stokes equations as one of the one million dollars problems . v. arnold @xcite realized that 2d euler equations are a hamiltonian system . recently , the author found lax pair structures for euler equations @xcite @xcite @xcite @xcite @xcite . understanding the structures of solutions to euler equations is of fundamental interest . of particular interest is the question on the global well - posedness of 3d navier - stokes and euler equations . our number one hope is that the lax pair structures can be useful in investigating the global well - posedness . our secondary hope is that the darboux transformation @xcite associated with the lax pair can generate explicit representation of homoclinic structures @xcite . the philosophical significance of the existence of lax pairs for euler equations is even more important . if one defines integrability of an equation by the existence of a lax pair , then both 2d and 3d euler equations are integrable . more importantly , both 2d and 3d navier - stokes equations at high reynolds numbers are singularly perturbed integrable systems . such a point of view changes our old ideology on euler and navier - stokes equations . the 2d euler equation can be written in the vorticity form , @xmath256 where the bracket @xmath257 is defined as @xmath258 @xmath259 is the vorticity , and @xmath260 is the stream function given by , @xmath261 and the relation between vorticity @xmath259 and stream function @xmath260 is , @xmath262 the lax pair of the 2d euler equation ( [ euler ] ) is given as @xmath263 where @xmath264 and @xmath250 is a complex constant , and @xmath265 is a complex - valued function . [ 2dlp ] consider the lax pair ( [ laxpair ] ) at @xmath266 , i.e. @xmath267 where we replaced the notation @xmath265 by @xmath68 . let @xmath268 be any fixed solution to the system ( [ d1 ] , [ d2 ] ) , we define the gauge transform @xmath269 : @xmath270\ , \label{gauge}\ ] ] and the transforms of the potentials @xmath259 and @xmath260 : @xmath271 where @xmath272 is subject to the constraints @xmath273 then @xmath274 solves the system ( [ d1 ] , [ d2 ] ) at @xmath275 . thus ( [ gauge ] ) and ( [ ptl ] ) form the darboux transformation for the 2d euler equation ( [ euler ] ) and its lax pair ( [ d1 ] , [ d2 ] ) . [ dt ] for kdv equation and many other soliton equations , the gauge transform is of the form @xcite , @xmath276 in general , gauge transform does not involve potentials . for 2d euler equation , a potential factor @xmath277 is needed . from ( [ d1 ] ) , one has @xmath278 the gauge transform ( [ gauge ] ) can be rewritten as @xmath279 the lax pair ( [ d1 ] , [ d2 ] ) has a symmetry , i.e. it is invariant under the transform @xmath280 . the form of the gauge transform ( [ gauge ] ) resulted from the inclusion of the potential factor @xmath277 , is consistent with this symmetry . our hope is to use the darboux transformation to generate homoclinic structures for 2d euler equation @xcite . the rossby wave equation is @xmath281 where @xmath282 is the vorticity , @xmath283 , and @xmath284 is the stream function . its lax pair can be obtained by formally conducting the transformation , @xmath285 , to the 2d euler equation @xcite , @xmath286 where @xmath265 is a complex - valued function , and @xmath250 is a complex parameter . the 3d euler equation can be written in vorticity form , @xmath287 where @xmath288 is the velocity , @xmath289 is the vorticity , @xmath290 , @xmath291 , and @xmath292 . @xmath111 can be represented by @xmath259 for example through biot - savart law . the lax pair of the 3d euler equation ( [ 3deuler ] ) is given as @xmath293 where @xmath294 @xmath250 is a complex constant , and @xmath209 is a complex scalar - valued function . another lax pair of the 3d euler equation ( [ 3deuler ] ) is given as @xmath295 where @xmath296 @xmath250 is a complex constant , and @xmath297 is a complex 3-vector valued function . our hope is that the infinitely many conservation laws generated by @xmath298 can provide a priori estimates for the global well - posedness of 3d navier - stokes equations , or better understanding on the global well - posedness @xcite . for more informations on the topics , see @xcite @xcite . to begin an infinite dimensional dynamical system study , we investigate the linearized 2d euler equation at a fixed point @xcite @xcite @xcite . we consider the 2d euler equation ( [ euler ] ) under periodic boundary condition in both @xmath4 and @xmath299 directions with period @xmath75 . expanding @xmath259 into fourier series , @xmath300 where @xmath301 , @xmath302 , and @xmath303 . the 2d euler equation can be rewritten as @xmath304 where @xmath305 is given by , @xmath306(p_1 q_2 -p_2 q_1 ) \nonumber \\ \label{af } \\ & = & { 1\over 2}[|q|^{-2}-|p|^{-2}]\left | \begin{array}{lr } p_1 & q_1 \\ p_2 & q_2 \\ \end{array } \right | \ , \nonumber\end{aligned}\ ] ] where @xmath307 for @xmath308 , similarly for @xmath68 . denote @xmath309 by @xmath8 . we consider the simple fixed point @xmath310 @xcite : @xmath311 of the 2d euler equation ( [ keuler ] ) , where @xmath312 is an arbitrary complex constant . the _ linearized two - dimensional euler equation _ at @xmath310 is given by , @xmath313 for any @xmath314 , we define the class @xmath315 to be the subset of @xmath316 : @xmath317 [ classify ] see fig.[class ] for an illustration of the classes . according to the classification defined in definition [ classify ] , the linearized two - dimensional euler equation ( [ le ] ) decouples into infinitely many _ invariant subsystems _ : @xmath318 ( 36,27.8)(-18,-12 ) ( -12,-11.5)(2,0)13(0,1)23 ( -16,-10)(0,2)11(1,0)32 ( 0,-14)(0,1)28 ( -18,0)(1,0)36 ( 0,15)(0,0)@xmath319 ( 18.5,0)(0,0)[l]@xmath320 ( -5.5,0)(-5.275,5.275)(0,5.5 ) ( 0,5.5)(5.275,5.275)(5.5,0 ) ( 5.5,0)(5.275,-5.275)(0,-5.5 ) ( 0,-5.5)(-5.275,-5.275)(-5.5,0 ) ( 4,4 ) ( 0,0)(1,1)3.7 ( 4.35,4.35)@xmath68 ( 4,-4 ) ( 8,0 ) ( -8,0 ) ( -8,-2 ) ( -12,-4 ) ( -12,-6 ) ( -4,2 ) ( -4,4 ) ( 0,6 ) ( 0,8 ) ( 4,10 ) ( 12,4 ) ( 0,-8 ) ( -4,-12)(1,1)17.5 ( -13.5,-7.5)(1,1)19.5 ( -13.5,-5.5)(1,1)17.5 ( -3.6,1.3)@xmath321 ( -7,12.1)(0,0)[b]@xmath322 ( -6.7,12)(1,-3)2.55 ( 6.5,13.6)(0,0)[l]@xmath323 ( 6.4,13.5)(-2,-3)2.0 ( 7,-12.1)(0,0)[t]@xmath324 ( 6.7,-12.25)(-1,3)2.62 ( -4.4,-13.6)(0,0)[r]@xmath325 ( -4.85,-12.55)(1,3)2.45 let @xmath326 be the linear operator defined by the right hand side of ( [ cle ] ) , and @xmath327 be the sobolev space where @xmath328 is an integer and @xmath329 . the eigenvalues of the linear operator @xmath326 in @xmath327 are of four types : real pairs ( @xmath330 ) , purely imaginary pairs ( @xmath331 ) , quadruples ( @xmath332 ) , and zero eigenvalues . the open disk of radius @xmath333 in @xmath316 , denoted by @xmath334 , is defined as @xmath335 and the closure of @xmath334 , denoted by @xmath325 , is defined as @xmath336 we have the following claims on the spectrum of the linear operator @xmath326 : 1 . if @xmath337 , then the entire @xmath327 spectrum of the linear operator @xmath326 is its continuous spectrum . see figure [ splb ] , where @xmath338 that is , both the residual and the point spectra of @xmath326 are empty . 2 . if @xmath339 , then the entire essential @xmath327 spectrum of the linear operator @xmath326 is its continuous spectrum . that is , the residual spectrum of @xmath326 is empty . the point spectrum of @xmath326 is symmetric with respect to both real and imaginary axes . see figure [ spla2 ] . [ sst ] ( 36,27.8)(-18,-12 ) ( 0,-14)(0,1)28 ( -18,0)(1,0)36 ( 0,15)(0,0)@xmath340 ( 18.5,0)(0,0)[l]@xmath341 ( 0.1,-7)(0,1)14 ( .2,-.2)(0,0)[tl]@xmath135 ( -0.2,-7)(1,0)0.4 ( -0.2,7)(1,0)0.4 ( 2.0,-6.4)(0,0)[t]@xmath342 ( 2.0,7.6)(0,0)[t]@xmath343 ( 36,27.8)(-18,-12 ) ( 0,-14)(0,1)28 ( -18,0)(1,0)36 ( 0,15)(0,0)@xmath340 ( 18.5,0)(0,0)[l]@xmath341 ( 0.1,-7)(0,1)14 ( .2,-.2)(0,0)[tl]@xmath135 ( -0.2,-7)(1,0)0.4 ( -0.2,7)(1,0)0.4 ( 2.0,-6.4)(0,0)[t]@xmath342 ( 2.0,7.6)(0,0)[t]@xmath343 ( 2.4,3.5 ) ( -2.4,3.5 ) ( 2.4,-3.5 ) ( -2.4,-3.5 ) ( 5,4 ) ( -5,4 ) ( 5,-4 ) ( -5,-4 ) ( 8,6 ) ( -8,6 ) ( 8,-6 ) ( -8,-6 ) denote by @xmath344 the right hand side of ( [ le ] ) , i.e. the whole linearized 2d euler operator , the spectral mapping theorem holds . @xmath345 let @xmath346 denote the number of points @xmath347 that belong to the open disk of radius @xmath333 , @xmath334 , and such that @xmath69 is not parallel to @xmath68 . the number of nonimaginary eigenvalues of @xmath344 ( counting the multiplicities ) does not exceed @xmath348 . another interesting discussion upon the discrete spectrum can be found in @xcite . since the introduction of continued fractions for calculating the eigenvalues of steady fluid flow , by meshalkin and sinai @xcite , this topics had been extensively explored @xcite @xcite @xcite @xcite @xcite @xcite @xcite @xcite . rigorous justification on the continued fraction calculation was given in @xcite @xcite . as an example , we take @xmath349 . when @xmath350 , the fixed point has @xmath351 eigenvalues which form a quadruple . these four eigenvalues appear in the invariant linear subsystem labeled by @xmath352 . one of them is @xcite : @xmath353 see figure [ figev ] for an illustration . the essential spectrum (= continuous spectrum ) of @xmath326 with @xmath352 is the segment on the imaginary axis shown in figure [ figev ] , where @xmath354 . the essential spectrum (= continuous spectrum ) of the linear 2d euler operator at this fixed point is the entire imaginary axis . ( 36,27.8)(-18,-12 ) ( 0,-14)(0,1)28 ( -18,0)(1,0)36 ( 0,15)(0,0)@xmath340 ( 18.5,0)(0,0)[l]@xmath341 ( 2.4,3.5 ) ( -2.4,3.5 ) ( 2.4,-3.5 ) ( -2.4,-3.5 ) ( 0.1,-10)(0,1)20 ( .2,-.2)(0,0)[tl]@xmath135 ( -0.2,-10)(1,0)0.4 ( -0.2,10)(1,0)0.4 ( 2.0,-9.4)(0,0)[t]@xmath342 ( 2.0,10.6)(0,0)[t]@xmath343 a rather well - known open problem is proving the existence of unstable , stable , and center manifolds . the main difficulty comes from the fact that the nonlinear term is non - lipschitzian . to simplify our study , we study only the case when @xmath355 is real , @xmath356 , i.e. we only study the cosine transform of the vorticity , @xmath357 to further simplify our study , we will study a concrete dashed - line model based upon the line of fixed points ( [ fixpt ] ) with the mode @xmath349 parametrized by @xmath312 , and the only unstable invariant linear subsystem labeled by @xmath352 . the so - called _ dashed - line model _ is given by @xcite , @xmath358 where @xmath359 @xmath360 the model is designed to model the hyperbolic structure of the 2d euler equation , connected to the line of fixed points ( [ fixpt ] ) with @xmath349 . figure [ model ] illustrates the collocation of the modes in this model , which has the _ `` dashed - line '' _ nature leading to the name of the model . ( 36,27.8)(-18,-12 ) ( -12,-11.5)(2,0)13(0,1)23 ( -16,-10)(0,2)11(1,0)32 ( 0,-14)(0,1)28 ( -18,0)(1,0)36 ( 0,15)(0,0)@xmath319 ( 18.5,0)(0,0)[l]@xmath320 ( -2.75,0)(-2.6375,2.6375)(0,2.75 ) ( 0,2.75)(2.6375,2.6375)(2.75,0 ) ( 2.75,0)(2.6375,-2.6375)(0,-2.75 ) ( 0,-2.75)(-2.6375,-2.6375)(-2.75,0 ) ( 2,2 ) ( 0,0)(1,1)1.85 ( 2.275,2.275)@xmath68 ( -12,-10 ) ( -10,-8 ) ( -8,-6 ) ( -6,-4 ) ( -4,-2 ) ( -2,0 ) ( 0,2 ) ( 2,4 ) ( 4,6 ) ( 6,8 ) ( 8,10 ) ( -14,-12)(1,1)24 ( -5.6,-5.4)@xmath321 ( -4.4,-13.6)(0,0)[r]@xmath325 ( -4.85,-12.55)(1,3)3.4 when @xmath361 , one can determine the stable and unstable manifolds of the fixed point @xmath362 : @xmath363 using the polar coordinates : @xmath364 we have the following explicit expressions for the stable and unstable manifolds of the fixed point @xmath310 and its negative @xmath365 represented through _ perverted heteroclinic orbits _ : @xmath366 \theta & = & - \ \frac{a_2}{2{\kappa } } \ \tau + \theta_0 \ , , \label{exus}\\[1ex ] \rho & = & \sqrt{\frac{-a_1}{a_2 } } \ r \ , , \nonumber \\[1ex ] \theta + \vartheta & = & \left\ { \begin{array}{ll } - \arcsin \left [ \frac{1}{2 } \sqrt{\frac{a_2}{-a_1}}\ , \right ] \ , & ( { \kappa}>0 ) \ , \\[2ex ] \pi + \arcsin \left [ \frac{1}{2 } \sqrt{\frac{a_2}{-a_1 } } \right ] \ , , & ( { \kappa}<0 ) \ , , \end{array } \right . \nonumber \end{aligned}\ ] ] where @xmath367 , @xmath368 are the two parameters parametrizing the two - dimensional stable ( unstable ) manifold , and @xmath369 the two auxilliary variables @xmath370 and @xmath371 have the expressions : @xmath372 - \frac{1}{\beta } \cos [ \beta \ \mbox{ln}\ \cosh \tau + \theta_0 ] \right\ } \ , , \\ \omega_5 & = & \frac{\alpha \beta}{1 + \beta^2 } \ \mbox{sech}\ \tau \left\ { \cos [ \beta \ \mbox{ln}\ \cosh \tau + \theta_0 ] + \frac{1}{\beta } \sin [ \beta \ \mbox{ln}\ \cosh \tau + \theta_0 ] \right\ } \ , , \end{aligned}\ ] ] where @xmath373 the explicit expression ( [ exus ] ) of the perverted heteroclinic orbits shows that the unstable manifold @xmath374 of the fixed point @xmath310 is the same as the stable manifold @xmath375 of the negative @xmath365 of @xmath310 , and the stable manifold of the fixed point @xmath310 is the same as the unstable manifold of the negative @xmath365 of @xmath310 . both @xmath374 and @xmath375 have the shapes of `` painted eggs '' ( figure [ het ] ) . @xmath374 and @xmath375 together form the `` lip '' ( figure [ het ] ) which is a higher dimensional generalization of the heteroclinic connection on plane . as the first step toward understanding the degeneracy v.s . nondegeneracy of the hyperbolic structure of 2d euler equation , we are interested in the @xmath48-homotopy deformation of such hyperbolic structure for the dashed - line model . we have calculated the melnikov functions to study the `` breaking '' or `` persistence '' of such hyperbolic structure when @xmath48 is small . it turns out that both the first and the second order melnikov functions are identically zero @xcite . results on chaos in partial differential equations are reported . results on lax pairs of euler equations of incompressible inviscid fluids are also reported . preliminary results on dynamical system studies of 2d euler equation are summarized . chow and j.k . hale , _ methods of bifurcation theory _ , springer - verlag , new york - berlin , 1982 . n . chow and j. k. hale and j. mallet - paret , an example of bifurcation to homoclinic orbits , _ j. differential equations _ * 37 , no.3 * ( 1980 ) , 351 . s. n. chow , x. b. lin , and k. j. palmer . a shadowing lemma with applications to semilinear parabolic equations . , 20 , no.3:547557 , 1989 . g. kovacic , dissipative dynamics of orbits homoclinic to a resonance band , _ phys lett a _ * 167 * ( 1992 ) , 143 . g. kovacic , hamiltonian dynamics of orbits homoclinic to a resonance band , _ phys lett a _ * 167 * ( 1992 ) , 137 . y. latushkin and y. li and m. stanislavova , the spectrum of a linearized 2d euler operator , _ submitted _ ( 2001 ) . y. li , smale horseshoes and symbolic dynamics in perturbed nonlinear schrdinger equations , _ journal of nonlinear sciences _ * 9 , no.4 * ( 1999 ) , 363 . y. li , bcklund - darboux transformations and melnikov analysis for davey - stewartson ii equations , _ journal of nonlinear sciences _ * 10 , no.1 * ( 2000 ) , 103 . y. li , on 2d euler equations . i. on the energy - casimir stabilities and the spectra for a linearized two dimensional euler equation , _ journal of mathematical physics _ * 41 , no.2 * ( 2000 ) , 728 . y. li , a lax pair for the 2d euler equation , _ journal of mathematical physics _ * 42 , no.8 * ( 2001 ) , 3552 . y. li , on 2d euler equations : part ii . lax pairs and homoclinic strucutres , _ communications on applied nonlinear analysis , to appear , available at : http://xxx.lanl.gov/abs/math.ap/0010200 , or http://www.math.missouri.edu/~cli_ ( 2002 ) . y. li , persistent homoclinic orbits for nonlinear schrdinger equation under singular perturbation , _ submitted , available at : http://xxx.lanl.gov/abs/math.ap/0106194 , or http://www.math.missouri.edu/~cli_ ( 2001 ) . y. li et al . , persistent homoclinic orbits for a perturbed nonlinear schrdinger equation , _ comm . pure appl . _ * xlix * ( 1996 ) , 1175 . y. li and d. w. mclaughlin , morse and melnikov functions for nls pdes , _ comm . * 162 * ( 1994 ) , 175 . y. li and d. w. mclaughlin , homoclinic orbits and chaos in discretized perturbed nls system , part i. homoclinic orbits , _ journal of nonlinear sciences _ * 7 * ( 1997 ) , 211 . s. wiggins , _ global bifurcations and chaos : analytical methods _ , springer - verlag , new york , 1988 . o. c. wright and m. g. forest . on the bcklund - gauge transformation and homoclinic orbits of a coupled nonlinear schrdinger system . , 141 , no.1 - 2:104116 , 2000 .

observations of ultraluminous infrared galaxies ( ulirgs ) with an achieved resolution approaching the diffraction limit in the mid - infrared from 8 - 25 m using the keck telescopes are reported . we find extremely compact structures , with spatial scales of ( diameter ) in six of the seven ulirgs observed . these compact sources emit between 30% and 100% of the mid - infrared energy from these galaxies . we have utilized the compact mid - infrared structures as a diagnostic of whether an agn or a compact ( 100 300 pc ) starburst is the primary power source in these ulirgs . in markarian 231 , the upper limit on the diameter of the 12.5 m source , 0.13 , shows that the size of the infrared source must increase with increasing wavelength , consistent with agn models . in iras 05189 - 2524 and iras 08572 + 3915 there is strong evidence that the source size increases with increasing wavelength . this suggests heating by a central source rather than an extended luminosity source , consistent with the optical classification as an agn . the compact mid - infrared sources seen in the other galaxies can not be used to distinguish the ultimate luminosity source . if these ulirgs are powered by compact starbursts , the star formation rates seen in the central few hundred parsecs far exceed the global rates seen in nearby starburst galaxies , and approach the surface brightness of individual clusters in nearby starburst galaxies .

astro-ph9911045

ultraluminous infrared galaxies ( ulirgs ; defined as those systems with luminosity from 8 - 1000 @xmath0 m l@xmath3 ) were discovered to be a significant class of objects in the local universe in the iras all sky survey ( soifer et al . 1987 , sanders et al . 1988 ) . these systems have recently been suggested as a major component in the universe at z@xmath4 2 through deep submillimeter surveys ( c.f . lilly et al . 1999 , blain et al . 1999 , barger et al . 1998 ) . whether ulirgs are powered by dust enshrouded agn or starbursts has been debated since their discovery(e.g . sanders 1999 , joseph 1999 ) . this question is made all the more important by the realization that ulirgs could produce a major fraction of the radiant energy in the universe ( lilly et al . 1999 , heckmann 1998 ) . a wide variety of observational techniques have been employed to probe the nature of the central luminosity source in ulirgs , ranging from optical spectroscopy ( sanders et al . 1988 ) to cm radio continuum imaging ( condon et al . 1991 , hereafter c91 ) , to vlbi observations ( smith , lonsdale and lonsdale 1998 ) and hard x - ray observations ( rieke 1988 , nakagawa et al . 1998 , isawawa 1999 ) . the nuclear environments of these galaxies have been effectively imaged with high spatial resolution in the optical and near infrared ( surace et al . 1998 , surace and sanders 1999 , scoville et al . 1999 , hereafter s99 ) . observations in the thermal infrared have the potential to address the question of the origin of the luminosity by establishing the size of the regions that are radiating the infrared luminosity . the iras observations showed that the mid and far infrared luminosity in these systems emerging predominantly between 50 and 200 @xmath0 m was due to thermal reradiation by dust . however , the spatial resolution of those observations , @xmath5 , was inadequate to place meaningful size constraints on the luminosity sources . because the mid - infrared wavelengths ( 825@xmath0 m ) carry a significant fraction of the total bolometric luminosity in ulirgs , ranging from @xmath610% to @xmath4 30% of the 81000@xmath0 m luminosity , observations at these wavelengths with a resolution @xmath7 have the potential to constrain the sizes of the emitting regions and thereby address the nature of the underlying sources that heat the radiating dust . observations in the 10@xmath0 m atmospheric window , with a resolution @xmath8 , have been reported for several ulirgs by miles et al . ( 1996 ) , and keto et al . ( 1992 ) . in most cases they found the sources to be unresolved implying that the sources were either very compact starbursts or dust enshrouded agn . with a diffraction limit of 0.24@xmath2 fwhm at 10@xmath0 m , the keck telescope provides a substantial improvement in spatial resolution over previous mid - infrared observations , probing the distribution of the thermal emission at the 100 - 300 pc scale in nearby ulirgs . with the introduction of mid - infrared imaging on the keck telescope , we have started a program to address this problem . in the first paper from this program we reported imaging of the nucleus of the closest ulirg - arp 220 ( soifer et al . the observations were able to spatially resolve the emission at 24.5@xmath0 m , and demonstrated that the central thermal source is optically thick at this wavelength . these observations placed severe constraints on the surface brightness in the infrared luminous nuclei . in this paper we report imaging observations from 8 - 24.5@xmath0 m of a sample of the closest ulirgs at spatial resolutions of 0.3 - 0.6@xmath2 . these data provide the highest spatial resolution yet achieved on the thermal emission from the closest ulirgs and trace the spatial distribution of the emergent luminosity in these systems . we adopt h@xmath9 = 75 km s@xmath10mpc@xmath10 . the objects observed were taken from the iras bright galaxy sample ( soifer et al . 1987 ) with the addition of iras 17208 - 0014 ( sanders et al . 1995 ) , which is at a low galactic latitude . the basic information for the seven objects in the sample is given in table 1 . the objects selected for observation were chosen on the basis of 1 ) meeting the luminosity definition of the ulirgs , i.e. @xmath11 @xmath12 , 2 ) having a 25@xmath0 m flux density @xmath4 1 jy based on iras measurements , 3 ) being at the lowest possible redshift for the best linear resolution , and 4 ) being available at the time of the observations . this sample contains the five closest ulirgs in the iras bright galaxy sample of soifer et al . , and seven of the ten ulirgs known with @xmath13 and z@xmath140.06 . the projected linear scales of these objects range from 360 pc arcsec@xmath10 to 1200 pc arcsec@xmath10 , so that the point spread function / angular resolution of the keck observations , 0.3@xmath2 at 12@xmath0 m and 0.6@xmath2 at 24.5@xmath0 m provide linear resolutions of 100 pc - 700 pc , depending on the object and the wavelength observed . three of the objects in the sample have been classified as seyfert nuclei , three as liners , and one as an hii region , based on optical spectroscopy . the observations were made primarily using the mirlin mid - infrared camera ( ressler et al . 1994 ) at the f/40 bent cassegrain visitor port of the keck ii telescope . the camera uses a 128@xmath15128 si : as array with a plate scale of 0.138@xmath2/pixel for a total field of view of 17@xmath1617@xmath2 . table 2 reports the central wavelengths for the filters through which each object was observed . at each wavelength the observing procedure was the same . a secondary with a square wave chop of amplitude 6@xmath2 in the north - south or east - west direction at 4 hz was employed for fast beam switching . the frames sampling each chop position were coadded separately in hardware , resulting in two images . after an interval of approximately a minute , the telescope was nodded perpendicular to the chop direction ( east - west or north - south ) by 6@xmath2 and a second pair of images was obtained in order to cancel residuals in the sky and to subtract telescope emission . this procedure was repeated a number of times at each wavelength . the data were reduced by differencing the two images obtained within the chop pairs at each nod location , and then coadding the resulting positive images , with the positions appropriately adjusted to a common location , to yield a positive image centered in a field approximately 6@xmath166@xmath2 . because of the chopper and telescope nod spacings employed for the observations , the data are not capable of measuring flux outside a 6@xmath2 diameter region . observations with mirlin of all the ulirgs except iras 05189 - 2524 were obtained in two nights in march 1998 . the observations of the targets were interleaved with observations of nearby bright stars that were used to establish the point spread function ( psf ) for the observations . the psf stars were observed only three times per night in each of the bands at 12.5 , 17.9 and 24.5 @xmath0 m , but an attempt was made to accompany the critical observations with measurements of a nearby psf star . at 24.5 @xmath0 m the psf was always equal to the diffraction limit of the telescope to be 0.62@xmath2 fwhm . at the shorter wavelengths , the size of the psf image was set by diffraction and the pixel sampling about half of the time ; at other times atmospheric seeing affected the image size . iras 05189 - 2524 was observed in october 1998 . during this run the psf stars were observed more frequently , but for wavelengths shorter than @xmath17 17.9 @xmath0 m significant variation in the measured size of the psf star was seen . the variation in the measured size of the psf calibration star is the largest uncertainty in the measurement of the source sizes ; this issue is addressed when observations of a particular object are described . the mirlin observations were made under photometric conditions . the photometry was calibrated based on observations of four bright stars , @xmath18 tau , @xmath18 boo , @xmath19 peg , and @xmath18 cet whose magnitudes , in turn , were based on iras photometry . the uncertainties in the photometry , based on the internal consistency of the observations , is estimated to be 5% at @xmath2017.9 @xmath0 m and 10% at 24.5 @xmath0 m . the flux density corresponding to 0.0 mag ( vega - based ) was taken to follow the prescription given in the explanatory supplement to the iras catalogs and atlases ( joint iras science team , 1989 ) , and is given in soifer et al . ( 1999 ) . in addition to the mirlin observations , ugc 5101 , markarian 231 and iras 08572 + 3915 were observed using the long wavelength spectrograph on the keck i telescope on the night of 01 april 1999 . the night was not photometric and so only measurements of the sizes of the compact sources were made ; all the observations were carried out at 12.5 @xmath0 m . the chopper was set to an amplitude of 5@xmath2 chopping north - south at a frequency @xmath6 5 hz . observations were made in a fashion similar to the mirlin observations , except that the telescope nodding was in the same direction as the chopping and the nodding amplitude was the same as the chopping amplitude . the data were reduced in a manner similar to the mirlin data . the pixel scale is 0.08@xmath2 pixel@xmath10 . observations of markarian 273 were made at 3.45 @xmath0 m using the near - infrared camera at the cassegrain focus of the 200-inch hale telescope at palomar observatory in may 1999 . the camera was read out in a 64 @xmath15 64 pixel subarray because of the high read rate required to avoid saturation . the scale is 0.125@xmath2 pixel@xmath10 . the chopper throw was 15@xmath2 at a frequency of @xmath61 hz . observations of stars given by elias et al . ( 1982 ) were used to establish the photometric scale . the mid - infrared observations of the ulirgs reported here show a variety of characteristics , generally consistent with their properties seen at high spatial resolution at radio wavelengths ( c.f . c91 ) and in the near infrared using nicmos on hst ( scoville et al . 1998 , s99 ) . the mid - infrared data on all the ulirgs are presented in figures 1 , 2 and 3 . in figure 1 we present the 12.5 @xmath0 m images of the ulirgs and , where the radio and near infrared data are available , compare these maps to the equivalent maps at 8.4 ghz from the data of c91 and to the nicmos 2.2 @xmath0 m data from s99 and scoville et al ( 1998 ) . in the case of three galaxies , iras 05189 - 2524 , iras 08572 + 3915 and markarian 231 , where there is a single compact nucleus that emits a substantial fraction of the 12.5@xmath0 m flux density , we have presented the corresponding psf star at 12.5@xmath0 m for comparison . in figure 2 we compare , for the five single nucleus sources , the integrated flux at 12.5 @xmath0 m as a function of beam radius with the same quantity for a corresponding point source to establish the spatial distribution of the mid - infrared radiation . figure 3 displays the spectral energy distributions ( seds ) obtained from the mid - infrared data and compares these data with the corresponding iras data and data at other wavelengths . the flux densities measured for the ulirgs in a 4@xmath2 diameter beam are presented in table 3 . the mid - infrared observations did not establish accurate absolute positions for the sources . in some cases , described below , we use arguments regarding the similarities of the structures seen at vastly different wavelengths to argue for the physical coincidence of the sources seen at different wavelengths . in the following sections we give results for the individual objects . mirlin observations of iras 05189@xmath212524 were made at 12.5 and 24.5 @xmath0 m . the shorter wavelength data have the higher signal - to - noise ratio , and so are presented in figure 1 . images at 12.5@xmath0 m ( mirlin ) , 2.2@xmath0 m ( nicmos , s99 ) and 8.4ghz ( vla , c91 ) are shown in figure 1a . the dominant central source is clear in all three maps . at 8.4ghz , the single source is concentrated with a measured size of 0.2@xmath220.17@xmath2(c91 ) . at 2.2 @xmath0 m , the unresolved source is @xmath23 and contains @xmath675% of the total flux measured in a 5@xmath2 diameter beam , and 70% of the flux contained in an 11@xmath2 diameter beam ( s99 ) . the source appears virtually unresolved in the mirlin image . as can be seen from figure 2 , there is excellent agreement between the curve of growth for iras 05189 - 2524 and the psf star observation . the apparent size of the psf star varied during the time of the observations , so the limit on the size of iras 05189 - 2524 is not as stringent as would otherwise be achieved . at 12.5@xmath0 m the size is @xmath240.2@xmath2 ( table 4 ) . the source size at 24.5 @xmath0 m is an upper limit as well , and is larger and thus consistent with that established at 12.5 @xmath0 m . as seen in figure 3 and table 3 , the flux densities measured by mirlin at 12.5 @xmath0 m and 24.5 @xmath0 m agree well with the iras observations of this source ; virtually all of the luminosity of this source at 12.5 @xmath0 m and 24.5 @xmath0 m is contained within the unresolved source measured here . the mirlin observations of iras 05189@xmath212524 lead to a limit on the source diameter of @xmath24170 pc in the mid - infrared ( table 5 ) . the combination of flux density and upper limit on the source sizes(table 4 ) , imply the brightness temperatures t@xmath25123 k and @xmath485 k at 12.5 @xmath0 m and 24.5 @xmath0 m respectively . the flux ratio between 12.5 @xmath0 m and 24.5 @xmath0 m corresponds to a color temperature t@xmath26169 k , while the color temperature derived from the 25@xmath0 m and 60@xmath0 m iras measurements is 83 k. a central source with the bolometric luminosity of iras 05189@xmath212524 would heat silicate or graphite grains to the apparent color temperature between 12 and 24.5@xmath0 m , t@xmath27 170 k , at a radius of 30 - 60 pc using the grain parameters of draine and lee ( 1984 ) . this is much smaller than the observational limit . alternatively , the size of an optically thin source which produces the 25@xmath2160 @xmath0 m flux ratio with a color temperature of 83 k is significantly larger , 170 - 250 pc in radius , again using the draine and lee grain parameters . mirlin observations of iras 08572 + 3915 were obtained at five wavelengths ( table 2 ) . in addition , lws observations of iras 08572 + 3915 were obtained at 12.5 @xmath0 m . optical and near infrared imaging ( sanders et al . 1988 , carico et al . 1990 , s99 ) show two galaxies separated by about 5@xmath2 in a northwest - southeast orientation . the southeast source is not detected in the radio ( c91 ) or in the present data . the 12.5@xmath0 m lws image in figure 1b shows a bright central source ( the northwest source in the near infrared ) that is unresolved . the size limit at 17.9 @xmath0 m is consistent with that found at 12.5 @xmath0 m . figure 1b and figure 2 show that the central source is dominant in the mid - infrared and radio , containing @xmath680% of the total flux measured in a 4@xmath2 diameter beam . at 2.2 @xmath0 m , the central unresolved source contains @xmath660% of the total flux in an 11.4@xmath2 diameter beam ( s99 ) . the most stringent observations constraining the size and spatial distribution of the source are the lws observations at 12.5@xmath0 m . five observations of the psf star ( bs3275 ) were interleaved with eight observations of the object . the average fwhm of the object ( 0.39 @xmath28 0.04@xmath2 ) was marginally greater than the average fwhm of the psf ( 0.34 @xmath28 0.01@xmath2 ) . formally this leads to a size of the 12.5@xmath0 m source of 0.19@xmath2 , but this is only a @xmath29 result . a @xmath30 limit on the size of the compact source is @xmath60.3@xmath2 . as illustrated in figure 2 , there appears to be a measureable flux , approximately 20% of the total at 12.5 @xmath0 m , that extends beyond the central core , and is contained within 4@xmath2 . the figure shows clearly that the flux measured in iras 08572 + 3915 continues to increase beyond the beam where the flux in the psf star has stopped increasing . figure 3 and table 3 compare the keck photometry in a 4@xmath2 diameter beam for iras 08572 + 3915 with the iras measurements . the mirlin photometry accounts for 70% of the iras flux density at 12.5@xmath0 m . the mirlin observations were taken with comparatively narrow filters , i.e. , @xmath31/@xmath32 0.1 , from 7.9@xmath0 m to 17.9@xmath0 m , while the iras 12@xmath0 m filter bandpass extends from 8 - 15@xmath0 m . given the uncertainties in the corrections for the different bandpasses , the mirlin data are consistent with all the flux detected by iras being confined to a 4@xmath2 diameter beam surrounding the detected source although it is possible that some of the flux could be accounted for by extended emission associated with the northwest source or from the southeastern source . neither the mirlin nor the lws observations detected the galaxy approximately 5@xmath2 southeast of the object centered in figure 1b . the mirlin data limit the contribution of the southeastern source to @xmath2430% of the total 12.5@xmath0 m flux ; it is likely that the southeast galaxy does not contribute significantly to the total infrared luminosity in this system . this is consistent with the optical and near infrared morphology and colors of this galaxy , which are indicative of quiescent galaxy nucleus ( surace et al . 1998 , surace and sanders 1999 ) . mirlin observations of ugc 5101 were obtained at 12.5 and 17.9 @xmath0 m . contour plots of this source in figure 1c show that at all wavelengths the structure is basically the same : a bright central core centered on an apparently elliptical structure having a major axis at a position angle of @xmath680@xmath33 . while the position angle of the elliptical structures agree well at 2.2@xmath0 m and 8.4 ghz , there is a suggestion that the position angle of the ellipse at 12.5 @xmath0 m is @xmath615@xmath33 smaller . at 8.4 ghz the central source contains 80% of the radio flux(c91 ) , while at 2.2@xmath0 m , the unresolved peak contains 35% of the flux in a 5@xmath2 diameter beam ( s99 ) . as shown in figure 2 , at 12.5 @xmath0 m approximately half of the total flux measured in a 4@xmath2 diameter beam is contained in the central compact source . the inner central compact source is indistinguishable from the psf ( figure 2 ) . from the dispersion in the sizes determined for the psf star ( bs3775 ) the compact source has a size ( 3@xmath34 ) less that 0.22@xmath2 . the size measured at 17.9@xmath0 m is consistent with that measured at 12.5 @xmath0 m . the data of figure 3 and table 3 show that at 12.5 @xmath0 m the flux measured by mirlin is @xmath660% of that measured by iras . this suggests that there is low surface brightness emission extended on a larger scale than 4@xmath2 . the 12.5@xmath0 m flux density contained in the compact source is @xmath630% of the total flux density at 12@xmath0 m measured by iras . markarian 231 has long been known both as an infrared luminous system ( rieke and low , 1972 ) and a seyfert 1 system ( boksenberg et al . the iras survey found that it is the most luminous object within 300 mpc , with a bolometric luminosity of @xmath35 . mirlin images of markarian 231 were obtained at seven wavelengths from 7.9 to 17.9@xmath0 m . in addition , lws imaging was obtained at 12.5@xmath0 m . figure 1d shows contour plots of markarian 231 at 12.5 @xmath0 m from the lws data and data at 8.4 ghz(c91 ) . the emission at all the mid - infrared wavelengths is dominated by the very bright central source , and is effectively unresolved . the radio continuum is entirely from the bright nucleus . lws imaging was used to establish the size of the central source at 12.5@xmath0 m . the lws observations consisted of eleven measurements of the size of markarian 231 , interleaved with nine measurements of a near - by psf star(bs 4905 ) over a period of two hours . the average fwhm of the psf star and markarian 231 were indistinguishable and were slightly greater than the diffraction limit of the telescope and the rms ( population ) scatter in the fwhm was 0.04@xmath2 . the central source thus is unresolved in these observations . we have chosen to set a 3@xmath34 limit on its fwhm of 0.13@xmath2 on the basis that a difference three times the average standard deviation of the fwhm would have been easily detected . the sed of markarian 231 is plotted in figure 3 , and shows clearly that all the flux measured by iras is detected in the 4@xmath2 diameter beam . the sed shows a dip at 10@xmath0 m apparently due to foreground silicate absorption of the central source . the mirlin observations of markarian 273 consist of imaging at seven wavelengths from 7.9@xmath0 m to 17.9@xmath0 m . figure 1e shows a comparision of the mirlin 12.5@xmath0 m data with the 8.4 ghz map ( c91 ) and the nicmos 2.2@xmath0 m image ( s99 ) , while a montage of images of markarian 273 is shown in figure 4 . at 12.5@xmath0 m the source structure is basically double , with the northern source brighter than the southwestern source at most wavelengths . a similar morphology is seen at 2.2@xmath0 m ( s99 ) , but not in the radio ( c91 ) . in comparing the 12.5@xmath0 m , nicmos and 8.4 ghz data we have assumed the spatial coincidence between the bright northern source in each image . the overall ageement between the 12.5@xmath0 m and nicmos images is excellent , while the discrepancy between the positions of the southwestern source in the infared and radio is real . the position angle between the northern and southwestern sources is the same ( to within the uncertainties ) in the 12.5@xmath0 m , 2.2@xmath0 m and 8.4 ghz images , while the source separations are 1.0@xmath2 , 1.0@xmath2 and 1.2@xmath2 at these same wavelengths . the apparent separation between the infrared and radio sources , 0.2@xmath2 , corresponds to 160pc at the distance of markarian 273 . note also that the flux ratio between the northern and southwestern source is considerably different between the infrared and radio . as shown in figure 4 , the structure seen in markarian 273 at all infrared wavelengths from 3.4@xmath0 m to 17.9@xmath0 m is basically the same . there is no evidence for a mid - infrared counterpart to the southeastern source seen at 8.4 ghz by c91 ( figure 1e ) and at 5 ghz by knapen et al . ( the bright source present to the southeast in the radio map is suggested to be a background radio object by knapen et al . , although the statistical likelihood of a background source falling within 1@xmath2 of the location of the radio core of markarian 273 is @xmath36 . there is a faint compact object found in the nicmos images ( s99 ) at the location of southeast radio source , although there is no evidence for any mid - infrared emission at this location . the upper limit on the 12.5@xmath0 m flux density at the location of the southeastern radio source of 15 mjy is 1/10 that of the northern source . this limit , coupled with the 8.4ghz flux density of this source of 3.2 mjy ( c91 ) , leads to a limit on the ratio of flux densities @xmath37 @xmath24 4.7 , consistent with the observed ratio @xmath38 @xmath394.5 for the northern source . by comparison , if the radio ( 0.2mjy at 8.4 ghz , c91 , knapen et al . ) and infrared ( 58 mjy at 12.5@xmath0 m ) flux densities of the southwestern source are assumed to be associated with the same physical source , the ratio @xmath40 @xmath39 290 , a factor of 65 times greater than that seen in the northern source , making this nucleus far more infrared bright relative to the radio than the northern source . if the southwestern radio and infrared sources are not related , the southwestern infrared source has an even larger @xmath37 ratio . as can be seen in figures 1e and 4 , the relative brightnesses of the nuclei change substantially as a function of wavelength . in order to get an estimate of the relative flux emitted by the two nuclei , the flux in two identically sized square beams was measured at the same locations on the images at all wavelengths presented ; the beams , 1.6@xmath2 on a side , are shown in figure 4 . the sky was taken northwest and southeast of each of the squares . the efficiency of the square beams was estimated by placing the psf of the appropriate wavelength at the locations of the nuclei within the square beam . the photometry of the nicmos image was arbitrarily normalized so that the sum of the fluxes from the two squares equals the sum of the fluxes give by knapen et al . ( 1997 ) and the flux density in a 4@xmath2 diameter beam equals that of knapen et al . the results of these photometric measurments are shown in figure 5 . at 3.4@xmath0 m and 10.3@xmath0 m the southwestern source is slightly ( 10% ) brighter than the northern source . at the other mid - infrared wavelengths the northern source is between a factor of 1.5 and 3 times brighter than the southwestern source . the different sed s of the components implied by their changing relative brightnesses is shown directly in figure 5 , where photometry of the two nuclei is shown . the northern source clearly shows a signficantly deeper absorption at 10@xmath0 m than the southwestern source . this is consistent with the distribution of co emission which strongly peaks on the northern source ( downes and solomon , 1998 ) , but is inconsistent with the near infrared colors found by s99 , where the northern source has a smaller j - k color than does the southwest source . the sed of figure 5 , particularly the relatively strong emission at 3.5 and 11.7 @xmath0 m , suggests aromatic hydrocarbon emission ( pah ) emission associated with the southwest nucleus . such emission is seen in the spectra of many ulirgs ( genzel et al . the emission from the southwest source could account for all of the 11.3@xmath0 m emission seen in the spectrum of the combined nuclei by dudley(1999).the northern source does not show a similar peak at 11.7@xmath0 m , but clearly shows the effects of substantial overlaying cold silicate absorption . the mirlin photometry shown in figure 3 was obtained using a 4@xmath2 diameter beam centered between the two nuclei . the comparison of the mirlin and iras data ( figure 3 ) demonstrates that the mid - infrared emission in this source emerges virtually entirely from the central region . because of the proximity of the two nuclei and the fairly low signal - to - noise ratio of the nuclei in some of the images , it is difficult to cleanly determine the sizes of the two sources . the data suggest that both nuclei have unresolved mid - infrared cores with sizes @xmath240.4@xmath2 . in the northern nucleus approximately 2/3 of the 12.5@xmath0 m flux is unresolved , while in the southwestern nucleus , about half the flux is unresolved . the data for this source were discussed in detail in soifer et al . ( 1999 ) and are included in figures 1f and 3 for completeness . mirlin imaging was obtained at 7.9 and 12.5@xmath0 m . the 12.5@xmath0 m image is shown in figure 1 g . the 2.2@xmath0 m nicmos image of iras 17208 - 0014 ( figure 1 g , s99 ) shows it to be an extended source of roughly 2@xmath2 diameter with a number of luminous starburst centers to the northeast of the nucleus . the mid - infrared image is apparently also similar , but the limitation due to the chop - nod spacing and the relatively low infrared surface brightness results in an image with low signal - to - noise ratio . the size of the 12.5@xmath0 m source of @xmath62@xmath2 or 1.6 kpc fwhm ( figure 2 ) is quite similar as well to that found in co by downs and solomon ( 1998 ) and jogee et al.(2000 ) . the best illustration of the spatial distribution of the mid - infrared emission is shown in figure 2 . the source flux smoothly increases up to 5@xmath2 diameter , with a fwhm of 2@xmath2 ; the apparent leveling off of the enclosed flux at a diamteter of 5@xmath2 is caused by the artificial imposition of the sky level between diameters of 5 and 6@xmath2 . evidence that the source is not extended much beyond a diameter of 5@xmath2 is provided by figure 3 which shows that , within the uncertainties , the enclosed flux within a 4@xmath2 diameter beam is the same as within the iras beam . the lack of a dominant compact component in the mid - infrared make this object very different from the other ulirgs in this sample . the radio emission from iras 17208 - 0014 is much more compact than the infrared emission . at least 90% of the 1.4ghz flux comes from a region smaller than 2@xmath162@xmath2 ( fwhm , condon et al . 1996 ) . martin et al . ( 1989 ) measure a radio source with a size of 0.26@xmath220.32@xmath2 at 1.6 ghz that accounts for @xmath690% of the 1.4 ghz flux density , suggesting a compact radio core . this core is clearly not dominant in the mid - infrared imaging . the near infrared nicmos imaging of this galaxy ( s99 ) suggests that the nucleus could be obscured at 2.2@xmath0 m , but the lack of detection of a strong compact core at 12.5@xmath0 m is quite peculiar , since the radio source is so compact . the data presented here , with angular resolution of @xmath240.3@xmath2 , represent the highest spatial resolution achieved to date in probing the central regions of ulirgs in the thermal infrared . table 4 summarizes the key observations , the sizes of the compact sources found in the ulirgs at 12.5@xmath0 m , 17.9@xmath0 m and 24.5@xmath0 m , as well as the fraction of the total flux at these wavelengths emitted from these cores . all of the ulirgs except iras 17208@xmath210014 have a nuclear source in the mid - infrared of size @xmath41200 pc that produces a substantial fraction , if not all , of the galaxy s mid - infrared luminosity . in three sources , iras 05189@xmath212524 , iras 08572 + 3915 and markarian 231 , a single unresolved nucleus produces between 80% and 100% of the total mid - infrared luminosity . in markarian 273 and arp 220 , two nuclei produce nearly all the mid - infrared emission in the galaxies . in the case of markarian 273 the nuclei are substantially unresolved , while in arp 220 at least one nucleus is resolved . because of its low redshift the linear size of the resolved nucleus in arp 220 is smaller than the limit on all the unresolved nuclei except markarian 231(see table 5 ) . in only one compact system , ugc 5101 , does the compact nucleus produce significantly less than half the total mid - infrared emission from the system . as noted previously , iras 17208@xmath210014 presents an interesting puzzle . there is no mid - infrared evidence for a compact nuclear source , but the radio data ( c91 , martin et al . 1989 ) strongly suggest a power source with a size @xmath240.3@xmath2 . in the following discussion , we will adopt the morphology implied by the 12.5@xmath0 m image , but we note that extremely high 12.5@xmath0 m extinction could be leading to an erroneous interpretation of the nuclear structure in iras 17208@xmath210014 . table 5 shows that the compact nuclear sources have linear sizes typically @xmath24200 pc in diameter . in arp 220 the diameter of the western source is 140 pc at 24.5@xmath0 m ( soifer et al . 1999 ) , while in markarian 231 , the diameter of the 12.5@xmath0 m source is @xmath24100 pc . one of the goals of the observations reported here is to distinguish the ultimate source of the infrared luminosity in these systems by means of the spatial extent of the infrared emission . if the dust is heated by a bright compact object such as an agn , the dust emission will be compact , and the size of the infrared source will increase with increasing wavelength roughly as @xmath42 , since the dust temperature will decrease with radius as @xmath43 and the wavelength of maximum emission @xmath44 . if , on the other hand , the heating is due to many hot stars in a starburst region , then the dust emission will be distributed on the same scale as the young , luminous stars . since the dust heating is due to the proximity of the dust to the luminous young stars there will be little or no temperature gradient seen . spatially extended 10 @xmath0 m emission therefore implies starburst activity , while spatially compact 10 @xmath0 m emission could be due to either a compact starburst or an agn . to quantify the compact mid - infrared emission we utilize the brightness temperatures derived from our observations and the color temperatures determined from the iras 25 and 60@xmath0 m observations of these objects . we assume that t@xmath45 measures the temperature of the grains heated by the luminosity source ; t@xmath46 should be @xmath14t@xmath45 if the optical depth @xmath47 . only in the limit of large optical depth should the brightness temperature approach the color temperature . based on the mirlin observations of arp 220 , soifer et al . ( 1999 ) showed that the dominant luminosity source , arp 220w , is optically thick with a brightness temperature at 25 @xmath0 m equal to the color temperature between 25 and 60 @xmath0 m . the observations reported here permit analogous constraints to be placed on some of the nuclear sources in other ulirgs in this sample . in markarian 231 , iras 05189@xmath212524 and iras 08572 + 3915 , the limits on the sizes of the central infrared sources are sufficiently small that it is worthwhile considering how these sizes constrain the far infrared sources . the size limit on markarian 231 of 0.13@xmath2 at 12.5@xmath0 m places the most severe limits on models of the infrared emitting region . in this case the observed flux density and size limit lead to a lower limit on the brightness temperature of t@xmath48 141k at 12.5@xmath0 m . if the infrared emitting source has the same size at 17.9 @xmath0 m , 25 @xmath0 m and 60 @xmath0 m , the corresponding brightness temperatures are 124 k , 128 k and 241 k. the brightness temperatures at 12.5@xmath0 m , 17.9@xmath0 m and 25 @xmath0 m are consistent with each other and with a nearly optically thick source over this wavelength range , but the 60 @xmath0 m brightness temperature is much too large to be consistent with dust models with reasonable grain properties . this implies that the 60@xmath0 m source size is much larger than the 12.5@xmath0 m size . the color temperature measured between 25 and 60 @xmath0 m , determined from the iras measurements , is 85 k. the size of a blackbody of 85 k required to produce the observed 60@xmath0 m flux density is 0.4@xmath2 , significantly larger than the limit on the source size at 12.5@xmath0 m . the source size determined if t@xmath26 t@xmath46 represents a lower limit on the size of the 60@xmath0 m source . for example , if the size of the 60@xmath0 m source were equal to the @xmath61@xmath2 size found in co by downs and solomon ( 1998 ) , the 60@xmath0 m brightness temperature would be 53k , much less than that inferred by assuming all the 25 and 60 @xmath0 m emission is emerging from the same region . the size limits are based on the assumption of large optical depth ( @xmath491 ) ; smaller optical depth would imply an even larger source . since significant emission is seen at @xmath50 m , either @xmath511 or we are viewing a tilted optically thick disk , as suggested for arp 220 ( scoville et al . 1998 , downs and solomon , 1998 , sakamoto et al . 1999 , soifer et al . 1999 ) . in either case , the size constraints show that in markarian 231 the infrared source increases in size with increasing wavelength . this is consistent with the picture of an agn with luminosity of 3@xmath1510@xmath52 l@xmath53 heating the surrounding dust and producing a radial temperature gradient . in this sample there are two additional ulirgs where significantly more than 50% of the mid - infrared luminosity emerges from a compact central source , iras 05189 - 2524 and iras 08572 + 3915 . in iras 05189 - 2524 , if the size of the 60@xmath0 m emitting source is the same as the limit on the 12.5@xmath0 m source measured here , t@xmath46(60@xmath0m)=101 k , which is significantly greater than t@xmath5485 k from the iras 25 and 60@xmath0 m data . for the same assumptions , in iras 08572 + 3915 t@xmath46(60@xmath0m)=77 k is slightly less than t@xmath5483 k determined from the iras data at 25 and 60@xmath0 m . for both sources , using dust properties described by draine and lee ( 1984 ) the sizes calculated for silicate and graphite dust grains heated by central sources of the observed bolometric luminosity , and having dust temperatures equal to the observed color temperatures are larger by a factor of 1.5 to 3 than the observed sizes of these sources at 12.5@xmath0 m . this suggests that the sizes of these infrared sources also increase with increasing mid and far - infrared wavelength . high apparent surface brightnesses and a radial temperature gradient are natural consequences of centrally heated dust emission surrounding a high luminosity source , i.e. , an agn . in markarian 231 there is direct evidence that the size of the infrared source increases with increasing wavelength in the mid and far infrared . in iras 05189 - 2524 and iras 08572 + 3915 , there is strong evidence for such radial temperature gradients based on the compact sources observed in the mid - infrared . the observed luminosities are adequate to quantitatively account for the inferred variation of dust temperature with radius if the dust is heated by a central source in these objects . the compact nature of the sources in the objects observed leads to very high luminosities produced in small volumes . by comparing these energy production rates with those found in local examples of starburst environments , we can determine whether known starburst environments can generate the apparent energy production in the ulirgs . in the core of orion the luminosity of @xmath55 emerges from a 0.3 pc size region ( werner et al . 1976 ) , while in the nuclear starburst in m82 , the nearest starburst galaxy , the luminosity of @xmath56 is emitted over a projected area of 450@xmath1575pc ( 30@xmath165@xmath2 ) ( rieke et al . these observations lead to apparent surface brightnesses of emergent luminosity of @xmath57/kpc@xmath58 . if , as seems likely , m82 is being viewed edge - on , then the mean surface brightness of this system , if viewed face on , is closer to 2@xmath59/kpc@xmath58 . a survey of the maximum surface brightness in nearby starburst galaxies in the uv , infrared and radio by meurer et al . ( 1997 ) leads to peak global averaged surface brightnesses in starburst galaxies of @xmath60/kpc@xmath58 . meurer et al . find that the peak surface brightnesses of clusters in starburst galaxies reaches @xmath61/kpc@xmath58 over regions @xmath610 pc in diameter . meurer et al . suggest that the global peak surface brightnesses is due to a mechanism whereby the pressure from stellar winds and supernovae is able to regulate the star formation rates in galaxy disks , while the formation of massive stars in giant hii regions is regulated by the `` kennicutt law '' , i.e. star formation occurs when the surface gas density exceeds a critical value that depends on the rotational velocity and shear ( kennicutt , 1989 ) . the surface brightnesses observed in the ulirgs can be estimated if we assume that the size ( or limit ) of the region measured in the mid - infrared is representative of the infrared emitting region as a whole . with this assumption , we have presented the surface brightnesses for the observed ulirgs in table 5 . we include two values of the surface brightness , a `` maximum '' which assumes that the fraction of the total infrared luminosity emerging from the compact core is the same as the fraction of the mid - infrared flux density emerging from the core , and a `` minimum '' that assumes that only the _ observed _ mid - infrared luminosity emerges from the compact core . in the case of markarian 231 , the former number is unphysically large ( section 5.2 ) . in the other systems , the two values represent likely upper and very conservative lower limits on the apparent surface brightness of the infrared emission . only in the case of iras 17208@xmath210014 , where the near and mid - infrared observations suggest the luminosity is due to star formation , is the measured infrared surface brightness comparable to the peak global value @xmath62/kpc@xmath58 found by meurer et al.(1997 ) . other than this case , the inferred surface brightnesses are much greater than found globally in disk starburst systems . the highest apparent surface brightnesses in the ulirgs are found in the western nucleus of arp 220 , which is most likely powered by a nuclear starburst ( smith et al . 1998b ) and the cores of markarian 231 and iras 05189@xmath212524 , which are probably powered by agn . in these cores the apparent surface brightness approaches or surpasses the surface brightness of clusters in starburst galaxies , while in the other systems ( excluding iras 17208@xmath210014 ) the surface brightness most probably exceeds @xmath63/kpc@xmath58 . in all these cases the surface brightness over regions of hundreds of parsecs across are comparable to the brightness found in @xmath610 pc sized clusters within starburst systems by meurer et al . for the ulirgs that are powered by star formation , the large surface brightnesses and sizes suggest that the global starburst intensity regulating mechanism of meurer et al . ( 1997 ) does not apply . gravitational instability allows for higher gas densities and hence higher star - formation rates in the solid - body ( no shear ) rotation nuclear regions . since the nuclei of most ulirgs are mergers , far from equilibrium , the gas densities may be very high , leading to high star formation rates and correspondingly high surface brightnesses for short times . ulirgs are clearly extraordinary events in the evolution of galaxies . several studies ( e.g. , sanders et al . 1988 , scoville , yun and bryant 1997 , downes and solomon 1998 ) suggest that the nuclear disks are unstable in some ulirgs . the large gas masses in these environments might then lead to nuclear star formation rates much greater than those found in less extreme systems . table 6 summarizes the characterization of the ulirgs applying many different diagnositics of the underlying energy sources . here we have summarized the structure in the mid - ir from the data presented here , in the near infrared from s99 and surace and sanders ( 1999 ) , and in the radio from the vlbi data of smith et al . the optical spectral classifications are from sanders et al . ( 1988 ) and veilleux et al ( 1995 ) . the near - infrared spectral classifications are from veilleux et al . ( 1999 ) and murphy ( private communication ) . the iso spectral classifications are based on the strength of the pah features and are taken from genzel et al . ( 1998 ) and lutz , veilleux and genzel ( 1999 ) . the x - ray classifications are from asca results and are taken from iwasawa ( 1999 ) and nakagawa et al . ( 1999 ) . in none of the objects in this sample , excepting perhaps markarian 273 and iras17208 - 0014 , is there a completely consistent set of `` classifications '' based on the wide range of wavelengths over which these objects are observed . in the object that is the clearest case for a central agn luminosity source , markarian 231 , the hard x - ray emission is not consistent with the simplest view of such power sources ( nakagawa et al . 1999 ) . in the predominantly seyfert nucleus iras 05189 - 2524 there is not a dominant compact radio core . in markarian 273 the two nuclei clearly seen in high resolution imaging complicates the classification , since none of the spectral classifications resolved the separate nuclei . in ugc 5101 , another source where the optical , infrared and x - ray indications point towards a starburst power source , a compact radio core characteristic of a central agn exists . compact mid - infrared sources occur in six of seven of these sources , where other evidence points strongly to both agn and starbursts as the dominant underlying power source . table 6 shows that most ulirgs do not present a picture over all wavelengths consistent exclusively with either an agn or starburst power source . in most objects observed here , it is likely that both contribute to some extent to their total bolometric luminosity , though markarian 231 and iras 17208 - 0014 appear to represent the extremes of dominant agn and starburst sources . diffraction limited mid - infrared observations of a sample of seven ulirgs obtained on the keck 10 m telescopes have shown : 1 - extremely compact structures , corresponding to emission on spatial scales of @xmath24100@xmath21200 pc , are seen in six of the seven systems observed . the mid - infrared emission in these galaxies generally emerges from a single region of diameter @xmath24 1 kpc , and in most cases from regions @xmath24200 pc in diameter . 2 - the upper limit on the diameter of the 12.5@xmath0 m source in markarian 231 , 0.13@xmath2 , is too small to be consistent with the observed 60@xmath0 m flux density emerging from a region this size , and demonstrates that in this agn the source size increases with increasing wavelength ( decreasing dust temperature ) . 3 - in iras 05189 - 2524 and iras 08572 + 3915 there is strong circumstantial evidence that the size of the infrared source increases with wavelength between 12.5 and 60 @xmath0 m , suggesting that these too are predominantly agn powered ulirgs . 4 - in the other objects observed , the angular size limits for the compact mid - infrared sources are not yet adequate to distinguish between agn and starburst luminosity sources , although the small sizes are consistent with a central luminosity source heating the surrounding dust as in agn models in all the sources except iras 17208 - 0014 . 5 - if star formation powers the luminosity in the very compact infrared sources seen in these systems , the star formation rates averaged over a few hundred parsecs is comparable to that seen in the brightest @xmath610 pc sized clusters in nearby starbursts , and exceeds by factors of up to 100 , the global star formation rates in nearby starburst galaxies . we thank j. aycock and r. campbell for assistance with the observations , r. goodrich , r. moskitis and the entire keck staff for their help establishing the visitor port and to barbara jones , rick puetter and the keck team that brought the lws into service , enabling these observations . the w. m. keck observatory is operated as a scientific partnership between the california institute of technology , the university of california and the national aeronautics and space administration . it was made possible by the generous financial support of the w. m. keck foundation . this research has made use of the nasa / ipac extragalactic database which is operated by the jet propulsion laboratory , caltech under contract with nasa . b.t.s , g.n . , k.m . and e.e . are supported by grants from the nsf and nasa . j.a.s . and b.t.s . are supported by the sirtf science center at caltech . sirtf is carried out at the jet propulsion laboratory . e.e.b . , a.j.w , n.z.s . and a.s.e . were supported by nasa grant nag5 - 3042 . this work was carried out in part ( m.w.w . , m.r . ) at the jet propulsion laboratory , operated by the california institute of technology , under an agreement with nasa . the development of mirlin was suported by nasa s office of space science . the national radio astronomy observatory ( j.j.c . ) is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . . images of iras 05189 - 2524 obtained at 2.2@xmath0 m ( s99 , upper left ) , 8.4 ghz ( c91 , lower left ) , and 12.5@xmath0 m ( this work , upper right ) . the lower right panel is the psf star image obtained at 12.5@xmath0 m contemporaneously with those of iras 05189 - 2524 and used to determine the 12.5@xmath0 m size of iras 05189 - 2524 . the peak of each image is at the location 0,0 in each panel . the contours in each panel are set so that the peak contour is 90% of the maximum surface brightness . the contours are spaced down by a factor of 1.34 so that the third contour down from the peak is at 50% of the peak surface brightness . the ninth coutour down is 8.5% of the peak . below this level the contours are spaced by a factor of two . the faintest contour is set at about 3 times the noise in the image . the mid - infrared data are from the mirlin observations . the hatched circle show the fwhm of the psf for the 2.2@xmath0 m observations . figure 1b . images of iras 08572 + 3915 . the format , source locations and contour levels are the same as in figure 1a . the mid - infrared data are from the lws observations . the 2.2@xmath0 m and 8.4 ghz data are from s99 and c91 respectively . the hatched circles show the fwhm of the psf for the appropriate observations . figure 1c . images of ugc 5101 . the format is 2.2@xmath0 m image(s99 ) - left , 12.5@xmath0mimage - center , 8.4 ghz image(c91 ) - right . the source locations and contours are set as in figure 1a . the mid - infrared data are from the mirlin observations . the 2.2@xmath0 m and 8.4 ghz data are from s99 and c91 respectively . the hatched circles show the fwhm of the psf for the observations . figure 1d . images of markarian 231 . the format is 12.5@xmath0 m image - left , 12.5@xmath0 m psf star - center , 8.4 ghz image - right . the source locations and contours are set as in figure 1a.the mid - infrared data are from the lws observations . the 8.4 ghz data are from c91 . the hatched circle in the 8.4 ghz panel shows the fwhm of the psf for this observation . figure 1e . images of markarian 273 . the format , source locations and contour levels are the same as in figure 1c . the mid - infrared data are from the mirlin observations . the 2.2@xmath0 m and 8.4 ghz data are from s99 and c91 respectively . the @xmath64 symbols are at the same coordinates in each panel , and are centered on the peaks of the 2.2@xmath0 m image . there is excellent agreement between the locations of the 2.2@xmath0 m and 12.5@xmath0 m peaks . the discrepancy between the positions of the infrared peaks and the radio peak is real ( see text ) . the hatched circles show the fwhm of the psf for the observations . figure 1f . images of arp 220 . the format , source locations and contour levels are the same as in figure 1c . the mid - infrared data are from the mirlin observations and are from soifer et al . the 2.2@xmath0 m and 8.4 ghz data are from s99 and c91 respectively . the hatched circle in the 12.5@xmath0 m panel shows the fwhm of the psf for the observation . figure 1 g . images of iras 17208 - 0014 . the left image is 2.2@xmath0 m , the right image is 12.5@xmath0 m . the source locations and contour levels are the same as in figure 1a . the mid - infrared data are from the mirlin observations . the 2.2@xmath0 m data are from s99 . the hatched circles show the fwhm of the psf for the observations . the curves of growth for the objects ( solid ) and psf s ( dashed ) for the five ulirgs which are single nuclei . in all five the psf s have been normalized so the enclosed flux of the psf within a radius of 2.5@xmath2 equals that of the object . for the objects , the sky level was set to be the value in an annulus between 2.5 and 3@xmath2 . in the plots for iras 08572 + 3915 and ugc 5101 two normalizations are used for the psf ; in the lower curve the psf is set to match the central profile of the galaxy , while the upper curve matches the enclosed flux from the psf and the galaxy at the outer radius of the measurement beams , as for the other galaxies . the data of iras0857 + 39 and markarian 231 were obtained with lws as outlined in the text ; the other three data sets are from mirlin observations . the spectral energy distributions of the seven ulirgs studied are shown as a function of the rest wavelength from @xmath61 @xmath0 m to 1 mm . filled circles represent the mirlin data , while open circles have been taken from carico et al . ( 1988 ) ( near infrared data , 5@xmath2 diameter beam ) and soifer et al . ( 1989 ) ( iras data @xmath61@xmath65 diameter beam ) . the mirlin photometry has been calculated for a 4@xmath2 diameter beam centered on the object or between the two nuclei if multiple nuclei are present . as discussed in the text , the photometry is based on iras standards . montage of images of markarian 273 . the images at wavelengths from 7.9@xmath0 m through 17.9@xmath0 m obtained with mirlin . the image at 3.4@xmath0 m was obtained on the 200-inch hale telescope using the cassegrain infrared camera . the images are centered so that the northern peak is at the location 0,0 in each panel . the locations of the boxes in the 12.5@xmath0 m image show where the photometry of the peaks was obtained in all the images . spectral energy distributions of the separate peaks in markarian 273 as determined from the images presented in figure 4 . the total flux ( also presented in figure 3b ) was measured in a 4@xmath2 diameter beam centered between the peaks and was distributed between the two sources by doing photometry in non - overlapping boxes ( displayed in the 12.5@xmath0 m panel of figure 4 ) centered on the two peaks . the uncertainties shown are statistical . the systematic uncertainty depends on the wavelength , but is probably in the range 15 - 20 % . iras05189 - 2524 & 0.043 & 12.08 & sey 1 & 830 + iras08572 + 3915 & 0.058 & 12.10 & liner & 1200 + ugc 5101 & 0.040 & 12.00 & liner & 820 + markarian 231 & 0.040 & 12.52 & sey 1 & 800 + markarian 273 & 0.037 & 12.13 & sey & 800 + arp 220 & 0.018 & 12.17 & liner & 360 + iras17208 - 0014 & 0.040 & 12.46 & hii & 800 + iras05189 & & & & & & @xmath68 & & @xmath68 + iras08572 & @xmath68 & @xmath68 & @xmath68 & & & @xmath68 & @xmath68 & + ugc 5101 & & & & & & @xmath68 & @xmath68 & + mrk 231 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & + mrk 273 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & + arp 220 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 & @xmath68 + iras17208 & @xmath68 & & & & & @xmath68 & & + iras05189 & & & & & & 720 & 740 & & 3.2@xmath280.2 & 3.5 & 13.9 + iras08572 & 615@xmath2818 & 434@xmath2810 & 221@xmath2810 & & & 218@xmath2810 & 340 & 552@xmath2846 & & 1.8 & 7.5 + ugc 5101 & & & & & & 185@xmath2810 & 260 & 256@xmath2842 & & 1.1 & 12.1 + mrk 231 & 1370@xmath2843 & 1488@xmath2818 & 1081@xmath2830 & 1075@xmath2821 & 1355@xmath6915 & 1836@xmath2820 & 1930 & 3249@xmath2849 & & 8.8 & 33.6 + mrk 273 & 267@xmath2815 & 141@xmath2810 & 92@xmath2810 & 38@xmath286 & 150@xmath285 & 210 @xmath2810 & 230 & 442@xmath2831 & & 2.3 & 23.7 + arp 220 & 521@xmath2815 & 155@xmath286 & 124@xmath2814 & 59@xmath2810 & 175@xmath288 & 404 @xmath2812 & 500 & 1170@xmath2850 & 9.8@xmath280.2 & 8.0 & 104.0 + iras17208 & 315@xmath2825 & & & & & 202@xmath2813 & 200 & & & 1.7 & 34.1 + iras05189&100 & @xmath70 0.20 & & & 100 & @xmath240.26 & @xmath71 & @xmath240.2 + iras08572 & @xmath680 & @xmath240.22 & @xmath6100 & @xmath240.25 & & & @xmath72 & @xmath240.2 + ugc 5101 & 50 & @xmath240.25 & 60 & @xmath240.25 & & & @xmath73 & @xmath240.2 + mrk 231 & 100 & @xmath240.13 & 85 & @xmath240.25 & & & @xmath74 & @xmath240.2 + mrk 273n & 67 & @xmath240.3 & & & & & @xmath75 & @xmath60.3 + mrk 273sw & 50 & @xmath240.3 & & & & & @xmath76 & @xmath240.2 + arp 220 e& & & & & 25 & & @xmath77 & @xmath240.2 + arp 220 w& & & & & 75 & @xmath78 & @xmath79 & 0.4 + iras17208 & 100 & 2.0 & & & & & @xmath80 & 2 + iras05189 & @xmath240.2 & @xmath24166 & @xmath81 & @xmath82 + iras08572 & @xmath240.22 & @xmath24260 & @xmath83 & @xmath84 + ugc 5101 & @xmath240.25 & @xmath24205 & @xmath85 & @xmath86 + mrk 231 & @xmath700.13 & @xmath24104 & @xmath87 & @xmath88 + mrk 273n & @xmath240.3 & @xmath24240 & @xmath89 & @xmath86 + arp 220 & 0.40 & 140 & @xmath90 & @xmath91 + iras17208 & 2.0 & 1600 & @xmath92 & @xmath93 + iras05189 & y / y / n & sey2 & sey & & agn + iras08572 & y / y / n & liner & hii & & sb + ugc 5101 & y / y / y & liner & & sb & sb + mrk 231 & y / y / y & sey1 & sey1 & agn & sb ? + mrk 273 & y / y / y & sey2 & sey ? & agn & agn + arp 220 & y / n / y & liner & & sb & sb ? + iras17208 & n / n/ ? & hii & hii & sb & +

we improve the recently developed functional renormalization group ( frg ) for impurities and boundaries in luttinger liquids by including renormalization of the two - particle interaction , in addition to renormalization of the impurity potential . explicit flow equations are derived for spinless lattice fermions with nearest neighbor interaction at zero temperature , and a fast algorithm for solving these equations for very large systems is presented . we compute spectral properties of single - particle excitations , and the oscillations in the density profile induced by impurities or boundaries for chains with up to lattice sites . the expected asymptotic power - laws at low energy or long distance are fully captured by the frg . results on the relevant energy scales and crossover phenomena at intermediate scales are also obtained . a comparison with numerical density matrix renormalization results for systems with up to 1000 sites shows that the frg with the inclusion of vertex renormalization is remarkably accurate even for intermediate interaction strengths .

cond-mat0403517

low dimensional electron systems exhibit a rich variety of surprising effects which are due to the cooperative interplay of impurities and interactions . in one dimension even clean metallic systems are always strongly affected by interactions : at low energy scales physical properties obey anomalous power - laws , known as luttinger liquid behavior , which is very different from the conventional fermi liquid behavior describing most higher dimensional metals.@xcite in luttinger liquids with repulsive interactions already a single static impurity is known to affect the low - energy properties drastically.@xcite the impurity potential in a repulsive luttinger liquid becomes dressed by long - range oscillations which suppress the spectral weight for single - particle excitations near the impurity and also the conductance through the impurity down to zero in the low - energy limit . the _ asymptotic _ low - energy properties of luttinger liquids with a single impurity are rather well understood . universal power - laws and scaling functions have been obtained by bosonization , conformal field theory , and exact solutions for the low - energy asymptotics in special integrable cases.@xcite what remains to be developed , however , is a many - body method for _ microscopic _ models of interacting fermions with impurities , which does not only capture correctly the universal low - energy asymptotics , but allows one to compute observables on all energy scales , providing thus also _ non - universal _ properties , and in particular an answer to the important question at what _ scale _ the ultimate asymptotics sets in . that scale can indeed be surprisingly low , and the properties above it very different from the asymptotic behavior . some of the non - universal properties can be computed numerically by the density matrix renormalization group ( dmrg),@xcite but this method is limited to lattice systems with about 1000 sites , and only a restricted set of observables can be evaluated with affordable computational effort . in the last few years it has been realized that the functional renormalization group ( frg ) is a source of powerful new computation tools for interacting fermi systems , especially for low - dimensional systems with competing instabilities and entangled infrared singularities . the starting point of this approach is an exact hierarchy of differential flow equations for the green or vertex functions of the system , which is obtained by taking derivatives with respect to an infrared cutoff @xmath1.@xcite approximations are then constructed by truncating the hierarchy and parametrizing the vertex functions with a manageable set of variables or functions.@xcite a relatively simple frg approximation for impurities in spinless luttinger liquids has been developed recently by some of us.@xcite the scheme starts from an frg hierarchy for one - particle irreducible vertex functions , as first derived in a field theoretical context by wetterich@xcite and morris,@xcite and for interacting fermi systems by salmhofer and honerkamp.@xcite a matsubara frequency cutoff is used as the flow parameter . the hierarchy is then truncated already at first order , such that the 2-particle vertex remains unrenormalized , and the flow of the self - energy , which describes the renormalized impurity potential , is determined by the bare 2-particle interaction . no simplified parametrization of the self - energy is necessary such that the full spatial dependence of the renormalized impurity potential can be obtained for very large lattice systems . in spite of the striking simplicity of this scheme it was shown that the effects of a single static impurity in a spinless luttinger liquid are fully captured qualitatively , and in the weak coupling limit also quantitatively.@xcite in particular , one obtains that impurity potentials in repulsive luttinger liquids become effectively stronger at lower energy scales , and act ultimately as a weak link between two otherwise separate wires , as predicted by kane and fisher.@xcite the frg also correctly yields the universal low energy power - laws with exponents that do not depend on the bare impurity strength . in addition , it was shown that the asymptotic behavior holds typically only at very low energy scales and for very large systems , except for very strong bare impurities.@xcite the frg approach to impurities in luttinger liquids was originally developed and tested for spectral densities of single - particle excitations , but has been applied very recently also to transport problems , such as persistent currents in mesoscopic rings@xcite and the conductance of interacting wires connected to non - interacting leads.@xcite the full power of the frg with its ability to deal naturally also with complex crossover phenomena emerges most convincingly in the multi - scale problem posed by the transport through a resonant double barrier.@xcite in the present work we further develop the frg approach for impurities in luttinger liquids by including 2-particle vertex renormalization , that is we go one step further in the hierarchy of flow equations than previously.@xcite for spinless fermions this extension does not matter qualitatively , as the lowest order is already qualitatively correct , but the quantitative accuracy of the results improves considerably in particular at intermediate interaction strengths . for spin-@xmath2 systems , which we will treat in a subsequent work,@xcite vertex renormalization is necessary to take into account that backscattering of particles with opposite spins at opposite fermi points scales to zero in the low energy limit . a crucial point is to devise an efficient parametrization of the vertex by a managable number of variables . here we focus on spinless lattice fermions with nearest neighbor interaction as a prototype model . the bulk model is supplemented by site or hopping impurities . we also analyze the influence of boundaries , which can be viewed as infinite barriers or infinitesimal weak links . we choose to parametrize the vertex by a renormalized nearest neighbor interaction , which allows us to capture various advantageous features : the low energy flow of the vertex at @xmath3 in the pure system is obtained correctly to second order in the renormalized couplings ; the non - universal contributions at finite energy scales are correct to second order in the bare interaction ; the algorithm for the flow of the self - energy remains as fast as in the absence of vertex renormalization , such that one can easily deal with up to @xmath0 lattice sites . we compute spectral properties of single - particle excitations near an impurity and the oscillations in the density profile induced by an impurity , or by a boundary . the accuracy of the calculation is checked by comparing with dmrg results for systems with up to 1000 lattice sites and with exact results for the asymptotic behavior , which can be obtained from the bethe ansatz and bosonization . the article is structured as follows . in sec . ii we introduce the microscopic model and various types of impurities . the frg formalism is developed in sec . iii , and worked out explicitly for the spinless fermion model with nearest neighbor interaction . iv is dedicated to results for the renormalized impurity potential , spectral properties , and the density profile . we finally conclude in sec . v with an outline of promising extensions of the present work . we consider the spinless fermion lattice model with nearest neighbor interaction and various types of impurity potentials . the hamiltonian has the form @xmath4 where the kinetic energy @xmath5 is given by nearest neighbor hopping with an amplitude @xmath6 , and @xmath7 is a nearest neighbor interaction of strength @xmath8 . we use standard second quantization notation , where @xmath9 and @xmath10 are creation and annihilation operators on site @xmath11 , respectively , and @xmath12 is the local density operator . the impurity is represented by @xmath13 where @xmath14 is a static potential . for site impurities @xmath15 this potential is local . for the important special case of a single site impurity @xmath16 the potential acts only on one site @xmath17 . we also consider hopping impurities described by the non - local potential @xmath18 for the special case of a single hopping impurity , @xmath19 the hopping amplitude @xmath6 is replaced by @xmath20 on the bond linking the sites @xmath17 and @xmath21 . in the following we will set the bulk hopping amplitude @xmath6 equal to one , that is all energies are given in units of @xmath6 . the clean spinless fermion model @xmath22 is exactly solvable via the bethe ansatz.@xcite the system is a luttinger liquid for all particle densities @xmath23 and any interaction strength except at half - filling for @xmath24 . for @xmath25 a charge density wave with wave vector @xmath26 forms , for @xmath27 the system undergoes phase separation . the luttinger liquid parameter @xmath28 , which determines all the critical exponents of the liquid , can be computed exactly from the bethe ansatz solution.@xcite at half - filling @xmath28 is related to @xmath8 by the simple explicit formula @xmath29 for @xmath30 . we will compute physical properties for the above models on finite chains with open boundary conditions . the frg equations can be solved easily for chains with up to @xmath0 sites , the dmrg can be applied for up to @xmath31 sites . to avoid oscillations emerging from the ends of the chain we will sometimes attach non - interacting semi - infinite leads to the finite interacting chain , with a smoothly decaying interaction at the contact . this section is dedicated to a detailed presentation of the frg equations for one - dimensional fermi systems with impurities , including vertex renormalization , at zero temperature . we use the one - particle irreducible version of the functional rg.@xcite the starting point is an exact hierarchy of flow equations for the irreducible vertex functions , which is obtained by cutting off the infrared part of the free propagator on a scale @xmath1 , and differentiating the generating functional for the vertex functions with respect to this scale.@xcite the cutoff can be imposed in many different ways . the only requirement is that the infrared singularities must be regularized such that the flow equations allow for a regular perturbation expansion in powers of the renormalized two - particle vertex . for our purposes a sharp matsubara frequency cutoff is the most efficient choice . a momentum cutoff is less suitable here since the impurity spoils momentum conservation . the cutoff is imposed by excluding modes with frequencies below scale @xmath1 from the functional integral representation of the system , or equivalently , by introducing a regularized bare propagator @xmath32 here @xmath33 is the bare propagator of the pure system , involving neither interactions nor impurities . instead of the sharp cutoff imposed by the step function @xmath34 one may also choose a smooth cutoff function , but the sharp cutoff has the advantage that it reduces the number of integration variables on the right hand side of the flow equations . note that we will frequently write expressions which are well defined only if the sharp cutoff is viewed as a limit of increasingly sharp smooth cutoff functions . the suppression of frequencies below scale @xmath1 affects all green and vertex functions of the interacting system , which become thus functions of @xmath1 . the original system is recovered in the limit @xmath35 . the first two equations in the exact hierarchy of flow equations for the one - particle irreducible @xmath36-particle vertex functions are depicted graphically in figs . 1 and 2 . the first equation yields the flow of the self - energy , which is related to the interacting propagator by the usual dyson equation @xmath37^{-1 } \ ; .\ ] ] here and below @xmath38 , @xmath39 etc . are operators , which do not refer to any particular single - particle basis , unless we write matrix indices explicitly . the right hand side of the flow equation involves the 2-particle vertex @xmath40 , and the so - called single scale propagator @xmath41 defined as @xmath42 g^{\lam } \ , = \ , - \ , \frac{1}{1 - g_0^{\lam } \ , \sg^{\lam } } \ ; \frac{\partial g_0^{\lam}}{\partial\lam } \ ; \frac{1}{1 - \sg^{\lam } \ , g_0^{\lam } } \ ; .\ ] ] note that @xmath41 has support only on a single frequency scale : @xmath43 . the precise flow equation for the self - energy ( sketched graphically in fig . 1 ) reads @xmath44 where @xmath45 is the inverse temperature . the numbers 1 , 2 , etc.are a shorthand for matsubara frequencies and labels for single - particle states such as site and spin indices . note that @xmath46 and @xmath47 due to matsubara frequency conservation . the exponential factor in the above equation is irrelevant at any finite @xmath1 , but is necessary to define the initial conditions of the flow at @xmath48 ( see below ) . the right hand side of the flow equation for the 2-particle vertex @xmath40 , shown in fig . 2 , involves @xmath40 itself , but also the 3-particle vertex @xmath49 . the latter could be computed from a flow equation involving the 4-particle vertex @xmath50 , and so on . to avoid this unmanagable proliferation of vertex functions we make our first approximation : we neglect the contribution of the 3-particle vertex to the flow of @xmath40 . the coupled system of flow equations for @xmath40 and @xmath39 is then closed . the contribution of @xmath49 to @xmath40 is small as long as @xmath40 is sufficiently small , because @xmath49 is initially ( at @xmath51 ) zero and is generated only from terms of third order in @xmath40.@xcite a comparison of the rg results to exact dmrg results and exact scaling properties shows that the truncation error is often surprisingly small even for rather large interactions.@xcite the explicit form of the truncated flow equation for the 2-particle vertex reads @xmath52 \times & \big [ & \ ! \gam^{\lam}(1',2';3,4 ) \ , \gam^{\lam}(3',4';1,2 ) \nonumber \\[2 mm ] & - & \ ! \gam^{\lam}(1',4';1,3 ) \ , \gam^{\lam}(3',2';4,2 ) - ( 3 \lra 4 , 3 ' \lra 4 ' ) \nonumber \\[2 mm ] & + & \ ! \gam^{\lam}(2',4';1,3 ) \ , \gam^{\lam}(3',1';4,2 ) + ( 3 \lra 4 , 3 ' \lra 4 ' ) \ , \big]\end{aligned}\ ] ] the various contributions are shown diagrammatically in fig . 3 . for a sharp frequency cutoff the frequency sums on the right hand side of the flow equations can be carried out analytically in the zero temperature limit , where the matsubara sums become integrals . at this point one has to deal with products of delta functions @xmath53 and expressions involving step functions @xmath54 . these at first sight ambiguous expressions are well defined and unique if the sharp cutoff is implemented as a limit of increasingly sharp broadened cutoff functions @xmath55 , with the broadening parameter @xmath56 tending to zero . the expressions can then be conveniently evaluated by using the following relation,@xcite valid for arbitrary continuous functions @xmath57 : @xmath58 \to \delta(x-\lam ) \int_0 ^ 1 f(t ) \ , dt \ ; , \ ] ] where @xmath59 . note that the functional form of @xmath55 for finite @xmath56 does not affect the result in the limit @xmath60 . instead of writing down the frequency - integrated flow equations in full generality , we first implement our second approximation : the frequency dependent flow of the renormalized 2-particle vertex @xmath40 is replaced by its value at vanishing ( external ) frequencies , such that @xmath40 remains frequency independent . as a consequence , also the self - energy becomes frequency independent . since the bare interaction is frequency independent , neglecting the frequency dependence leads to errors only at second order ( in the interaction strength ) for the self - energy , and at third order for the vertex function at zero frequency . besides the quantitative errors we miss qualitative properties related to the frequency dependence of the self - energy , such as the suppression of the one - particle spectral weight in the bulk of a pure luttinger liquid . on the other hand , a comparison with exact numerical results and asymptotic analytical results shows that the impurity effects are not qualitatively affected by the frequency dependence of @xmath61 , at least for weak interactions . carrying out the frequency integration in the flow equation for the self - energy , and inserting a frequency independent 2-particle vertex , one obtains @xmath62 where the lower indices @xmath63 , @xmath64 , etc . label single - particle states ( not frequencies ) and @xmath65^{-1 } \ ; .\ ] ] note that @xmath66 has no jump at @xmath43 , in contrast to @xmath38 . the frequency integrated flow equation for the 2-particle vertex , evaluated at vanishing external frequencies , has the form @xmath67 \frac{1}{2\pi } \ , \sum_{\om = \pm\lam } \ , \sum_{3,3 ' } \sum_{4,4 ' } \ ! & \big [ & \!\ ! \frac{1}{2 } \ , \tg^{\lam}_{3,3'}(i\om ) \ , \tg^{\lam}_{4,4'}(-i\om ) \ , \gam^{\lam}_{1',2';3,4 } \ , \gam^{\lam}_{3',4';1,2 } \nonumber \\[2 mm ] & + & \!\ ! \tg^{\lam}_{3,3'}(i\om ) \ , \tg^{\lam}_{4,4'}(i\om ) \ , \left ( - \gam^{\lam}_{1',4';1,3 } \ , \gam^{\lam}_{3',2';4,2 } + \gam^{\lam}_{2',4';1,3 } \ , \gam^{\lam}_{3',1';4,2 } \right ) \big]\end{aligned}\ ] ] the flow is determined uniquely by the differential flow equations and the initial conditions at @xmath68 . for @xmath68 the vertex functions are given by the bare interactions of the system . in particular , the flow of the 2-particle vertex starts from the antisymmetrized bare 2-particle interaction while @xmath36-particle vertices of higher order vanish at @xmath68 , in the absence of @xmath36-body interactions with @xmath69 . the self - energy at @xmath68 is given by the bare one - particle potential , that is by those one - particle terms which are not included already in @xmath33 . we choose to include the translationally invariant bulk kinetic energy and the chemical potential in @xmath33 , while the impurity ( site or hopping ) potential @xmath70 is assigned to the initial condition of the self - energy . in a numerical solution the flow starts at some large finite initial cutoff @xmath51 . here one has to take into account that , due to the slow decay of the right hand side of the flow equation for @xmath39 at large @xmath1 , the integration of the flow from @xmath68 to @xmath71 yields a contribution which does not vanish in the limit @xmath72 , but rather tends to a finite constant . since @xmath73 for @xmath74 , this constant is easily determined as @xmath75 where @xmath76 is the bare antisymmetrized interaction . in summary , the initial conditions for the self - energy and the 2-particle vertex at @xmath48 are @xmath77 \gam^{\lam_0}_{1',2';1,2 } & = & i_{1',2';1,2 } \ ; , \end{aligned}\ ] ] where @xmath78 is the bare impurity potential . for the flow at @xmath79 the factor @xmath80 in eq . ( 16 ) can be discarded . for a finite lattice system with @xmath81 sites the flow of the 2-particle vertex @xmath82 involves @xmath83 independent flowing variables , if translation invariance is assumed , and @xmath84 variables , if the influence of the impurity on the flow of the 2-particle vertex is taken into account . for a treatment of large systems it is therefore necessary to reduce the number of variables by a suitable approximate parametrization of the vertex . in the low energy limit ( small @xmath1 ) the flow is dominated by a very small number of variables anyway , the others being irrelevant according to standard rg arguments.@xcite in particular , the frequency dependence of the vertex , discarded already above , is irrelevant for the flow of @xmath40 at small @xmath1 . for larger @xmath1 one can use perturbation theory as a guide for a simple but efficient parametrization of @xmath40 . we neglect the feedback of the self - energy into the flow of @xmath40 and replace @xmath66 by @xmath33 in eq . the renormalization of @xmath40 includes thus only bulk , not impurity , contributions , such that @xmath40 remains translation invariant . while this turns out to be sufficient for capturing the effects of isolated impurities in otherwise pure systems , it is known that impurity contributions to vertex renormalization become important in macroscopically disordered systems.@xcite for a more concrete treatment of the vertex renormalization , we have to focus on a specific model . in this article we restrict ourselves to the spinless fermion model with nearest neighbor interaction , leaving an extension to spin-@xmath2 systems with more general interactions for subsequent work.@xcite for spinless fermions the 2-particle vertex and the self - energy are fully characterized by either site or momentum variables . in the low energy limit , the flow of the vertex is dominated by contributions with momenta close to the fermi points , such that the right hand side of the flow equation is determined by momentum components of the vertex @xmath85 with @xmath86 . due to the antisymmetry of the vertex , there is only one such component which is non - zero : @xmath87 in the low energy limit the momentum dependence of the vertex away from @xmath88 is irrelevant . there are therefore many possible choices for the functional form of @xmath85 , which all lead to the correct low energy asymptotics . for a model with a bare nearest neighbor interaction @xmath8 , a natural and efficient choice is to parametrize the flowing vertex simply by a renormalized nearest neighbor interaction @xmath89 , which leads to a real space vertex of the form @xmath90 with @xmath91 . this yields the following structure in momentum space : @xmath92 \ , \delta^{(2\pi)}_{k_1+k_2,k'_1+k'_2 } \ ; , \ ] ] where the kronecker delta implements momentum conservation ( modulo @xmath93 ) . the flowing coupling constant @xmath89 is linked to the value of the vertex at the fermi points by the relation @xmath94 \ ; .\ ] ] the flow equation for @xmath95 becomes @xmath96 with the particle - particle and particle - hole contributions @xmath97 { \rm ph } & = & - [ g^0_p(i\om)]^2 \ , \gam^{\lam}_{k_f , p;k_f , p } \ , \gam^{\lam}_{p ,- k_f;p ,- k_f } \nonumber \\[2 mm ] { \rm ph ' } & = & g^0_{p - k_f}(i\om ) \ , g^0_{p+k_f}(i\om ) \ , \gam^{\lam}_{-k_f , p+k_f;k_f , p - k_f } \ , \gam^{\lam}_{p - k_f , k_f;p+k_f ,- k_f } \ ; , \end{aligned}\ ] ] where @xmath40 on the right hand side of the flow equation is given by eq . ( 24 ) . using eq . ( 25 ) to replace @xmath98 by @xmath99 on the left hand side of eq . ( 26 ) , one obtains a flow equation for @xmath89 of the simple form @xmath100 the function @xmath101 depends only on the cutoff @xmath1 and the fermi momentum @xmath3 . an explicit formula for @xmath101 can be obtained by carrying out the momentum integral in eq . ( 26 ) using the residue theorem , as sketched in appendix a. for finite systems the momentum integral should be replaced by a discrete momentum sum ; however , this leads to sizable corrections only for very small systems . the flow equation ( 28 ) can be integrated to @xmath102 where @xmath103 is the primitive function of @xmath101 with @xmath104 for @xmath105 . integrating @xmath101 one obtains @xmath106 & & + \ , \frac{\mu_0 ^ 4}{2(4-\mu_0 ^ 2)^{3/2 } } \ , \tanh^{-1 } \frac{4 + \mu_0 ^ 2 + i\mu_0\lam } { \sqrt{(4 + \mu_0 ^ 2 + i\mu_0\lam)^2 + 4(\lam-2i\mu_0)^2 } } \big ] \ ; , \end{aligned}\ ] ] where @xmath107 , while @xmath108 and @xmath109 denote the main branch of the inverse of the complex functions @xmath110 and @xmath111 , respectively . at half - filling , corresponding to @xmath112 , the function @xmath101 is particularly simple ( see appendix a ) , such that also @xmath89 has a relatively simple form : @xmath113 in fig . 4 we show results for the renormalized nearest neighbor interaction @xmath89 as obtained from the flow equation at various densities @xmath23 , for a bare interaction @xmath114 . while the renormalization does not follow any simple rule at intermediate scales @xmath1 , all curves saturate at a finite value @xmath115 in the limit @xmath35 , corresponding to a finite @xmath116 , as expected for a luttinger liquid fixed point.@xcite the above simplified flow equation yields not only the correct low energy asymptotics to second order in the renormalized vertex , but contains also all _ non - universal _ second order corrections to the vertex at @xmath88 at higher energy scales . hence the resulting fixed point coupling @xmath116 , from which the luttinger liquid parameter @xmath28 can be computed , is obtained correctly to second order in @xmath8 . the calculation of @xmath28 , presented in appendix b , illustrates how luttinger liquid parameters can be related to the fixed point couplings obtained from the frg . a comparison of the rg result for @xmath28 with exact results from the bethe ansatz solution of the spinless fermion model shows that the vertex renormalization scheme described above is not only very simple , but also surprisingly accurate . parametrizing @xmath40 by a renormalized nearest neighbor interaction has the enormous advantage that the self - energy , as determined by the flow equation ( 16 ) , is a tridiagonal matrix in real space , that is only the matrix elements @xmath117 and @xmath118 are non - zero . inserting @xmath40 from eq . ( 23 ) into ( 16 ) , one obtains the following simple coupled flow equations for the diagonal and off - diagonal matrix elements : @xmath119 \frac{\partial}{\partial\lam } \ , \sg^{\lam}_{j , j \pm 1 } \ ; = & & \frac{u^{\lam}}{2\pi } \sum_{\om = \pm\lam } \tg^{\lam}_{j , j \pm 1}(i\om ) \ ; .\end{aligned}\ ] ] note that the self - energy enters also the right hand side of these equations , via @xmath120 . since @xmath39 and @xmath121 are both tridiagonal in real space , the matrix inversion required to compute the diagonal and first off - diagonal elements of @xmath66 from @xmath39 can be carried out very efficiently , and thus for very large systems . in appendix c we describe an algorithm for the numerical solution of the flow equation for @xmath61 which scales only linearly with the size of the system . in the next section we will present results for spectral properties of single - particle excitations near an impurity or boundary , and for the density profile . here we describe how the relevant observables are computed from the solution of the flow equations . integrating the flow equation for the self - energy @xmath39 down to @xmath122 yields the physical ( cutoff - free ) self - energy @xmath61 , and the single - particle propagator @xmath123 . from the latter spectral properties of single - particle excitations can be extracted . we focus on local spectral properties , which are described by the _ local spectral function _ @xmath124 where @xmath125 is the local propagator , analytically continued to the real frequency axis from above . in our approximation the self - energy is frequency - independent and can therefore be viewed as an effective single - particle potential . the propagator @xmath126 is thus the green function of an effective single - particle hamiltonian . in real space representation this hamiltonian is given by the tridiagonal matrix @xmath127 , where the matrix elements of @xmath128 are the hopping amplitudes in @xmath129 , eq . ( 2 ) . for a lattice with @xmath81 sites this matrix has @xmath81 ( including possible multiplicities ) eigenvalues @xmath130 and an orthonormal set of corresponding eigenvectors @xmath131 . for the spectral function @xmath132 one thus obtains a sum of @xmath133-functions @xmath134 where @xmath135 , and the _ spectral weight _ @xmath136 is the squared modulus of the amplitude of @xmath131 on site @xmath11 . for large @xmath81 the level spacing between neighboring eigenvalues is usually of order @xmath137 , except for one or a few levels outside the band edges which correspond to bound states . due to even - odd effects etc . the spectral weight @xmath136 generally varies quickly from one eigenvalue to the next one . a smooth function of @xmath138 which suppresses these usually irrelevant finite - size details can be obtained by averaging over neighboring eigenvalues . in addition it is useful to divide by the level spacing between eigenvalues to obtain the conventional local _ density of states _ , which we denote by @xmath139 . boundaries or impurities induce a density profile with long - range friedel oscillations , which are expected to decay with a power - law at long distances.@xcite the expectation value of the local density @xmath140 could be computed from the local one - particle propagator @xmath141 , if @xmath126 was known exactly . however , the approximate flow equations for @xmath61 can be expected to describe the asymptotic behavior of @xmath126 correctly only at long distances between creation and annihilation operator in time and/or space , while in the local density operator time and space variables coincide . in the standard rg terminology @xmath140 is a _ composite _ operator , which has to be renormalized separately.@xcite to derive a flow equation for @xmath140 , we follow the usual procedure for the renormalization of correlation functions involving composite operators : we add a term @xmath142 with a small field @xmath143 to the hamiltonian , and take derivatives with respect to @xmath143 in the flow equations . the local density is given by @xmath144 where @xmath145 is the grand canonical potential of the system in the presence of the field @xmath143 . note that we use the same symbol @xmath140 for the density operator and its expectation value . in the presence of a cutoff @xmath1 the grand canonical potential obeys the exact flow - equation @xmath146 \ , [ g^{\lam}(i\om ) - g_0^{\lam}(i\om ) ] \right\}\ ] ] in a short - hand matrix notation . this flow - equation follows from the functional flow equation for the generating functional for vertex functions @xcite and the relation between the grand canonical potential and the zero - particle vertex . at zero temperature the matsubara frequency sum becomes an integral which , for the sharp frequency cutoff ( 10 ) , can be carried out analytically . this yields @xmath147 \right\ } \ ; .\ ] ] we attribute @xmath143 to the interaction part of the hamiltonian , not to @xmath129 , such that @xmath33 remains independent of @xmath143 . the self - energy is modified via the additional local and frequency - independent contribution @xmath148 to its initial value @xmath149 at scale @xmath51 . the density profile can be obtained from the above equations and the flow equation for @xmath39 by computing the shift of @xmath150 generated by a small finite perturbation @xmath143 , that is by numerical differentiation . alternatively one may carry out the @xmath143-derivative analytically in the flow equations , which yields a flow equation for the density in terms of the density response vertex . taking the @xmath143 derivative in ( 37 ) yields @xmath151\ ] ] with the density response vertex @xmath152 where the propagator @xmath66 is computed as previously , that is in the absence of @xmath143 . we compute the self - energy @xmath39 in the presence of @xmath143 within the same approximation as previously . it is thus determined from the flow equation ( 16 ) with a frequency - independent two - particle vertex @xmath40 . taking a derivative of that equation with respect to @xmath143 at @xmath153 yields the flow equation for the response vertex @xmath154 note that @xmath155 is frequency - independent in our approximation and that there is no contribution from the @xmath143-derivative of @xmath40 , since we neglect self - energy contributions in the flow of the two - particle vertex . for spinless fermions with a ( renormalized ) nearest neighbor interaction , the matrix @xmath155 is tridiagonal , that is only the components @xmath156 and @xmath157 are non - zero , and their flow is given by @xmath158 \frac{\partial}{\partial\lam } r_{j;l , l\pm 1}^{\lam } & = & \frac{u^{\lam}}{2\pi } \sum_{\om=\pm\lam } \sum_{l ' } \sum_{r ' = 0,\pm 1 } \tg_{l , l'}^{\lam}(i\om ) \ , r_{j;l',l'+r'}^{\lam } \ , \tg_{l'+r',l\pm 1}^{\lam}(i\om ) \ ; .\end{aligned}\ ] ] the right hand sides involve only one unrestricted lattice summation , and can be computed in @xmath159 time ( see appendix c ) . the initial condition for the response vertex is @xmath160 . the initial condition for the density is @xmath161 , for any filling , due to the slow convergence of the flow equation ( 38 ) at large frequencies , which yields a finite contribution to the integrated flow from @xmath162 to @xmath51 for arbitrarily large finite @xmath51 , as in the case of the self - energy discussed in more detail above . to avoid the interference of friedel oscillations emerging from the impurity or one boundary with those coming from the ( other ) boundaries of our systems we suppress the influence of the latter by coupling the finite chain to semi - infinite non - interacting leads , with a smooth decay of the interaction at the contacts . the presence of leads modifies the self - energy in the interacting region only via a boundary term . in particular , a lead coupled to the first site of the interacting region via nearest neighbor hopping ( with amplitude @xmath163 ) yields the additional contribution @xmath164 to the first diagonal element of self - energy matrix in real space.@xcite note that this term is frequency dependent and independent of @xmath1 . a lead coupled to the last site yields an analogous contribution to the last diagonal element of the self - energy . we now present and discuss explicit results for the self - energy , spectral properties near the impurity or boundary , and the density profile , as obtained from the frg . a comparison with exact dmrg results is made for the spectral weight at the fermi level and for the density profile . some frg results for the self - energy and spectral properties of the spinless fermion model have been published already previously,@xcite but only for half - filling and without any vertex renormalization . here we present results also away from half - filling and demonstrate the quantitative improvement of accuracy obtained by including vertex renormalization . in addition we present for the first time frg results for the density profile . using a faster algorithm ( see appendix c ) than previously we are now able to solve the flow equations for much larger systems . the typical shape of the self - energy in the vicinity of an impurity can be seen in fig . 5 , where we plot the diagonal elements @xmath165 and the off - diagonal elements @xmath166 near a site impurity of strength @xmath167 added to the spinless fermion model with interaction strength @xmath168 at quarter - filling . recall that the self - energy is tridiagonal in real space and frequency independent within our treatment . the diagonal elements can be interpreted as a local effective potential , the off - diagonal elements as a non - local effective potential which renormalizes the hopping amplitudes . at long distances from the impurity both @xmath165 and @xmath166 tend to a constant . the former describes just a bulk shift of the chemical potential , the latter a bulk renormalization of the hopping amplitude toward larger values . the oscillations around the bulk shifts are generated by the impurity . the wave number of the oscillations is @xmath169 , where @xmath3 is the fermi wave vector of the bulk system at quarter - filling . the amplitude of the oscillations of @xmath61 decays slower than the inverse distance from the impurity at intermediate length scales , but approaches a decay proportional to @xmath170 for @xmath171 . this can be seen most clearly by plotting an effective exponent @xmath172 for the decay of the oscillations , defined as the negative logarithmic derivative of the oscillation amplitude with respect to the distance @xmath173 . in fig . 6 we show the effective exponent resulting from the oscillations of @xmath165 as a function of the distance from a site impurity , for @xmath168 and half - filling . the impurity is situated at the center of a long chain with @xmath174 sites . to avoid interferences with oscillations from the boundaries we have attached semi - infinite non - interacting leads to the ends of the interacting chain , as described in the last paragraph of iii.c.2 . only for relatively large impurity strengths the asymptotic regime corresponding to @xmath175 is reached before finite size effects set in . for small @xmath70 one can see that @xmath172 increases from values below one , but the asymptotic long - distance behavior is cut off by the boundaries of the interacting region . for very small @xmath70 ( for example @xmath176 in fig . 6 ) we observe a plateau in @xmath172 for intermediate distances from the impurity site . in this regime @xmath172 is close to @xmath177 which can be understood by analytically solving our flow equations for small @xmath70.@xcite the long - range @xmath178-oscillations of the self - energy lead to a marked suppression of the spectral weight for single - particle excitations at the fermi level , that is at @xmath179 . in fig . 7 we show the local density of states @xmath139 on the site next to a site impurity of strength @xmath167 for the spinless fermion model at half - filling . the result for the interacting system at @xmath168 is compared to the non - interacting case . even - odd effects have been eliminated by averaging over neighboring eigenvalues ( see sec . iii.c.1 ) . @xmath133-peaks outside the band edges corresponding to bound states are not plotted . the interaction leads to a global broadening of the band , which is due to an enhancement of the bulk hopping amplitude , and also to a strong suppression of @xmath139 at low frequencies which is not present in the non - interacting system . for a finite system ( here @xmath180 ) the spectral weight at the fermi level remains finite , but tends to zero with increasing system size . in fig . 8 we show results for the density of states choosing the same parameters as in fig . 7 , but now for densities away from half - filling : @xmath181 and @xmath182 . in addition to the dip near @xmath183 a second singularity appears at a finite frequency . this effect is due to the fact that a long - range potential with a wave number @xmath178 does not only strongly scatter states with momenta near @xmath3 , but also those with momenta close to @xmath184 . indeed the singularity is situated at @xmath185 , where @xmath186 is the renormalized ( bulk ) dispersion . in the half - filled case only one singularity is seen simply because @xmath187 for @xmath112 . the spectral weight at the fermi level is expected to vanish asymptotically as a power - law @xmath188 with increasing system size , where @xmath189 is the _ boundary exponent _ describing the power - law suppression of the density of states at the boundary of a semi - infinite chain.@xcite that exponent depends only on the bulk parameters of the model , not on the impurity strength . for the spinless fermion model it can be computed exactly from the bethe ansatz solution.@xcite we therefore analyze the asymptotic behavior of the spectral weight at the fermi level by defining an effective exponent @xmath190 as the negative logarithmic derivative of the spectral weight with respect to the system size , such that @xmath190 tends to a ( positive ) constant in case of a power - law suppression . in fig . 9 we show results for @xmath190 as obtained from the frg for the spinless fermion model at quarter - filling with up to about @xmath0 sites , for a weak ( @xmath191 ) and an intermediate ( @xmath192 ) interaction parameter . the spectral weight has been computed either at a boundary , or near a hopping impurity of strength @xmath193 . results obtained from the rg without ( upper panel ) and with ( lower panel ) vertex renormalization are compared to exact dmrg results ( for up to 512 sites ) and the exact boundary exponents @xmath194 , plotted as horizontal lines . the frg results follow a power - law for large @xmath81 , with the same asymptotic exponent for the boundary and impurity case , confirming thus the expected universality . however , the asymptotic regime is reached only for fairly large systems , even for the intermediate interaction strength @xmath192 . the comparison with the exact dmrg results and exact exponents shows that the frg is also quantitatively rather accurate , and that the inclusion of vertex renormalization leads to a substantial improvement at intermediate coupling strength . results for the effective exponent @xmath195 in the case of a site impurity are shown in fig . 10 , at quarter - filling and for an interaction strength @xmath168 . the comparison of the different curves obtained for different impurity strengths confirms once again the expected asymptotic universality , and also how the asymptotic regime shifts rapidly toward larger systems as the bare impurity strength decreases . we finally discuss results for the density profile @xmath140 . boundaries and impurities induce friedel oscillations of the local density with a wave vector @xmath178 . in a non - interacting system these oscillations decay proportionally to the inverse distance from the boundary or impurity . in an interacting luttinger liquid the friedel oscillations are expected to decay as @xmath196 at long distances @xmath173 . for a very weak impurity one expects a slower decay proportional to @xmath197 at intermediate distances , and a crossover to the asymptotic power - law with exponent @xmath28 at very long distances.@xcite at intermediate distances the response of the density to a weak impurity can be treated in linear response theory , such that the density modulation is determined by the density - density response function at @xmath178 , which leads to the power - law decay with exponent @xmath198 . in fig . 11 we show frg and dmrg results for the density profile @xmath140 for a spinless fermion chain with 128 sites and interaction strength @xmath168 at half - filling . the friedel oscillations emerge from both boundaries and interfere in the center of the chain . the accuracy of the rg results is excellent for all @xmath11 . for incommensurate filling factors the density profile looks more complicated . this can be seen in fig . 12 , where functional rg results are shown for the density modulation @xmath199 near the boundary of a system with an average density @xmath200 and @xmath201 sites . for long distances from the boundary the oscillation amplitude has a well - defined envelope which fits to a power - law as a function of @xmath11 . we now analyze the long - distance behavior of the amplitudes more closely for the half - filled case , and compare to exact results for the asymptotic exponents . in fig . 13 we show frg results for the amplitude of density oscillations emerging from an open boundary , for a very long spinless fermion chain with @xmath202 sites and various interaction strengths @xmath8 at half - filling . the other end of the chain ( opposite to the open boundary ) is smoothly connected to a non - interacting lead . in a log - log plot ( upper panel of fig . 13 ) the amplitude follows a straight line for almost all @xmath11 , corresponding to a power - law dependence . deviations from a perfect power - law can be seen more neatly by plotting the effective exponent @xmath203 , defined as the negative logarithmic derivative of the amplitude with respect to @xmath11 ( see the lower panel of fig . the effective exponent is almost constant except at very short distances or when @xmath11 approaches the opposite end of the interacting chain , which is not surprising . from a comparison with the exact exponent ( horizontal lines in the figure ) one can assess the quantitative accuracy of the frg results . effective exponents describing the decay of friedel oscillations generated by site impurities of various strengths are shown in fig . 14 , for a half - filled spinless fermion chain with @xmath204 sites and interaction @xmath168 . both ends of the interacting chains are coupled to non - interacting leads to suppress oscillations coming from the boundaries . for strong impurities the results are close to the boundary result ( cf . 13 ) , as expected . for weaker impurities the oscillations decay more slowly , that is with a smaller exponent , and approach the boundary behavior only asymptotically at large distances ( beyond the range of our chain for @xmath205 ) . for very weak impurities ( @xmath176 in fig . 14 ) the oscillation amplitude follows a power - law corresponding to the linear response behavior with exponent @xmath206 at intermediate distances . we finally present some results for the effective exponent of the density oscillation decay in the case of an _ attractive _ interaction @xmath207 , see fig . 15 . in that case the effective impurity strength should scale to zero at low energies and long distances.@xcite indeed , for weak and moderate bare impurity potentials the effective exponent in fig . 15 approaches the linear density response exponent @xmath206 . only for very strong impurities the density oscillations decay with the smaller boundary exponent over several length scales in summary , we have shown that the frg provides an ideal tool for computing the intriguing properties of luttinger liquids with static impurities . the method yields ab initio results for microscopic model systems at all energy scales from the fermi energy to the ultimate low energy limit . we have demonstrated the power of the method by computing spectral properties of single - particle excitations as well as the oscillations in the density profile induced by impurities or boundaries for a spinless fermion model with nearest neighbor interaction . with the inclusion of vertex renormalization , in addition to the renormalization of the effective impurity potential , our results agree remarkably well with exact asymptotic results and numerical dmrg data even for intermediate interaction strength . there is a broad range of interesting further applications and extensions of the frg for impurities in luttinger liquids beyond the scope of the present article : _ spin-@xmath2 fermions _ : the inclusion of spin degrees of freedom requires a parametrization of the two - particle vertex with several coupling constants . we have already derived flow equations for the ( extended ) hubbard model , with an effective vertex parametrized by local and nearest neighbor interactions.@xcite _ feedback of impurities on vertex _ : for isolated impurities the influence of impurities on the vertex renormalization is irrelevant for the asymptotic low energy or long distance behavior , although it may contribute quantitatively at intermediate scales . in the case of disordered systems with a finite impurity density the influence of the latter on the two - particle vertex is crucial and must be taken into account.@xcite in principle this is possible by computing the vertex flow with full propagators , which contain the renormalized impurity potential via the self - energy . _ finite temperature _ : the frg approach can be extended without major complications to finite temperature . this is particularly useful for studying the temperature dependence of transport properties.@xcite _ transport _ : in the absence of inelastic processes ( @xmath208 ) the conductance of the interacting wire can be computed from the one - particle green function in the presence of leads.@xcite several results have already been presented in short articles.@xcite _ inelastic processes _ : inelastic processes appear at second order in the interaction and can be included in the flow equations by inserting the second order vertex into the flow equation for the self - energy without neglecting its frequency dependence . this procedure would also capture the anomalous dimension of the bulk system , which is missing in the present work . * acknowledgements : * + we are grateful to manfred salmhofer for useful discussions . this work benefitted from the workshop realistic theories of correlated electron materials in fall 2002 at itp - ucsb . thanks the bundesministerium fr bildung und forschung for financial support . here we sketch some details concerning the explicit evaluation of the flow equations for the 2-particle vertex . inserting the momentum structure of @xmath40 , ( 24 ) , into the flow equation ( 26 ) and replacing @xmath95 by @xmath89 on the left hand side yields @xmath209 where @xmath210 ^ 2 } { ( i\om - \xi^0_{p - k_f})(i\om - \xi^0_{p+k_f } ) } \ ; .\ ] ] here @xmath211 with @xmath212 is the bare dispersion relation relative to the bare fermi level . since @xmath213 can be written as a rational function of @xmath214 and @xmath215 , the @xmath216-integral can be carried out analytically using the substitution @xmath217 and the residue theorem . after a lengthy but straightforward calculation one obtains the following result for the coefficient @xmath101 in ( 28 ) : @xmath218 & & \times \ ; \frac { 3i\mu_0 ^ 4 - 10\mu_0 ^ 3\lam - 12i\mu_0 ^ 2(\lam^2 + 1 ) + 6\lam^3\mu_0 + 18\lam\mu_0 + 6i\lam^2 + i\lam^4 } { \pi ( 2\mu_0 + i\lam)(4 - \mu_0 ^ 2 + \lam^2 - 2i\lam\mu_0)^2 } \ , \bigg ] \ ; .\end{aligned}\ ] ] at half - filling ( @xmath219 ) , this reduces to @xmath220 \ ; .\ ] ] in this appendix we show how the luttinger liquid parameter @xmath28 , which determines the critical exponents of luttinger liquids , can be computed from the fixed point couplings as obtained from the rg . a comparison of the rg result for @xmath28 with the exact bethe ansatz result for the bulk model ( without impurity ) serves also as a check for the accuracy of our vertex renormalization . a relation between the fixed point couplings and @xmath28 can be established via the exact solution of the fixed point hamiltonian of luttinger liquids , the luttinger model . for spinless fermions , @xmath28 is determined by the luttinger model parameters @xmath221 and @xmath222 as @xmath223 where @xmath221 is the interaction between left and right movers and @xmath222 the effective fermi velocity of the model , that is the slope of the ( linear ) dispersion relation , with a possible shift due to interactions between particles moving in the same direction ( @xmath224-coupling ) already included.@xcite we therefore need to extract @xmath221 and @xmath222 from the rg flow in the limit @xmath35 . in order to obtain @xmath28 correctly to order @xmath225 , it is sufficient to obtain @xmath222 correctly to linear order in @xmath8 . the luttinger model interaction @xmath221 and the fixed point coupling @xmath226 from the rg are not simply identical , in contrast to what one might naively expect . to find the true relation between @xmath221 and @xmath116 , one has to take into account that the forward scattering limit of the dynamical 2-particle vertex is generally not unique ( in the absence of cutoffs ) , and depends on whether momentum or frequency transfers tend to zero first . this ambiguity is well - known in fermi liquid theory , where it leads to the distinction between quasi - particle interactions and scattering amplitudes,@xcite but is equally present in luttinger liquids , for the same reason in all cases : the ambiguity of the small momentum , small frequency limit of particle - hole propagators contributing to the vertex function . in the _ dynamical limit _ , where the momentum transfer @xmath227 vanishes first , the singular particle - hole propagators do not contribute . in fermi liquids this limit yields the quasi - particle interaction . in the opposite _ static limit _ the frequency transfer @xmath228 vanishes first and particle - hole propagators yield a finite contribution . in the presence of an infrared cutoff @xmath229 the forward scattering limit of the vertex function is unique , since the ambiguity in the particle - hole propagator is due to the infrared pole of the single - particle propagator . hence @xmath230 is well defined . however , @xmath230 and also its limit for @xmath35 depend on the choice for the cutoff function . for a momentum cutoff , which excludes states with excitation energies below @xmath1 around the fermi points , particle - hole excitations with small momentum transfers @xmath227 are impossible . hence particle - hole propagators with infinitesimal @xmath227 do not contribute to the vertex at any @xmath229 , such that @xmath230 converges to the dynamical forward scattering limit , which is simply given by the bare coupling constant @xmath221 in the luttinger model . for a frequency cutoff the particle - hole propagators with vanishing momentum and frequency transfer yield a finite contribution at @xmath229 , which tends to the static limit for @xmath35 . this can be seen directly by integrating @xmath231 ^ 2 $ ] over @xmath1 from infinity to zero . hence the vertex @xmath230 obtained from our frequency cutoff rg tends to the _ static _ forward scattering limit . for the luttinger model , the static forward scattering limit of the vertex can be obtained from the dynamical effective interaction between left and right movers @xmath232 , which is defined as the sum of particle - hole chains @xmath233 where @xmath234 is the bare particle - hole bubble for right ( + ) and left ( - ) movers . note that only odd powers of @xmath221 contribute to the effective interaction between left and right movers . this effective interaction appears naturally in the exact solution of the luttinger model via ward identities.@xcite for the static limit one obtains @xmath235 ^ 2}\ ] ] which we identify with our fixed point coupling @xmath116 as obtained from the rg with frequency cutoff . inverting this relation between @xmath221 and @xmath116 we obtain @xmath236\ ] ] for spinless fermions the difference between @xmath221 and @xmath116 appears only at third order in the coupling , but for models with spin the distinction becomes important already at second order . the fermi velocity @xmath222 can be computed from the ( frequency independent ) self - energy in momentum space as @xmath237 where @xmath238 is the bare fermi velocity . the self - energy is computed from the flow equation ( 32 ) , which can be rewritten in momentum space as @xmath239 where @xmath240 . the chemical potential @xmath241 has to be fixed by the final condition @xmath242 , where @xmath243 depends only on the density , not the interaction . from the tridiagonal structure of @xmath61 in real space , but also from the above expression it follows that @xmath244 has the form @xmath245 . the functional flow equation for @xmath244 yields a coupled set of ordinary flow equations for the coefficients @xmath246 and @xmath247 , with initial conditions @xmath248 and @xmath249 . the momentum integrals can be evaluated analytically via the residue theorem , such that the remaining set of two coupled differential equations ( with @xmath89 as input ) can be easily solved numerically . the result for @xmath222 is correct at least to first order in @xmath8 , but not necessarily to second order , since our simplified parametrization captures the 2-particle vertex correctly to second order only at the fermi points . inserting @xmath221 and @xmath222 into the luttinger model formula ( b1 ) we can now compute @xmath28 as a function of @xmath8 and density for the microscopic spinless fermion model . in fig . 16 we show results for @xmath250 for various fixed densities , which are compared to the exact result from the bethe ansatz solution @xcite in the inset . the rg results are indeed correct to order @xmath225 for small @xmath8 , as expected , and they are surprisingly accurate also for larger values of @xmath8 . for @xmath252 the flow equation for the self - energy ( 16 ) can be written as @xmath253 \ ; .\ ] ] in order to compute its right - hand side , one needs to invert the tridiagonal matrix @xmath254 where @xmath255 is complex symmetric ( _ not _ hermitean ) with diagonal elements @xmath256 , @xmath257 , and first off - diagonal elements @xmath258 , @xmath259 . note that @xmath260 such that @xmath255 is non - singular and its inverse well - defined . the inverse @xmath261 is not tridiagonal but a full matrix which can be computed by standard methods in @xmath262 time . however , for an interaction that does not extend beyond nearest neighbors on the lattice , only the tridiagonal part of @xmath263 is required , which can be computed in @xmath264 time , such that much larger lattices can be treated . we shall first explain how this is done and then present the resulting algorithm that can directly be incorporated into a computer program . under certain assumptions ( see below ) , a matrix can be uniquely factorized into a lower unit triangular matrix @xmath81 , a diagonal matrix @xmath265 , and an upper unit triangular matrix @xmath8 ( `` ldu factorisation '' ) : @xmath266.@xcite for a tridiagonal matrix @xmath255 the unit triangular matrices @xmath81 and @xmath8 are in fact unit bidiagonal : their matrix elements are one on the diagonal , and only the first off - diagonal is nonzero . since our @xmath255 is symmetric we have @xmath267 . thus we obtain a factorization of the form @xmath268 where the label `` @xmath269 '' distinguishes this factorization from another one used below . the prescription to compute the elements @xmath270 and @xmath271 is well known and can be found for example in ref . @xcite . starting in the upper left corner one proceeds to increasing row and column numbers until one arrives at the lower right corner of @xmath255 : @xmath272 this works well since in our case @xmath273 , such that one never divides by zero . to compute the inverse @xmath274 , one could directly calculate @xmath275 . it is however easier and more accurate to find the inverse by solving the linear system of equations @xmath276 , where @xmath63 is the identity matrix , by `` backsubstitution '' . to be specific , consider the @xmath277 column vector @xmath278 of @xmath263 : @xmath279 where @xmath280 is the @xmath277 unit vector . the first step is to solve the linear system @xmath281 for @xmath282 , and the second step to solve @xmath283 for @xmath278 . to solve a tridiagonal linear system for one vector takes @xmath264 time , so solving for the full inverse matrix @xmath263 takes @xmath262 time . now we shall derive an algorithm to compute the elements of @xmath282 and @xmath278 . begin with the last column @xmath284 : @xmath285 can be solved from the first to the last row and gives @xmath286 . next @xmath287 can be solved starting from the last row , @xmath288 . from there one can work upwards by backsubstitution , @xmath289 ( @xmath290 ) . for the other columns @xmath291 , one can not take the shortcut and has to solve both linear systems for @xmath282 and @xmath278 . but it is now important to realize that for any column vector @xmath292 , if we somehow know the diagonal element @xmath293 , the next element above the diagonal is @xmath294 thus we have a prescription how to go up one row in @xmath263 . together with the symmetry of @xmath263 , that is @xmath295 , which follows from the symmetry of @xmath255 , we get the first off - diagonal element one column to the left _ without _ solving the two linear systems in ( [ eq : appc : backsubst ] ) . so it is possible to compute directly the tridiagonal part of the inverse . however , there is another algorithm which is much more accurate for near - singular matrices at the end of the rg flow : the double factorization.@xcite it does not rely on the symmetry of @xmath263 but uses the complementary `` udl '' factorization @xmath296 where the matrix elements are obtained as @xmath297 we proceed as for the ldu factorization above and get @xmath298 we can combine equations ( [ eq : appc : fromdiag ] ) and ( [ eq : appc : todiag ] ) to relate consecutive diagonal elements : @xmath299 thus we start with ( [ eq : appc : backsubst0 ] ) and use the @xmath300 decomposition to go one matrix element to the right in the inverse matrix , from the diagonal to the first off - diagonal ( [ eq : appc : todiag ] ) , while the @xmath301 decomposition allows to go down by one , back to the next diagonal element ( [ eq : appc : fromdiag ] ) . there is no need to compute the full inverse matrix . one can implement the algorithm without knowing the derivation by using equations ( [ eq : appc : ldu ] ) and ( [ eq : appc : udl ] ) - ( [ eq : appc : backsubst2 ] ) . one can further eliminate the @xmath8 s using equations ( [ eq : appc : ldu ] ) and ( [ eq : appc : udl ] ) and implement the algorithm such that only the input vectors @xmath302 , @xmath303 and the output vectors @xmath304 , @xmath305 enter the temporary storage . this double factorization is numerically accurate to more than 10 significant digits ( using double precision ) even for large lattices ( @xmath0 sites ) and almost singular matrices with @xmath306 which appear at the end of the flow for half filling . the right hand side of the flow equations ( 41 ) for the density response vertex @xmath307 can be computed in @xmath159 time using the fact that the upper triangular part of the inverse of a tridiagonal matrix ( here @xmath308 ) is the upper triangular part of the outer product of two vectors.@xcite first computational applications of the frg to fermi systems were performed for the 2d hubbard model , see d. zanchi and h.j . schulz , phys . b * 61 * , 13609 ( 2000 ) ; c.j . halboth and w. metzner , _ ibid _ * 61 * , 7364 ( 2000 ) ; c. honerkamp , m. salmhofer , n. furukawa , and t. m. rice , _ ibid . _ * 63 * , 035109 ( 2001 ) . see , for example , t. giamarchi and h. maurey , in _ correlated fermions and transport in mesoscopic systems _ , edited by t. martin , g. montambaux , and j. tran thanh van ( editions frontieres , paris , 1997 ) . the derivation will be given in an upcoming paper containing a detailed discussion of transport properties , see t. enss , s. andergassen , v. meden , w. metzner , u. schollwck , and k. schnhammer ( unpublished ) .

the class of-leaf powers consists of graphs that have a-leaf root , that is , a tree with leaf set , where , if and only if the-distance between and is at most . structure and linear time recognition algorithms have been found for- ,- ,- , and , to some extent ,-leaf powers , and it is known that the union of all-leaf powers , that is , the graph class , forms a proper subclass of strongly chordal graphs . despite from that , no essential progress has been made lately . in this paper , we use the new notion of clique arrangements to suggest that leaf powers are a natural special case of strongly chordal graphs . the clique arrangement of a chordal graph is a directed graph that represents the intersections between maximal cliques of by nodes and the mutual inclusion of these vertex subsets by arcs . recently , strongly chordal graphs have been characterized as the graphs that have a clique arrangement without bad-cycles for . we show that the clique arrangement of every graph of is free of bad-cycles . the question whether this characterizes the class exactly remains open .

1402.1425

leaf powers are a family of graph classes that has been introduced by nishimura etal . @xcite to model the problem of reconstructing phylogenetic trees . in particular , a given finite simple graph @xmath2 is called the @xmath1-leaf power of a tree @xmath3 for some @xmath17 , if @xmath4 is the set of leaves in @xmath3 and any two distinct vertices @xmath18 are adjacent , that is @xmath5 , if and only if the distance of @xmath6 and @xmath7 in @xmath3 is at most @xmath1 . for all @xmath17 , the class of graphs that are a @xmath1-leaf power of some tree , is simply called @xmath1-leaf powers and denoted by @xmath0 . the general problem , from a graph theoretic point of view , is to structurally characterize @xmath0 for all fixed @xmath17 and to provide efficient recognition algorithms . obviously , a graph @xmath14 is a @xmath8-leaf power , if and only if it is the disjoint union of cliques , that is , @xmath14 does not contain a chordless path of length @xmath8 . dom etal . @xcite prove that @xmath9-leaf powers are exactly the graphs that do not contain an induced bull , dart , or gem . brandstdt etal . @xcite contribute to the characterization of @xmath9-leaf powers by showing that they are exactly the graphs that result from substituting cliques into the nodes of a tree . moreover , they give a linear time algorithm to recognize @xmath9-leaf powers building on their characterization . a characterization of @xmath10-leaf powers in terms of forbidden subgraphs is yet unknown . however , basic @xmath10-leaf powers , the @xmath10-leaf powers without true twins , are characterized by eight forbidden subgraphs @xcite . the structure of basic @xmath10-leaf powers has further been analyzed by brandstdt etal . @xcite , who provide a nice characterization of the two - connected components of basic @xmath10-leaf powers that leads to a linear time recognition algorithm even for @xmath10-leaf powers . for @xmath11-leaf powers , a polynomial time recognition was given in @xcite . however , no structural characterization is known , even for basic @xmath11-leaf powers . only for distance - hereditary basic @xmath11-leaf powers a characterization in terms of @xmath19 forbidden induced subgraphs has been discovered @xcite . except from the result in @xcite that @xmath20 is not true for every @xmath1 , there have not been any more essential advances in determining the structure of @xmath1-leaf powers for @xmath21 since 2007 . instead , research has focused on generalizations of leaf powers @xcite , which also turned into dead ends , very soon . on the other hand , if we push @xmath1 to infinity , then it turns out that not every graph is a @xmath1-leaf power for some @xmath17 . in particular , a @xmath1-leaf power is , by definition , the subgraph of the @xmath1-th power of a tree @xmath3 induced by the leaves of @xmath3 . since trees are sun - free chordal and as taking powers and induced subgraphs do not destroy this property , it follows trivially that every @xmath1-leaf power , despite the value of @xmath1 , is strongly chordal @xcite . but even not every strongly chordal graph is a @xmath1-leaf power for some @xmath17 . in fact , we are aware of exactly one counter example , which has been found by bibelnieks etal . @xcite and is shown as @xmath22 in figure [ fig : counterexamples ] . insofar , it is reasonable to ask for a precise characterization of the graphs that are not a @xmath1-leaf power for any @xmath17 . this problem can equivalently be formulated as to describe the graphs in the class @xmath23 , which we call leaf powers , for short . interestingly , brandstdt etal . @xcite show that @xmath16 coincides with the class of fixed tolerance nest ( neighborhood subtree tolerance ) graphs , a well - known graph class with an absolutely different motivation given by bibelnieks etal . @xcite . naturally , characterizations and an efficient recognition algorithms for this class are also open questions today . however , by brandstdt etal . @xcite , it is know that @xmath16 is a superclass of ptolemaic graphs , that is , gem - free chordal graphs @xcite , and even a superclass of directed rooted path graphs , introduced by gavril @xcite . recently , we introduced the clique arrangement in @xcite , a new data structure that is especially valuable for the analysis of strongly chordal graphs . the clique arrangement @xmath24 of a chordal graph @xmath14 is a directed acyclic graph that has certain vertex subsets of @xmath14 as a node set and describes the mutual inclusion of these sets by arcs . in particular , for every set @xmath25 of maximal cliques of @xmath14 there is a node in @xmath26 for @xmath27 and two nodes @xmath28 are joined by an arc @xmath29 , if @xmath30 and there is no @xmath31 with @xmath32 . in @xcite , we give a new characterization of strongly chordal graphs in terms of a forbidden cyclic substructure in the clique arrangement , called bad @xmath1-cycles for @xmath33 , and we show how to construct the clique arrangement of a strongly chordal graph in nearly linear time . it is known that the clique arrangements of ptolemaic graphs are even directed trees @xcite . since all ptolemaic graphs are leaf powers and all leaf powers are strongly chordal , it appears likely that the degree of acyclicity in clique arrangements of leaf powers is between forbidden bad @xmath1-cycles , @xmath34 , and the complete absence of cycles . this paper describes a cyclic substructure that is forbidden in the clique arrangement of leaf powers . for convenience , we call these substructures bad @xmath8-cycles , although they are not the obvious continuation of the concept of bad @xmath1-cycles for @xmath34 . as the main result of this paper , we show that bad @xmath8-cycles occur in @xmath13 , if and only if @xmath14 contains at least one of seven induced subgraphs @xmath35 depicted in figure [ fig : counterexamples ] . we leave it as an open question , if these seven graphs are sufficient to characterize @xmath16 in terms of forbidden subgraphs . however , we conjecture that this is the case . this would imply a polynomial time recognition algorithm for @xmath16 , by using the possibility of efficiently recognizing strongly chordal graphs and checking the containment of a finite number of forbidden induced subgraphs . we refer to several graph classes which are not explicitly defined due to space limitations . for a comprehensive survey on graph classes we would like to refer to @xcite . throughout this paper , all graphs @xmath2 are simple , without loops and , with the exception of clique arrangements , undirected . we usually denote the vertex set by @xmath4 and the edge set by @xmath36 , where the edges are also called arcs in a directed graph . we write @xmath37 , respectively @xmath38 in the directed case , for @xmath5 and @xmath39 for @xmath40 . for all vertices @xmath41 in an undirected graph , we let @xmath42 denote the _ open neighborhood _ and @xmath43 = n(x ) \cup \{x\}$ ] the _ closed neighborhood _ of @xmath6 in @xmath14 . in a directed graph , @xmath44 denotes the set of neighbors that are reachable from @xmath6 by a single arc and @xmath45 are the neighbors that reach @xmath6 by a single arc . if @xmath46 then @xmath6 is a _ source _ and if @xmath47 then @xmath6 is a _ sink_. an _ independent set _ in @xmath14 is a set of mutually nonadjacent vertices . a _ clique _ @xmath48 is a set of mutually adjacent vertices and @xmath49 is called maximal , if there is no clique @xmath50 with @xmath51 . the set of all maximal cliques of @xmath14 is denoted by @xmath52 . a ( simple ) path in a graph @xmath14 is a sequence @xmath53 of non - repeating vertices in @xmath14 , such that @xmath54 for all @xmath55 . if @xmath36 is clear from the context , then we denote the path by @xmath56 in an undirected graph . in a directed graph , @xmath57 specifies a directed path and we say that @xmath58 _ reaches _ @xmath59 . the distance @xmath60 between two vertices @xmath61 of an ( un- ) directed graph @xmath14 is the minimum number of edges in an ( un- ) directed path starting in @xmath6 and ending in @xmath7 . if the edge @xmath62 is additionally present in @xmath36 , then we talk of a ( simple ) cycle in @xmath14 , and as for paths , an undirected cycle is denoted by @xmath63 . an undirected cycle is called _ induced @xmath1-cycle @xmath64 _ , if @xmath14 contains @xmath65 , if and only if @xmath66 or @xmath67 and @xmath68 . a tree @xmath3 is an undirected connected acyclic graph , that is , for all pairs @xmath61 of vertices there exists a path @xmath69 , and @xmath3 is free of cycles . directed graphs are acyclic , if they are free of directed cycles . a vertex subset @xmath70 induces a _ @xmath1-sun _ in @xmath14 , if @xmath71 is a clique and @xmath72 is an independent set and for every edge @xmath73 between @xmath74 and @xmath75 , either @xmath76 or @xmath77 , where the indices are counted modulo @xmath1 . by definition , a graph is _ chordal _ , if and only if it does not contain induced @xmath1-cycles for all @xmath78 , and by farber @xcite a graph is _ strongly chordal _ , if and only if it does not contain induced @xmath1-suns for all @xmath15 . beside the many useful properties of ( strongly ) chordal graphs , see for example @xcite , this paper uses in particular the following two properties , that are folklore but nevertheless have been shown in @xcite : [ lma_vertex_node_none_adjacency ] if @xmath14 is a chordal graph and @xmath79 are maximal cliques of @xmath14 , then there is a vertex @xmath80 such that @xmath39 for all @xmath81 . [ lma_stronglychordal_cliquecutstructure ] if @xmath14 is a strongly chordal graph and @xmath82 any nonempty subset of @xmath52 , then there are two maximal cliques @xmath83 such that @xmath84 . a strongly chordal graph @xmath2 is the @xmath1-leaf power of a tree @xmath3 for @xmath17 , if @xmath4 is the set of leaves in @xmath3 and for all @xmath18 there exists @xmath5 , if and only if @xmath85 . the tree @xmath3 is called a @xmath1-leaf root of @xmath14 , in this case . notice that @xmath1-leaf roots are not necessarily unique for given @xmath1-leaf powers . for all @xmath17 , the class @xmath0 consists of all graphs that are a @xmath1-leaf power for some tree and @xmath12 is the class of leaf powers . the clique arrangement @xmath86 of a chordal graph @xmath14 , as introduced in @xcite , is a directed acyclic graph with node set @xmath87 that contains exactly all intersections of the maximal cliques of @xmath14 , and arc set @xmath88 that describes their mutual inclusion . clearly , the set of sinks in @xmath13 corresponds exactly to @xmath52 . the following simple facts for clique arrangements are also introduced in @xcite : [ obs_sink_intersection ] if @xmath89 is a node in the clique arrangement @xmath24 of a chordal graph @xmath14 and if @xmath90 , then @xmath91 . moreover , if @xmath92 are the sinks of @xmath13 that are reached from @xmath74 by directed paths , then @xmath93 . [ obs_node_intersection ] if @xmath94 are nodes in the clique arrangement @xmath24 of a chordal graph @xmath14 such that their intersection @xmath95 is not empty , then @xmath89 . although @xmath13 is acyclic by definition , we call the following structure a cycle in @xmath13 for the lack of a better term . for any @xmath96 , a _ @xmath1-cycle _ of @xmath13 is a set of nodes @xmath97 such that for all @xmath98 there is a directed path from @xmath99 to @xmath100 and a directed path from @xmath99 to @xmath101 ( counted modulo @xmath1 ) . the nodes @xmath102 are called _ starters _ of the cycle and the nodes @xmath103 are called _ terminals _ of the cycle . note that by definition , @xmath104 for all @xmath98 . in @xcite , we call a @xmath1-cycle _ bad _ , if @xmath15 and for all @xmath105 there is a directed path from @xmath99 to @xmath106 , if only if @xmath107 ( counted modulo @xmath1 ) . [ thm_badcycles ] let @xmath2 be a chordal graph and @xmath24 be the clique arrangement of @xmath14 . then @xmath14 is strongly chordal , if and only if @xmath13 is free of bad @xmath1-cycles for all @xmath15 . in this paper we apply two other properties of clique arrangements for strongly chordal graphs : [ lma_node_to_sink_intersection ] let @xmath14 be a strongly chordal graph with clique arrangement @xmath24 and let @xmath108 be three distinct nodes such that @xmath109 . there are sinks @xmath110 such that @xmath111 is reachable from @xmath75 and @xmath112 is reachable from @xmath113 and @xmath114 . [ lma_cliquearrangement_inducedsubgraphs ] let @xmath2 be a chordal graph with clique arrangement @xmath24 that occurs as an induced subgraph of a chordal graph @xmath115 with clique arrangement @xmath116 , that is , @xmath117 $ ] . there exists a function @xmath118 that fulfills the following two conditions for all @xmath119 : 1 . @xmath120 , and 2 . @xmath13 has a directed path from @xmath74 to @xmath75 , if and only if @xmath121 has a directed path from @xmath122 to @xmath123 . bibelnieks etal . @xcite are the first to find a strongly chordal graph , namely @xmath22 , that is not in @xmath16 and , consequently , show that the classes are not equivalent . in fact , they were looking for a strongly chordal graph that is not a fixed tolerance nest graph , but by brandstdt etal . @xcite , we know that @xmath16 and this class are equal . since then , it has been conjectured that @xmath22 is the smallest forbidden induced subgraph of leaf powers . to show that @xmath22 is not in @xmath16 , bibelnieks etal . @xcite use a lemma of broin etal . the basic idea of the proof of this lemma is to show for certain pairs of edges @xmath124 and @xmath125 in @xmath14 that the path between @xmath58 and @xmath126 is disjoint from the path between @xmath127 and @xmath128 in every leaf root of @xmath14 . in particular , this happens , if vertices @xmath129 exist in @xmath14 with @xmath130 $ ] and @xmath131 $ ] . the graph @xmath22 has a cycle @xmath132 , where the condition is fulfilled for many pairs of edges in the cycle . it follows that every leaf root of @xmath22 would have a cycle , which is a contradiction . in this section , we want to show that there are at least six other strongly chordal graphs @xmath133 that are not in @xmath16 . interestingly , every of these six graphs is smaller than @xmath22 . for our proof , we generalize the argument of bibelnieks etal . @xcite for pairs of edges @xmath124 and @xmath125 that correspond to disjoint paths in leaf roots . the following lemma provides three corresponding conditions : [ lma_criticaledges ] let @xmath2 be a @xmath1-leaf power of a tree @xmath3 for some @xmath17 and let @xmath124 and @xmath125 be two edges of @xmath14 on distinct vertices @xmath134 . the paths @xmath135 and @xmath136 in @xmath3 are disjoint , that is , do not share any node , if at least one of the following conditions holds : 1 . at most one of the edges @xmath137 is in @xmath36 . 2 . there is a vertex @xmath138 such that @xmath139 , and @xmath140 $ ] , and @xmath141 , and @xmath142 . 3 . there are distinct vertices @xmath143 such that @xmath144 $ ] , and @xmath145 $ ] . 1 . assume that the two paths are not disjoint . then @xmath3 contains ( not necessarily distinct ) nodes @xmath146 and @xmath147 such that ( i ) the path @xmath135 consists of three subpaths , firstly @xmath148 , secondly @xmath149 , and thirdly @xmath150 and ( ii ) the path @xmath136 consists of three subpaths , too , without loss of generality , the first is @xmath151 and the last is @xmath152 . hence , the path between @xmath146 and @xmath147 is the intersection between the two paths . because @xmath153 and @xmath154 in @xmath14 we get the following inequations by definition : @xmath155 as at most one of the edges @xmath156 is in @xmath36 , we know that at least one of @xmath157 and @xmath158 is true . if @xmath159 and @xmath160 , then we get @xmath161 such that combining ( [ eqn_lem_criticaledges_a ] ) and ( [ eqn_lem_criticaledges_c ] ) yields @xmath162 and combining ( [ eqn_lem_criticaledges_b ] ) and ( [ eqn_lem_criticaledges_d ] ) yields @xmath163 , a contradiction . otherwise , if @xmath164 and @xmath165 , we get the inequations @xmath166 such that combining equation ( [ eqn_lem_criticaledges_a ] ) and ( [ eqn_lem_criticaledges_e ] ) yields @xmath167 . putting this estimate of @xmath168 into ( [ eqn_lem_criticaledges_b ] ) yields @xmath169 . by ( [ eqn_lem_criticaledges_f ] ) we can conclude that @xmath170 , which is a contradiction to the preconditions . \2 . as the edges @xmath171 and @xmath125 are joined in @xmath14 by at most one edge , @xmath172 or @xmath173 , it follows from 1 . that @xmath174 is disjoint from @xmath136 in @xmath3 . analogously , the edges @xmath175 and @xmath125 are joined by at most one edge in @xmath14 , either @xmath176 or @xmath177 . hence , in @xmath3 , the path @xmath178 is disjoint from @xmath136 , too . because @xmath3 is a tree , it follows that the nodes on @xmath135 are a subset of the combined nodes of the paths @xmath174 and @xmath178 . consequently , there is no node that simultaneously belongs to @xmath135 and @xmath136 . \3 . if @xmath179 then @xmath180 for all @xmath181 and @xmath182 . otherwise , @xmath183 is an induced @xmath184 in @xmath14 . hence , in this case @xmath164 , @xmath159 , @xmath160 and @xmath165 and we are done . if @xmath185 , then for all @xmath181 and @xmath182 , the edges @xmath186 and @xmath187 are joined at most by the edge @xmath188 in @xmath14 . this means by 1 . that @xmath189 is disjoint from @xmath190 in @xmath3 . again , as @xmath3 is a tree , it follows that the nodes on @xmath135 are a subset of the accumulated nodes on @xmath174 and @xmath178 and , similarly , the nodes on @xmath136 are a subset of the nodes on @xmath191 and @xmath192 . consequently , there can not be a node that simultaneously belongs to @xmath135 and @xmath136 . based on this more general concept , we can find a cycle @xmath132 in every graph from @xmath35 such that many pairs of edges in the cycle fulfill at least one of the three conditions . the following theorem states that this is never compatible with the existence of a leaf root . [ thm_counterexamples ] the graphs @xmath35 are not in @xmath16 . = [ draw = black , fill = black , circle , minimum size=3pt , inner sep=0 ] + @xmath193 + + @xmath194 + + @xmath195 + @xmath196 + + @xmath197 + @xmath198 + + @xmath22 = [ shape = rounded rectangle , text width=1.1 cm , text centered , draw = black , scale=0.9 ] = [ shape = rounded rectangle , text width=1.1 cm , text centered , draw = black , densely dashed , scale=0.9 ] = [ shape = rounded rectangle , text width=1.1 cm , text centered , draw = black , double , scale=0.9 ] = [ shape = rounded rectangle , text width=1.1 cm , text centered , draw = black , line width=0.85pt , scale=0.9 ] = [ color = black ] = [ color = black , densely dashed ] = [ color = black , line width=1pt ] this implies that @xmath35 are forbidden induced subgraphs for @xmath16 . in the following section , we analyze the clique arrangement of these seven graphs and show that they share one particular cyclic property , related to bad @xmath1-cycles . as shown in @xcite , strongly chordal graphs can be characterized by forbidden bad @xmath1-cycles in their clique arrangements , where @xmath15 . but by theorem [ thm_counterexamples ] , this does not fully capture the cyclic structure that is forbidden in leaf powers . in this section , we show that there are certain kinds of @xmath8-cycles which may not occur as a subgraph in the clique arrangement of a leaf power . in particular , we call a @xmath8-cycle _ bad _ , if for all @xmath199 there is a directed path from starter @xmath99 to terminal @xmath106 that does not contain a node @xmath74 which fulfills @xmath200 . the following theorem provides the main argument of this paper : [ thm_bad2cyc_counterexp ] let @xmath2 be a strongly chordal graph with clique arrangement @xmath24 . the graph @xmath13 contains a bad @xmath8-cycle , if and only if @xmath14 contains one of the graphs @xmath35 as an induced subgraph . the proof starts by showing the first direction , that is , if @xmath13 contains a bad @xmath8-cycle , then @xmath14 contains one of the graphs @xmath35 as an induced subgraph . among the bad @xmath8-cycles of @xmath13 we select a cycle with starters @xmath201 and terminals @xmath202 that primarily minimizes the summed cardinalities of the terminals @xmath203 and secondarily maximizes the summed cardinalities of the starters @xmath204 . because @xmath205 and @xmath206 have a non - empty intersection , which contains at least @xmath207 , lemma [ obs_node_intersection ] provides a node @xmath208 . in the following we provide a number of claims to support our arguments . the proofs of all these claims are found in section [ sec : techproofs ] . we start by shaping the bad @xmath8-cycle : [ cla_pnodes ] for all @xmath199 there is a path @xmath209 from @xmath99 to @xmath106 that does not contain a node @xmath74 with @xmath200 , in particular @xmath209 does not contain @xmath3 , such that @xmath209 contains a node @xmath210 with * @xmath211 , * @xmath212 , * @xmath213 , and * there exists a sink @xmath214 in @xmath13 with @xmath215 that fulfills @xmath216 . in the following we refer to the nodes @xmath210 by the @xmath217-nodes and we call @xmath214 the @xmath218-nodes . the pure existence of the @xmath218-nodes does not directly imply that they are different : [ cla_qnode : intersection ] for all @xmath219 with @xmath220 , the sinks @xmath214 and @xmath221 differ . for the pairwise intersection between the @xmath217-nodes , claim [ cla_pnodes ] directly implies for all @xmath222 that @xmath223 . we can now infer the following two additional statements about the intersections between the @xmath217-nodes and the intersections between the @xmath218-nodes : [ cla_pnode : intersection : vertical ] for all @xmath224 it is true that @xmath225 . [ cla_pqnode : intersection : hor : dia ] for all @xmath226 it is true that @xmath227 and @xmath228 . we deduce that @xmath229 for all @xmath219 with @xmath220 . following the construction of the @xmath217-nodes , we also know for all @xmath199 that the set @xmath230 is not empty . using the collected facts about the mentioned nodes on the bad @xmath8-cycle , the next two claims start selecting vertices to construct one of the induced subgraphs @xmath35 : [ cla_uvertices ] for all @xmath224 , the starter @xmath99 contains a vertex @xmath231 such that @xmath232 . [ cla_pwvertices ] for all @xmath199 , there is a vertex @xmath233 such that * for all @xmath234 it is true that @xmath235 and * @xmath236 is neither adjacent to @xmath237 , nor to any vertex in @xmath238 , in @xmath239 or in @xmath240 . depending on the edges between the six central vertices of @xmath35 , there exist up to two additional vertices in @xmath241 . this dependency is also visible in the clique arrangement . consider the sets @xmath242 , @xmath243 , @xmath244 and @xmath245 and moreover , for all @xmath199 let @xmath246 . if one of the sets @xmath247 induces a clique in @xmath14 , then it follows that @xmath205 or @xmath206 are proper subsets of maximal cliques in @xmath14 : [ cla_tsinks ] for all @xmath199 and @xmath248 , the node @xmath249 is not a sink in @xmath13 , if @xmath250 is a clique in @xmath14 . in such a case , if @xmath250 is a clique , we select an additional vertex from the sink that is reachable from @xmath249 : [ cla_twvertices ] for all @xmath199 and @xmath248 , if @xmath250 is a clique in @xmath14 , then there is a sink @xmath251 which is reachable from @xmath249 and contains a vertex @xmath252 such that * @xmath253 is not one of the vertices @xmath254 , * @xmath253 is neither adjacent to @xmath255 nor to any vertex in @xmath256 , and * @xmath253 is adjacent to at most one vertex of @xmath257 and @xmath258 . in the remainder of the proof we select the central vertices @xmath259 from @xmath260 for all @xmath199 to ultimately induce a forbidden subgraph . but before explaining how to select these four vertices , we briefly summarize the results gathered in the proof so far . by claim [ cla_uvertices ] , we know that there are vertices @xmath261 and , from the construction of the @xmath217-nodes in claim [ cla_pnodes ] , it follows that @xmath262 and @xmath263 are cliques in @xmath14 , regardless of the choice of @xmath264 . moreover , by claim [ cla_pwvertices ] , there exists an independent set @xmath265 in @xmath14 such that for all @xmath199 , the vertex @xmath236 is adjacent to @xmath231 and @xmath259 but not to any of the vertices @xmath266 . finally , claim [ cla_twvertices ] states that certain circumstances imply the existence of two non - adjacent vertices @xmath267 and @xmath268 in @xmath14 that are both adjacent to @xmath269 and @xmath270 and such that for all @xmath271 it is true that @xmath253 is adjacent to @xmath272 and @xmath273 but not adjacent to @xmath274 and @xmath275 . the claim leaves it open , if @xmath253 can be adjacent to either @xmath257 or @xmath258 and , consequently , we cope with this problem during the following vertex selection . these facts are subsequently used without explicit mentioning . moreover , in the following vertex selection we write @xmath276)\ ] ] to state that @xmath14 contains an induced @xmath277 for @xmath278 on vertices @xmath279 , @xmath58 , @xmath280 , @xmath281 , @xmath282 , @xmath283 , @xmath284 , @xmath285 , @xmath286 , @xmath287 , optionally including @xmath288 . hence , both vertex sets @xmath289 and @xmath290 form a clique in @xmath14 and every @xmath291 is exactly adjacent to @xmath292 . depending on @xmath293 the vertices @xmath294 and @xmath295 are present and @xmath296 is exactly adjacent to @xmath297 for @xmath224 . the adjacency between @xmath298 depends on @xmath293 , too . to find suitable vertices for the forbidden induced subgraphs , we have to distinguish between three cases : 1 . * assume that at most one of the sets @xmath299 is a clique in @xmath14 : * because of symmetry we just have the following two subcases : 1 . * assume that at most @xmath300 is a clique in @xmath14 : * because @xmath301 are not cliques , we can select vertices @xmath302 for all @xmath199 such that @xmath303 and @xmath304 . at most one of the edges @xmath305 or @xmath306 is present in @xmath36 because otherwise @xmath307 is an induced @xmath184 in @xmath14 . if @xmath308 and @xmath309 , then @xmath310 clearly , if @xmath300 is a clique , then @xmath311 and @xmath309 and then we have @xmath312 2 . * assume that at most @xmath313 is a clique in @xmath14 : * analogously to the previous case , we can select @xmath302 for all @xmath199 such that @xmath308 and @xmath309 and again , either @xmath314 or @xmath315 as otherwise @xmath316 is an induced @xmath184 in @xmath14 . the case of @xmath308 and @xmath309 yields an induced @xmath193 and has already been handled in the first case . if without loss of generality @xmath314 , then @xmath317 2 . * assume that exactly two of the sets @xmath318 are cliques in @xmath14 : * if @xmath300 and @xmath319 are cliques but not @xmath313 and @xmath320 , then vertices @xmath302 exist for all @xmath199 such that @xmath303 and @xmath304 and consequently , @xmath307 is an induced @xmath184 . analogously , @xmath313 and @xmath320 being the cliques implies @xmath316 as an induced @xmath184 . because of this and symmetry , we have only one remaining case , namely @xmath300 and @xmath313 are the cliques and this implies that @xmath321 is a clique . + next we show that there exist vertices @xmath302 for all @xmath199 such that @xmath304 and @xmath309 . for that purpose , assume that every vertex in @xmath322 , that is adjacent to some vertex in @xmath323 for @xmath271 , is also adjacent to all vertices in @xmath324 . then , as @xmath319 and @xmath320 are not cliques , there are vertices @xmath325 such that there is @xmath326 and @xmath327 with @xmath328 and @xmath329 . by our assumption , it follows that @xmath330 and @xmath331 and hence , there is @xmath332 , an induced @xmath184 in @xmath14 . consequently , the assumption was wrong and we can select the vertices such that @xmath304 and @xmath309 . + because @xmath321 is a clique , it follows by claim [ cla_twvertices ] that @xmath268 exists , and if @xmath268 is neither adjacent to @xmath333 nor to @xmath334 , then @xmath335 otherwise , if @xmath268 is adjacent to @xmath336 , then we get @xmath337 and if @xmath338 , then @xmath339 3 . * assume that at least three of the sets @xmath318 are cliques in @xmath14 : * in this case , we select any vertex @xmath302 for all @xmath199 . as at least one of the sets @xmath340 or @xmath341 is a clique , it follows from claim [ cla_twvertices ] that @xmath268 exists . analogously , @xmath342 or @xmath343 is a clique and thus , @xmath267 exists . + assume first that @xmath267 and @xmath268 are completely disjoint from @xmath344 . by symmetry we just have to consider the cases of ( i ) @xmath304 , which leads to @xmath345 ( ii ) @xmath309 , which yields @xmath346 and ( iii ) @xmath264 are a clique where @xmath347 + finally , we have to check all the cases where @xmath267 or @xmath268 are adjacent to one of the vertices @xmath344 . because of symmetry we can simply assume that @xmath267 is adjacent to @xmath336 . + assume that @xmath268 is neither adjacent to @xmath333 nor to @xmath334 . then ( iv ) @xmath311 and @xmath303 implies @xmath348 ( v ) @xmath308 and @xmath314 yields @xmath349 and ( vi ) @xmath311 and @xmath314 gives @xmath350 + if @xmath338 , then ( vii ) @xmath308 implies @xmath351 and ( viii ) @xmath311 yields @xmath352 moreover , if @xmath353 , then ( ix ) @xmath303 results in @xmath354 and ( x ) @xmath314 provides @xmath355 the following shows the converse direction , that is , if @xmath14 contains one of @xmath35 as an induced subgraph , then @xmath13 has a bad @xmath8-cycle . we basically use lemma [ lma_cliquearrangement_inducedsubgraphs ] . the clique arrangement of all graphs @xmath35 contains a bad @xmath8-cycle with starters @xmath356 , @xmath357 and terminals @xmath358 , @xmath359 . moreover , there are nodes @xmath360 , @xmath361 such that @xmath362 and @xmath363 for all @xmath199 . if @xmath14 contains an induced subgraph @xmath35 , then there is a function @xmath364 , that maps these nodes to some nodes of the clique arrangement @xmath13 such that @xmath365 and @xmath366 for all @xmath199 . assume that at least one of these four paths in @xmath13 , say @xmath367 , contains a node @xmath74 with @xmath368 . if @xmath74 is situated on the subpath @xmath369 , then it follows that @xmath370 and , hence , @xmath371 , a contradiction . hence , @xmath74 is on the subpath @xmath372 . here , @xmath373 is a subset of @xmath374 and thus , also a subset of @xmath375 . this means that @xmath376 , which implies @xmath377 and @xmath378 . consequently , we are in the case were the induced subgraph in @xmath14 is one of @xmath241 . the clique arrangement of all these graphs contains a sink @xmath379 that is reached from @xmath206 . in @xmath13 , we have @xmath380 , thus , @xmath381 , which finally means that @xmath382 , a contradiction . hence , @xmath74 does not exist and @xmath13 contains a bad @xmath8-cycle with starters @xmath383 and terminals @xmath384 . the main theorem , presented in this section , and theorem [ thm_counterexamples ] lead to the following conclusion : [ cor_bad2cyc_leafp ] let @xmath2 be a graph in @xmath16 with clique arrangement @xmath24 . the graph @xmath13 does not contain a bad @xmath8-cycle . hence , leaf powers fit naturally into the hierarchy of chordal graphs , right between strongly chordal graphs , which have clique arrangements without bad @xmath1-cycles for @xmath15 , and ptolemaic graphs , whose clique arrangements are entirely free of cycles . in this paper , we were able to indicate that leaf powers @xmath16 are a natural subclass of strongly chordal graphs by showing that their clique arrangements are not only free of bad @xmath1-cycles for @xmath15 but also for @xmath385 . moreover , we proved that the clique arrangement of a strongly chordal graph @xmath14 comprises a bad @xmath8-cycle , if and only if @xmath14 contains at least one of @xmath35 as an induced subgraph . this means that , beside the forbidden induced subgraphs of strongly chordal graphs , that is , the family of suns , this finite number of graphs describe a cyclic composition of cliques that is not realizable by a @xmath1-leaf root for any @xmath17 . it remains for future work to find a complete characterization of @xmath16 in terms of forbidden subgraphs . during our deep analysis of leaf powers we have considered a huge variety of graphs and their clique arrangements . we have not a single example of a graph @xmath14 that has a clique arrangement @xmath13 without bad @xmath1-cycles for @xmath17 , where a corresponding leaf root of @xmath14 is unknown . therefore , we conjecture that a strongly chordal graph @xmath14 has a @xmath1-leaf root for some @xmath17 , if and only if @xmath13 is free of bad @xmath8-cycles . if this was true , a polynomial time recognition algorithm is straight found by the efficient recognition of strongly chordal graphs and the possibility to check for a finite number of induced subgraphs in polynomial time . answering this question implies the challenge of constructing leaf roots from bad - cycle - free clique arrangements . this turns out to be sophisticated , especially if the clique arrangement has @xmath8-cycles that are not bad . first of all , we know that @xmath386 and @xmath387 as otherwise @xmath388 are not distinct nodes . let @xmath389 be the set of sinks reachable from @xmath75 and @xmath390 be the set of sinks reachable from @xmath113 . if @xmath75 and @xmath113 are sinks themselves then @xmath391 and @xmath392 and trivially we set @xmath393 and @xmath394 . if just one of the nodes , @xmath75 or @xmath113 , is a sink , without loss of generality , @xmath391 , then lemma [ lma_stronglychordal_cliquecutstructure ] implies that there are sinks @xmath395 such that @xmath396 and @xmath397 . clearly , @xmath398 and @xmath399 as otherwise @xmath400 . by lemma [ obs_node_intersection ] , @xmath401 and @xmath402 are nodes in @xmath26 because both contain @xmath74 . it is straight forward that @xmath403 and moreover , @xmath404 and @xmath405 as otherwise @xmath400 . we have a @xmath9-cycle with starters @xmath406 and terminals @xmath407 . by theorem [ thm_badcycles ] , the cycle is not bad and as @xmath387 it follows that @xmath408 or @xmath409 . if @xmath408 , then @xmath410 and @xmath411 imply that @xmath412 . from @xmath413 and @xmath414 we obtain @xmath415 and , as @xmath416 , we have @xmath417 . hence , in this case @xmath418 and we set @xmath419 and @xmath420 . in an analogous fashion we get @xmath421 , if @xmath409 and then we set @xmath422 and @xmath420 . if none of the nodes @xmath423 is a sink , then lemma [ lma_stronglychordal_cliquecutstructure ] implies that there are sinks @xmath424 and @xmath425 such that @xmath426 and @xmath427 . if @xmath75 reaches one of the sinks @xmath49 and @xmath428 or @xmath113 reaches one of the sinks @xmath429 and @xmath430 , without loss of generality , @xmath113 reaches @xmath430 , then @xmath431 . notice that @xmath113 can not reach @xmath429 in this case , as otherwise @xmath400 . by lemma [ obs_sink_intersection ] , @xmath113 is exactly the intersection of all cliques in @xmath432 which includes @xmath430 . by definition , @xmath433 and , hence , we also have @xmath434 . as @xmath435 , we conclude that @xmath436 . the node @xmath437 exists by lemma [ obs_node_intersection ] because it contains @xmath74 as a subset . clearly , if @xmath438 or @xmath439 then @xmath440 . otherwise , we have a @xmath9-cycle with starters @xmath441 and terminals @xmath442 . by theorem [ thm_badcycles ] , the cycle is not bad and as @xmath443 or @xmath444 implies @xmath445 or @xmath400 , we have @xmath446 , again . because @xmath447 and @xmath436 , this implies that @xmath413 and @xmath412 , and because @xmath109 , we have @xmath415 and thus , @xmath417 . hence , @xmath448 and we set @xmath419 and @xmath449 . if neither @xmath75 reaches the sinks @xmath49 or @xmath428 nor @xmath113 reaches the sinks @xmath429 or @xmath430 , then we consider the nodes @xmath437 and @xmath450 , which exist by lemma [ obs_node_intersection ] , as they all contain @xmath74 as a subset . clearly , if @xmath451 or @xmath452 coincides with @xmath75 , then @xmath75 reaches one of the sinks @xmath49 or @xmath428 and analogously , @xmath451 and @xmath452 are not @xmath113 . moreover , @xmath453 implies that @xmath451 is a subset of @xmath75 and @xmath113 and , by that , a subset of @xmath74 , which means that @xmath454 and that @xmath419 and @xmath449 . otherwise , as @xmath455 , we get a @xmath10-cycle with starters @xmath456 and terminals @xmath457 . thm_badcycles ] states that the cycle is not bad and as @xmath458 and @xmath459 , it must be true , without loss of generality , that @xmath408 and hence , we get that @xmath413 and that there is a @xmath9-cycle with starters @xmath406 and terminals @xmath460 . this cycle is not bad , either , and hence , either @xmath461 or @xmath462 . if @xmath461 , then @xmath412 , which implies @xmath415 and thus , @xmath463 and then @xmath419 and @xmath449 . the case @xmath462 implies that @xmath464 and then we consider the node @xmath465 , which exists because it contains @xmath74 as a subset . if @xmath466 , @xmath452 is contained in @xmath109 , which means that @xmath467 and that @xmath422 and @xmath468 . otherwise , we have a @xmath9-cycle with starters @xmath469 and terminals @xmath470 . because the cycle is not bad , it is true that @xmath409 or @xmath471 . if @xmath409 , then @xmath472 , which implies @xmath473 and thus , @xmath474 and then @xmath422 and @xmath468 . in the other case , @xmath50 is contained in @xmath429 and @xmath430 and thus , in @xmath75 . moreover , @xmath50 is a subset of @xmath475 , thus , a subset of @xmath452 , and , consequently , contained in @xmath476 , which means that @xmath50 is also in @xmath113 . together , this implies that @xmath477 and that @xmath419 and @xmath468 . we first fix a function @xmath364 . for that purpose notice that for every maximal clique @xmath49 of @xmath478 , there is at least one maximal clique @xmath50 in @xmath14 such that @xmath479 and we define @xmath480 . notice that @xmath481 . moreover , for every node @xmath89 that is not a maximal clique , there is the subset @xmath92 of maximal cliques in @xmath478 that are reached in @xmath482 by a directed path from @xmath74 . as @xmath93 , we define @xmath483 for the node @xmath484 that fulfills @xmath485 . 1 . since @xmath364 is a function , @xmath486 implies @xmath487 . conversely , if @xmath487 but @xmath488 , then there are non - adjacent vertices @xmath489 and @xmath490 , which are , by definition , both in @xmath122 , a contradiction . 2 . by definition , there is a directed path from @xmath122 to @xmath123 in @xmath121 , if @xmath491 . as @xmath492 and @xmath493 this implies @xmath494 , which , by definition , means that there is a directed path from @xmath74 to @xmath75 in @xmath13 . + conversely , let there be a directed path from @xmath74 to @xmath75 in @xmath13 , thus , let @xmath494 . this means that in @xmath13 the set @xmath92 of maximal cliques reached from @xmath75 is a subset of the maximal cliques @xmath495 reached from @xmath74 , hence , @xmath496 . consequently , @xmath497 , which implies that there is a directed path from @xmath122 to @xmath123 in @xmath121 . the proof works basically the same as in @xcite . assume that at least one of the graphs @xmath35 is a @xmath1-leaf power of a tree @xmath3 for some @xmath17 and that @xmath498 are the parent nodes of the leaves @xmath499 in @xmath3 . consider for all @xmath199 the path @xmath500 in @xmath3 as well as , for all @xmath224 , the path @xmath501 in @xmath3 . from lemma [ lma_criticaledges ] we get that @xmath502 and @xmath503 . similarly , lemma [ lma_criticaledges ] implies that @xmath504 and @xmath505 . this means that the subtree @xmath205 of @xmath3 given by the union @xmath506 is disjoint from the subtree @xmath206 of @xmath3 given by @xmath507 . as @xmath3 is a tree , there is a node @xmath508 situated on the path connecting the subtrees @xmath205 and @xmath206 such that @xmath508 is on every path @xmath69 in @xmath3 that connects a node @xmath6 from @xmath205 and a node @xmath7 from @xmath206 . in particular , that also means that @xmath508 is on @xmath509 , if @xmath510 and @xmath511 , and that @xmath508 is on @xmath512 , if @xmath513 and @xmath514 . hence , @xmath515 , as both paths contain @xmath508 , which is a contradiction to lemma [ lma_criticaledges ] . the claims proved in the following are stated in a general and simple fashion , and they often use indices @xmath199 for the occurring nodes . however , because the bad @xmath8-cycle is symmetric , the proofs always show the individual statements just for the case @xmath516 without explicit indication . we start by choosing an arbitrary path @xmath517 from @xmath518 to @xmath205 that does not contain a node @xmath74 with @xmath200 , which exists by the definition of bad @xmath8-cycles . obviously , this implies that the node @xmath3 is not on @xmath517 . firstly , there are nodes @xmath519 on the path @xmath520 that are joined by an arc @xmath521 such that @xmath522 and @xmath523 and @xmath524 , hence , on @xmath517 , the node @xmath217 is the last exit to @xmath3 . clearly , we have @xmath525 and thus , if such arc does not exist , then every node on the path , except @xmath205 itself , would be a subset of @xmath3 . because @xmath3 is not on @xmath526 , even the predecessor @xmath218 of @xmath205 reaches @xmath3 by a directed path . hence , as @xmath527 , there is a directed path @xmath528 and , consequently , the arc @xmath529 is transitive , a contradiction . next we show that @xmath530 implies also that @xmath531 . this can be seen by the use of the intersection node @xmath532 , which entirely contains @xmath533 because @xmath534 and @xmath535 . as @xmath523 and @xmath536 , it follows that @xmath74 is not equal to the node @xmath537 . moreover , since @xmath538 and @xmath522 , it follows that @xmath539 and hence , there is a path @xmath540 . but @xmath74 can not be a node on that path , unless @xmath541 , because otherwise @xmath521 would be a transitive arc . but @xmath541 implies that @xmath542 passes a node that fulfills @xmath543 , which is a contradiction to the selection of the bad @xmath8-cycle . hence , @xmath544 must be true . however , @xmath537 is not necessarily the node @xmath545 we are looking for . particularly , it may happen that no sink @xmath218 of @xmath13 fulfills @xmath546 . for that reason , let @xmath547 be the sinks reachable from @xmath537 by directed paths and let @xmath548 . because @xmath549 , lemma [ obs_sink_intersection ] implies that , if @xmath550 for all @xmath551 , then @xmath530 . hence , we can select @xmath552 such that @xmath553 and we set @xmath554 and @xmath555 . of course , it may happen that @xmath545 is not on the path @xmath517 , but now we have a new path @xmath556 . we use @xmath452 as a replacement for @xmath517 , because it is easy to see that it does not contain a node @xmath74 with @xmath557 , too . if such a node @xmath74 was on the subpath @xmath558 , then @xmath559 , and , if it was on the subpath @xmath560 , then @xmath561 , which both contradicts the construction of @xmath545 . the case @xmath566 is impossible for all @xmath567 , because then @xmath568 implies @xmath569 , which is forbidden by claim [ cla_pnodes ] . if we assume that @xmath570 , then we get a @xmath9-cycle with starters @xmath571 and terminals @xmath572 . certainly , @xmath545 is not contained in @xmath206 as otherwise @xmath573 implies @xmath574 , which is forbidden by claim [ cla_pnodes ] . similarly , we get that @xmath575 . that @xmath563 is a direct consequence of claim [ cla_pnodes ] . hence , the @xmath9-cycle is bad , a contradiction to theorem [ thm_badcycles ] . let @xmath576 be the node representing the intersection @xmath577 , which exists by lemma [ obs_node_intersection ] as @xmath578 . if we assume that @xmath579 , then we have two paths @xmath580 and @xmath581 and we obtain a @xmath8-cycle with starters @xmath582 and terminals @xmath202 . we show that this cycle is bad by arguing that none of @xmath583 , @xmath584 , @xmath585 , and @xmath586 contains a node @xmath74 that fulfills @xmath587 . clearly , the existence of @xmath74 on one of @xmath585 , @xmath586 , @xmath588 , and @xmath589 contradicts to the choice of @xmath517 , @xmath590 , @xmath585 , and @xmath586 . if @xmath74 was on the path @xmath591 , then we would get @xmath559 , which has been eliminated in claim [ cla_pnodes ] . similarly , the path @xmath592 does not contain @xmath74 , and hence , the new @xmath8-cycle is bad . but this contradicts to the choice of the primal bad @xmath8-cycle , because , by @xmath593 , we obtain @xmath594 . if @xmath595 , then clearly @xmath596 . otherwise , let @xmath218 be the intersection node for @xmath597 , which exists by lemma [ obs_node_intersection ] . we get a @xmath9-cycle with starters @xmath598 and terminals @xmath599 . if @xmath600 then the @xmath9-cycle is bad , because @xmath601 and @xmath563 by claim [ cla_pnodes ] . this contradicts theorem [ thm_badcycles ] . if @xmath269 does not exist , then @xmath518 is a subset of @xmath604 . as @xmath518 can not be entirely contained in a single set , @xmath605 or @xmath606 , we find two distinct nodes @xmath607 and @xmath608 by lemma [ obs_node_intersection ] . the same lemma reveals the existence of a node @xmath609 , because @xmath113 contains at least as @xmath533 . we get a @xmath9-cycle with starters @xmath388 and terminals @xmath610 . we show that this cycle is bad by the help of @xmath611 . firstly , @xmath74 can not be a subset of @xmath606 , because otherwise @xmath606 , which already contains @xmath612 , contains also @xmath613 , which would imply that @xmath614 . secondly and similarly , @xmath75 can not be a subset of @xmath605 , because otherwise @xmath615 . finally , by @xmath616 , it follows that @xmath617 . this bad @xmath9-cycle contradicts theorem [ thm_badcycles ] , hence , the node @xmath269 exists . as the @xmath218-nodes represent distinct maximal cliques in @xmath14 , lemma [ lma_vertex_node_none_adjacency ] allows to select vertices @xmath618 and @xmath619 such that @xmath6 is not adjacent to any vertex in @xmath620 and @xmath7 is not adjacent to any vertex in @xmath621 . we show that at least one of @xmath6 and @xmath7 is not adjacent to all vertices in @xmath622 . if @xmath623 we are done . otherwise , assume that @xmath6 has a neighbor @xmath624 and that @xmath7 has a neighbor @xmath625 . as @xmath626 and @xmath627 and @xmath37 , it follows that @xmath628 as otherwise @xmath14 contains @xmath629 as an induced @xmath184 . now consider the vertex @xmath270 , which is at the same time in @xmath620 and in @xmath621 according to claim [ cla_uvertices ] . hence , according to the choice of @xmath6 and @xmath7 , we have @xmath630 and @xmath631 . moreover , as @xmath270 and @xmath632 are both in @xmath621 and because @xmath270 and @xmath633 are both in @xmath620 , we get @xmath634 and @xmath635 , which implies that @xmath14 has @xmath636 as an induced @xmath637 . this means , our assumption was wrong and we let @xmath638 be a vertex in @xmath639 that has no neighbors in @xmath620 and in @xmath621 . first of all , we have already seen that @xmath638 is not adjacent to @xmath270 . therefore , @xmath638 is in @xmath640 , as every vertex in @xmath545 is adjacent to @xmath270 by @xmath641 . moreover , this means that @xmath642 and @xmath643 as both , @xmath336 and @xmath334 , are adjacent to @xmath270 , which follows from @xmath644 and @xmath645 . as @xmath646 and @xmath647 , it follows also that @xmath638 is not adjacent to @xmath336 and @xmath334 . from claim [ cla_pqnode : intersection : hor : dia ] we know that @xmath648 and @xmath649 are subsets of @xmath3 . because @xmath650 and @xmath651 , this means also that @xmath652 and @xmath653 . hence , from @xmath654 and @xmath655 it follows already that @xmath638 is not adjacent to vertices in @xmath322 or in @xmath656 . it remains to show that @xmath657 , that @xmath658 and that @xmath638 is not adjacent to any vertex in @xmath659 . if @xmath660 is an empty set , then @xmath638 is not adjacent to vertices in @xmath659 otherwise , if @xmath661 is not empty , assume that there are vertices @xmath662 and @xmath663 such that @xmath39 . recall that @xmath638 is adjacent to all vertices in @xmath664 including @xmath7 and not adjacent to @xmath270 . unlike @xmath638 , the vertices @xmath6 and @xmath7 are adjacent to @xmath270 , because @xmath665 and @xmath666 . this means that @xmath14 has @xmath667 as an induced @xmath184 , a contradiction . consequently , @xmath668 is a clique in @xmath14 and there is a maximal clique of @xmath14 represented by a sink @xmath669 of @xmath13 such that @xmath670 . because @xmath208 , it follows from lemma [ lma_node_to_sink_intersection ] that there are distinct sinks @xmath671 , reachable from @xmath205 , and @xmath672 , reachable from @xmath206 , such that @xmath673 . clearly , @xmath674 and because @xmath675 and @xmath676 , we have @xmath677 and @xmath678 . we let @xmath679 and @xmath680 and obtain a @xmath9-cycle with starters @xmath681 and terminals @xmath682 . by construction , there are vertices @xmath683 and @xmath684 that are also contained in @xmath669 . because @xmath685 and @xmath686 , it follows that @xmath687 and @xmath688 and this in turn means that @xmath689 and @xmath690 . consequently , @xmath691 , as otherwise @xmath692 , which is a contradiction to the construction @xmath693 . analogously , we have @xmath694 . finally , @xmath695 implies that all vertices in @xmath3 , including @xmath270 , are adjacent to @xmath638 , which is impossible . this means that the @xmath9-cycle is bad and , hence , @xmath661 has to be empty and @xmath638 is not adjacent to any vertex in @xmath664 . as @xmath333 is adjacent to all vertices in @xmath659 , it follows that @xmath657 . assume that @xmath638 and @xmath333 are adjacent and select any vertices @xmath683 and @xmath684 . if @xmath37 , then we obtain @xmath696 as an induced @xmath184 in @xmath14 , and otherwise , we get @xmath697 as an induced @xmath637 in @xmath14 . hence , @xmath698 can not be true . let @xmath699 be a sink of @xmath13 that represents one of the maximal cliques of @xmath14 with @xmath700 . consider the node @xmath672 that results from the intersection @xmath701 , which exists by lemma [ obs_node_intersection ] , as both , @xmath699 and @xmath206 , contain @xmath702 . if @xmath703 , we get a new @xmath8-cycle with starters @xmath201 and terminals @xmath704 . assume that there is a node @xmath74 with @xmath705 on one of @xmath517 , @xmath585 , @xmath706 , and @xmath707 . if @xmath74 is neither on the path @xmath708 nor on the path @xmath709 , then @xmath74 is located on one of the paths @xmath517 , @xmath590 , @xmath585 , and @xmath586 , a contradiction to the choice of these paths . if @xmath710 , then it follows that @xmath711 , which is impossible due to the construction of @xmath712 . the same holds if @xmath74 is on the path @xmath709 . hence , the new @xmath8-cycle is bad . this is a contradiction to the choice of the primal cycle , because , by @xmath703 , we have @xmath713 . consequently , @xmath672 equals @xmath206 , and thus , @xmath714 , which means that @xmath206 is not a sink in @xmath13 . lemma [ lma_node_to_sink_intersection ] implies the existence of two distinct sinks @xmath671 , reachable from @xmath205 , and @xmath672 , reachable from @xmath206 , such that @xmath715 . because @xmath321 is a clique , claim [ cla_tsinks ] implies that @xmath206 is not a sink , hence , @xmath716 . from lemma [ lma_vertex_node_none_adjacency ] it follows that @xmath717 contains at least one vertex @xmath268 that is not adjacent to any vertex in @xmath718 . as @xmath321 is a clique , all vertices in @xmath712 and in @xmath719 are adjacent to all vertices in @xmath545 . consequently , @xmath268 is not in @xmath702 . if @xmath267 exists , then it can neither be the same vertex as @xmath268 nor be adjacent to @xmath268 , because @xmath720 . clearly , by claim [ cla_pwvertices ] , @xmath268 is not one of the vertices @xmath721 , because , unlike @xmath268 , they are adjacent to vertices in @xmath722 . moreover , claim [ cla_pwvertices ] implies that @xmath268 is not @xmath333 , because , unlike @xmath333 , the vertex @xmath268 is adjacent to all vertices in @xmath656 . similarly @xmath268 is not @xmath334 . it remains to show that @xmath268 is not adjacent to @xmath638 and @xmath336 and adjacent to at most one vertex @xmath336 or @xmath334 . if @xmath723 , then we can select any vertex @xmath724 and get @xmath725 as an induced @xmath184 in @xmath14 . analogously , if @xmath726 , then we select @xmath727 to find @xmath728 as induced @xmath184 in @xmath14 . finally , if @xmath268 is adjacent to @xmath333 and @xmath334 , then we select @xmath683 and get an induced @xmath9-sun in @xmath14 with central clique @xmath729 and independent set @xmath730 .

observational and theoretical arguments suggest that the momentum carried in mass outflows from agn can reach several times , corresponding to outflow rates of hundreds of solar masses per year . radiation pressure on resonant absorption lines alone may not be sufficient to provide this momentum deposition , and the transfer of reprocessed ir radiation in dusty nuclear gas has been postulated to provide the extra enhancement . the efficacy of this mechanism , however , will be sensitive to multi - dimensional effects such as the tendency for the reprocessed radiation to preferentially escape along sightlines of lower column density . we use monte carlo radiative transfer calculations to determine the radiation force on dusty gas residing within approximately 30 parsecs from an accreting super - massive black hole . we calculate the net rate of momentum deposition in the surrounding gas and estimate the mass - loss rate in the resulting outflow as a function of solid angle for different black hole luminosities , sightline - averaged column densities , clumping parameters , and opening angles of the dusty gas . we find that these dust - driven winds carry momentum fluxes of 1 - 5 times and correspond to mass - loss rates of 10 - 100 per year for a black hole radiating at or near its eddington limit . these results help to explain the origin of high velocity molecular and atomic outflows in local ulirgs , and can inform numerical simulations of galaxy evolution including agn feedback .

1204.0063

the nature of the interaction between an accreting super - massive black hole ( smbh ) and its host galaxy remains a challenging problem in the study of galaxy evolution . numerical simulations reveal that gas can be drawn inward toward the nucleus by gravitational torques arising from a series of gravitational instabilities @xcite . this gas typically forms a dusty structure at small radii with a characteristic length scale of @xmath41 - 10 parsecs which in some cases has been imaged directly @xcite . phenomenologically , this structure can be modelled as a torus @xcite , but there is an ongoing theoretical effort to provide a detailed , self - consistent explanation of its configuration and what supports it . if a sufficiently strong poloidal magnetic field is present at the parsec scale , one possible explanation is that the dusty gas is launched as a hydromagnetic wind @xcite . heating of the ism from stellar feedback might support the dusty gas in a puffy disk @xcite . the disk might be simultaneously supported by infrared radiation pressure @xcite , or the infrared radiation pressure may generate a failed wind @xcite . regardless of what supports the torus , gas continues to be drawn in to the black hole accretion disk at small radii ( @xmath5 cm ) , where it powers an active galactic nucleus ( agn ) . the radiation emitted from smbh accretion disks influences the dynamics of the torus itself , along with the dynamics of the host galaxy . this feedback may act through a number of channels that include radiative heating ( e.g. * ? ? ? * ) , jets @xcite , and winds driven by radiation pressure on resonant ultraviolet lines @xcite and dust @xcite . our challenge is to understand the combined effect of all these modes of interaction . improving our understanding of this connection will be crucial for answering questions about the growth of smbhs , observations of agn , and the star formation histories in galaxies . recent observations have begun to reveal the violent impact that agn may have on their host galaxies . observations of obscured quasars such as sdss j1356 + 1026 have revealed outflows extending out to tens of kiloparsecs from the galactic nucleus @xcite . the estimated mechanical luminosity of these outflows ( @xmath6 ergs s@xmath7 ) is too large to be explained by the inferred star formation activity . other obscured quasars possess more massive outflows , with mass - loss rates of hundreds of solar masses per year @xcite . meanwhile , observations of local ultra - luminous infrared galaxies ( ulirgs ) have led to the discovery of outflows with velocities that are correlated with the agn bolometric luminosity @xcite . these outflows also have mass - loss rates equal to several times the star formation rate and in some cases exceeding 1000 solar masses per year , depleting the gas on timescales as short as @xmath8 years . adding to our picture are studies of post - starburst galaxies , exhibiting outflows with median velocity of approximately 1000 km s@xmath7 , suggesting that past agn activity played a role in launching the gas @xcite . these observations are complemented by numerical simulations of agn feedback @xcite that account for deposition of both energy and momentum from the accretion radiation , including a combination of heating by x - rays and photoionizations , radiation pressure at the kiloparsec scale , and winds driven from within a radius of less than 100 parsecs . taken together , these effects can help to explain both the @xmath9 - @xmath10 relation @xcite and the existence of galactic outflows observed at speeds of thousands of km s@xmath7 . the results , particularly those of @xcite , also suggest that line - driven winds may be insufficient to drive observed outflows , and that a large amount of momentum ( @xmath11 ) may need to be deposited via absorption by dust grains during the period when the smbh is optically thick to both ultraviolet and far - infrared radiation , the time when most black hole growth is believed to occur @xcite . a large uncertainty in the numerical calculations referenced above is the amount of radiative momentum deposited within the central unresolved radius . the velocity and mass - loss rate of the resulting wind depend sensitively on this coupling . moreover , in those studies the momentum was deposited in a spherically symmetric manner . in reality , multidimensional effects , such as the tendency for radiation to escape out the rarefied , polar regions of the gas distribution , will be crucial . these effects have been considered by several previous studies . @xcite computed the radiation forces exerted on a torus modelled as a constant density cylindrical shell , and @xcite extended that work to account for a more self - consistent rearrangement of the gas under the influence of the radiation . a radiation - hydrodynamics study that linked the effects of compton scattering and broad absorption line winds at the parsec scale with inflow processes on galactic ( kiloparsec ) scales in two spatial dimensions was undertaken by @xcite , and this was extended in order to capture the radiative transfer through dusty gas in @xcite . our study extends this work further by performing three dimensional monte carlo radiative transfer calculations for dusty gas , including both smooth and clumpy gas distributions , and by integrating the force on columns of gas in order to quantify the mass outflow rate from agn radiating at high luminosity . the momentum flux in radiation from a smbh accretion disk with luminosity @xmath12 is @xmath1 . generally @xmath12 will not exceed @xmath13 , the eddington luminosity set by the electron scattering ( thomson ) opacity . dust will contribute to the opacity seen by the radiation at large radii , but only at distances greater than the radius @xmath14 at which its temperature drops below the sublimation temperature @xmath15 . although the sublimation temperature varies for each grain depending on its composition and its size , we choose to adopt the simplification of assigning a uniform sublimation temperature to all the dust in our calculations . the sublimation radius may be estimated as @xmath16 when the gas distribution surrounding the smbh is not isotropic , @xmath14 may vary with angle . within this radius , electron scattering dominates the opacity , and the usual eddington limit applies . once the intrinsic photons from the accretion disk encounter dust in the surrounding gas , they are absorbed and the energy is re - emitted at infrared wavelengths . if the gas is also optically thick to the infrared , then the re - emitted radiation will continue to be absorbed and re - emitted in a random - walk pattern until it exits the optically thick region . along the way , momentum will be imparted by the photons to the gas multiple times . in this scenario it is possible for the radiation to transfer momentum to the gas at a rate that exceeds @xmath17 . for a spherically symmetric problem , this `` boost '' factor to the infrared radiation force is exactly the infrared optical depth of the gas , which can be shown as follows : in steady - state , when radiative equilibrium holds and the luminosity as a function of radius is constant , we may compute the radiation force per volume @xmath18 as @xmath19 the total outward force exerted by the radiation is @xmath20 where @xmath21 is the radial optical depth for the infrared photons . in a gas rich galactic nucleus with a column density of @xmath22 @xmath23 , a mean mass per particle of 1.5 times the proton mass , and an infrared dust opacity of 10 @xmath24 per gram of gas , an initial guess for the optical depth would be approximately 250 . there are two primary effects that will reduce the actual radiation force from such a high value . the first is the lack of spherical symmetry : a torus obscures only a fraction of the solid angle surrounding the accretion disk , and the presence of clumps and voids in the torus can increase the photon mean free path for certain sightlines . the second effect is dust sublimation : dust will be absent from the innermost regions of the nucleus that contribute a substantial fraction to the gas column density , and the force integral can be well - approximated by setting its lower limit to @xmath14 . to get a sense of the sort of momentum deposition rates that have been observed , consider the case of mrk 231 . this system features an outflow of neutral gas with velocities in the range 360 - 900 km s@xmath7 and a mass - loss rate estimated at 420 solar masses per year @xcite . the momentum flux in the outflow , estimated by multiplying the mass loss rate by the velocity , is between 2.6 to 6.5 times @xmath1 where @xmath12 is measured to be @xmath25 ergs s@xmath7 . the kinetic luminosity of the outflow , on the other hand , is estimated at @xmath26 ergs s@xmath7 , less than 1% of the bolometric agn luminosity . modeling the force from radiation pressure , and predicting by what factor it exceeds @xmath1 , becomes a difficult problem to tackle analytically in the absence of spherical symmetry , the presence of clumps , and with an accounting for dust sublimation . for these reasons , we turn here to three - dimensional radiative transfer calculations using the wavelength - dependent monte carlo radiative transfer code sedona @xcite . given that the radiative diffusion time in these systems is shorter than the dynamical times , we restrict ourselves to steady - state configurations that do not include an explicit coupling to hydrodynamics . in section [ methodologysection ] , we describe how we parametrize the gas configurations surrounding the black hole and how we treat the key physical processes in the radiative transfer . in section [ results ] , we present our results for a series of calculations in which we vary the opening angle of the torus , the amount of gas present , and the accretion disk luminosity . we also examine how our dynamical conclusions are affected by accounting for a clumpy rather than smooth distribution of dust and gas . finally , in section [ conclusion ] we present our conclusions . although the specific region we are studying is difficult to observe directly , gravito - hydrodynamic simulations @xcite provide information about its configuration before the effects of radiative feedback are felt . the gas and stars form a thick disk roughly in vertical hydrostatic equilibrium ( our usage of the word `` disk '' throughout the remainder of this study refers to what is usually labelled as the torus and should not be confused with a reference to the black hole accretion disk , which is unresolved at our scales of interest ) . the puffiness of the disk in the @xcite simulations is to some extent determined by the sub - grids turbulent velocity dispersion when strong stellar feedback in the ism is included , but also by bending modes ( firehose instabilities ) driven by resolved velocities when less stellar feedback is included . while further accretion of the gas at this scale will rely on non - axisymmetric torques , we first adopt a simple axisymmetric , hydrostatic disk model analogous to one used in @xcite . this parametrization captures the key features of the gas configuration seen in the hydrodynamics simulations , but allows us greater control over free parameters and removes unnecessary complications in our attempt to isolate the effects of the radiation . such a parametrization also allows us to systematically introduce clumpiness into the gas for certain calculations ( which , among other effects , breaks axisymmetry ) , as will be described in section [ clumpysetup ] the vertical structure of the smooth disk model may be calculated by solving the equation of hydrostatic equilibrium in the normal ( @xmath27 ) direction , assuming an isothermal equation of state with effective sound speed @xmath28 set by both the resolved and sub - grid velocity dispersion , along with any contribution from the thermal pressure of the gas , @xmath29 with solution @xmath30 \right\ } \ ; .\ ] ] here @xmath31 denotes the gravitational potential , @xmath32 denotes the density of the gas , and @xmath33 is the cylindrical radius . if we assume that the gravitational potential is dominated by the mass of the central black hole @xmath9 at these scales , then the density distribution is @xmath34 \right \}\;.\ ] ] in the limit of small @xmath35 , this yields a gaussian vertical structure . in this limit , the ratio of the squared sound speed to the squared keplerian velocity @xmath36 functions as the ratio of the disk scale height to the cylindrical radius , and for convenience we choose to define a parameter that makes this identification universal : @xmath37 moderately large values of @xmath38 ( @xmath39 ) are suggested by the observed fraction of obscured versus unobscured quasars , although generally this fraction correlates strongly with luminosity @xcite . meanwhile , typical values of @xmath40 found in @xcite range from 0.1 to 0.5 . in this study we will consider @xmath40 in the range 0.1 to 0.35 . mid - ir interferometric observations find the mid - plane density may be well - fit with a power - law @xmath41 where @xmath42 lies within a range of approximately 0.4 to 1.4 , with a tendency toward larger values for more luminous agn @xcite . this is also in agreement with the simulations presented in @xcite in which @xmath43 . we find that the results for varying @xmath42 correlate strongly with the corresponding change in torus mass within the computational domain , and so we choose to capture variations in torus mass and column density by varying the normalization of the radial density profile ( the parameter @xmath44 described below ) , while fixing @xmath42 to @xmath45 for all calculations presented in this paper . converting to spherical polar coordinates @xmath46 and @xmath47 , where @xmath47 is taken to be zero along the z - axis , we obtain @xmath48 \ ; .\ ] ] here @xmath49 represents some inner cut - off radius where the density is @xmath44 , and to prevent the radial column density from diverging we take @xmath50 . one undesirable aspect of this model is that it leads to an accumulation of mass in the polar region of the disk , where @xmath51 is small . to correct for this , we allow the density profile to drop as a power law in the spherical radius @xmath46 rather than in the cylindrical radius @xmath33 . this amounts to dropping the factor of @xmath52 , which is only significant far from the disk mid - plane . this leaves @xmath53 \;,\ ] ] which is quite similar to the phenomenological models of @xcite and @xcite that were used to explain the properties of spectral energy distributions observed in dusty agn . the results from @xcite indicate that @xmath40 does not change by more than a factor of order unity for all @xmath33 . for simplicity , we take @xmath40 to be a constant for all @xmath33 and allow it to vary as a free parameter for different disk models . for all calculations in this study we assume a black hole mass @xmath9 of @xmath3 @xmath2 , and we parametrize the luminosity as a fraction of the electron - scattering eddington luminosity for that mass . as mentioned above , we also vary @xmath44 , which sets the sightline - averaged column density @xmath54 . unless stated otherwise , @xmath54 corresponds to the column density integrated to a distance of @xmath55 parsecs from the central black hole . also , unless @xmath54 is being varied explicitly , @xmath44 is set so that the sightline - averaged column density is @xmath56 @xmath23 , with a mid - plane column density of @xmath57 @xmath23 . these values are consistent with the calculations from @xcite of surface densities of @xmath58 @xmath59 @xmath2 kpc@xmath60 for the central 10 parsecs surrounding the black hole . the fiducial parameters are summarized in table [ fiducialparameters ] . for this smooth density model we use a two - dimensional grid with spherical polar @xmath61 coordinates , with logarithmic spacing in the radial coordinate and linear spacing in the angular coordinate . our resolution is 192 radial zones and 64 @xmath47 zones for @xmath47 ranging from @xmath62 to @xmath63 , with an assumed symmetry for @xmath64 . the radial zones span radii ranging from @xmath55 pc to an outer radius @xmath65 cm ( @xmath66 32.4 pc ) . the 0.1 pc scale was chosen because it is a larger scale than the typical black hole accretion disk , but also small compared to the typical dust sublimation radius . we ignore all momentum deposition inside the 0.1 pc radius , and since nearly all the momentum deposition occurs at and beyond the sublimation radius , the exact choice of innermost radius has little effect on our results . slices of the gas density for the model developed in this section , along with a simulation from @xcite , are shown in figure [ densityslices ] . ccccc 0.3 & @xmath56 & 1.5 & 1 & @xmath3 + [ fiducialparameters ] , except @xmath67 @xmath23 ( chosen to match the simulation in the bottom panel ) . bottom : a density slice taken from a hydrodynamical simulation of gas accretion onto a central black hole ( see @xcite ) . note that the color map is truncated at @xmath68 @xmath23 , and the density in the model distribution continues to drop below this value . , scaledwidth=50.0% ] , except @xmath67 @xmath23 ( chosen to match the simulation in the bottom panel ) . bottom : a density slice taken from a hydrodynamical simulation of gas accretion onto a central black hole ( see @xcite ) . note that the color map is truncated at @xmath68 @xmath23 , and the density in the model distribution continues to drop below this value . , scaledwidth=50.0% ] it has long been predicted on theoretical grounds that the dusty gas surrounding an accreting smbh will not be smoothly distributed , but will instead form clumps @xcite . this prediction has been supported by observations such as the variability of x - ray absorbing column densities in type 2 seyferts @xcite as well as ir spectroscopy @xcite . a vast literature exists concerning radiative transfer through clumpy torus models , with many prescriptions for generating clumpy density distributions from an underlying smooth density model and comparing the results to observations @xcite . our method for generating the clumpy gas distributions most closely resembles those of @xcite and @xcite . we use a three dimensional grid and spherical - polar @xmath69 coordinates , with logarithmic spacing in the radial coordinate and linear spacing in the angular coordinates . our resolution is 128 radial zones , 96 @xmath47 zones for all @xmath47 ranging between @xmath62 and @xmath70 , and 192 @xmath71 zones for all @xmath71 ranging from @xmath62 to @xmath72 . the density of each clump in a given simulation is the same , and a preset number of clumps are placed on the grid . the clump positions are sampled from a probability distribution derived from a smooth density distribution as described in section [ diskgeometry ] . if two clumps overlap in position , their densities are added . each clump s radius is set to a fixed number of grid zones in a given simulation , and the logarithmic radial spacing of the grid causes the size and optical depth of the clumps to grow with increasing distance from the smbh . overlaid on the clumps is a diffuse , smooth background gas distribution that is generated by multiplying the density distribution from section [ diskgeometry ] by 10@xmath60 . an example is pictured in figure [ buildingupsnapshot ] . and the clumping parameters for model 3 in table [ clumpingparameters ] . not pictured is the diffuse background gas . the white cube drawn at the center has a side length of 2 parsecs.,scaledwidth=50.0% ] for each clumpy gas distribution , the total mass in the computational domain was set equal to that of our fiducial smooth density model . the parameters varied between each clumpy gas distribution are the ratio of the clump diameter to the radial distance of the clump from the black hole ( @xmath73 ) , the gas density in the clump ( @xmath74 ) , and the number of clumps in the simulation volume @xmath75 . the choices of @xmath75 and @xmath76 set the average number of clumps per line of sight , which we compute by averaging over all ( @xmath47 , @xmath71 ) sightlines with a weighting to account for the solid angle subtended by each sightline . we have chosen four combinations of clumping parameters to allow the average number of clumps per line of sight to take on values as low as 6 ( in line with the results from @xcite ) to as large as 105 in order to demonstrate a transition to the smooth density models . these parameter choices are listed in table [ clumpingparameters ] . ccccc 1 & .24 & @xmath77 & 36864 & 105 + 2 & .12 & @xmath78 & 36864 & 25 + 3 & .24 & @xmath78 & 4608 & 13 + 4 & .49 & @xmath78 & 576 & 6.5 + [ clumpingparameters ] ultimately we find that it is the volume filling fractions of the clumpy gas models that correlate most strongly with the integrated force exerted by the accretion radiation . if we let @xmath79 denote the ratio of the volume occupied by at least one clump to the total volume within a sphere of radius @xmath46 centered on the black hole , then we find that it can be well approximated via broken power laws . for model 1 , @xmath80 for models 2 through 4 , @xmath81 figure [ columndensityhistograms ] shows the distribution of column density along randomly sampled sightlines for both our smooth and clumpy models . all column density values quoted in this study assume a mean mass per particle of 1.5 times the proton mass . making the gas clumpy leads to a larger number of sightlines with lower column densities compared to the smooth gas distribution , and spreads out the peak on the higher end of the column density distribution . both of these effects are more in line with observational surveys of agn @xcite . at the same time , our clumping prescription tends to make the column density distribution bimodal , with a division between sightlines that intersect no clumps versus those that intersect at least one clump . this bi - modality , which is not present in the observations , persists for all clumping parameters considered in this study , although it can be avoided if a larger fraction of the mass is allocated to the diffuse phase . . middle : clumpy gas column density distribution with same clumping parameters as those used in figure [ buildingupsnapshot ] , also integrated for all gas from 0.1 pc to large radii . bottom : all parameters are the same as the panel above , but this time the column density is integrated from the sublimation radius outward ( i.e. these are the dusty gas columns ) . ] . middle : clumpy gas column density distribution with same clumping parameters as those used in figure [ buildingupsnapshot ] , also integrated for all gas from 0.1 pc to large radii . bottom : all parameters are the same as the panel above , but this time the column density is integrated from the sublimation radius outward ( i.e. these are the dusty gas columns ) . ] . middle : clumpy gas column density distribution with same clumping parameters as those used in figure [ buildingupsnapshot ] , also integrated for all gas from 0.1 pc to large radii . bottom : all parameters are the same as the panel above , but this time the column density is integrated from the sublimation radius outward ( i.e. these are the dusty gas columns ) . ] the monte carlo technique partitions the luminosity of the accreting black hole into equal - energy photon packets that probabilistically interact with the surrounding gas . the packets were transported in three dimensions for all calculations in this study . we improve our statistics by mapping the energy and momentum deposited by the packets into a two - dimensional array of zones a photon that scatters at spherical coordinates ( @xmath46,@xmath71,@xmath47 ) is mapped to position ( @xmath46,@xmath82 ) where @xmath83 if @xmath84 and @xmath85 if @xmath86 . in monte carlo radiative transfer , the specific intensity of the radiation @xmath87 is constructed by counting the number of photon packets with wavelength @xmath88 that enter into each grid zone at position @xmath89 and with direction vector @xmath90 in a given interval of time . specifically , the radiation force per volume @xmath18 at a given position is defined as @xmath91 to compute @xmath18 in a given zone of our computational domain with volume @xmath92 over a time interval @xmath93 , we perform a sum a sum over all photon packets entering the zone . each photon packet carries with it an energy @xmath94 , a direction of travel @xmath95 , and a wavelength @xmath96 . associated with that wavelength is an opacity @xmath97 , measured per gram of gas , and which depends on whether dust is present at location @xmath89 . the packet traverses a path of length @xmath98 within a zone at position @xmath89 . the force is then @xmath99 the radiative acceleration @xmath100 is simply defined as @xmath101 . our calculations apply the stationarity approximation , in which we solve the steady - state radiative transfer problem for a fixed gas density distribution . this approximation is justified if the radiative heating time scale and the radiative diffusion time scale are much shorter than the dynamical time scale . for a sound speed of 200 km s@xmath7 and a characteristic length scale of 10 pc , the dynamical time is approximately @xmath59 seconds . meanwhile , the photon diffusion time through the disk never exceeds @xmath58 seconds , and for many disk parameters the diffusion time is substantially shorter than that . the radiative heating time , estimated by dividing the thermal energy of the gas by the rate of radiative energy deposition , is @xmath102 which is also much shorter than the dynamical time . in this case , the condition of radiative equilibrium allows us to compute the dust temperatures by balancing radiative heating and cooling , @xmath103 for most calculations , the photons are emitted isotropically at the edge of the 0.1 pc sphere surrounding the origin . the effect of anisotropic emission is treated in section [ anisotropicresults ] . we follow the photon propagation for time intervals of @xmath104 seconds , at which point we update the temperature of the dust in each grid zone . we treat dust as present everywhere where the dust temperature is below 1400 kelvin . the dust temperatures are updated until convergence is obtained at the one percent level , which typically takes fewer than 40 iterations if the initial dust temperature is set to 100 kelvin in every zone . finally , for estimating the dynamics of the gas based on the radiation pressure on the dust , we assume perfect hydrodynamical coupling between the dust and the gas , as justified in @xcite . we use the `` intrinsic '' ( unreddened ) agn spectral energy distribution described in @xcite . the majority of the spectral energy is found in the optical and near - uv and originates from the accretion disk , which resembles a @xmath105 kelvin black body emitter . the spectrum also contains a sizable x - ray component . intentionally absent from this spectrum is any infrared component , which we will calculate self - consistently based on the reprocessing of the radiation by dust . [ intrinsicspectrum ] we use tabulated dust opacities and albedos based on @xcite for wavelengths greater than 10 angstroms , and @xcite for shorter wavelengths , all corresponding to visual extinction ratio @xmath106 and assuming a fixed dust - to - gas mass ratio of 1/125 . these values were interpolated between 48 reference wavelengths . in practice , the difference between scattering and absorption is that for an absorption interaction , the wavelength of the re - emitted photon packet will be sampled from a probability distribution that depends on the dust s temperature , whereas the wavelength will remain unchanged for a scattering interaction . for wavelengths less than 100 angstroms , we ignore scattering by dust since it will be almost entirely in the forward direction and hence will not lead to a net transfer of momentum , although we still allow for absorption by dust . electron scattering is only relevant for photons with wavelengths less than @xmath107 angstroms , when the dust absorption cross section drops below that of the thompson cross section , and when the photons are energetic enough to scatter equally well off of both bound and free electrons . we account for anisotropic , inelastic electron scattering in accordance with the klein - nishina formula . figure [ dusttempslices ] shows slices of the equilibrium dust temperature and the radiative acceleration vector field for disks of varying opening angles and with a smooth gas distribution . the color scheme is set so that all temperatures above the dust sublimation temperature appear as solid gray . arrows representing the acceleration are plotted in zones where the dust is not sublimated and where the gas density exceeds @xmath108 g @xmath109 . we find that the dust sublimation region has an aspherical , hour - glass shape . sublimation extends to larger radii in the polar regions where the dusty gas is optically thin in the infrared . there , the dust absorbs ultraviolet radiation but emits in the infrared , forcing it to reach a higher temperature to maintain radiative equilibrium . , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . regions in gray indicate where dust is sublimated ( dust temperature that exceeds 1400 k ) . acceleration arrows are present in zones where the dust is not sublimated and the gas density exceeds @xmath108 g / @xmath110 . the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . ] , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . regions in gray indicate where dust is sublimated ( dust temperature that exceeds 1400 k ) . acceleration arrows are present in zones where the dust is not sublimated and the gas density exceeds @xmath108 g / @xmath110 . the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . ] , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . regions in gray indicate where dust is sublimated ( dust temperature that exceeds 1400 k ) . acceleration arrows are present in zones where the dust is not sublimated and the gas density exceeds @xmath108 g / @xmath110 . the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . ] interestingly , nearly all the angular redistribution of the radiation occurs near the surface of the dust sublimation region . light from the central source initially travels isotropically to the inner edge of the dusty gas , and a large fraction of the photons are absorbed at the dust interface . when photons are re - emitted in the infrared , many are sent back into the sublimation region . it is through this re - emission that the net flux becomes anisotropic at small radii . when infrared photons succeed in penetrating deep into the dusty gas , they generate a nearly radial radiative flux , as they would in a spherically symmetric problem ( see figure [ fluxslice ] ) . . arrows with significant deviation from the radial direction are colored black , while the boundary of the dust sublimation region is marked with black cells . the arrow lengths are proportional to log@xmath111(@xmath114 net flux ( cgs ) ) . through a process of absorption of uv light and re - emission in the ir at the inner wall of the dusty gas , flux is channeled toward the poles in the outermost part of the dust sublimation region . the radiation travels radially in the dusty portion of the gas.,scaledwidth=50.0% ] figure [ radaccelinradiusnormalized ] displays how the radiative acceleration varies with radius and polar angle for the fiducial simulation . the behavior of the acceleration is quite different inside and outside the dust sublimation region the presence of dust raises the opacity of the gas and therefore raises the radiative acceleration ( as in equation [ radiativeacceleration ] ) . in a given solid angle the acceleration is highest just beyond the edge of the dust sublimation region , where ultraviolet and optical photons can push on optically thick , dusty gas . the acceleration rapidly drops as the radiation penetrates farther into the dusty gas and ultraviolet / optical light is converted into infrared , to which the dust is less opaque . for all solid angles , the acceleration settles to a constant ratio above gravity at sufficiently large radius , indicating that the acceleration eventually becomes proportional to @xmath115 , further evidence that the infrared radiation diffuses primarily in the radial direction . in addition to the radial dependence of the acceleration , there is an angular dependence that arises from the diversion of flux from the mid - plane to the polar regions of the disk at the surface of the sublimation region . slices of the net acceleration with gravitational acceleration included are shown in figure [ combinedaccelslices ] . in all cases the acceleration is primarily radial in direction , either outward or inward . note that for opening angles @xmath116 there is a critical polar angle below which radiation dominates over gravity and above which gravity dominates . in these cases inflow may persist in the equatorial region while gas is blown out at angles directed farther away from the mid - plane , potentially leading to a steady state outflow . however , the radiative acceleration dominates over gravity everywhere when @xmath117 for this agn luminosity and column density . , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . inward - directed arrows are colored black , and the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . for small opening angles , the gravitational acceleration exceeds that of the radiation in the equatorial region up to a critical angle above the mid - plane , and beyond this angle radiation dominates . as the opening angle increases , photons deposit more momentum in the dusty gas , and for sufficiently large polar angle the radiation force can exceed gravity in all directions.,title="fig : " ] [ combinedaccelslices : hoverrp1 ] , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . inward - directed arrows are colored black , and the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . for small opening angles , the gravitational acceleration exceeds that of the radiation in the equatorial region up to a critical angle above the mid - plane , and beyond this angle radiation dominates . as the opening angle increases , photons deposit more momentum in the dusty gas , and for sufficiently large polar angle the radiation force can exceed gravity in all directions.,title="fig : " ] [ combinedaccelslices : hoverrp2 ] , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . inward - directed arrows are colored black , and the arrow lengths are proportional to log@xmath111(@xmath112 ) where @xmath113 is in cgs units . for small opening angles , the gravitational acceleration exceeds that of the radiation in the equatorial region up to a critical angle above the mid - plane , and beyond this angle radiation dominates . as the opening angle increases , photons deposit more momentum in the dusty gas , and for sufficiently large polar angle the radiation force can exceed gravity in all directions.,title="fig : " ] [ combinedaccelslices : hoverrp3 ] for another perspective , in figure [ combinedaccelinangle ] we plot the integrated radiative acceleration for columns of gas as a function of polar angle ( without gravitational acceleration included ) . we assume there are no forces in the tangential directions ( i.e. , each column is accelerated independently ) , and that the radiation force is shared along the whole column as the inner gas pushes on outer gas . to compute this net acceleration , we first compute the integrated net force in each solid angle ( including the effects of both gravity and radiation ) , @xmath118 along with the mass in that solid angle , @xmath119 where @xmath14 denotes the edge of the dust sublimation region for each value of @xmath47 . then , the net integrated acceleration is simply @xmath120 \big/ \left[\frac{d m_\mathrm{gas}}{d \omega}\right ] \;.\ ] ] the value of @xmath113 depends on the choice of @xmath121 . however , we will show in section [ mdotscaling ] that the dependence of the rate of mass outflow on @xmath121 is very small . from figure [ combinedaccelinangle ] we see that as the opening angle of the parsec - scale disk becomes smaller , the radiative acceleration becomes more sharply divided between the optically thin and optically thick portions of the disk . this primarily reflects the sharper density gradients present for smaller opening angles . as we will show in section [ effectivetauopeningangles ] , even though the radiative force is greater in the optically thick portion of the disk , the force does not rise as quickly as the mass . this causes the acceleration to decrease toward the mid - plane . , except for opening angles which vary as indicated ( while conserving mass in the calculation domain ) . the acceleration is lowest in the equatorial region , even though the force from radiation pressure is highest there , because the force does not rise as quickly as the mass as the polar angle increases.,scaledwidth=50.0% ] the ability of radiation to clear away ambient gas is enhanced by the fact that diffusing photons deposit their momentum multiple times as they random walk outwards . we can quantify this effect in each solid angle by dividing the integrated force on the gas column in that solid angle by the radiative momentum per time per solid angle leaving the inner source . we call the resulting quantity @xmath122 , which is computed as @xmath123 \big/ \left[\left(\frac{1}{4 \pi } \right ) \frac{l_\mathrm{bh}}{c } \right ] \ ; , \label{taueffdefinition}\ ] ] and we extend the lower limit of the integral defining @xmath124 from @xmath14 to @xmath62 when computing this quantity . we may also compute an average value of this quantity averaged over all lines of sight , @xmath125 figure [ tauopeningangles ] summarizes our results for @xmath126 for various disk opening angles while holding the other parameters at their fiducial values as listed in table [ fiducialparameters ] . increasing the opening angle boosts @xmath122 for all @xmath47 , up to a maximum value of approximately 5 - 6 for these parameters . in the polar region , this effect can be understood simply in terms of the presence of more mass in that region when the opening angle is larger . meanwhile , even though there is less mass present in the equatorial region as the opening angle increases , the radiative flux in that region increases such that @xmath122 is able to increase there as well . if we calculate the radiation force on spherically distributed gas with the same @xmath54 , we find that @xmath126 = 13 . thus , even though the effective radiation force exceeds @xmath1 in figure [ tauopeningangles ] , the enhancement is not as large for a realistic disk geometry as it is in the spherically symmetric case . for the largest opening angle considered in this study ( @xmath127 ) , @xmath128 is smaller than the spherically symmetric value by a factor of @xmath129 . for various opening angles ( see equation [ taueffdefinition ] ) . parameters correspond to the fiducial values in table [ fiducialparameters ] , except for opening angles which vary as indicated ( while conserving the total mass in the calculation domain ) . @xmath130 is a measure of the radiative force on the gas column at a given polar angle , and it reaches its highest values in the equatorial region.,scaledwidth=50.0% ] figure [ tauclumpyspherecomparison ] shows how @xmath131 varies with the clumpiness of the gas . the shape of the momentum deposition as a function of @xmath47 appears generally the same for the clumpy and smooth cases . this suggests that the smooth density distributions employed throughout most of this study provide accurate approximations to the behavior of more realistic clumpy density distributions , although they should systematically overestimate the radiation force on the dusty gas by a factor of @xmath129 in the most extreme cases of clumping ( fewest clumps per line of sight ) considered here . for various clump densities @xmath132 and clump sizes . clump sizes are specified by the ratio of the clump diameter @xmath133 to clump radial position @xmath46 . in all simulations the number of clumps is varied such that the total mass of the gas in the simulations domain is held constant . the diffuse background makes up 1% of the mass in all simulations . the number of clumps per line of sight , listed in the legend , is computed by averaging over all ( @xmath47 , @xmath71 ) sightlines with a weighting to acount for the solid angle subtended by each sightline . , scaledwidth=50.0% ] figure [ forceonclumps ] illustrates how the radiation force acts on portions of individual clumps . note how the force remains radially directed even in the presence of clumps , and how clumps shadow gas behind them . ( cgs).,scaledwidth=50.0% ] if the black hole accretion disk is aligned with the mid - plane of the gas present at the scale of our calculation , one might expect that there would be more flux emitted in the polar directions than in the mid - plane direction . according to one prescription @xcite , the emitted flux should obey @xmath134 where the first factor accounts for projected surface area and the second factor accounts of limb - darkening in an optically thick atmosphere . there is reason to doubt the validity of this model when relativistic effects are taken into account which tend to redirect radiation back toward the mid - plane @xcite . moreover , it remains unclear whether the black hole accretion disk is aligned with the torus . we nevertheless choose to explore the scenario described by equation [ anisotropicemission ] in order to test the sensitivity of our results to variations in the emission pattern of the accretion disk . unlike the case of isotropic emission , we find that for anisotropic emission @xmath122 peaks at an intermediate angle @xmath135 . the peak arises because at small polar angles there is hardly any gas present to provide optical depth , whereas hardly any light penetrates into the dusty gas at large polar angles . for a calculation with our fiducial parameters , the ratio of @xmath128 in the case of anisotropic emission versus @xmath128 for the case of isotropic emission is 0.72 , indicating that photons tend to escape from the disk with fewer interactions when they are emitted in an anisotropic manner . this ratio will be even smaller for smaller disk opening angles . if , instead of being diverted away form the torus plane as prescribed by equation [ anisotropicemission ] , the radiation is beamed toward the plane , @xmath136 would _ increase _ compared to the fiducial simulation . from this point on , we will only consider models with a smooth density distribution and isotropic central emission . nevertheless , it is important to keep in mind that the integrated force and @xmath137 are likely to be modified due to the effects of gas clumping and anisotropic emission of radiation . we can not determine precisely the dynamics of the gas without coupling the radiative transfer calculation to a hydrodynamic solver in a time - dependent calculation . however , we may apply an eddington - type argument to approximate whether gas will be blown away in a given solid angle and to estimate the mass - loss rate . this argument considers the gravitational and radiation forces but ignores centrifugal acceleration of the gas , viscous or gravitational torques , and shocks . the neglect of centrifugal support will not significantly alter the results in the cases when the radiation force on a column of gas is much less than or much greater than the corresponding force of gravity , but it will contribute to an under - estimation of the mass outflow rate in the intermediate range . let @xmath138 denote the time taken to accelerate the gas in a given column with mass @xmath139 to a distance @xmath121 at constant acceleration @xmath140 . then @xmath141 we define a differential mass outflow rate per solid angle @xmath142 , @xmath143 we can also define a mass outflow rate integrated over the entire volume ( all of @xmath47 ) , @xmath144 where we have taken advantage of the assumed symmetry for @xmath64 . whenever @xmath145 is less than zero for a particular value of @xmath47 , we do not add it to the total reported value for the total volume - integrated @xmath137 , since we are only interested in the gas that gets blown away . our gas density prescription ( section [ diskgeometry ] ) indicates that for any given polar angle , the density goes as @xmath146 . this allows us to compute @xmath147 in terms of @xmath121 and the sublimation radius @xmath148 , @xmath149 \ ; . \label{massinsolidangle}\end{aligned}\ ] ] in section [ mdotopeningangleresults ] we will present values for @xmath150 calculated using equations [ mdotdefinition ] , [ massinsolidangle ] , and the values of @xmath151 calculated using the monte carlo . for the rest of this section , we present a simple scaling argument to demonstrate that our estimates of the mass outflow rate depend only weakly on our choice of the outer radius @xmath121 ( which is somewhat arbitrary ) . our results from section [ temperatureandaccelerationresults ] indicate that we can think of the radiative acceleration as being divided into two parts : a spike in acceleration at the sublimation radius that arises from the absorption of ultraviolet and optical photons , and acceleration due to absorption of infrared photons that goes as @xmath152 at large radii . only the second type of acceleration is sensitive to our choice of @xmath121 . we may approximate the infrared radiation force as @xmath153 r^2dr } \nonumber \\ & = \frac{1}{\gamma - 1 } \ , \rho(r_\mathrm{sub})\left [ a_\mathrm{rad}(r_\mathrm{sub } ) - \frac{g \ , m_\mathrm{bh}}{r_\mathrm{sub}^2 } \right ] r_\mathrm{sub}^3 \nonumber \\ & \times \left [ 1 - \left ( \frac{r_\mathrm{sub}}{r_\mathrm{out}}\right)^{\gamma - 1 } \right ] \;,\end{aligned}\ ] ] where @xmath154 refers to the value of the radiative acceleration at the sublimation radius that provides the correct normalization for the inverse - square law acceleration at large radii . using equation [ mdotdefinition ] , dropping factors of order unity , and assuming @xmath155 , we finally arrive at @xmath156^{1/2 } r_\mathrm{sub}^{5/2 } \left(\frac{r_\mathrm{out}}{r_\mathrm{sub}}\right)^{1 - \frac{1}{2}\gamma } \;.\end{aligned}\ ] ] from the simulations performed in @xcite , @xmath42 tends to fall between 1.5 and 2 , and as already noted we have fixed @xmath42 at 1.5 for all numerical calculations in this study . we expect that the density profile will ultimately truncate at about 1 kpc . so , the ratio of the @xmath142 due to absorption of infrared photons that we would calculate using @xmath121 of 1 kpc versus @xmath121 of 32.4 pc would be , for @xmath157 , only 2.4 . for @xmath158 , @xmath142 would be invariant with respect to choice of @xmath121 , and for @xmath159 the ratio would be 0.42 . the fact that a significant portion of the radiative acceleration in the monte carlo calculations comes from the spike near the dust sublimation radius further reduces the sensitivity of our results to our choice of @xmath121 . a final quantity that will be useful to us is the velocity of the gas in a solid angle @xmath160 , @xmath161 once again focusing on the infrared acceleration at large radii and making the same approximations as we did for estimating @xmath142 , we find @xmath162 r_\mathrm{sub } } \left(\frac{r_\mathrm{out}}{r_\mathrm{sub}}\right)^{\frac{1}{2}\gamma - 1 } \;. \label{fullvoutresult}\end{aligned}\ ] ] therefore our calculations for the velocity of the gas will have a similarly weak dependence on @xmath121 as that of the mass - loss rate . figure [ mdotopeningangles ] shows the differential mass outflow rate @xmath145 calculated using equation [ mdotdefinition ] for disks with various opening angles . the densities of the innermost radial grid zones have been re - scaled to keep the total mass in the calculation domain constant in each case . for various opening angles . parameters correspond to the fiducial values in table [ fiducialparameters ] , except for opening angles which vary as indicated ( while conserving the total mass in the calculation domain ) . , scaledwidth=50.0% ] the overall mass outflow rate declines with smaller @xmath40 due to the increased funneling of radiation into the low - density polar regions . for @xmath163 the differential mass outflow rate peaks at @xmath164 , while for @xmath165 the peak is at an intermediate polar angle . this is due to an interplay between the amount of mass available to be cleared away , its inertia , and the gravitational force acting upon it . near the equator , however , the large amount of gas can not be unbound by the radiative acceleration and so there is not outflow , even though the force due to radiation is strongest there . we emphasize that figure [ mdotopeningangles ] represents only a snapshot in time of the mass outflow rate for an accreting smbh . the evolution of @xmath166 with time is not calculated here and requires a fully coupled radiation - hydrodynamics calculation . depending on the resulting rearrangement of the gas , the long - term mass loss rate could conceivably be either larger or smaller than the rate calculated here . one possible scenario is that an initially optically thick and puffy disk will blow away gas in the polar region . in the absence of a replenishing mechanism that operates on a timescale shorter than @xmath167 , this might cause the disk to become thinner , reducing the tendency for radiation to blow out more gas . alternatively , the removal of gas from above the near - midplane could change the geometry of the flux sufficiently so as to induce a stronger vertical component , drawing up more gas from the midplane and leading to yet more mass loss . yet another possibility , already suggested in section [ temperatureandaccelerationresults ] , is that a steady - state inflow / outflow develops with a mass loss rate that hovers close to the instantaneous value computed here . figure [ mdotcolumndensities ] shows the differential mass outflow rate @xmath145 for disks with varying column densities ( the column density is averaged over all lines of sight , and includes both dusty and non - dusty gas ) . the variation in column density is directly proportional to variation in the total mass present in the calculation domain . for various sightline - averaged column densities ( measured to 0.1 pc from the black hole ) . parameters correspond to the fiducial values in table [ fiducialparameters ] , except for the average column densities which vary as indicated . the total mass in the calculation domain varies proportionally with the average column density.,scaledwidth=50.0% ] as expected , more mass can be ejected when there is more mass present to begin with . however , we find the scaling to be sub - linear : @xmath168 for the range of column densities included in this study ( power - law scaling relations for all the free parameters in the problem will be summarized in section [ summaryscalings ] ) . the slow growth of @xmath137 with column density is due to the fact that at higher column densities , the radiative force on the gas in the densest portions of the disk does not rise as quickly as the mass present there , and so gravity becomes increasingly effective at limiting the outflow rate . finally , figure [ mdotluminosities ] shows the differential mass outflow rate @xmath145 for calculations with varying black hole luminosities . for higher agn luminosities , not only is there a higher net force on a column at a given value of @xmath47 for which the net force was already outward ( positive ) , but also the net force becomes positive on columns at larger polar angles . for all other parameters held constant , there exists a critical luminosity at which the radiation force exceeds gravity for all polar angles , and all the gas would be blown away . for opening angle @xmath40 = 0.3 and our fiducial mean column density @xmath169 @xmath23 , radial density power - law @xmath157 and black hole mass @xmath170 @xmath2 , this critical luminosity is @xmath171 . the existence of such a critical luminosity might help to explain the dearth of quasars observed to be radiating at the full value of their inferred eddington limit , although the precise value of the limiting luminosity presented here should be considered a rough estimate , and may vary with time as the gas rearranges itself following the initial outflow we have estimated . for various luminosities . parameters correspond to the fiducial values in table [ fiducialparameters ] , except for the luminosities which vary as indicated . , scaledwidth=50.0% ] the scalings of @xmath128 with the parameters of the problem , varied one at a time from the fiducial values listed , for a black hole with mass @xmath3 @xmath2 , can be summarized as follows : @xmath172 for these fits , six data points were used for @xmath40 spanning 0.1 to 0.35 , five data points were used for @xmath12 spanning 0.03 to 1 , and four data points were used for @xmath54 spanning @xmath173 to @xmath174 @xmath23 . all of these calculations used @xmath157 , and the results for changing @xmath42 generally correlate with the results for the corresponding change in @xmath54 . we may also present a summary scaling relation for the volume - integrated mass outflow rate @xmath137 calculated over the same range of parameters : @xmath175 note that @xmath54 in these scaling relations corresponds to column densities integrated from 0.1 pc to large radii . if instead we use the column density integrated from the edge of the dust sublimation radius outward , then the fiducial column density becomes @xmath176 @xmath23 , the column density power - law in equation [ effectivetauscalingresults ] changes to @xmath177 , and the column density power - law in equation [ mdotscalingresults ] changes to @xmath178 . also note that the rates in equation [ mdotscalingresults ] correspond to a radius of 32 parsecs from the central smbh , and there is a weak dependence on radius ( no stronger than @xmath179 when @xmath157 ; refer to section [ mdotscaling ] . ) the fiducial value for the mass outflow rate of 144 @xmath2 per year may seem surprisingly large . that value was computed for a black hole radiating at its full eddington luminosity , and at that luminosity the radiative acceleration beats out gravity at all solid angles . therefore , there is reason to suspect that such a large outflow rate is not sustainable for many gas dynamical times at the parsec scale , as the amount of mass present and the opening angle of the disk readjust during the outflow . the quoted outflow rate also does not incorporate the effects of making the gas distribution clumpy , and as was demonstrated in section [ clumpyresults ] , this reduces the integrated force by a factor of @xmath129 for significant clumping . since the mass loss rate should roughly scale as the integrated force to the @xmath180 power ( as argued in section [ mdotscaling ] ) , the mass outflow rate will be reduced approximately by a factor of @xmath181 in the case of significant clumping . on the other hand , the mass - loss rates calculated above correspond to the mass swept up within a radius of approximately 32 parsecs from the central black hole . extrapolating our results out to 1 kpc , as discussed in section [ mdotscalingresults ] , might boost the outflow rates by roughly a factor of 2 for @xmath157 . with those caveats in mind , the important conclusions to be drawn from equations [ effectivetauscalingresults ] and [ mdotscalingresults ] are that the radiation force may reach several times ( @xmath182 ) @xmath183 , and that the mass outflow rates can easily reach tens of solar masses per year for parameters close to our fiducial values . in the proper circumstances ( @xmath184 @xmath23 , @xmath185 , and @xmath186 ) , the mass outflow rates can reach up to 100 @xmath2 per year . it is also interesting to note that the momentum enhancement and mass outflow rate depend relatively strongly on the agn luminosity and disk opening angle @xmath40 . the fact that the dependence of the mass outflow rate on luminosity is steeper than the dependence of the radiation force on luminosity may at first seem surprising . this scaling has its origins in an effect noted in section [ massoutflowotherparameters ] , specifically that increasing the luminosity allows the radiation force to exceed gravity for a larger fraction of the solid angle , adding more mass to the outflow than was present at lower luminosities . finally , to drive home the point that momentum deposition , not heating , is responsible for the large computed outflow rates , we can estimate the corresponding kinetic luminosities . we use our estimate of the gas velocity as a function of solid angle @xmath160 ( equation [ voutdefinition ] ) to compute the fraction of the accretion luminosity @xmath12 that goes into kinetic luminosity for the same parameter range used above : @xmath187\,\left [ \int{\frac{1}{2}\ , \frac{d\dot{m}}{d\omega}(\theta ) \ , v_\mathrm{out}^2(\theta)\,d\omega } \right ] \nonumber \\ & = \left[\frac{2(2 \pi)}{l}\right]\,\left [ \int_0^{\pi/2}{\frac{1}{2}\ , \frac{d\dot{m}}{d\omega}(\theta ) \ , v_\mathrm{out}^2(\theta)\,\sin \theta \ , d\theta } \right ] \ ; .\end{aligned}\ ] ] once again by varying each parameter one at a time with respect to the fiducial values , our results for a @xmath188 solar mass black hole can be summarized as @xmath189 if the column density is computed by integrating from the dust sublimation radius outward , the corresponding power - law in equation [ kineticlumscaling ] barely changes at all , increasing to 0.21 . the mass - weighted average velocity of the gas in the outflow will be approximately equal to @xmath190 , although a more accurate value can be obtained by integrating @xmath160 weighted by @xmath191 and only counting contributions from solid angles and radii for which gas can be blown out . for our fiducial parameters this average velocity at 32 parsecs from of the computational domain is approximately 1000 km s@xmath7 . it is important to compare and contrast the results of our calculation to previous studies that addressed the same physical problem , including @xcite , @xcite and @xcite . the first of those studies included a calculation of steady - state , multi - wavelength radiative transfer through a static dusty torus , but for a different torus geometry and slightly different boundary conditions for the radiative transfer than those used here . by modifying our density distribution to match that of @xcite , and preventing dust from sublimating within the pre - determined boundaries of the torus , we find results that are in qualitative agreement with theirs , and quantitatively the values for the components of the radiation force each agree to within 32% in the central region of the torus ( our computed radiation forces are smaller ) . the primary difference between the studies comes in the range of luminosities considered . @xcite point out that for a large range of torus parameters , if @xmath192 the radiation force will overwhelm gravity and lead to an outflow , possibly halting accretion . this conclusion is consistent with our results , although we have sought to quantify the rate of mass outflow and have considered the possibility of simultaneous inflow and outflow processes . the work of @xcite and @xcite generate self - consistent density distributions for a torus that is dynamically influenced by radiation pressure , by assuming dynamical equilibrium in the first case and with a numerical radiation - hydrodynamics solver in the second and third . again , these studies find that disruptive outflows should occur when @xmath192 for compton - thick torii , consistent with the present work . furthermore , we find encouraging quantitative agreement with @xcite when we use a gray radiative opacity of @xmath193 @xmath194 per gram of gas to match theirs . for a @xmath195 black hole with accretion luminosity of @xmath196 erg s@xmath7 surrounded by a torus of mass @xmath197 enclosed within a radius of 3 parsecs , we find @xmath137 using equations [ mdotdefinition ] and [ integratedmdotdefinition ] to be 4.6 @xmath2 per year , in agreement with their value of 5 @xmath2 per year . if we extend the estimation of @xmath137 to include gas out to a radius of 32 parsecs , then our value of @xmath137 rises to 27 @xmath2 per year . interestingly , when we perform the same calculation with wavelength - dependent radiative opacity , this value drops slightly to 25 @xmath2 per year . this decrease is due in part to a larger region of dust sublimation in the wavelength dependent case . to summarize , our work is consistent with previous studies but seeks to quantify the mass lost in outflows driven by highly luminous ( @xmath198 ) accretion events , accounting for mass driven away at large radii ( @xmath199 parsecs ) . by integrating the momentum deposited in columns of gas , we can make a statement of how the outflows can affect the host galaxy . we have calculated how radiation pressure from a luminous accretion disk around a smbh drives a powerful outflow of gas via continuum radiation pressure on dust at distances of 0.1 - 30 pc from the black hole . using ambient gas conditions motivated by observational constraints on nuclear obscuration in agn ( @xmath185 , @xmath184 @xmath23 ) we find that a @xmath200 smbh radiating at eddington can drive a wind with velocities of @xmath201 s of km s@xmath7 and an instantaneous mass loss rate of @xmath4 10 - 100 @xmath2 per year ( see equation [ mdotscalingresults ] ) . for smbhs with masses @xmath202 , the outflow rates could approach @xmath203 per year . radiative heating sublimates the dust out to distances of roughly 0.5 to 1 pc in the mid - plane , and radiation pressure drives away the gas and dust in the polar regions , leaving behind what may constitute the observed dusty torus . the wide - angle bipolarity of these outflows corresponds well to observations of obscured quasars @xcite and seyfert 2s @xcite . although the radiative acceleration is greatest in the polar regions , the majority of the ejected mass comes from oblique angles where there is a more significant reservoir of gas . by contrast , gas in the equatorial plane is more difficult to unbind because of its large inertia and large integrated gravitational attraction . the net momentum flux in the resulting outflow can exceed @xmath0 by factors of up to 5 for the parameters studied , as infrared photons interact multiple times during their outward diffusion . as recently demonstrated in the calculations of @xcite , outflows with these properties have a significant impact on gas in the surrounding host galaxy . our results for the outflows match reasonably well the observed outflows in local ulirgs such as mrk 231 @xcite . the mass - loss rates and kinetic luminosity fractions we calculate also provide a reasonable match to observations of obscured quasars @xcite , although our model does not provide a mechanism for launching large amounts of gas at the high velocities ( @xmath204 km s@xmath7 ) observed in these systems at small radii . one possibility is that these quasars are exhibiting both line and continuum radiation pressure driven outflows . we find that the net effect of the agn radiation on the surrounding gas is a strong function of the opening angle of the accreting gas at the parsec - scale ( the torus ) . increasing the opening angle allows more momentum to be deposited in all directions because the mass distribution and emergent radiative flux become more isotropic . we also find a steep dependence of the outflow rate on the luminosity of the accretion disk , because at higher luminosities gas becomes unbound over a greater range of solid angles . this result is also in agreement with the observed anti - correlation between obscured agn fraction and agn luminosity @xcite , although we are restricting our attention to a single black hole mass . keeping all of our parameters at the fiducial values listed in table [ fiducialparameters ] but varying the luminosity , we find that outward radiative acceleration begins to exceed gravity at all angles once @xmath205 reaches a value of @xmath206 . this value is subject to uncertainty given our approximate treatment of the gas dynamics , but it may nevertheless help to aid understanding of the relative dearth of broad - line quasars observed to be radiating at their full eddington luminosity @xcite . the effects of dust sublimation play a crucial rule in determining the angular dependence of the radiative force on the torus . the redistribution of flux between polar angles takes place almost entirely in the region of gas in which dust has been sublimated , where infrared radiation is re - emitted from the edge of the dusty gas at angles deviating from the radial direction . once photons penetrate into the dusty gas , they tend to diffuse radially and deposit momentum almost entirely in the radial direction . all of the results presented above must be considered in light of the approximations and assumptions that we have used . in particular , while allowing for a range of torus scale heights and masses , we have focused on an initial distribution of gas close to hydrostatic equilibrium , without accounting for the self - consistent dynamical rearrangement of the torus as the outflow takes place , perhaps accompanied by inflow when possible . we have also discounted centrifugal support of the torus , which may lead to an underestimate of the mass loss rate when the integrated radiation force is close to that of gravity . another effect we do not capture in these calculations is the potential for the outflowing gas to develop radiative rayleigh - taylor instabilities , which might provide more avenues of photon leakage and reduce the coupling of the radiative momentum to the gas as studied by @xcite . concerning this last point , our clumpy gas simulations can provide a preliminary estimate of the extent to which the radiation force would be reduced in the presence of such instabilities . also , due to the high opacity encountered by the direct ultraviolet radiation from the accretion disk , the direct radiation field plays a much more important role in launching the gas in our calculation than in @xcite . as demonstrated in @xcite , the radiative rayleigh - taylor instability can be suppressed when the direct radiation field is sufficiently strong . a fully coupled radiation - hydrodynamic calculation will be needed to fully understand the subsequent behavior of the gas in time . future work will focus on incorporating the results of this study into hydrodynamic simulations of black hole accretion . in addition to the coupling to the hydrodynamics , more details pertinent to the radiative physics may be addressed in such calculations , including line absorption , anisotropic scattering off of dust , metallicity gradients , and variations in the average dust - to - gas ratio . we thank jason dexter , claude - andr faucher - gigure , and nahum arav for helpful conversations , along with the referee for thoughtful recommendations . nr is supported by the department of energy office of science graduate fellowship program ( doe scgf ) , made possible in part by the american recovery and reinvestment act of 2009 , administered by orise - orau under contract no . de - ac05 - 06or23100 . eq was supported in part by the david and lucile packard foundation . support for pfh was provided by nasa through einstein postdoctoral fellowship award number pf1 - 120083 issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 03060 . this work is supported by the director , office of energy research , office of high energy and nuclear physics , and divisions of nuclear physics , of the u.s . department of energy under contract no . de - ac02 - 05ch11231 . this research used resources of the national energy research scientific computing center , which is supported by the office of science of the u.s . department of energy under contract no . de - ac02 - 05ch11231 .

-decay and positron decay are believed to play a consequential role during the late phases of stellar evolution of a massive star culminating in a supernova explosion . the-decay contributes in maintaining a respectable lepton - to - baryon ratio , , of the core prior to collapse which results in a larger shock energy to produce the explosion . the positron decay acts in the opposite direction and tends to decrease the ratio . the structure of the presupernova star is altered both by the changes in and the entropy of the core material . recently the microscopic calculation of weak - interaction mediated rates on key isotopes of iron was introduced using the proton - neutron quasiparticle random phase approximation ( pn - qrpa ) theory with improved model parameters . here i discuss in detail the improved calculation of decay rates for iron isotopes (fe ) in stellar environment . the pn - qrpa theory allows a microscopic `` state - by - state '' calculation of stellar rates as explained later in text . excited state gamow - teller distributions are much different from ground state and a microscopic calculation of decay rates from these excited states greatly increases the reliability of the total decay rate calculation specially during the late stages of stellar evolution . the reported decay rates are also compared with earlier calculations . the positron decay rates are in reasonable agreement with the large - scale shell model calculation . the main finding of this work includes that the stellar-decay rates offe are around 3 5 orders of magnitude smaller than previously assumed and hence irrelevant for the determination of the evolution of during the presupernova phase of massive stars . the current work discourages the inclusion offe in the list of key stellar-decay nuclei as suggested by former simulation results . beta decay , gt strength distributions , pn - qrpa , core - collapse supernovas , massive stars 21.60.jz , 23.40.-s , 26.50.+x , 97.10.cv = 0.5 cm

1408.4407

the classical and pioneering works in the field of stellar physics include dynamics of supernova explosion by baade @xmath5 zwicky @xcite , energy production in stars and stellar evolution by bethe @xcite , and , synthesis of elements in stars by burbidge et al . since then the microphysics of supernova explosion has come a long way . whereas the strong interactions ( fusion reactions ) are responsible for providing the fuel to the stars empowering them throughout their life cycles , it is precisely the weak interactions that play a decisive role in determining both the presupernova stellar structure and the nucleosynthesis . the weak interaction reactions lead to the initiation of the gravitational collapse of the core of a massive star triggering a supernova explosion , control the lepton - to - baryon fraction of the core throughout the course of stellar evolution , play a key role in neutronisation of the core material , and , affect the formation of heavy elements above iron via the r - process at the final stage of the supernova explosion . much to the advantage of the astrophysicists , the temperature during the late phases of stellar evolution is high enough for the matter composition to be given by nuclear statistical equilibrium . this means that one can get away with the need of reaction networks for the strong and electromagnetic interactions and the composition of the matter is given by the saha equation . weak interactions in presupernova stars are known to be dominated by allowed fermi ( vector - type ) and gamow - teller ( axial - vector type ) transitions . the calculation of weak - interaction rates is very sensitive to the distribution of the gt@xmath6 strength function . in the gt@xmath7 strength a proton is changed into a neutron whereas the gt@xmath8 strength is responsible for transforming a neutron into a proton . it was fuller , fowler , and newman ( ffn ) @xcite who first performed an extensive calculation of stellar weak rates including the capture rates , decay rates , neutrino energy loss rates and gamma heating rates for a wide density and temperature domain . they performed their detailed calculation for 226 nuclei in the mass range @xmath9 . the authors recognized the key role played by the gt giant resonance and noted that measured decay rates exploited only a small fraction of the total available strength . the centroids of the gt@xmath6 distribution functions determine the effective energy of the capture and decay reactions . ffn estimated the gt contributions to the rates by a parametrization based on the independent particle model . aufderheide et al . @xcite later extended the ffn work for heavier nuclei with a @xmath10 60 and took into consideration the quenching of the gt strength neglected by ffn . authors in ref . @xcite also stressed on the importance of @xmath0-decay rates in the iron core prior to the collapse . they tabulated the 71 top @xmath0-decay nuclei averaged throughout the stellar trajectory for @xmath11 ( see table 26 of ref . the measured data from various @xmath12 and @xmath13 experiments later revealed the misplacement of the gt centroids adopted in the parameterizations of ffn and subsequently used in the calculation of weak rates by ref . since then theoretical efforts were concentrated on the microscopic calculations of weak - interaction mediated rates of iron - regime nuclide . two such widely used models are the large - scale shell model ( lssm)(e.g . @xcite ) and the proton - neutron quasiparticle random phase approximation ( pn - qrpa ) theory ( e.g. @xcite ) . the pn - qrpa theory is an efficient way to generate gt strength distributions which constitute a primary and nontrivial contribution to the weak - interaction mediated rates among iron - regime nuclide . the shell model calculations perform a detailed study of the nuclear spectroscopy , however , the pn - qrpa model has two important advantages . it can handle any arbitrarily heavy system of nucleons since the calculation is performed in a luxurious model space of up to 7 major oscillator shells . the second advantage is even more important for the calculation of weak - interaction rates in stellar matter . because of the high temperatures prevailing during the presupernova and supernova phases of a massive star , there is a reasonable probability of occupation of parent excited states and the total weak interaction rates have a finite contribution form these excited states . thus , in calculating a stellar rate , one must know the gt strength distributions of the excited states of the parent nucleus . as experimental information about excited state strength functions seems inaccessible , aufderheide @xcite stressed much earlier the need to probe these strength functions theoretically . the pn - qrpa model calculates the gt strength distribution strengths of _ all _ excited states of parent nucleus in a microscopic fashion and this feature of the pn - qrpa model greatly enhances the reliability of the calculated rates in stellar matter . these excited states are like resonances having finite band width and contributions from many discrete states calculated microscopically within the pn - qrpa framework . the construction of these excited states and calculation of the relevant nuclear matrix elements will be shown in the next section . in this sense the pn - qrpa model allows a fully microscopic `` state - by - state '' calculation of stellar weak rates . the calculation of stellar weak rates on iron - regime nuclei are sensitive to both the weak low - energy and strong high - energy ground and parent excited state gt strength distributions . the low energy part is more important during the earlier phase and the high energy part becomes important during the late phases of stellar evolution @xcite . previous pn - qrpa calculations have shown that for certain nuclei the excited state rates can command the total weak rate ( e.g. @xcite ) . this clearly endorses the contribution of excited states in the calculation of total weak rates . other calculations revert to approximations like the so - called brink s hypothesis ( in the electron capture direction ) and back - resonances ( in the @xmath0-decay direction ) . brink s hypothesis states that gt strength distribution on excited states is _ identical _ to that from ground state , shifted _ only _ by the excitation energy of the state . gt back resonances are the states reached by the strong gt transitions in the inverse process ( electron capture ) built on ground and excited states . nabi and klapdor - kleingrothaus @xcite first reported the calculation of weak interaction rates for 709 nuclei with a = 18 to 100 in stellar environment using the pn - qrpa theory . these included capture rates , decay rates , gamma heating rates , neutrino energy loss rates , probabilities of beta - delayed particle emissions and energy rate of these particle emissions . the authors then presented a detailed calculation of stellar weak interaction rates over a wide range of temperature and density scale for sd- @xcite and fp / fpg - shell nuclei @xcite . these also included the weak interaction rates for nuclei with a = 40 to 44 ( not yet calculated by shell model ) . since then these calculations were further refined with use of more efficient algorithms , computing power , incorporation of latest data from mass compilations and experimental values , and fine - tuning of model parameters both in the sd- shell @xcite and fp - shell @xcite region . all theoretical calculations of stellar weak interactions have inherent uncertainties . the uncertainties associated with the pn - qrpa model were discussed in detail in ref . the reliability of the pn - qrpa calculation was discussed in length by nabi and klapdor - kleingrothaus @xcite . there the authors compared the measured data ( half lives and b(gt@xmath6 ) strength ) of thousands of nuclide with the pn - qrpa calculation and got fairly good comparison . earlier half - lives of @xmath14 decays were calculated systematically for about 6000 neutron - rich nuclei between the beta stability line and the neutron drip line using the pn - qrpa model @xcite . similarly half - lives for @xmath15/ec ( electron capture ) decays for neutron - deficient nuclei with atomic numbers z = 10 - 108 were calculated up to the proton drip line for more than 2000 nuclei using the same model @xcite . these microscopic calculations gave a remarkably good agreement with the then existing experimental data ( within a factor of two for more than 90@xmath16 ( 73@xmath16 ) of nuclei with experimental half - lives shorter than 1 s for @xmath14 ( @xmath15/ec ) decays ) . most nuclei of interest of astrophysical importance are the ones far from stability and one has to rely on theoretical models to estimate their beta decay properties . the accuracy of the pn - qrpa model increases with increasing distance from the @xmath0-stability line ( shorter half - lives ) @xcite . this is a promising feature with respect to the prediction of experimentally unknown half - lives ( specially those present in the stellar interior ) , implying that the predictions are made on the basis of a realistic physical model . the isotopes of iron , @xmath3fe , are mainly responsible for decreasing the electron - to - baryon ratio during the oxygen and silicon burning phases of massive stars through electron capture and positron decay processes . the electron captures on these iron isotopes are the dominant process . nevertheless , the @xmath0-decay rates for these isotopes of iron are also argued to be relevant during the presupernova evolution of massive stars in literature . because of their astrophysical importance @xmath4fe were included in the list of key @xmath0-decay nuclei that have a significant impact on the presupernova evolution of massive stars after core silicon burning for @xmath17 compiled by aufderheide and collaborators ( see tables 19 , 20 and 26 of ref . @xcite ) . later heger and collaborators @xcite studied the presupernova evolution of massive stars ( of masses @xmath18 and @xmath19 ) and found @xmath20fe in the list of top five nuclei that increase @xmath1 via @xmath0-decay and positron capture during the silicon burning phases of these massive stars . the authors employed the lssm @xmath0-decay rates in their simulation code . the calculations of gt@xmath6 strength distributions and stellar weak rates for these isotopes of iron were introduced earlier using the pn - qrpa theory with improved model parameters @xcite . there the author was able to reproduce fairly well the experimental centroids and the total strength distributions in both directions for the even - even iron isotopes , @xmath21fe where measurements were available . in this paper i discuss in detail the calculation of @xmath2 decay rates of @xmath3fe in stellar environment and its astrophysical implications . the main finding of this work is that @xmath0-decay rates on @xmath3fe are around 3 5 orders of magnitude smaller than previously assumed and are irrelevant for the determination of the evolution of @xmath1 during the presupernova phases of massive stars . in the following section i briefly describe the theoretical formalism used to calculate the stellar electron and positron decay rates . the stellar @xmath2-decay rates of @xmath3fe are presented in section 3 . here i also compare the pn - qrpa decay rates with earlier calculations . summary and conclusions are finally presented in section 4 . the calculation may be divided into three main steps . a bogoliubov transformation was used to define quasiparticle states in terms of nucleon states and then the rpa equation was solved in the basis of proton - neutron quasiparticle pairs . the last step was to calculate the decay rates in stellar interior . the hamiltonian for the calculation was of the form @xmath22 the first step was approximated by a nilsson + bcs calculation . single - particle energies and wave functions were calculated in the nilsson model , which takes into account nuclear deformation @xcite . pairing was treated in the bcs approximation . in the second step regarding the rpa calculation , both particle - hole ( ph ) and particle - particle ( pp ) gt forces were employed . the gt forces are appropriate for calculation of gt strength functions , since the transition amplitudes are closely connected to the forces . the particle - particle interaction , first considered by cha @xcite , has usually been neglected in @xmath14 decay ( e.g. @xcite ) but is of decisive importance in @xmath15 decay ( e.g. @xcite ) . both the particle - hole and particle - particle interaction can be given a separable form . the interactions were characterized by two interaction constants : @xmath23 ( for particle - hole interaction ) and @xmath24 ( for particle - particle interaction ) . in this work , the values of @xmath23 and @xmath24 were taken as 0.15 mev and 0.07 mev , respectively . the value of the strength parameters was determined by a fit to the measured data ( gt strengths and centroids ) available for the isotopic chain . other parameters required for the calculation of weak rates are the pairing gaps , the nuclear deformations and the q - value of the nuclear reactions . i applied the traditional choice of @xmath25 in this project @xcite . the deformation parameter was recently argued as an important parameter for qrpa calculations at par with the pairing parameter by stetcu and johnson @xcite . as such rather than using deformations from some theoretical mass model ( as used in earlier calculations of pn - qrpa weak rates @xcite ) the experimentally adopted values of the deformation parameters for @xmath21fe , extracted by relating the measured energy of the first @xmath26 excited state with the quadrupole deformation , were taken from raman et al . for the case of @xmath20fe ( where such measurement lacks ) the deformation of the nucleus was calculated as @xmath27 where @xmath28 and @xmath29 are the atomic and mass numbers , respectively , and @xmath30 is the electric quadrupole moment taken from ref . q - values were taken from the recent mass compilation of audi et al . @xcite . the incorporation of measured deformations for @xmath21fe and a smart choice of strength parameters led to an improvement of the calculated gt@xmath6 distributions compared to the measured ones . excited state calculation of gt@xmath6 strengths was performed for a total of 246 parent states in @xmath31fe , 297 states in @xmath20fe and 266 states in @xmath32fe , covering excitation energies in the vicinity of 15 mev ( as explained earlier ) . for each parent excited state , transitions were calculated to 150 daughter excited states . daughter excitation energies up to around 20 mev were considered in the calculation . the band widths of energy states were chosen according to the density of states of the concerned nucleus . in this way contribution from all excited states was incorporated in the calculation . a large model space assists in reproducing low - lying spectrum and higher excitations @xcite . the use of a separable interaction assisted in the incorporation of a luxurious model space of up to 7 major oscillator shells which in turn made possible to consider these many excited states in both parent and daughter nuclei . in order to further increase the reliability of the calculated rates experimental data were incorporated in the calculation wherever possible . the calculated excitation energies were replaced with measured levels when they were within 0.5 mev of each other . missing measured states were inserted and inverse transitions ( along with their log@xmath33 values ) were also taken into account . no theoretical levels were replaced with the experimental ones beyond the excitation energy for which experimental compilations had no definite spin and/or parity . the beta decay ( bd ) and positron decay ( pd ) rates of a transition from the @xmath34 state of the parent to the @xmath35 state of the daughter nucleus are given by @xmath36\left[b(f)_{ij } + \left({\raise0.7ex\hbox{$ g_{a } $ } \!\mathord{\left/ { \vphantom { g_{a } g_{v } } } \right . \kern-\nulldelimiterspace}\!\lower0.7ex\hbox{$ g_{v } $ } } \right)^{2}_{eff } b(gt)_{ij } \right]\left[f_{ij}^{bd(pd ) } ( t,\rho , e_{f } ) \right ] . \label{phase space}\ ] ] the value of d was taken to be 6295s @xcite . b(f ) and b(gt ) are reduced transition probabilities of the fermi and gamow - teller ( gt ) transitions , respectively , @xmath37 @xmath38 here @xmath39 is the spin operator and @xmath40 stands for the isospin raising and lowering operator . details of the calculation of reduced transition probabilities can be found in ref . . the effective ratio of axial and vector coupling constants , @xmath41 , which takes into account the observed quenching of the gt strength was taken to be ( from ref . @xcite ) : @xmath42 with @xmath43 = -1.254 @xcite . interestingly , vetterli and collaborators @xcite and rnnqvist et al . @xcite predicted the same quenching factor of 0.6 for the rpa calculation in the case of @xmath31fe when comparing their measured strengths to rpa calculation . the @xmath44 are the phase space integrals and are functions of stellar temperature ( @xmath45 ) , density ( @xmath46 ) and fermi energy ( @xmath47 ) of the electrons . they are explicitly given by @xmath48 and by @xmath49 in eqs . ( [ bd ] ) and ( [ pd ] ) , @xmath50 is the total energy of the electron including its rest mass . @xmath51 is the total @xmath0-decay energy , @xmath52 where @xmath53 and @xmath54 are masses and excitation energies of the parent nucleus , and @xmath55 and @xmath56 of the daughter nucleus , respectively . z , w ) are the fermi functions and were calculated according to the procedure adopted by gove and martin @xcite . g@xmath6 is the fermi - dirac distribution function for positrons ( electrons ) . @xmath58^{-1},\ ] ] @xmath59^{-1},\ ] ] here @xmath60 is the kinetic energy of the electrons and @xmath61 is the boltzmann constant . the total decay rate per unit time per nucleus is finally given by @xmath62 where @xmath63 is the probability of occupation of parent excited states and follows the normal boltzmann distribution . after the calculation of all partial rates for the transition @xmath64 the summation was carried out over all initial and final states until satisfactory convergence was achieved in the rate calculation . stellar decay rates are fragile functions of the available phase space , ( @xmath65 ) . under terrestrial conditions , @xmath0-decay of @xmath3fe is not possible ( @xmath66 = -8.2430 mev , -3.4520 mev and -4.5660 mev , for @xmath31fe , @xmath20fe and @xmath32fe , respectively @xcite ) . however , for stellar conditions the phase space can become positive depending on the calculated energy eigenvalues of the underlying theoretical model . construction of parent and daughter excited states and calculation of nuclear matrix elements in the pn - qrpa model is treated separately in the next section . the excited states in the pn - qrpa model can be constructed as phonon - correlated multi - quasi - particle states . the rpa is formulated for excitations from the @xmath67 ground state of an even - even nucleus . the model extended to include the quasiparticle transition degrees of freedom yields decay half - lives of odd - mass and odd - odd parent nuclei with the same quality of agreement with experiment as for even - even nuclei ( where only qrpa phonons contribute to the decays ) @xcite . for the odd - a nucleus , @xmath20fe , the ground state can be expressed as a one - quasiparticle state , in which the odd quasiparticle ( q.p ) occupies the single - q.p . orbit of the smallest energy . then there exits two different type of transitions : phonon transitions with the odd neutron acting only as a spectator and transition of the odd neutron itself . for the later case , phonon correlations were introduced to one - quasiparticle states in first - order perturbation @xcite . the transition amplitudes between the multi - quasi - particle states can be reduced to those of single - quasi - particle states as shown below . excited states of even - even nucleus are obtained by one - proton ( or one - neutron ) excitations . they are described , in the quasiparticle ( q.p . ) picture , by adding two - proton ( two - neutron ) q.p.s to the ground state @xcite . excited states of @xmath21fe are two - proton q.p . states and two - neutron q.p . states . transitions from these initial states are possible to final proton - neutron q.p . pair states in the odd - odd daughter nucleus . the transition amplitudes and their reduction to correlated ( @xmath68 ) one - q.p . states are given by @xmath69 @xmath70 where @xmath71 = -1 , 0 , 1 , are the spherical components of the spin operator . four - proton ( four - neutron ) q.p . states and higher q.p . states were not considered in the model for the construction of excited states of @xmath21fe . higher q.p . states may affect the total decay rates specially at parent excitation energies well in excess of 5 mev . however the higher q.p . states were not considered in this calculation for consistency reasons . the model used for the calculation of weak - interaction rates was briefly introduced in ref . this model , along with the recipe to construct excited states as described in ref . @xcite , was used to calculate electron and positron capture rates ; electron and positron decay rates ; and finally the neutrino and anti - neutrino cooling rates due to @xmath3fe . it was reported in ref . @xcite that the calculated electron capture rates and neutrino cooling rates , using this model , was up to a factor four bigger than the corresponding lssm rates @xcite for the even - even isotopes of iron . using the same model the calculated @xmath0-decay rates were found to be 3 - 5 orders of magnitude smaller @xcite than the corresponding lssm decay rates . the total decay rates are commanded mainly by the gt forces operating in the particle - particle and particle - hole channels . these forces were characterized by the interaction constants @xmath23 and @xmath24 as described previously . work is currently in progress to include higher q.p . states in the construction of parent excited states to account for possible complicated configurations expected at high excitation energies . any effect on the calculated @xmath0-decay rates due to inclusion of higher q.p . states would be reported in future . when a nucleus has an odd nucleon ( a proton and/or a neutron ) , some low - lying states are obtained by lifting the q.p . in the orbit of the smallest energy to higher - lying orbits @xcite . for @xmath20fe nucleus , the excited states can be constructed + ( 1 ) by lifting the odd neutron from ground state to excited states ( one - q.p . state ) , + ( 2 ) by three - neutron states , corresponding to excitation of a neutron ( three - q.p . states ) , or , + ( 3 ) by one - neutron two - proton states , corresponding to excitation of a proton ( three - q.p . states ) . the formulae for multi - q.p . transitions and their reduction to correlated ( @xmath68 ) one - q.p . states are given by , @xmath72 @xmath73 -\delta ( p_{1}^{f},p_{1}^{i } ) [ \delta ( n_{1}^{f},n_{1}^{i } ) < n_{2c}^{f } \mid t_{\pm}\sigma_{-\mu } \mid p_{2c}^{i } > \nonumber\\ - \delta ( n_{2}^{f},n_{1}^{i } ) < n_{1c}^{f } \mid t_{\pm}\sigma_{-\mu } \mid p_{2c}^{i}>]\end{aligned}\ ] ] @xmath74 next i describe the construction of excited states of the daughter nuclei and the calculation of the respective matrix elements in the model . the low - lying states of an odd - proton even - neutron nucleus ( daughter of @xmath20fe ) can be constructed , + ( 1 ) by exciting the odd proton from ground state ( one - q.p . states ) , + ( 2 ) by three - proton states , corresponding to excitation of a proton ( three - q.p . states ) , or , + ( 3 ) by one - proton two - neutron states , corresponding to excitation of a neutron ( three - q.p . states ) . the multi - q.p . transitions can again be reduced to correlated ( @xmath68 ) one - q.p . states , @xmath75 @xmath76 -\delta ( n_{1}^{f},n_{1}^{i } ) [ \delta ( p_{1}^{f},p_{1}^{i } ) < p_{2c}^{f } \mid t_{\pm}\sigma_{\mu } \mid n_{2c}^{i } > \nonumber \\ - \delta ( p_{2}^{f},p_{1}^{i } ) < p_{1c}^{f } \mid t_{\pm}\sigma_{\mu } \mid n_{2c}^{i}>]\end{aligned}\ ] ] @xmath77 for an odd - odd nucleus the ground state is assumed to be a proton - neutron q.p . pair state of smallest energy . low - lying states in an odd - odd nucleus are expressed in the q.p . picture by proton - neutron pair states ( two - q.p . states ) or by states which are obtained by adding two - proton or two - neutron q.p.s ( four - q.p . states ) @xcite . therefore states in an odd - odd nucleus ( daughter of @xmath21fe ) are expressed in q.p . transformation by two - q.p . states ( proton - neutron pair states ) or by four - q.p . states ( two - proton or two - neutron q.p . reduction of two - q.p . states into correlated ( @xmath68 ) one - q.p . states is given as @xmath78 @xmath79 while the four - q.p . states are simplified as @xmath80 -\delta ( n_{1}^{f},n_{1}^{i } ) [ \delta ( p_{1}^{f},p_{2}^{i})\delta ( p_{2}^{f},p_{3}^{i } ) < n_{2c}^{f } \mid t_{\pm}\sigma_{-\mu } \mid p_{1c}^{i } > \nonumber\\ -\delta ( p_{1}^{f},p_{1}^{i } ) \delta ( p_{2}^{f},p_{3}^{i } ) < n_{2c}^{f } \mid t_{\pm}\sigma_{-\mu } \mid p_{2c}^{i } > + \delta ( p_{1}^{f},p_{1}^{i } ) \delta ( p_{2}^{f},p_{2}^{i})\nonumber\\ < n_{2c}^{f}\mid t_{\pm}\sigma_{-\mu } \mid p_{3c}^{i}>]\end{aligned}\ ] ] @xmath81 @xmath82 -\delta ( p_{2}^{f},p_{1}^{i } ) [ \delta ( n_{1}^{f},n_{2}^{i})\delta ( n_{2}^{f},n_{3}^{i } ) < p_{1c}^{f } \mid t_{\pm}\sigma_{\mu } \mid n_{1c}^{i } > \nonumber\\ -\delta ( n_{1}^{f},n_{1}^{i } ) \delta ( n_{2}^{f},n_{3}^{i } ) < p_{1c}^{f } \mid t_{\pm}\sigma_{\mu } \mid n_{2c}^{i } > + \delta ( n_{1}^{f},n_{1}^{i } ) \delta ( n_{2}^{f},n_{2}^{i } ) \nonumber\\ < p_{1c}^{f } \mid t_{\pm}\sigma_{\mu } \mid n_{3c}^{i}>]\end{aligned}\ ] ] @xmath83 for all the given q.p . transition amplitudes [ eqs . ( [ first])- ( [ last ] ) ] , the antisymmetrization of the single- q.p . states was taken into account : + @xmath84 , + @xmath85 , + @xmath86 , + @xmath87 . + gt transitions of phonon excitations for every excited state were also taken into account . here i assumed that the quasiparticles in the parent nucleus remained in the same quasiparticle orbits . a detailed description of the formalism can be found in ref . the improved pn - qrpa model , as described in the previous section , was used to calculate the gt@xmath6 strength distributions of iron isotopes in astrophysical conditions . the model incorporated experimental excitation energies , log@xmath33 values and deformations as discussed earlier . the gt strength distributions ( in both directions ) were calculated for ground and 245 excited states in @xmath31fe , 296 excited states in @xmath20fe and 265 excited states in @xmath32fe . the model was tested for the case of @xmath21fe where measured distributions were available . it was shown in ref . @xcite that the comparison of gt@xmath6 strength distributions for @xmath21fe with measured data was very good ( e.g. the total b(gt@xmath8 ) strength for @xmath31fe calculated within the framework of ref . @xcite was 9.33 which was narrowed down to 7.56 using the improved model of pn - qrpa @xcite in comparison with the measured value of 7.5 @xmath88 0.7 @xcite ) . the gt@xmath6 strength distributions for ground and all excited states of iron isotopes are available as ascii files and can be requested from the author . in order to highlight the important role of parent excited states in the calculation of total decay rates , i present in table 1 the values of the total beta decay ( @xmath89 ) and positron decay ( @xmath90 ) rates with contributions from all excited states in units of @xmath91 . these rates are calculated at selected values of astrophysical temperature and density shown in first column ( the first number within the parenthesis gives the value of density in units of @xmath92 and the second number denotes the stellar temperature in units of @xmath93 k ) . it is to be noted that the corresponding ground - state beta and positron decay rates are energetically forbidden . the excited states play a key role in the calculation of total decay rates which are very sensitive to the available phase space ( @xmath94 ) . the construction of excited states in the pn - qrpa model was discussed in the previous section . the phase space integrals given in eq . ( [ phase space ] ) increase considerably at higher stellar temperatures and cause orders of magnitude enhancement in the calculated decay rates . fine grid calculation of positron and @xmath0-decay rates for iron isotopes as a function of stellar temperature , density and fermi energy , suitable for core - collapse simulations and interpolation purposes , is available as ascii files and can be requested from the author . figures [ fig1 ] , [ fig2 ] and [ fig3 ] show the calculated @xmath0-decay rates for @xmath3fe , respectively . each figure shows four panels depicting the calculated @xmath0-decay rates at selected temperature and density domain . the upper left panel shows the decay rates in low - density region ( @xmath95 = 10^{0.5 } , 10^{1.5}$ ] and @xmath96 ) , the upper right in medium - low density region ( @xmath95 = 10^{3.5 } , 10^{4.5}$ ] and @xmath97 ) , the lower left in medium - high density region ( @xmath98 = 10^{6.5 } , 10^{7.5}$ ] and @xmath99 ) and finally the lower right panel depicts the calculated rates in high density region ( @xmath95 = 10^{9.5 } , 10^{10.5}$ ] and @xmath100 ) of stellar core of massive stars . the decay rates are given in logarithmic scales ( to base 10 ) in units of @xmath91 . t@xmath101 gives the stellar temperature in units of @xmath93 k. one should note the orders of magnitude increment in @xmath0-decay rates as the stellar temperature increases . the rates are almost superimposed on one another as a function of stellar density in the first three panels . this means that there is no appreciable change in the @xmath0-decay rates when increasing the density by an order of magnitude . however as the stellar matter moves from the medium high density region to high density region these rates start to peel off from one another . in high density regions the rates start to decrease appreciably due to a decrease in the phase space . there is a sharp exponential increase in the decay rates as the stellar temperature increases up to @xmath102 t@xmath101 = 5 . beyond this temperature the slope of the rates reduces drastically with increasing density . for a given density the @xmath0-decay rates increase monotonically with increasing temperatures . the calculated stellar @xmath0-decay rates are smallest for @xmath31fe and biggest for @xmath20fe . the calculated @xmath0-decay rates are smaller in magnitude compared to the corresponding positron capture rates ( in the same direction ) on these iron isotopes . figures [ fig4 ] , [ fig5 ] and [ fig6 ] again show four panels depicting the calculated positron decay rates at selected temperature and density domain for @xmath3fe , respectively . the upper left panel shows the positron decay rates in low - density region ( @xmath95 = 10^{0.5 } , 10^{1.5}$ ] and @xmath96 ) , the upper right in medium - low density region ( @xmath95 = 10^{3.5 } , 10^{4.5}$ ] and @xmath97 ) , the lower left in medium - high density region ( @xmath95 = 10^{6.5 } , 10^{7.5}$ ] and @xmath99 ) and finally the lower right panel depicts the calculated positron decay rates in high density region ( @xmath95 = 10^{9.5 } , 10^{10.5}$ ] and @xmath100 ) of the stellar core . the rates are given in logarithmic scales ( to base 10 ) in units of @xmath91 . the calculated positron decay rates are greater than the corresponding @xmath0-decay rates by orders of magnitude . due to an increase in phase space with increasing temperature , the positron decay rates increase by orders of magnitude as t@xmath101 increases . the increase is exponential up to @xmath102 t@xmath101 = 5 . as temperature rises ( the degeneracy parameter is negative for positrons ) , more and more high - energy positrons are created leading in turn to higher decay rates . it can be seen from these figures that the positron decay rates are almost the same as functions of the core density . the positron decay rates are biggest for @xmath20fe and smallest for @xmath32fe . it is worth mentioning that the beta and positron decay rates are very small numbers and can change by orders of magnitude by a mere change of 0.5 mev , or less , in parent or daughter excitation energies ( as the rates are very sensitive to the available q@xmath103 window ) and are more reflective of the uncertainties in the calculation of energies . it would be interesting to know how the reported decay rates of iron isotopes compare with earlier calculations for temperature and density domains of astrophysical interest . for the sake of comparison i took into consideration the pioneer calculations of ffn @xcite and those performed using the large - scale shell model ( lssm ) @xcite . the ffn rates were used in many simulation codes ( e.g. kepler stellar evolution code @xcite ) while lssm rates were employed in recent simulation of presupernova evolution of massive stars in the mass range 11 - 40 @xmath104 @xcite . positron decay rates of iron isotopes may decrease the @xmath1 value of massive stars . as mentioned before , the positron decay rate calculations do not differ appreciably by changing densities . figure [ fig7 ] depicts the comparison of positron decay rates for @xmath31fe with earlier calculations . the upper panel displays the ratio of calculated rates to the lssm rates , @xmath105 , while the lower panel shows a similar comparison with the ffn calculation , @xmath106 . the comparison is made for the selected temperature and density domain . here one sees that the pn - qrpa and lssm calculations are in overall reasonable agreement . the comparison is good at low temperatures while at higher temperatures , @xmath107 5 , the lssm rates are bigger around an order of magnitude . the ffn rates are much bigger than the reported positron decay rates by 1 2 orders of magnitude . it is reminded that ffn neglected the quenching of the gt strength in their rate calculation . further ffn did not take into effect the process of particle emission from excited states and their parent excitation energies extended well beyond the particle decay channel . these high lying excited states began to show their cumulative effect at high temperatures and densities . the comparison of reported positron decay rates of @xmath20fe with lssm calculation is somewhat different . for the first time one sees that the pn - qrpa calculated positron decay rate is bigger than lssm ( figure [ fig8 ] ) . this however is true only for @xmath108 = 1 . during the early phases of presupernova evolution the reported positron decay rates are bigger than lssm rates by an order of magnitude . for successive stages , 1.5 @xmath109 5 , the two calculations are in good agreement . at still higher temperatures the lssm rates exceed by more than an order of magnitude . excited state gt distributions in @xmath20fe play a key role in determining the total decay rate at various stages of presupernova evolution . it is reminded that the pn - qrpa makes a microscopic assessment of gt strength distribution for all such parent excited states . comparison with the ffn rates is fairly constant . it is to be noted that the ratios are shown on a linear scale ( bottom panel of figure [ fig8 ] ) . at @xmath108 = 10 , the ffn rates surpass the reported rates for reasons already mentioned . the comparison of reported positron decay rates of @xmath32fe with previous calculations is shown in figure [ fig9 ] . the lssm decay rates are bigger by as much as a factor of 20 ( the comparison is better at low temperatures ) . the lower panel of figure [ fig9 ] shows a whopping enhancement in the calculated positron decay rates of @xmath32fe of around 13 orders of magnitude compared to ffn rates at low temperatures . however there are reasons for this unusual suppression of ffn positron decay rates . unmeasured matrix elements for allowed transitions were assigned an average value of @xmath1105 in ffn calculations . later , @xmath13 experiments were performed for @xmath32fe @xcite which revealed a too small @xmath111 value assignment for allowed transitions employed by ffn . consequently the ffn rates are much too smaller than the lssm and pn - qrpa calculations at low temperatures ( lssm rates are also bigger than ffn rates by roughly 13 orders of magnitude at @xmath108 = 1.5 ) . the lssm positron decay rates for iron isotopes are bigger up to a factor of 20 as compared to the pn - qrpa rates . the overall comparison of positron decay rates of @xmath3fe shows that the calculated rates are in reasonable agreement with the lssm calculated rates . major differences are seen in the case of the @xmath0-decay rate calculations and are discussed in detail below . authors in ref . @xcite reported that , for even - even and odd - a nuclei , ffn systematically placed the back resonance at much lower energies and concluded that contribution of the back resonances to the @xmath0-decay rates for these nuclei decreases . they estimated that lssm @xmath0-decay rates as a result were smaller , on the average , by a factor of 20 ( 40 ) as compared to the ffn @xmath0-decay rates for even - even ( odd - a ) nuclei . figure [ fig10 ] depicts the comparison of @xmath0-decay rates for @xmath31fe with earlier calculations . as before the upper panel displays the ratio of calculated rates to the lssm rates , @xmath112 , while the lower panel shows a similar comparison with the ffn calculation , @xmath113 . the density scale is shown in the inset . a mutual comparison of lssm and ffn @xmath0-decay rates for the case of @xmath31fe reveals that for low temperatures ( t@xmath114 2 ) and densities @xmath95 \sim 10^{6}-10^{7}$ ] the lssm @xmath0-decay rates are bigger than the ffn rates by as much as four orders of magnitude . only at higher temperatures and densities are the lssm rates smaller than the ffn rates ( by around an order of magnitude ) . the lssm @xmath0-decay rates for @xmath31fe are 3 4 orders of magnitude bigger than the pn - qrpa rates for the physical conditions depicted in figure [ fig10 ] . the enhancement ratio increases with increasing densities . the reasons for bigger lssm @xmath0-decay rates are not very clear the q - values and phase space formulation for the two calculations appear the same . perhaps the contribution from back resonances needs a further cut back in the lssm calculation for the @xmath0-decay of even - even and odd - a nuclei ( i.e. the back resonances for these nuclei should be put at even higher energies than calculated by lssm ) . as mentioned before the reported rates do not employ these approximations and calculate all the excited state gt strength distributions in a microscopic fashion . the comparison with the ffn rates reveals that the reported decay rates are smaller by as much as four orders of magnitude ( at higher temperatures ) . at temperature t@xmath101 = 1 and density @xmath95 = 10^{6}$ ] , where the lssm @xmath0-decay rates are bigger than the ffn rates by roughly four orders of magnitude , the reported rates are bigger than the ffn rates by an order of magnitude . for all other points the ffn rates are much bigger than pn - qrpa rates for reasons mentioned before . the @xmath0-decay rates of @xmath20fe are believed to be important during the silicon burning stages of massive stars as per the simulation results of ref . figure [ fig11 ] shows a comparison of the three calculations for @xmath0-decay rates of @xmath20fe during the relevant temperature and density domain of stellar core . only at low temperatures and densities are the lssm decay rates smaller than the corresponding ffn numbers . at higher temperatures ( t@xmath115 5 ) and densities ( @xmath95 > 10^{7}$ ] ) the lssm rates surpass the ffn decay rates by more than two orders of magnitude . the comparison of pn - qrpa rates with lssm suggests that the lssm rates are bigger by 2 3 orders of magnitude ( at lower densities the comparison is relatively better ) . in fact the lssm @xmath0-decay rates are even bigger than their calculated positron capture rates ( only at higher temperatures their positron capture rates surpass the @xmath0-decay rates ) . in contrast the pn - qrpa calculated @xmath0-decay rates are suppressed as compared to the corresponding positron capture rates for all temperature and density scales . at higher densities the lssm @xmath0-decay rates are bigger by 3 5 orders of magnitude . the ffn rates are up to four orders of magnitude bigger than the pn - qrpa rates for reasons mentioned before . a study of lssm and the ffn rates reveals that lssm @xmath0-decay rates of @xmath32fe are much bigger than the ffn rates at high temperatures and densities ( by more than an order of magnitude ) . figure [ fig12 ] shows how the reported @xmath0-decay rates of @xmath32fe compare with earlier calculations for relevant physical conditions . it can be seen from the figure that the reported rates are much smaller ( up to four orders of magnitude ) than previous calculations . comparison with lssm calculation shows that at higher densities the reported @xmath0-decay rates are suppressed by up to three orders of magnitude . at higher temperatures ( t@xmath101 = 30 ) the comparison ratio improves . however the lssm rates are still bigger by around an order of magnitude . on the other hand the ffn rates are bigger than the reported @xmath0-decay rates by around four orders of magnitude at lower densities and temperatures . the comparison ratio improves with increasing temperatures and densities . the pn - qrpa rates are bigger than the ffn rates by around a factor of 8 at t@xmath101 = 30 . the comparison study of stellar @xmath0-decay rates of iron isotopes suggests that the lssm @xmath0-decay rates are bigger than the corresponding pn - qrpa rates by 3 5 orders of magnitude . earlier using the same model the electron capture rates due to @xmath3fe were calculated and were found to be in overall good comparison with the lssm electron capture rates @xcite . in fact the pn - qrpa electron capture rates on @xmath31fe were around a factor three bigger than lssm rates in the relevant astrophysical conditions ( see table 3 of ref . @xcite ) . using the same model , however , the calculated @xmath0-decay rates are smaller by 3 5 orders of magnitude . this suppression in the pn - qrpa @xmath0-decay rates comes from the excited state gt distributions ( smaller nuclear matrix elements ) which are much different from the ground state distribution . in order to gain a detailed insight of these distributions i took a representative temperature and density point of t@xmath101 = 10 and @xmath95 = 10^{7}$ ] ( weak interaction rates due to iron isotopes are argued to contribute effectively around such physical conditions during the presupernova evolution of massive stars ) . table 2 shows the data of the ground and first four excited states that have a finite partial decay rate contribution to the total @xmath0-decay rate of @xmath3fe within the framework of the model of magnitude greater than @xmath116 . the first column shows the calculated parent excited energy state in units of mev , the second column gives the product of occupation probability and partial @xmath0-decay rate from this state in units of @xmath91 . the third and fourth column give the centroid and total b(gt ) strength , respectively , from this parent state . the cut - off energy in daughter nuclei is 12 mev . it can be seen clearly from table 2 that for the even - even isotopes the centroids of the excited state gt strength distributions are shifted to much higher energies in the daughter . the @xmath117 strengths are also considerably smaller from the corresponding ground state strengths . these are mainly responsible for the smaller total @xmath0-decay rates . the excited state gt strength distributions are also much different from the ground state distribution for the case of @xmath20fe . the excited state gt distributions for @xmath3fe are shown graphically in figures [ fig13 ] , [ fig14 ] and [ fig15 ] , respectively . the ground state gt strength distributions were presented earlier in ref . @xcite . here i have shown only the first three excited state distributions that have a finite partial decay rate contribution of magnitude greater than @xmath116 . note the different scales of the b(gt ) strength values in the vertical panels of these figures . from these figures it is clear that the brink s hypothesis ( and back resonances for the calculation of @xmath0-decay ) is not a good approximation for calculation of stellar weak interaction rates of iron isotopes . these and similar finite partial decay rates lead to an overall suppression in the total @xmath0-decay rate which happens to be 5.44e-09 @xmath91 , 4.32e-07 @xmath91 and 2.62e-05 @xmath91 for @xmath31fe , @xmath20fe and @xmath32fe , respectively , under the physical conditions stated above . capture and decay rates are considered to be very important in controlling the lepton - to - baryon ratio and entropy of the core of massive stars during the presupernova evolutionary phases . these are the two key parameters which later play a key role in the dynamics of supernova explosion . a reliable and microscopic calculation of these weak interaction rates can assist us in a better understanding of the explosion process . the pn - qrpa model has a very good rapport in calculation of beta decay rates . the model has access to a huge model space of seven major shells and is the only available model to perform a fully microscopic state - by - state calculation of weak rates in stellar interior . the stellar electron and positron @xmath0-decay rates of iron isotopes were presented using the pn - qrpa model . incorporation of latest experimental data and an optimum selection of model parameters increased the reliability of the calculated rates . the calculated rates were also compared with previous calculations . the key findings of this work include that the brink s hypothesis and back resonances used in previous calculations of decay rates for iron isotopes are not a good approximation to use . a microscopic calculation of excited state gt strength distributions greatly increases the reliability of calculated rates . the @xmath0-decay rates calculated by large scale shell model calculation are 3 5 orders of magnitude bigger than the reported rates in astrophysical conditions . the microscopic calculation of excited state gt strength distribution calculated within the framework of the pn - qrpa model , which are responsible for the reduced @xmath0-decay rates , was also discussed . during the early phases of presupernova evolution the lssm calculated @xmath0-decay rates are even bigger than the competing positron capture rates ( both occur in the same direction and tend to decrease @xmath1 ) . the pn - qrpa calculated @xmath0-decay rates are smaller than the competing positron capture rates . ffn @xmath0-decay rates are up to four orders of magnitude bigger than the reported rates under the same conditions . the pn - qrpa calculation validates the finding by authors in ref . @xcite that ffn places the so - called back resonances at too low excitation energies in even - even and odd - a nuclei . the reported calculation also suggests that the placement of back resonances employed by lssm needs further push towards higher excitation energies . the lssm positron decay rates are , comparatively , in reasonable agreement with the pn - qrpa rates . what may be the astrophysical implications of the reported decay rates of iron isotopes ? the @xmath0-decay rates of iron nuclei are much smaller than previously assumed and this news is important for core - collapse simulators world - wide . the current study suggests that @xmath0-decay rates of iron isotopes are irrelevant for the determination of the evolution of @xmath1 during the presupernova phase of massive stars . a review of inclusion of iron isotopes @xmath4fe in the list of key stellar @xmath0-decay nuclei as suggested by previous simulation results @xcite is in order . * acknowledgments : * the author would like to acknowledge the kind hospitality provided by the abdus salam ictp , trieste , where part of this project was completed . the author also wishes to acknowledge the support of research grant provided by the higher education commission pakistan , through the hec project nos . 20 - 1171 and 20 - 1283 . baade w. & zwicky f. , on supernovae , _ proc . _ , 20 , 254 - 259 , 1934 . bethe h. a. , energy production in stars , _ phys . rev . _ , 55 , 434 - 456 , 1939 . burbidge e. m. , burbidge g. r. , fowler w. a. & hoyle f. , synthesis of the elements in stars , _ rev . phys _ , 29 , 547 - 650 , 1957 . fuller g. m. , fowler w. a. & newman m. j. , stellar weak - interaction rates for sd - shell nuclei . i. nuclear matrix element systematics with application to @xmath118al and selected nuclei of importance to the supernova problem , _ astrophys . _ , 42 , 447 - 473 , 1980 ; stellar weak interaction rates for intermediate mass nuclei . ii . a = 21 to a = 60 , _ astrophys . j. _ , 252 , 715 - 740 , 1982 ; stellar weak interaction rates for intermediate mass nuclei . rate tables for the free nucleons and nuclei with a = 21 to a = 60 , _ astrophys . _ , 48 , 279 - 320 , 1982 ; stellar weak interaction rates for intermediate mass nuclei . iv . interpolation procedures for rapidly varying lepton capture rates using effective log ( ft)- values , _ astrophys . j. _ , 293 , 1 - 16 , 1985 . aufderheide m. b. , fushiki i. , woosley s. e. , stanford e. & hartmann d. h. , search for important weak interaction nuclei in presupernova evolution , _ astrophys . ser _ , 91 , 389 - 417 , 1994 . langanke k. & martnez - pinedo g. , shell - model calculations of stellar weak interaction rates : ii . weak rates for nuclei in the mass range a = 45 - 65 in supernovae environments , _ nucl . _ , a673 , 481 - 508 , 2000 . nabi j .- un & klapdor - kleingrothaus h. v. , microscopic calculations of stellar weak interaction rates and energy losses for fp- and fpg - shell nuclei , _ at . data nucl . data tables _ , 88 , 237 - 476 , 2004 . aufderheide m. b. , stellar electron capture rates and the @xmath31fe(n , p ) experiment , _ nucl . _ , a526 , 161 - 187 , 1991 . nabi j .- un & sajjad m. , comparative study of gamow - teller strength distributions in the odd - odd nucleus @xmath119v and its impact on electron capture rates in astrophysical environments , _ phys . , 76 , 055803 - 1 - 055803 - 8 , 2007 . nabi j .- un & klapdor - kleingrothaus h. v. , microscopic calculations of weak interaction rates of nuclei in stellar environment for a = 18 to 100 , _ eur . j. a _ , 5 , 337 - 339 , 1999 . nabi j .- un & klapdor - kleingrothaus h. v. , weak interaction rates of sd - shell nuclei in stellar environments calculated in the proton - neutron quasiparticle random - phase approximation , _ at . data nucl . data tables _ , 71 , 149 - 345 , 1999 . nabi j .- un & rahman m .- ur , gamow - teller transitions from @xmath120 mg and their impact on the electron capture rates in the o+ne+mg cores of stars , _ phys . , 75 , 035803 - 1 - 035803 - 5 , 2007 . nabi j .- un , stellar neutrino loss rates due to @xmath120 mg suitable for o+ne+mg core simulations , _ phys . rev . , 78 , 045801 - 1 045801 - 11 , 2008 . nabi j .- un , detailed microscopic calculation of stellar electron and positron capture rates on @xmath120 mg for o+ne+mg core simulations , _ phys . , 78 , 035201 - 1 - 035201 - 10 , 2008 . nabi j .- un & rahman m .- ur , gamow - teller strength distributions and electron capture rates for @xmath20co and @xmath32ni , _ phys . lett _ , b612 , 190 - 196 , 2005 . nabi j .- un , sajjad m. & rahman m .- ur , electron capture rates on titanium isotopes in stellar matter , _ acta phys . pol . b _ , 38 , 3203 - 3223 , 2007 . nabi j .- un , rahman m .- ur & sajjad m. , electron and positron capture rates on @xmath20co in stellar matter , _ braz . _ , 37 , 1238 - 1245 , 2007 . nabi j .- un , rahman m .- ur & sajjad m. , gamow - teller ( gt@xmath6 ) strength distributions of @xmath32ni for ground and excited states , _ acta phys . b _ , 39 , 651 - 669 , 2008 . nabi j .- un & sajjad m. , neutrino energy loss rates and positron capture rates on @xmath20co for presupernova and supernova physics , _ phys . , 77 , 055802 - 1 - 055802 - 13 , 2008 . nabi j .- un & sajjad m. , expanded calculations of proton - neutron quasiparticle random phase approximation ( pn - qrpa ) electron capture rates on @xmath20co for presupernova and supernova physics _ can . j. phys _ , 86 , 819 - 828 , 2008 . staudt a. , bender e. , muto k. & klapdor - kleingrothaus h. v. , second - generation microscopic predictions of beta - decay half - lives of neutron - rich nuclei , _ at . data nucl . data tables _ , 44 , 79 - 132 , 1990 . hirsch m. , staudt a. , muto k. & klapdor - kleingrothaus h. v. , microscopic predictions of @xmath15/ec - decay half - lives , _ at . data nucl . data tables _ , 53 , 165 - 193 , 1993 . heger a. , woosley s. e. , martnez - pinedo g. & langanke k. , presupernova evolution with improved rates for weak interactions , _ astrophys . j. _ , 560 , 307 - 325 , 2001 . nabi j .- un , weak - interaction - mediated rates on iron isotopes for presupernova evolution of massive stars , _ eur . phys . j. a _ , 40 , 223 - 230 , 2009 . nilsson s. g. , binding states of individual nucleons in strongly deformed nuclei , _ mat . selsk _ , 29 , 1 - 68 , 1955 . cha d. , @xmath121 strength in nuclei , _ phys . , 27 , 2269 - 2281 , 1983 . staudt a. , hirsch m. , muto k. & klapdor - kleingrothaus h. v. , calculation of @xmath0-delayed fission of @xmath122tl and application of the quasiparticle random - phase approximation to the prediction of @xmath15-decay half - lives of neutron - deficient isotopes , _ phys . _ , 65 , 1543 - 1546 , 1990 . hirsch m. , staudt a. , muto k. & klapdor - kleingrothaus h. v. , microscopic calculation of @xmath15/ec - decay half - lives with atomic numbers z = 10 - 30 , _ _ , a535 , 62 - 76 , 1991 . stetcu i. & johnson c. w. , gamow - teller transitions and deformation in the proton - neutron random phase approximation , _ phys . , 69 , 024311 - 1 - 024311 - 7 , 2004 . raman s. , malarkey c. h. , milner w. t. , nestor , jr . c. w. & stelson p. h. , transition probability , b(e2)@xmath123 , from the ground to the first - excited 2@xmath124 state of even - even nuclides , _ at . data nucl . data tables _ , 36 , 1 - 96 , 1987 . mller p. & nix j. r. , atomic masses and nuclear ground - state deformations calculated with a new macroscopic - microscopic model , _ at . data nucl . data tables _ , 26 , 165 - 196 , 1981 . audi g. , wapstra a. h. & thibault c. , the ame2003 atomic mass evaluation ( ii ) . tables , graphs and references , _ nucl . _ , a729 , 337 - 676 , 2003 . haxton w. c. & johnson c. , weak - interaction rates in @xmath125o , _ phys . rev . _ , 65 , 1325 - 1328 , 1990 . yost g. p. , barnett r. m. , hinchliffe i. , _ et al . _ ( particle data group ) , review of particle properties , _ phys . _ , b204 , 1 - 476 , 1988 . gaarde c. , gamow - teller and m1 resonances , _ nucl . _ , a396 , 127c-144c , 1983 . rodin v. , faessler a. , simkovic f. & vogel p. , uncertainties in the 0-decay nuclear matrix elements , _ czech . _ , 56 , 495 - 503 , 2006 . vetterli m. c. , husser o. , abegg r. , alford w. p. , celler a. , frekers d. , helmer r. , henderson r. , hicks k. h. , jackson k. p. , jeppesen r. g. , miller c. a. , raywood k. & yen s. , gamow - teller strength deduced from charge exchange reactions on @xmath31fe at 300 mev , _ phys . , 40 , 559 - 569 , 1989 . rnnqvist t. , cond h. , olsson n. , ramstrm e. , zorro r. , blomgren j. , hkansson a. , ringbom a. , tibell g. , jonsson o. , nilsson l. , renberg p .- u . , van der werf s. y. , unkelbach w. & brady f. p. , the @xmath21fe(n , p)@xmath21mn reactions at e@xmath126 = 97 mev , _ nucl . _ , a563 , 225 - 246 , 1993 . gove n. b. & martin m. j. , log - f tables for beta decay , _ at . data nucl . data tables _ , 10 , 205 - 317 , 1971 . muto k. , bender e. , oda t. & klapdor h. v. , proton - neutron quasiparticle rpa with separable gamow - teller forces , _ z. phys . a _ , 341 , 407 - 415 , 1992 . muto k. , bender e. & klapdor h. v. , proton - neutron quasiparticle rpa and charge - changing transitions , _ a _ , 333 , 125 - 129 , 1989 . anderson b. d. , lebo c. , baldwin a. r. , chittrakarn t. , madey r. & watson j. w. , gamow - teller strength in the @xmath31fe(p , n)@xmath31co reaction at 135 mev , _ phys . , 41 , 1474 - 1485 , 1990 . weaver t. a. , zimmerman g. b. & woosley s. e. , presupernova evolution of massive stars , _ astrophys . , 225 , 1021 - 1029 , 1978 . el - kateb s. , jackson k. p. , alford w. p. , abegg r. , azuma r. e. , brown b. a. , celler a. , frekers d. , husser o. , helmer r. , henderson r. s. , hicks k. h. , jeppesen r. , king j. d. , raywood k. , shute g. g. , spicer b. m. , trudel a. , vetterli m. & yen s. , spin - isospin strength distributions for fp shell nuclei : results for the @xmath20mn(n , p ) , @xmath32fe(n , p ) , and @xmath127ni(n , p ) reactions at 198 mev , _ phys . , 49 , 3128 - 3136 , 1994 .

model of a passive scalar field advected by the compressible gaussian strongly anisotropic velocity field with the covariance is studied by using the field theoretic renormalization group and the operator product expansion . the inertial - range stability of the corresponding scaling regime is established . the anomalous scaling of the single - time structure functions is studied and the corresponding anomalous exponents are calculated . their dependence on the compressibility parameter and anisotropy parameters is analyzed . it is shown that , as in the isotropic case , the presence of compressibility leads to the decreasing of the critical dimensions of the important composite operators , i.e. , the anomalous scaling is more pronounced in the compressible systems . all calculations are done to the first order in . * combined effects of small scale anisotropy and compressibility on anomalous scaling of a passive scalar * + e. juriinov , m. juriin , r.remecky , and m.scholtz _ institute of experimental physics , slovak academy of sciences , + watsonova 47 , 04001 koice , slovakia _ + _ laboratory of information technologies , joint institute for nuclear research , + 141 980 dubna , moscow region , russian federation _ + _ laboratory of theoretical physics , joint institute for nuclear research , + 141 980 dubna , moscow region , russian federation _ + _ department of physics and astrophysics , institute of physics , + p.j . afrik university , park angelinum 9 , 04001 koice , slovakia _ e - mail : jurcisin@thsun1.jinr.ru , eva.jurcisinova@post.sk , + richard.remecky@student.upjs.sk , scholtzzz@pobox.sk pacs numbers : 47.10.+g ; 47.27.-i keywords : passive scalar ; renormalization group ; anomalous scaling

nlin0607066

during the last two decades the so - called `` rapid change model '' of a passively advected scalar by a self - similar gaussian @xmath8correlated in time velocity field introduced by kraichnan @xcite and number of its extensions have played the central role in the theoretical investigation of intermittency and anomalous scaling , the problems which stay in the center of attention in the framework of the inertial range investigation of fully developed turbulence @xcite , i.e. , in the range characterized by the scales which are far away from the largest scales at which energy is pumping into the system and , at the same time , far away form the smallest scales which are related to the dissipation processes . there are at least three reasons for this interest : first of all , it is well - known from experimental and also theoretical studies that the deviations from the statements of the famous classical kolmogorov - obukhov phenomenological theory ( see , e.g. , @xcite ) is surprisingly more noticeable and visible for the simpler models of passively advected scalar quantity ( scalar field ) than for the velocity field itself ( see , e.g. , @xcite ) ; second , at the same time , the problem of passive advection of scalar field ( as well as vector field ) is much easier from theoretical point of view than the original problem of anomalous scaling in the framework of navier - stokes velocity field , and , in the end , third , even very simplified models with given gaussian statistics of velocity field lead to the anomalous behavior which describe many features of the real turbulent advection , see , e.g. , refs.@xcite and references cited therein . as was already mentioned , the crucial role in these studies was played ( and is still played ) by the aforementioned rapid change model of a passive scalar advection , where , for the first time , the systematic analysis of the corresponding anomalous exponents was done on the microscopic level . for example , within the so - called `` zero - mode approach '' to the rapid change model @xcite ( see also survey paper @xcite ) the anomalous exponents are found from the homogenous solutions ( zero modes ) of the closed equations for the single - time correlations . on the other hand , one of the most effective approach for studying self - similar scaling behavior is the method of the field theoretic renormalization group ( rg ) @xcite . it can be also used in the theory of fully developed turbulence @xcite and related problems , e.g. , in the problem of a passive scalar advection by the turbulent environment @xcite ( see also refs.@xcite for details ) . in refs.@xcite the field theoretic rg and operator - product expansion ( ope ) were used in the systematic investigation of the rapid - change model of a passive scalar . it was shown that within the field theoretic approach the anomalous scaling is related to the very existence of so - called `` dangerous '' composite operators with negative critical dimensions in ope ( see , e.g. , refs.@xcite for details ) . in the subsequent papers @xcite the anomalous exponents of the model were calculated within the @xmath1 expansion up to order @xmath9 ( three - loop approximation ) . here @xmath1 is a parameter which describes a given equal - time pair correlation function of the velocity field ( see next section ) . important advantages of the rg approach are its universality and calculational efficiency : a regular systematic perturbation expansion for the anomalous exponents was constructed , similar to the well - known @xmath10-expansion in the theory of phase transitions . besides various generalizations of the kraichnan model towards more realistic ones , namely , models with inclusion of small scale anisotropy @xcite , compressibility @xcite , and finite correlation time of the velocity field @xcite were studied by the field theoretic approach . general conclusion of all these investigations is that the anomalous scaling , which is the most intriguing and important feature of the kraichnan rapid change model , remains valid for all generalized models . in ref.@xcite the field theoretic rg and ope were applied to the rapid change model of passive scalar advected by gaussian strongly anisotropic velocity field where the anomalous exponents of the structure functions were calculated to the first order in @xmath1 expansion . it was shown that in the presence of small - scale anisotropy the corresponding exponents are nonuniversal , i.e. , they are functions of the anisotropy parameters , and they form the hierarchy with the leading exponent related to the most `` isotropic '' operator . the importance of these investigations is related to the question of the influence of anisotropy on inertial - range behavior of passively advected fields as well as the velocity field itself @xcite ( see also the survey paper @xcite and references cited therein , as well as recent astrophysical investigations , e.g , in refs.@xcite ) . on one hand , it was shown that for the even structure ( or correlation ) functions the exponents which describe the inertial - range scaling exhibit universality and they are ordered hierarchically in respect to degree of anisotropy with leading contribution given by the exponent from the isotropic shell but , on the other hand , the survival of the anisotropy in the inertial - range is demonstrated by the behavior of the odd structure functions , namely , the so - called skewness factor decreases down the scales slower than expected earlier in accordance with the classical kolmogorov - obukhov theory . on the other hand , in ref.@xcite the influence of compressibility and large - scale anisotropy on the anomalous scaling behavior was studied in the aforementioned model in the order @xmath11 in the @xmath1 expansion , and anomalous exponents of higher - order correlation functions were calculated as functions of the parameter of compressibility . it was shown that the exponents exhibit hierarchy related to the degree of anisotropy . again , the existence of small - scale anisotropy effects were demonstrated by the odd dimensionless ratios of correlation functions : the skewness and hyperskewness factors , and it was shown that the persistent of small - scale anisotropy is more pronounced for larger values of the compressibility parameter @xmath12 . from this point of view , compressible systems are very interesting to be studied . in present paper , we shall continue in the investigation of the model , namely , for the first time , we shall study the model with compressible velocity field together with assumption about its small - scale uniaxial anisotropy in the first order in @xmath1 , i.e. , we shall extend the model studied in ref.@xcite to the compressible case . in this situation , in general , two types of diffusion - advection problems of scalar particles exist in the compressible case which are identical in the incompressible case . in what follows , we shall study only one of them , namely , the advection of so - called `` tracer '' ( see , e.g. , ref . @xcite ) , i.e. , for example , concentration of a scalar impurity , temperature , entropy , etc . details will be shown in the next section . first of all we shall establish stability of the scaling regime of the model and coordinates of the corresponding infrared ( ir ) stable fixed point will be found analytically as functions of the compressibility and anisotropy parameters . these results will be then used in the analysis of the asymptotic behavior of the single - time structure functions of a passively advected scalar field . the paper is organized as follows : in sec.[sec : model ] we present definition of the model and introduce the compressibility and small - scale uniaxial anisotropy to the given pair correlation function of the velocity field . in sec.[sec : field ] we give the field theoretic formulation of the original stochastic problem and discuss the corresponding diagrammatic technique . the analysis of the ultraviolet ( uv ) divergences of the model is given , the multiplicative renormalizability of the model is established , and the renormalization constants are calculated in one - loop approximation . in sec.[sec : scalreg ] we analyze infrared ( ir ) asymptotic behavior of the model which is governed by the ir stable fixed point . explicit expressions for the coordinates of the ir fixed point are found . in sec.[sec6 ] the renormalization of needed composite operators is done and their critical dimensions are found as functions of parameters of the model . in sec.[sec : conc ] discussion of results is present . we shall consider the problem of the advection of a passive `` tracer '' @xmath13 which is described by the following stochastic equation @xmath14 where @xmath15 , @xmath16 , @xmath17 is the laplace operator , @xmath18 is the coefficient of molecular diffusivity ( hereafter all parameters with a subscript @xmath19 denote bare parameters of unrenormalized theory ; see below ) , @xmath20 is the @xmath21-th component of the compressible velocity field @xmath22 , and @xmath23 is a gaussian random noise with zero mean and correlation function @xmath24 where parentheses @xmath25 hereafter denote average over corresponding statistical ensemble . the noise defined in eq.([correlator ] ) maintains the steady - state of the system but the concrete form of the correlator will not be essential in what follows . the only condition which must be fulfilled by the function @xmath26 is that it must decrease rapidly for @xmath27 , where @xmath28 denotes an integral scale related to the stirring . as was already mentioned in introduction , another type of the diffusion - advection problem exists in the case when compressibility of the velocity field is supposed which is given by the following more general stochastic equation @xcite @xmath29 it describes the passive advection of a density field but , in what follows , we shall not study this problem . in the case of incompressible velocity field ( it is given mathematically by the divergence - free condition @xmath30 ) both models are equivalent . in real problems it is traditionally assumed that the velocity field @xmath22 satisfies stochastic navier - stokes equation @xcite . in spite of , in what follows , we shall suppose that the velocity field obeys a gaussian distribution with zero mean and two - point correlator @xmath31 , \label{corv}\ ] ] i.e , we shall work in the framework of the so - called rapid - change model @xcite . here , @xmath32 denotes the dimension of the @xmath33 space , and @xmath34 is an amplitude factor related to the coupling constant @xmath35 of the model ( expansion parameter in the perturbation theory , see next section ) by the relation @xmath36 , where @xmath37 is the characteristic uv momentum scale . the parameter of the energy spectrum of the velocity field @xmath38 is taken in such a way that its `` kolmogorov '' value ( the value which corresponds to the kolmogorov scaling of the velocity correlation function in developed turbulence ) is @xmath39 , and @xmath40 is another integral scale . in general , the scale @xmath40 may be different from the integral scale @xmath28 introduced in eq.([correlator ] ) but , in accordance with ref.@xcite , we suppose that @xmath41 . in the incompressible isotropic case the second - rank tensor @xmath42 in eq.([corv ] ) has the simple form of the ordinary transverse projector @xmath43 , where @xmath44 . this tensor is changed when one incorporates small - scale anisotropy or compressibility . let us briefly discuss these questions . first of all , in the case of incompressible anisotropic case a new second - rank tensor must be again a transverse operator because the velocity field is still divergence - free ( @xmath30 ) . the simplest way how to introduce small - scale uniaxial anisotropy is to take the operator @xmath42 in the following way @xcite @xmath45 where @xmath46 is the usual transverse projection operator ( as defined above ) , the unit vector @xmath47 determines the distinguished direction , and @xmath48 , @xmath49 are parameters characterizing the anisotropy . the positive definiteness of the correlation function ( [ corv ] ) imposes the following restrictions on their values : @xmath50 . the operator ( [ t - ij ] ) is a special case of the general transverse structure that possesses uniaxial anisotropy : @xmath51 where @xmath52 denotes the angle between the vectors @xmath47 and @xmath53 ( @xmath54 ) . using gegenbauer polynomials @xcite the scalar functions in representation ( [ generalt ] ) may be expressed in the form @xmath55 it was shown in ref.@xcite that all main features of the general model with the anisotropy structure represented by eq.([generalt ] ) are included in the simplified model with the special form of the transverse operator given by eq.([t - ij ] ) . second , in the case of the compressible isotropic velocity field the second - rank tensor @xmath42 in eq.([corv ] ) is not longer a transverse projector but it is rather a combination of the transverse projector @xmath56 and of the longitudinal projector @xmath57 as a result of the fact that the velocity field is not longer solenoidal one . thus , in this case , the correlator ( [ corv ] ) contains the tensor structure of the following general form @xmath58 where @xmath59 is a free parameter of the compressibility . the value @xmath60 corresponds to the divergence - free ( incompressible ) advecting velocity field . it was used in refs.@xcite for determination of the influence of the compressibility on the anomalous scaling of the correlation functions of scalar density , as well as tracer fields in two - loop level . in what follows , we shall study the combined effects given by the small - scale anisotropy and compressibility and our aim will be to study possible deviations from the conclusions given in ref.@xcite , where the rapid - change model of passively advected scalar field with small scale uniaxial anisotropy was studied . to do this , it is necessary to introduce into the velocity field correlator ( [ corv ] ) corresponding second - rank tensor which will have needed properties . the simplest way how to do this is to add longitudinal projector @xmath61 to the uniaxial anisotropic transverse tensor structure given in eq.([t - ij ] ) . thus , the result tensor structure which will be used in our investigations is defined as @xmath62 it means that we have introduced the compressible term to the isotropic component of the axially anisotropic tensor structure ( [ t - ij ] ) only . therefore , the contributions of compressibility and anisotropy are given by a simple sum of the corresponding terms . nevertheless , as we shall see , the combined effects of anisotropy and compressibility on the results will not be a simple sum of them . according to the well - known general theorem ( see , e.g. , refs.@xcite ) the stochastic problem ( [ stoch1 ] ) and ( [ correlator ] ) is equivalent to the field - theoretic model of the set of three fields @xmath63 with the following action functional @xmath64^{-1 } v_j(t_2,{\bf x_2 } ) \nonumber \\ & & + \frac{1}{2 } \int dt_1\,d^d{\bf x_1}\,dt_2\,d^d{\bf x_2 } \ , \theta^{\prime}(t_1,{\bf x_1 } ) d^f(t_1,{\bf x_1};t_2,{\bf x_2 } ) \theta^{\prime}(t_2,{\bf x_2 } ) \nonumber \\ & & + \int dt\,d^d{\bf x}\,\ , \theta^{\prime}\left[-\partial_t - v_i\partial_i+\nu_0\triangle \right]\theta , \label{action1}\end{aligned}\ ] ] where @xmath65 is an auxiliary field , and all summations over the vector indices are implied . the second and the third integral in eq.([action1 ] ) correspond to the martin - siggia - rose action @xcite for the stochastic problem ( [ stoch1 ] ) , ( [ correlator ] ) at fixed velocity field @xmath66 , and the first integral describes the gaussian averaging over @xmath67 defined by the correlator @xmath68 in eq.([corv ] ) with tensor @xmath69 given in eq.([tg - ij ] ) . action ( [ action1 ] ) is given in a form convenient for application of the field theoretic perturbation analysis with the standard feynman diagrammatic technique . from the quadratic part of the action one obtains the matrix of bare propagators . the wave - number - frequency representation of propagators of the fields @xmath70 and @xmath65 is @xmath71 where @xmath72 is the fourier transform of the function @xmath73 form eq.([correlator ] ) . on the other hand , the bare propagator of the velocity field @xmath74 is defined by eq.([corv ] ) with the tensor structure given by eq.([tg - ij ] ) . in what follows , we shall need only the propagators @xmath75 and @xmath76 . their graphical representation is shown in fig.[fig1 ] . the interaction in the model is given by the nonlinear term @xmath77 with the vertex factor which in the wave - number - frequency representation has the form ( the momentum flows into the vertex via the scalar field @xmath70 ) : @xmath78 its graphical representation is given in fig.[fig1 ] . epsf = 6.0 cm = 6.0 cm standard power counting @xcite leads to the identification of correlation functions with superficial uv divergences . in the framework of the rapid - change passive advection models detail analysis of this question was done , e.g. , in ref.@xcite , where it was shown that the only one - particle - irreducible ( 1pi ) green function which possesses superficial uv - divergences is the function @xmath79 ( within the rapid - change model the situation is unchanged when compressibility of the system is assumed ) and , in the isotropic case , this green function leads only to the renormalization of the term @xmath80 of action ( [ action1 ] ) and the corresponding uv divergences may be fully absorbed in the adequate redefinition of the existing parameters @xmath35 , @xmath18 . thus , all correlation functions calculated in terms of the renormalized parameters @xmath81 are uv finite . the situation becomes , however , more complicated when anisotropy is introduced . it is related to the fact that in this case the 1pi green function @xmath82 contains divergences corresponding to the structure @xmath83 ( the only possible anisotropic structure @xmath84 ) which is not present in original unrenormalized action ( [ action1 ] ) . it leads to the non - renormalizability of the model in the anisotropic case . to make the model multiplicatively renormalizable it is necessary to extend original action ( [ action1 ] ) by including needed term with corresponding new parameter . as a result the extended model is described by the following action @xmath64^{-1 } v_j(t_2,{\bf x_2 } ) \nonumber \\ & & + \frac{1}{2 } \int dt_1\,d^d{\bf x_1}\,dt_2\,d^d{\bf x_2 } \ , \theta^{\prime}(t_1,{\bf x_1 } ) d^f(t_1,{\bf x_1};t_2,{\bf x_2 } ) \theta^{\prime}(t_2,{\bf x_2 } ) \label{action2 } \\ & & + \int dt\,d^d{\bf x } \theta^{\prime}\left[-\partial_t - v_i\partial_i+\nu_0\triangle + \chi_0 \nu_0 ( { \bf n}\cdot{\bf \partial})^2\right]\theta , \nonumber\end{aligned}\ ] ] where a new unrenormalized parameter @xmath85 has been introduced . as was pointed out in ref.@xcite the stability of the system requires the positivity of the total viscous contribution @xmath86 , i.e. , the inequality @xmath87 must be fulfilled . this modification leads , of course , also to the modification of the corresponding isotropic propagators of the fields @xmath88 given in eqs.([thetathetap ] ) and ( [ thetatheta ] ) which are now defined as ( see also ref.@xcite ) @xmath89 after this modifications the model defined by action ( [ action2 ] ) becomes multiplicatively renormalizable and the standard rg analysis can be now applied . the corresponding renormalized action has the form @xmath90^{-1 } v_j(t_2,{\bf x_2 } ) \nonumber \\ & & + \frac{1}{2 } \int dt_1\,d^d{\bf x_1}\,dt_2\,d^d{\bf x_2 } \ , \theta^{\prime}(t_1,{\bf x_1 } ) d^f(t_1,{\bf x_1};t_2,{\bf x_2 } ) \theta^{\prime}(t_2,{\bf x_2 } ) \label{action3 } \\ & & + \int dt\,d^d{\bf x } \theta^{\prime}\left[-\partial_t - v_i\partial_i+\nu z_1 \triangle + \chi \nu z_2 ( { \bf n}\cdot{\bf \partial})^2\right]\theta , \nonumber\end{aligned}\ ] ] where @xmath91 and @xmath92 are the renormalization constants ( they absorb the uv divergent parts of the 1pi function @xmath82 ) . it is equivalent to the multiplicative renormalization of the bare parameters @xmath93 , and @xmath85 , namely @xmath94 where @xmath81 , and @xmath95 are renormalized counterparts of the corresponding bare parameters , and @xmath96 is a scale setting parameter or the reference mass ( it has the same canonical dimension as the wave number ) . in what follows , we shall work in minimal subtraction scheme ( ms ) , therefore , in one - loop approximation , the renormalization constants @xmath97 have the form @xmath98 . thus , @xmath99 is a function of dimensionless parameters but it is independent of @xmath1 . by comparison of the corresponding terms in action ( [ action3 ] ) with definitions of the renormalization constants @xmath97 for parameters ( [ zetka ] ) , one obtains the following relations between renormalization constants : @xmath100 where last relation in eq.([zetka1 ] ) is a consequence of the fact that @xmath34 defined in eq.([corv ] ) is not renormalized , i.e. , @xmath101 . standardly , the formulation through the action functional ( [ action1 ] ) ( or through ( [ action3 ] ) in the anisotropic case ) replaces the statistical averages of random quantities in the stochastic problem defined by eqs.([stoch1 ] ) and ( [ correlator ] ) with equivalent functional averages with weight @xmath102 . generating functionals of total green functions g(a ) and connected green functions w(a ) are then defined by the functional integral @xmath103 where @xmath104 represents a set of arbitrary sources for the set of fields @xmath105 , @xmath106 denotes the measure of functional integration , and linear form @xmath107 is defined as @xmath108.\label{form}\ ] ] let us continue with renormalization of the model . the relation @xmath109 , where @xmath110 stands for the complete set of bare parameters and @xmath111 stands for renormalized one , leads to the relation @xmath112 for the generating functional of connected green functions . by application of the operator @xmath113 at fixed @xmath110 on both sides of the latest equation one obtains the basic rg differential equation @xmath114 where @xmath115 represents operation @xmath116 written in the renormalized variables . its explicit form is @xmath117 where we denote @xmath118 for any variable @xmath119 and the rg functions ( the @xmath120 and @xmath121 functions ) are given by well - known definitions ( see , e.g. , refs.@xcite ) and , in our case , by using relations ( [ zetka1 ] ) for renormalization constants , they have the following form @xmath122 the renormalization constants @xmath123 are determined by the requirement that the one - particle irreducible green function @xmath124 must be uv finite when is written in renormalized variables . in our case it means that they have no singularities in the limit @xmath125 . the one - particle irreducible green function @xmath126 is related to the self - energy operator @xmath127 by the dyson equation @xmath128 thus @xmath123 are found from the requirement that the uv divergences are canceled in eq.([dyson ] ) after substitutions @xmath129 and @xmath130 . this determines @xmath123 up to an uv finite contributions , which are fixed by the choice of the renormalization scheme . in the ms scheme , as was mentioned already above , all the renormalization constants have the form : 1 + _ pole in @xmath1_. in one - loop approximation the self - energy operator @xmath127 is represented by the corresponding one - particle irreducible diagram which is shown in fig.[fig3 ] but it must be stressed that , at the same time , it is an exact result because in rapid - change model all higher - loop diagrams contain at least one closed loop which is built on by retarded or advanced propagators only , thus they are automatically equal to zero . epsf the diagram in fig.[fig3 ] has the following analytical representation @xmath131 where @xmath132 denotes the d - dimensional sphere , and the functions @xmath133 and @xmath134 have the following simple explicit form @xmath135 in the end , the renormalization constants @xmath91 and @xmath92 are given as follows @xmath136 because of the second rank tensor ( [ tg - ij ] ) is taken in the simple form of the sum of the uniaxial anisotropic transverse projector and the longitudinal projector then , as a result of this fact , the self - energy operator @xmath127 is also a simple sum of the results obtained in refs.@xcite and @xcite . using the definition of anomalous dimensions @xmath137 in eq.([gammanu ] ) one comes to the following expressions @xmath138 where we denote @xmath139 , and @xmath140 are given in eqs.([aa1 ] ) and ( [ aa2 ] ) . let us note once more that the expressions ( [ zz1])-([gammas ] ) are exact as a result of non - existence of the higher - loop corrections . possible scaling regimes of a renormalizable model are directly given by the ir stable fixed points of the corresponding system of rg equations @xcite . the fixed point of the rg equations is defined by @xmath120-functions , namely , by requirement of their vanishing . in our model , the coordinates @xmath141 of a fixed point are found from the system of two equations @xmath142 the beta functions @xmath143 and @xmath144 are defined in eqs.([betag ] ) , and ( [ betau ] ) . to investigate the ir stability of a fixed point it is enough to analyze the eigenvalues of the matrix @xmath145 of the first derivatives : @xmath146 the ir asymptotic behavior is governed by the ir stable fixed points , i.e. , those for which both eigenvalues are positive . using the explicit expressions given in eq.([gammas ] ) together with the definitions of @xmath120 functions in eqs.([betag ] ) and ( [ betau ] ) leads to the explicit expressions for the coordinates of the non - trivial ir stable fixed point @xmath147 in the incompressible limit @xmath148 one comes to the results of ref.@xcite , namely @xmath149 on the other hand , in isotropic limit @xmath150 one has ( see , e.g. , ref.@xcite ) @xmath151 thus , as one can see , our definition of the second rank tensor ( [ tg - ij ] ) leads to the simple relation among @xmath152 , and @xmath153 , namely @xmath154 the values of anomalous dimensions @xmath155 and @xmath156 are found exactly at fixed point and are defined as follows @xmath157 where @xmath158 . the matrix of the first derivatives taken at the fixed point has simple diagonal form @xmath159 i.e. , the fixed point is ir stable if @xmath160 . this is also the only condition to have @xmath161 ( together , of course , with earlier discussed physical assumptions , namely : @xmath162 and @xmath59 , see sec.[sec : model ] ) . the physical condition @xmath163 is also fulfilled without further restrictions on the parameter space . the issue of interest are especially multiplicatively renormalizable equal - time two - point quantities @xmath164 ( see , e.g. , @xcite ) . the example of such quantity are the equal - time structure functions @xmath165^{n}\rangle,\,\ , r=|{\bf x}-{\bf x^{\prime}}| \label{struc}\ ] ] in the inertial range specified by the inequalities @xmath166 ( @xmath167 is an internal length ) . here parentheses @xmath168 mean functional average over fields @xmath169 with weight @xmath170 . first let us describe briefly ir scaling behavior in general on the example of an equal - time function @xmath164 which is multiplicatively renormalizable . the ir scaling behavior of the function @xmath164 ( for @xmath171 and any fixed @xmath172 ) @xmath173 is related to the existence of ir stable fixed points of the rg equations ( see above ) . in eq.([frscaling ] ) @xmath174 and @xmath175 are corresponding canonical dimensions of the function @xmath176 : @xmath174 is the frequency dimension , and @xmath175 is the total canonical dimension . they are related by the relation @xmath177 , where @xmath178 is corresponding momentum canonical dimension . the existence of two independent canonical dimensions is related to the fact that our model belongs among two - scale dynamical models ( details see , e.g. , in refs.@xcite ) . in eq.([frscaling ] ) @xmath179 is so - called scaling function which can not be determined by rg equation ( see , e.g. , @xcite ) , and @xmath180 is the critical dimension defined as @xmath181 here @xmath182 is the fixed point value of the anomalous dimension @xmath183 , where @xmath184 is renormalization constant of multiplicatively renormalizable quantity @xmath176 , i.e. , @xmath185 @xcite , and @xmath186 is the critical dimension of frequency with @xmath187 as it is shown in eq.([gammyfix ] ) . now , let us apply the above discussion to the inertial - range analysis of the equal - time structure functions as defined in eq.([struc ] ) . it is well - known that , in the isotropic case , the odd functions @xmath188 vanish , while for @xmath189 simple dimensional considerations give @xmath190 where @xmath191 are some functions of dimensionless variables . first , the multiplicative renormalizability of the model leads to the existence of differential rg equations for these structure functions and their asymptotic behavior for @xmath192 and any fixed @xmath172 is given by ir stable fixed point of the rg equations and the structure functions can be written in the following form @xmath193 with unknown scaling functions @xmath194 . in the theory of critical phenomena @xcite the quantity @xmath195 is known as `` anomalous dimension '' and @xmath196\equiv-2n + \gamma_{n}$ ] is termed the `` critical dimension '' ( see above ) , where @xmath197 is the corresponding `` canonical dimension '' . in our case , @xmath198 @xcite , i.e. , representation ( [ strucdim2 ] ) implies the existence of a scaling in the ir region ( @xmath199 , @xmath172 fixed ) with definite critical dimensions of all `` ir relevant '' parameters , @xmath196 = -2n(1- \varepsilon)$ ] , @xmath200 , @xmath201 and fixed `` irrelevant '' parameters @xmath18 and @xmath167 regardless of the form of the functions @xmath202 . the second stage of rg analysis is associated with the investigation of small @xmath172 behavior of the functions @xmath203 in eq.([strucdim2 ] ) using the ope . it shows that , in the limit @xmath204 , the functions @xmath202 have the asymptotic form @xmath205 where @xmath206 are some coefficients regular in @xmath172 and the summation is implied over certain renormalized composite operators @xmath207 with critical dimensions @xmath208 @xcite . in the case under consideration , the leading operators @xmath207 have the form @xmath209 . in sec.[sec6 ] we shall consider them in detail where the complete calculation of the critical dimensions of the composite operators @xmath210 will be present for arbitrary values of @xmath211 , @xmath32 , @xmath12 and @xmath212 . epsf = 3.6 cm let us now study the behavior of the scaling function in eq.([frscaling ] ) . according to the ope @xcite , the equal - time product @xmath213 of two renormalized composite operators . constructions @xmath214 and @xmath215^n$ ] are typical examples ] at @xmath216 and @xmath217 can be written in the following form @xmath218 where summation is taken over all possible renormalized local composite operators @xmath219 allowed by symmetry with definite critical dimensions @xmath220 , and the functions @xmath221 are the corresponding wilson coefficients regular in @xmath222 . the renormalized correlation function @xmath223 can be now found by averaging eq.([fff ] ) with the weight @xmath224 with @xmath225 from eq.([action3 ] ) . the quantities @xmath226 appear on the right - hand side , and their asymptotic behavior in the limit @xmath227 is then found from the corresponding rg equations and has the form @xmath228 . from the ope ( [ fff ] ) one can find that the scaling function @xmath179 in the representation ( [ frscaling ] ) for the correlation function @xmath213 has the form @xmath229 where the coefficients @xmath221 are regular in @xmath230 . the principal feature of the turbulence models is the existence of the so - called `` dangerous '' operators with negative critical dimensions @xcite . their presence in the ope determines the ir behavior of the scaling functions and leads to their singular dependence on @xmath28 when @xmath231 . therefore , the turbulence models are crucially different from the models of critical phenomena , where the leading contribution to the representation ( [ ff1 ] ) is given by the simplest operator @xmath232 with the dimension @xmath233 , and the other operators determine only the corrections that vanish for @xmath231 . epsf = 8.5 cm = 8.5 cm if the spectrum of the dimensions @xmath220 for a given scaling function is bounded from below , the leading term of its behavior for @xmath234 is given by the minimal dimension . it will be our case and for the investigation of the anomalous scaling of the structure functions ( [ struc ] ) the leading contribution of the taylor expansion is given by the tensor composite operators constructed solely of the scalar gradients ( see , e.g. , ref.@xcite for details ) @xmath235\equiv \partial_{i_1}\theta \cdots \partial_{i_p}\theta ( \partial_{i}\theta \partial_{i } \theta)^n , \label{composite}\ ] ] where @xmath236 is the total number of the fields @xmath70 entering into the operator and @xmath237 is the number of the free vector indices . epsf = 8.5 cm = 8.5 cm here we shall briefly discuss the renormalization of the composite operators ( [ composite ] ) which play central role in our investigation ( complete and detailed discussion of the renormalization of this composite operators can be found in ref.@xcite ) but first we describe the basis of general theory . epsf = 8.5 cm = 8.5 cm the necessity of additional renormalization of the composite operators ( [ composite ] ) is related to the fact that the coincidence of the field arguments in green functions containing them leads to additional uv divergences . these divergences must be removed by special kind of renormalization procedure which can be found , e.g. , in refs.@xcite , where their renormalization is studied in general . the renormalization of composite operators in the models of turbulence is discussed in refs.@xcite . besides , typically , the composite operators are mixed under renormalization . epsf = 8.5 cm = 8.5 cm let @xmath239 be a closed set of composite operators which are mixed only with each other in renormalization . then the renormalization matrix @xmath240 and the matrix of corresponding anomalous dimensions @xmath241 for this set are given as follows @xmath242 renormalized composite operators are subject to the following rg differential equations @xmath243 which lead to the following matrix of critical dimensions @xmath244 @xmath245 where @xmath246 a @xmath247 are diagonal matrices of corresponding canonical dimensions and @xmath248 is the matrix of anomalous dimensions ( [ 2.2 ] ) taken at the fixed point . in the end , the critical dimensions of the set of operators @xmath249 are given by the eigenvalues of the matrix @xmath250 . the so - called `` basis '' operators that possess definite critical dimensions have the form @xmath251 where the matrix @xmath252 is such that @xmath253 is diagonal . epsf = 8.5 cm = 8.5 cm epsf = 8.5 cm = 8.5 cm as was already mentioned , in our case of a scalar admixture with anisotropy and compressibility the central role is played by the tensor composite operators @xmath254 , constructed solely of the scalar gradients @xcite . it is convenient to deal with the scalar operators obtained by contracting the tensors with the appropriate number of the vectors @xmath47 , namely @xmath235\equiv[(\mathbf{n\partial})\theta]^{p}(\partial_{i}\theta\partial_{i}\theta)^{n},\quad n\equiv2n+p.\label{fnp}\ ] ] detail analysis shows that the composite operators ( [ fnp ] ) with different @xmath255 are not mixed with each other in renormalization , that is why the corresponding infinite renormalization matrix @xmath235=z_{[n , p][n^{\prime},p^{\prime}]}f_r[n^{\prime},p^{\prime}]\,\label{renfk}\ ] ] is block - triangular , i.e. , @xmath256[n^{\prime},p^{\prime}]}=0 $ ] for @xmath257 . thus , the critical dimensions associated with the operator @xmath238 $ ] are completely determined by the eigenvalues of the subblocks with @xmath258 . epsf = 8.5 cm = 8.5 cm in the isotropic case , as well as in the case when large - scale anisotropy is present , the elements @xmath256\,[n , p']}$ ] vanish for @xmath259 , therefore the block @xmath256\,[n , p']}$ ] is triangular . the same is valid for the matrices @xmath260 and @xmath261 defined in eqs.([2.5 ] ) and ( [ 32b ] ) . thus , the presence of large - scale anisotropy does not affect critical dimensions of the operators ( [ fnp ] ) . situation is radically different in the case when small - scale anisotropy is present , where the operators with different values of @xmath237 mix in renormalization in such a way that the matrix @xmath256\,[n , p']}$ ] is not triangular and one can write @xmath235=\sum_{l=0}^{\lfloor n/2 \rfloor } z_{[n , p]\,[n , n-2l ] } f^r[n , n-2l]\ , , \label{fnl}\ ] ] where @xmath262 means the integer part of the @xmath263 . therefore , each block of renormalization constants with given @xmath255 is an @xmath264 matrix . of course , the matrix of critical dimensions ( [ 32b ] ) , whose eigenvalues at ir stable fixed point are the critical dimensions @xmath265 $ ] of the set of operators @xmath238 $ ] , has also dimension @xmath264 . epsf = 8.5 cm = 8.5 cm epsf = 8.5 cm = 8.5 cm in what follows , we shall calculate the matrix @xmath256[n^{\prime},p^{\prime}]}$ ] in one - loop approximation ( note that in spite of the uv renormalization of our model , where one - loop result is the complete solution of the problem because of nonexistence of higher - loop corrections , the renormalization of the composite operators has two- and higher - loop corrections ) . it was analyzed in detail in ref.@xcite , therefore we shall not repeat it here , and we confine ourself only to the necessary information . we are interested in @xmath255-th term , denoted as @xmath266 $ ] , of the expansion of the generating functional of one - particle - irreducible green functions with one composite operator @xmath238 $ ] and any number of fields @xmath70 , denoted as @xmath267 $ ] . it has the following form @xmath268(x)\theta(x_{1})\cdots\theta(x_{n})\rangle_{\textrm{1-ir}}.\label{gamma1}\ ] ] in one - loop approximation the function ( [ gamma1 ] ) is given as @xmath269+\gamma^{(1)}\ , , \label{gamma2}\ ] ] where @xmath270 is given by the analytical calculation of the diagram in fig.[fig4 ] ( see , e.g. , ref.@xcite for details ) and the first term in eq . ( [ gamma2 ] ) represents `` tree '' approximation . after cumbersome but straightforward calculations ( see ref.@xcite ) one finds the uv divergent part of @xmath270 from eq.([gamma2 ] ) in the following appropriate form @xmath271 , \label{oper1}\ ] ] with @xmath272 where @xmath273 are corresponding hypergeometric functions , and coefficients @xmath274 and @xmath275 for @xmath276 and @xmath277 are given in appendix a. epsf = 8.5 cm = 8.5 cm epsf = 8.5 cm = 8.5 cm the renormalization matrix @xmath256\,[n , p']}$ ] is then found from the condition that @xmath278 $ ] from eq.([gamma2 ] ) must be free of uv divergences when are written in renormalized variables , i.e. , it does not contain poles in @xmath1 . in the ms scheme one obtains @xmath279[n , p+2(i-2 ) ] } = \delta_{2i}+\frac{g\ , s_{d}}{(2\pi)^d}\frac{a}{8\varepsilon}\ , q_{i } \label{znp}\ ] ] where @xmath276 , @xmath280 , and coefficients @xmath281 are given in eq.([oper2 ] ) . in the end , from eq.([znp ] ) and using definitions ( [ 2.2 ] ) , one finds the following expressions for the elements of the matrix of anomalous dimensions @xmath282\,[n',p']}$ ] @xmath283[n , p+2(i-2 ) ] } = -\frac{g\ , s_{d}}{(2\pi)^d}\frac{a}{4}\ , q_{i}\ ] ] for @xmath276 and the matrix of critical dimensions ( [ 32b ] ) has the form @xmath284[n , p']}=n\,\varepsilon+\gamma_{[n , p][n , p']}^{*},\label{dnp}\ ] ] where the asterisk means that @xmath282[n , p']}$ ] is taken at the fixed point given by eqs.([gaa1 ] ) and ( [ haa2 ] ) . eq.([dnp ] ) , which depends on the anisotropy parameters @xmath285 , as well as on the compressibility parameter @xmath12 , is desired one - loop expression for the matrix of critical dimensions of the composite operators ( [ fnp ] ) . the critical dimensions of the operators @xmath238 $ ] , which we denote as @xmath265 $ ] , are after all equal to the eigenvalues of the matrix of critical dimensions ( [ dnp ] ) . in the isotropic case or in the case with large - scale anisotropy , where the matrix ( [ dnp ] ) is triangular , the eigenvalues are equal directly to the diagonal elements of the matrix . this fact allows us to assign uniquely the concrete critical dimension to the corresponding composite operator even in the case with small - scale anisotropy and study their hierarchical structure as a functions of @xmath237 ( see @xcite for details ) . in ref.@xcite it was shown that some of the critical dimensions are negative , therefore they lead to the anomalous scaling ( singular behavior of the scaling functions ) . at the same time , the leading role is played by the operators with the most negative critical dimensions which are the operators with @xmath286 for the structure functions ( [ struc ] ) with even @xmath255 , and the operators with @xmath287 for the structure functions with odd @xmath255 . thus , the combination of the rg representation with the ope ( [ ope ] ) leads to the final asymptotic expression for the structure functions ( [ struc ] ) within the inertial range @xmath288}+\dots\}\,,\label{struc10}\ ] ] where @xmath237 obtains all possible values for given @xmath289 , @xmath290 are numerical coefficients which are functions of the parameters of the model , and dots means contributions by the operators others than @xmath238 $ ] ( see , e.g. , @xcite for details ) . in what follows , we shall investigate the influence of compressibility on the picture found in ref.@xcite . our aim is to find the answer on the question whether the presence of compressibility leads to the more pronounced anomalous scaling of structure functions of the scalar field or not . another purpose of the present analysis is to check whether the combine effects defined by compressibility and small - scale anisotropy can lead to the more complicated structure of critical dimensions than it was studied in ref.@xcite ( see also ref.@xcite ) . one possibility is that the pairs of complex conjugate eigenvalues of the matrix of critical dimensions can exist . it leads to the oscillation behavior of the corresponding scaling function @xcite , i.e. , the scaling functions in eq .. ([struc10 ] ) would contain terms of the following form @xmath291 + c_2 \sin\left[\delta_i ( r / l ) \right]\right\},\ ] ] where @xmath292 and @xmath293 are real and imaginary part of @xmath294 , and @xmath295 are constants . another possibility is related to the situation if the matrix of critical dimensions can not be diagonalized and has only the jordan form . then a logarithmic correction would be involved to the powerlike behavior ( see ref.@xcite ) . epsf = 8.5 cm = 8.5 cm behavior of the eigenvalues of the matrix of critical dimensions @xmath265 $ ] for various values of @xmath255 for different values of the compressibility parameter @xmath12 and as the functions of the anisotropy parameters @xmath296 and @xmath297 are shown in figs.[fig56]-[fig2728 ] . it can be immediately seen that only real eigenvalues of the corresponding matrix exist as in the incompressible case studied in ref.@xcite . at the same time , the hierarchical structure of the critical dimensions for different values of @xmath237 and the same @xmath255 shown in ref.@xcite is also preserved . the dependence of the critical dimension @xmath298 $ ] is not shown because it is identically equal to zero . it can be shown either by direct calculation or by using the schwinger equation ( see , e.g. , ref.@xcite ) . all the figures show that compressibility leads to smaller values of the critical dimensions of the corresponding composite operators , thus , at the same time , to the more noticeable anomalous scaling . as it is shown in the figures , the effects of compressibility are the most pronounced for small values of the anisotropy parameters , as well as for the negative values of these parameters . on the other hand , in the limit of large positive values of anisotropy parameters @xmath296 and @xmath297 the effects of compressibility tend to zero . especially , in the limit @xmath299 the influence of compressibility is vanished completely which can be shown by direct analytical investigation . in this paper we have analyzed asymptotic behavior of the single - time structure functions @xmath300 of a passively advected scalar field by the compressible velocity field with small - scale anisotropy by using the field - theoretic renormalization group and the operator product expansion in a minimal - subtraction scheme of analytical renormalization . it is shown that the leading - order powerlike asymptotic behavior of the single - time structure functions of a scalar field within the inertial range in our compressible anisotropic case is given by the critical dimensions of the same composite operators as in the incompressible anisotropic case but now they acquire rather strong dependence on the compressibility parameter . further , it is shown that when the parameter of compressibility is increasing then the critical dimensions of relevant composite operators become smaller , therefore , within our model , we can conclude that the anomalous scaling of the structure functions of a passive scalar quantity advected by a given velocity field is more pronounced in the compressible stochastic environment than in the incompressible one ( see figs.[fig56]-[fig2728 ] ) . concrete calculations were done up to the structure functions of order @xmath301 . in our calculations we have not found possible oscillatory modulation ( related to the possible existence of the complex conjugate eigenvalues of the matrix of critical dimensions ) , as well as we have not found possible logarithmic corrections ( related to the fact that the corresponding matrix can has only jordan form ) to the leading powerlike asymptotic . thus , all calculated corrections have had purely powerlike behavior . it is also shown that the critical dimensions are ordered hierarchically as in the incompressible case @xcite , i.e. , the compressibility does not disturb the anisotropic structure of the critical dimensions rather it shifts them toward smaller values . the largest shift of the critical dimensions is present for negative and small positive values of the anisotropy parameters . on the other hand , when the anisotropy parameters tend to infinity the effects of compressibility vanish . the explicit form of the coefficients @xmath274 and @xmath275 with @xmath276 and @xmath277 ( see eq.([oper2 ] ) ) is @xmath302 @xmath303 @xmath304 @xmath305 @xmath306 @xmath307 @xmath308 @xmath309 @xmath310 @xmath311 @xmath312 @xmath313 99 r. h. kraichnan , phys . fluids * 11 * , 945 ( 1968 ) . a. s. monin , a. m yaglom , _ statistical fluid mechanics _ ( mit press , cambridge , ma , 1975 ) , vol . 2 . , edited by j. c. r. hunt , o. m. philips , and d. williams [ proc . london , ser . a * 434 * , 1 - 240 ( 1991 ) ] . w. d. mccomb , _ the physics of fluid turbulence _ ( clarendon , oxford , 1990 ) . u. frisch , _ turbulence : the legacy of a.n . ( cambridge university press , cambridge , 1995 ) . r. a. antonia , e. j. hopfinger , y. gagne , and f. anselmet , phys . a * 30 * , 2704 ( 1984 ) . k. r. sreenivasan , proc . london , ser . a * 434 * , 165 ( 1991 ) . m. holzer and e. d. siggia , phys . fluids * 6 * , 1820 ( 1994 ) . a. pumir , phys . fluids * 6 * , 2118 ( 1994 ) . c. tong and z. warhaft , phys . fluids * 6 * , 2165 ( 1994 ) . t. elperin , n. kleeorin , and i. rogachevskij , phys . e * 52 * , 2617 ( 1995 ) ; phys . . lett . * 76 * , 224 ( 1996 ) ; phys . e * 53 * , 3431 ( 1996 ) . k. gawedzki and a. kupiainen , phys . lett . * 75 * ( 1995 ) 3834 ; d. bernard , k. gawedzki , and a. kupiainen , phys . rev . e * 54 * ( 1996 ) 2564 ; m. chertkov , g. falkovich , i. kolokov , and v. lebedev , phys . e * 52 * ( 1995 ) 4924 ; m. chertkov and g. falkovich , phys . lett . * 76 * ( 1996 ) 2706 . b. i. shraiman and e. d. siggia , phys . rev . lett . * 77 * , 2463 ( 1996 ) ; a. pumir , b. i. shraiman , and e. d. siggia , phys . rev . e * 55 * , r1263 ( 1997 ) . a. pumir , europhys . lett . * 34 * , 25 ( 1996 ) ; * 37 * , 529 ( 1997 ) ; phys . e * 57 * , 2914 ( 1998 ) . m. avellaneda and a. majda , commun . phys . * 131 * , 381 ( 1990 ) ; * 146 * , 139 ( 1992 ) ; a. majda , j. stat . phys . * 73 * , 515 ( 1993 ) ; d. horntrop and a. majda , j. math . tokyo * 1 * , 23 ( 1994 ) . q. zhang and j. glimm , commun . phys . * 146 * , 217 ( 1992 ) . r. h. kraichnan , phys . lett . * 72 * , 1016 ( 1994 ) . r. h. kraichnan , v. yakhot , and s. chen , phys . lett . * 75 * , 240 ( 1995 ) . g. falkovich , k. gawdzki , m. vergassola , rev . mod . phys . * 73 * , 913 ( 2001 ) . j. zinn - justin , _ quantum field theory and critical phenomena _ ( clarendon , oxford , 1989 ) . a. n. vasilev , _ quantum - field renormalization group in the theory of critical phenomena and stochastic dynamics _ ( st . petersburg institute of nuclear physics , st . petersburg , 1998 ) [ in russian ; english translation : gordon & breach , 2004 ] . c. de domoinicis and p.c . martin , phys . a * 19 * , 419 ( 1979 ) . l. ts . adzhemyan , a. n. vasilev , and yu . m. pismak , theor . phys . * 57 * , 1131 ( 1983 ) . l. ts . adzhemyan , a. n. vasilev , and m. hnatich , theor . math . phys . * 58 * , 47 ( 1984 ) . l. ts . adzhemyan , n. v. antonov , and a. n. vasilev , usp . nauk * 166 * , 1257 ( 1996 ) [ phys . usp . * 39 * , 1193 ( 1996 ) ] . l. ts . adzhemyan , n. v. antonov , and a. n. vasilev , _ the field theoretic renormalization group in fully developed turbulence _ ( gordon @xmath318 breach , london , 1999 ) . l. ts . adzhemyan , n. v. antonov , and a. n. vasilev , phys . e * 58 * , 1823 ( 1998 ) ; theor . 120 * , 1074 ( 1999 ) . l. ts . adzhemyan , n. v. antonov , v. a. barinov , yu . s. kabrits , and a. n. vasilev , phys . e * 64 * , 056306 ( 2001 ) ; _ ibid . _ * 63 * , 025303(r ) ( 2001 ) . adzhemyan , n. v. antonov , m. hnatich , and s. v. novikov , phys . e * 63 * , 016309 ( 2000 ) . l. ts . adzhemyan , and n. v. antonov , phys . e * 58 * , 7381 ( 1998 ) . n. v. antonov , and j. honkonen , phys . e * 63 * , 036302 ( 2001 ) . n. v. antonov , phys . e * 60 * , 6691 ( 1999 ) . n. v. antonov , physica d * 144 * , 370 ( 2000 ) ; zap . pomi * 269 * , 79 ( 2000 ) . l. ts . adzhemyan , n. v. antonov and j. honkonen , phys . e * 66 * , 036313 ( 2002 ) ; m. hnati , e. juriinov , m. juriin and m. repaan , j. phys . a : math * 39 * , 8007 ( 2006 ) ; o. g. chkhetiani , m. hnatich , e. juriinov , m. juriin , a mazzino , and m. repaan , czech . j. phys . * 56 * , 827 ( 2006 ) ; j. phys . a : math . gen . * 39 * , 7913 ( 2006 ) ; phys . e * 74 * , 036310 ( 2006 ) . a. lanotte and a. mazzino , phys . e * 60 * , r3483 ( 1999 ) ; n. v. antonov , a. lanotte , and a. mazzino , _ ibid . _ * 61 * , 6586 ( 2000 ) ; a. celani , a. lanotte , a. mazzino , and m. vergassola , phys . rev . lett . * 84 * , 2385 ( 2000 ) . i. arad , l. biferale , and i. procaccia , phys . e * 61 * , 2654 ( 2000 ) ; i. arad , v. lvov , e. podivilov , and i. procaccia , _ ibid . _ * 62 * , 4904 ( 2000 ) . s. kurien , k. g. aivalis , and k. r. sreenivasan , j. fluid mech . * 448 * , 279 ( 2001 ) ; m. m. afonso and m. sbragaglia , j. turbulence * 6 * , 10 ( 2005 ) ; s. g. saddoughi and s. v. veeravalli , j. fluid mech . * 268 * , 333 ( 1994 ) ; v. borue and s. a. orszag , _ ibid . _ * 306 * , 293 ( 1996 ) . i. arad , b. dhruva , s. kurien , v. s. lvov , i. procaccia , and k. r. sreenivasan , phys . lett . * 81 * , 5330 ( 1998 ) ; i. arad , l. biferale , i. mazzitelli , and i. procaccia , _ ibid . _ * 82 * , 5040 ( 1999 ) ; s. kurien , v. s. lvov , i. procaccia , and k. r. sreenivasan , phys . e * 61 * , 407 ( 2000 ) ; s. kurien and k. r. sreenivasan , _ ibid . _ * 62 * , 2206 ( 2000 ) ; i. arad , v. s. lvov , and i. procaccia , physica a * 288 * , 280 ( 2000 ) ; l. biferale and f. toschi , phys . lett . * 86 * , 4831 ( 2001 ) ; i. arad and i. procaccia , phys . e * 63 * , 056302 ( 2001 ) ; l. biferale , i. daumont , a. lanotte , and f. toschi , _ ibid . _ * 66 * , 056306 ( 2002 ) ; v. s. lvov , i. procaccia , and v. tiberkevich , _ ibid . _ * 67 * , 026312 ( 2003 ) ; l. biferale , g. boffetta , a. celani , a. lanotte , and f. toschi , phys . fluids * 15 * , 2105 ( 2003 ) . k. yoshida and y. kaneda , phys . e * 63 * , 016308 ( 2001 ) ; k. yoshida , t. ishihara , and y. kaneda , phys . fluids * 15 * , 2385 ( 2003 ) . l. biferale and i. procaccia , phys . rep . * 414 * , 43 ( 2005 ) . a. bigazzi , l. biferale , s. m. a. gama , and m. velli , astrophys . j. * 638 * , 499 ( 2006 ) . l. sorriso - valvo , v. carbone , r. bruno , and p. veltri , europhys . lett . * 75 * , 832 ( 2006 ) . l. d. landau and e. m. lifshitz , _ fluid mechanics _ , 2nd ed . ( pergamon press , oxford , 1987 ) . a. m. obukhov , izv . nauk sssr , geogr . geofiz . * 13 * , 58 ( 1949 ) . g. eyink , phys . e * 54 * , 1497 ( 1996 ) ; g. eyink and j. xin , phys . rev . lett . * 77 * , 2674 ( 1996 ) . r. h. kraichnan , phys . lett . * 78 * , 4922 ( 1997 ) . i. s. gradshtejn and i. m. ryzhik , _ tables of integrals , series and products _ ( academic , new york , 1965 ) . p. c. martin , e. d. siggia , h. a. rose , phys . rev . a * 8 * , 423 ( 1973 ) . j. c. collins , _ renormalization . an introduction to renormalization , the renormalization group , and the operator product expansion _ ( cambridge university press , cambridge , 1984 ) .

the mass function ( himf ) for galaxies in the local universe is constructed from the results of the arecibo strip survey , a blind extragalactic survey in the 21 cm line . the survey consists of two strips covering in total square degrees of sky , with a depth of and was optimized to detect column densities of neutral gas ( ) . the survey yielded 66 significant extragalactic signals of which approximately 50% are cataloged galaxies . no free floating clouds without stars are found . vla follow - up observations of all signals have been used to obtain better measurements of the positions and fluxes and allow an alternate determination of the achieved survey sensitivity . the resulting himf has a shallow faint end slope ( ) , and is consistent with earlier estimates computed for the population of optically selected gas rich galaxies . this implies that there is not a large population of gas rich low luminosity or low surface brightness galaxies that has gone unnoticed by optical surveys . the influence of large scale structure on the determination of the himf from the arecibo strip survey is tested by numerical experiments and was not found to affect the resulting himf significantly . the cosmological mass density of at the present time determined from the survey , , is in good agreement with earlier estimates . we determine lower limits to the average column densities of the galaxies detected in the survey and find that none of the galaxies have , although there are no observational selection criteria against finding lower density systems . eight percent of the signals detected in the original survey originated in groups of galaxies , whose signals chanced to coincide in frequency .

astro-ph9707109

the distribution function of neutral hydrogen masses among galaxies and intergalactic clouds ( the mass function or himf ) , and more generally , the neutral hydrogen density in the nearby universe , @xmath9 , are important inputs into models of cosmology and galaxy evolution . different attempts have been made to construct an himf by using optically selected galaxies ( rao & briggs 1993 , hereafter rb , solanes , giovanelli & haynes 1996 ) . these studies are based on the assumption that is always associated with optically bright galaxies . a major concern is whether the himf is complete when it is computed for these galaxies . for example , the population of low surface brightness ( lsb ) galaxies might hypothetically constitute a substantial portion of the population of nearby extragalactic objects ( disney 1976 , mcgaugh 1996 , dalcanton , spergel & summers 1997 , sprayberry et al . the lsb population easily escapes detection optically and would not be included in the samples that are commonly used to evaluate the himf . this would be particularly problematic since lsb galaxies are generally found to be rich in neutral gas ( schombert et al . 1992 , de blok , mcgaugh & van der hulst 1996 ) and could therefore contribute substantially to the neutral gas content . gas rich dwarf galaxies may also play an important part . for example tyson & scalo ( 1988 ) have argued that the majority of these dwarf galaxies remain undetected because only the small portion that is currently undergoing a rapid phase of star formation is presently observed with optical telescopes . a final possible population of gas rich systems that would escape inclusion in optical catalogs is a class of intergalactic clouds without stars . so far , only a few such systems have been discovered , and they are always found to be gravitationally bound to a galaxy or a group of galaxies ( for example the leo ring , schneider 1989 ) . clearly , the himf and the content of the local universe should be measured directly , in such a way that they suffer no bias against gas rich galaxies or intergalactic clouds which are difficult to detect optically . this is possible by means of surveys in the line . several of these surveys have been carried out over the last two decades . the majority were single dish observations using on / off techniques , since these surveys were done in conjunction with observations targeted on cataloged galaxies ( fisher & tully 1981b , giovanelli & haynes 1985 , 1989 ) . some surveys were concentrated on groups of galaxies ( haynes & roberts 1979 , lo & sargent 1979 , fisher & tully 1981a , hoffman , lu & salpeter 1992 ) to specifically search for clouds in the vicinity of known galaxies . other surveys were designed to find signals in voids ( krumm & brosch 1984 , szomoru et al 1996 ) , or to compare voids and superclusters ( weinberg et al . 1991 , szomoru et al . since these surveys are not pointed at randomly chosen regions of sky , they may not provide fair tests of the shape of the himf or an unbiased measure of the average density . a few truly blind surveys have been conducted , the first one in driftscan mode ( shostak 1977 ) and one by observing a series of pointings on lines of constant declination ( kerr & henning 1987 , henning 1992 ) . it is worrisome that this latter survey could not reproduce the himf defined for optically selected galaxies , possibly because the survey was targeted toward large volumes of known voids . another possibility is that the achieved survey sensitivity is not well understood , leading to an underestimation of the true number density of rich galaxies and intergalactic clouds . more recently , several surveys have been made using the arecibo telescope ( sorar 1994 , spitzak 1996 , schneider 1997 ) . surveys are in progress at dwingeloo ( dogs ) and at parkes , where the survey will cover the entire southern sky ( staveley - smith et al . 1996 ) . in this paper we analyze the arecibo strip survey ( sorar 1994 ) , a blind survey for extragalactic covering @xmath0 square degrees of sky , out to a redshift of 7400 @xmath2 . the analysis of the survey results will concentrate specifically on the understanding of the achieved survey sensitivity and the vulnerability to large scale structure . we describe the details of the survey and the optical and 21 cm follow - up observations in section [ observations.sect ] . section [ sensitivity.sect ] gives a detailed analysis of the achieved survey sensitivity . we present the himf in section [ himf.sect ] , the possible influence of large scale structure on the determination of the himf is examined by performing numerical experiments , and the cosmological mass density of at the present time is calculated in this section . in section [ discussion.sect ] we compare our findings with previous estimates of the himf based on 21 cm surveys and optically selected galaxy samples and discuss the implications of our results . section [ conclusions.sect ] summarizes the results . the distances used in this paper are based on a hubble constant @xmath10 . the strip survey was carried out on the arecibo 305 m telescope in the period of august 1993 until february 1994 . the survey was designed to take advantage of periods of construction when the telescope pointing was immobilized but the receiving systems were still operational . therefore , the data were taken in driftscan mode and the telescope beam traced strips of constant declination . the same strips were retraced for as many as 30 days in order to obtain very sensitive observations that are capable of detecting of low surface density . the main survey was divided in two sessions : one survey covering 10.5 hours of ra at @xmath11 , and the second survey covering 9.7 hours of ra at @xmath12 . all observations were made at night . the limiting column density was @xmath13 ( at a @xmath4 level ) for gas filling the telescope beam , which subtends 3 @xmath14 kpc at 3 @xmath14 mpc and 70 @xmath14 kpc at 74 @xmath14 mpc . the total sky coverage was approximately 65 square degrees , with a depth of @xmath1 @xmath2 . the survey was capable of detecting masses of @xmath15 at 7 @xmath14 mpc and @xmath16 at the full depth of the survey , in the main beam which has a fwhm of 3 arcmin . the first sidelobe of the telescope beam pattern could detect @xmath17 at the full depth of the survey , which makes the sidelobes effective in detecting high mass galaxies over a 15 arcmin wide strip . the details of the ahiss and the data reduction are described by sorar ( 1994 ) . a summary of the reduction path is described by briggs et al . ( 1997 ) . the survey yielded a total of 61 detections , of which approximately half could be associated with cataloged galaxies listed in the nasa extragalactic database ( ned ) . five detections have no obvious counterparts on the digitized sky survey ( dss ) although they are more than @xmath18 away from the galactic plane . fig . [ slices.fig ] shows slice diagrams of the location of all selected objects compared with galaxies in the cfa catalog ( geller & huchra 1989 ) . galaxies within a @xmath19 strip centered on the declination of the strip are plotted . different symbols are used to distinguish between cataloged ( circles ) and uncataloged ( boxes ) galaxies . the size of the symbols reflects the mass of the galaxies , in the sense that bigger symbols indicate larger masses . the galactic plane intersects the strips at @xmath20 and @xmath21 as indicated by the dashed lines . the cfa catalog is obviously incomplete in these regions of the sky due to galactic extinction , whereas the survey suffers no bias against detection in these regions . it is clear from this figure that selected galaxies generally follow the same structures as the optical galaxies . in particular , if we only consider those regions of the sky where the cfa catalog is complete , we find that more than 80% of the selected galaxies lie in regions where the average galaxies density is higher than the mean density . this is consistent with the finding that lsb galaxies and gas - rich dwarfs lie on structures delineated by normal , high surface brightness galaxies ( bothun et al . 1986 , thuan , gott & schneider 1987 , eder et al . 1989 , thuan et al . 1991 , mo , mcgaugh & bothun 1994 ) . furthermore , none of the selected galaxies in the sections where the cfa is complete are found in regions where the galaxy density is less than one fifth of the cosmic mean . this is in agreement with the results of szomoru et al . ( 1994 ) and weinberg et al . ( 1991 ) that selected galaxies are not found in selected void fields . a more detailed analysis of the large scale distribution of the selected galaxies will be presented elsewhere . = 11.5 cm follow - up 21 cm synthesis observations on all signals found in this survey have been performed with the nrao very large array ( vla ) in d - configuration . these follow - up observations are essential for three reasons : + 1 ) as stated before , the survey was capable of detecting signals as far as 7 arcmin from the center of the survey strip . consequently , the coordinates yielded by the survey have uncertainties on the order of several arcminutes . the association with a cataloged galaxy or a galaxy on the dss is therefore not always unambiguous , especially in those cases where there are several prominent galaxies visible within a 15 arcmin wide strip . 21 cm observations with spatial resolution of @xmath22 arcmin , are sufficient to obtain unique identifications . + 2 ) flux measurements from the ahiss can be poor if a signal is detected at large distance from the center of the beam . in principle , a correction to the flux could be made since the response function of the telescope is known with reasonable accuracy , but this is only possible if the positional accuracy is sufficient . furthermore , measurements from the survey spectra might underestimate the integral flux if the source is more extended than the primary beam . thus , a rough measure of the distribution increases confidence in the analysis . + 3 ) some signals can be caused by pairs or small groups of galaxies , whose line emission might stack up in the same channels . it is not obvious from the arecibo survey spectra which signals are caused by more than one galaxy . in fact , this situation was found by the vla observations to occur in five cases . short vla observations ( @xmath23 20 min ) of all 61 detected galaxies were performed during the d - configuration sessions in may 1995 and september 1996 . the signal of three systems fell below the detection limit of the snap - shot observation . these systems were re - observed in the d - configuration during the second session , but with longer integration times ( @xmath23 3 hours ) , resulting finally in confirmation of all 61 detections at levels consistent with the ahiss sensitivity . the three weaker signals originated in galaxies whose declinations are at the center of the ahiss . the vla observations were performed using 63 channels over a 3.125 mhz bandwidth , corresponding to a velocity range of approximately 660 @xmath2 . on - line hanning smoothing was used , resulting in a velocity resolution of @xmath24 @xmath2 . phase calibrators were observed once for each source . only seven different phase calibrators were used . each time that the phase calibrator was changed , a primary flux calibrator ( 3c48 or 3c286 ) was observed in order to tie the flux scales together . therefore , the primary calibrators were not observed with the same correlator settings as each galaxy , but only with six correlator settings . the overhead time due to slewing of the telescopes and observations of primary calibrators is significantly decreased this way . this technique provided adequate passband calibration for these short observations , where high dynamical range or accurate channel - to - channel flux density calibrations are not needed . there were no strong continuum sources in these fields . since the observing conditions were generally quite good , little editing was necessary to remove interference and bad baselines . continuum was removed from the data in the @xmath25-plane , by making linear fits to the real and imaginary components for each visibility in the line - free channels and subtracting the appropriate values from all channels . the @xmath25 data were calibrated and transformed to datacubes using natural weighting . using natural weighting rather than uniform weighting results in a slightly lower spatial resolution , but higher sensitivity . the resolution in the transformed datacubes is @xmath26 . the final r.m.s noise was approximately 1.2 mjy / beam for each channel ( corresponding to a minimal detectable column density of @xmath27 [ @xmath4 ] ) for the short integrations and 0.8 mjy / beam ( limiting column density @xmath29 [ @xmath4 ] ) for the longer integrations . since the observed galaxies were generally barely resolved , cleaning of the data was not really necessary . nonetheless , because the synthesized beam of these short observations has strong deviations from a gaussian shape , we chose to clean the datacubes and restore them with a gaussian beam in order to make an effective search for companion objects throughout the primary beam . total masses were calculated using @xmath30 , where @xmath31 is the distance to the source in mpc , and @xmath32 is the flux density in mjy over profile width @xmath33 in @xmath2 . total maps were constructed by adding the regions in the channel maps that contain line emission . contour maps of emission and global profiles will be presented in a future paper . [ compmhi.fig ] shows the masses as measured with the vla plotted against the masses that are derived from the ahiss spectra . different symbols are used to distinguish between single detections ( filled circles ) and multiple detections ( open circles ) . arecibo measurements of these multiple detections are of course always unreliable estimates of the real masses . this figure clearly illustrates the fact that the masses derived from the arecibo spectra generally underestimate the true masses , and that follow - up observations were essential to determine accurate measurements . in the analysis of the ahiss results we will use masses calculated from the vla observations , except for galaxies with integrated vla fluxes less than @xmath34 . galaxies with these low fluxes are only found at small distances from the center of the arecibo beam , where the normalized response function is close to unity . arecibo measurements for these galaxies are therefore reliable estimates of their real fluxes . table [ tabel ] summarizes the global parameters of the detected galaxies which are derived from the vla observations . the following information is contained : column 1 : identification number of the arecibo detection . indices indicate multiple detections . column 2 : name of galaxy if already cataloged . columns 3 and 4 : b1950 coordinates . column 5 : logarithm of mass . column 6 : heliocentric velocity , calculated by taking the mean of the velocities at 20% of peak flux density . column 7 : declination offset from center of the survey strip . column 8 : identification code for multiple detections . we have also started a program of optical @xmath35-band imaging of the selected galaxies with the 2.5 m isaac newton telescope ( int ) on la palma . during four observing runs in the period between october 1995 and march 1997 we have been able to obtain images of all target galaxies . the first noteworthy result of the follow - up optical observations is that _ all _ of the sources more than @xmath18 away from the galactic plane are found to be associated with galaxies in the @xmath35-band images . we have seen no indication that any of the sources detected in the arecibo survey are anything other than ` ordinary ' galaxies having both gas and stars . put another way , we have so far failed to find any sources that are pure clouds without stars . the analysis of the optical observations will be presented in a subsequent paper . the improved coordinates and fluxes derived from the vla observations help to verify the completeness limit of the survey . the detectability is determined in the following way : the mass in a detected signal can be expressed as @xmath36 , where @xmath31 is the distance to the source in mpc , and @xmath32 is the flux density in mjy that here is considered to be constant over a rectangular profile of width @xmath33 km / s . the sensitivity of an observation is optimal when the spectra are smoothed to the velocity width of the source . with optimal smoothing the noise becomes @xmath37 , where @xmath38 is the spectral resolution of the receiving system and @xmath39 is the noise level in the unsmoothed spectra . the limiting flux density @xmath40 for a @xmath41 detection is then given by @xmath42 . for the ahiss , the average noise level after coadding spectra taken at different days was 0.75 mjy for a velocity resolution of 16 km / s . the normalized response function of the survey telescope , @xmath43 , describes the relative response to a source which is detected at an offset @xmath44 from the center of the survey strip . in other words , @xmath43 is the integral of the flux density sensed by the telescope as a source makes a cut through the beam pattern , missing the center of the beam by angle @xmath44 , normalized in such a way that @xmath45 . for the arecibo telescope this function falls off to @xmath46 at @xmath47 and reaches a second maximum of @xmath48 at @xmath49 due to the high sidelobe level . we define a correction factor @xmath50 that accounts for the shape of the response function by @xmath51 . the limiting flux density of a galaxy at an offset @xmath44 from the center of the survey strip can then be expressed as @xmath52 . in general , an source should be detected by the survey if its flux density exceeds this limiting flux density , that is @xmath53 . this can be rewritten as @xmath54 where we define @xmath55 as the detectability . the factor @xmath56 represents the normalized feed gain of the telescope , which is a function of frequency . @xmath56 can be approximated by an analytical expression of the form @xmath57/w)^2 $ ] , where @xmath58 is the frequency in mhz , @xmath59 is the center of the survey band and @xmath60 is a parameter which determines the width of the band . during the observations at @xmath61 the shape of the normalized feed gain remained unchanged and could be satisfactorily fit by @xmath62 mhz and @xmath63 mhz . during the @xmath64 observations , the gain was retuned a few times to the settings : @xmath62 mhz and @xmath65 mhz and @xmath66 mhz and @xmath67 mhz . the @xmath56 dependence of @xmath55 leads to a ` distance dependence ' for the flux density sensitivity . the sensitivity can now be verified by plotting the detectability @xmath55 of the objects against declination offset from the center of the survey strip . this is shown in fig . [ sens.fig ] . all signals that were detected are shown in this plot . the horizontal errorbars are the result of a combination of positional accuracy in the vla maps and the spatial extent of the galaxies in the direction orthogonal to the survey strip . the signals corresponding to pair or group detections are indicated by open circles and letters a to e are used to identify these related or confused signals in table [ tabel ] . filled circles mark all single object detections . the solid line represents @xmath68 , the limit to the detectability . in principle , all filled circles should lie above the solid line , the area below this line is ` undetectable ' . the line is a satisfactory limit to the data points , especially if we consider the naive character of eq.[detect.eq ] . that is , this equation assumes that the detected profiles are symmetric and featureless . since the profiles are generally heavily smoothed , this is a reasonable assumption , but lop - sided profiles or strong double horned profiles might exceed the detection limit while they are formally ` undetectable ' according to eq.[detect.eq ] . the limiting depth @xmath69 , the maximum distance to which the object could be placed and still remain within the sample , can now be expressed as a function of @xmath70 , @xmath44 and @xmath33 : @xmath71 the variation of feed gain with frequency imposes a minor correction to the limiting depth @xmath69 which can be compensated by solving for @xmath69 in @xmath72 that is , the true flux density @xmath73 of a galaxy located at the distance limit @xmath69 multiplied by the feed gain at a frequency corresponding to that distance has to be equal to the limiting flux density @xmath40 obtained at the center frequency , where the feed gain is optimized and the nominal noise level of the system is normally calculated . the next step is to calculate the effective search volume of the survey . the volume @xmath74 of a slice of the total survey volume at declination @xmath75 and with length @xmath76 ( the total length of the strips in radians , @xmath77 ) and width @xmath78 is given by @xmath79 where @xmath80 is the limiting depth of the survey imposed by the bandwidth of the receiving system . the total survey volume that is sensitive to a galaxy with mass @xmath70 and velocity spread @xmath81 can then be calculated by taking the integral over @xmath44 . for galaxies in the flux limited regime , for which @xmath82 for each @xmath44 , this integral simply becomes : @xmath83 for galaxies in the bandwidth limited regime , @xmath84 for @xmath44 smaller than a certain critical value @xmath85 . the integral must now be split up in separate parts for the flux limited regimes and the band width limited regime : @xmath86 where we made use of the symmetry of the beam shape in the last step . this effective search volume is still dependent on two variables : @xmath70 and @xmath33 . the most convenient parameterization of @xmath87 for computing an himf is to express @xmath87 as a function of @xmath70 only . this can be achieved by adopting a relation between @xmath70 and @xmath33 . such a relation is known to exist since optical luminosity @xmath88 is related to @xmath33 via the tully - fisher relation ( tully & fisher 1977 ) , and @xmath88 is related to @xmath70 as @xmath89 ( see briggs 1990 ) . briggs & rao ( 1993 , hereafter br ) determined the @xmath70-@xmath33 relation empirically by plotting @xmath33 against log@xmath70 for 1139 optically selected galaxies from the catalog by fisher & tully ( 1981b ) . a fit to these points gives @xmath90 . recently , salpeter & hoffman ( 1996 ) analyzed observations of 70 dwarf galaxies and find a similar trend : @xmath91 . this relation is therefore valid over a wide range in mass . our data are also in good agreement with this relation . note that the velocity widths in these relations are not the inclination corrected maximum rotational velocities , but just the observed velocity spreads . the effective survey volume @xmath92 can now be calculated by substituting the relation between @xmath33 and @xmath70 found by br and @xmath93 into eq.[volume1.eq ] and eq.[volume2.eq ] . this volume as a function of @xmath70 is shown by the solid line in fig . [ volume.fig ] . in the flux limited regime , for masses @xmath94 , the search volume is @xmath95 . ( a proportionality often used in the literature is @xmath96 [ e.g. , henning 1995 , schneider 1997 ] . this power 3/2 arises if the dependence of @xmath33 on @xmath70 is discarded , effectively assuming that all galaxies have the same profile width . ) in the high mass region ( @xmath97 ) , the limiting depths are no longer determined by the detectability of the signals , but simply by the bandwidth of the receiving system . therefore , the effective survey volume for high mass systems is not dependent on mass or @xmath33 . the total survey volume in this regime , is @xmath98 . for lower masses the volume decreases rapidly and is only @xmath99 for @xmath100 . the dashed and the dotted line show the effective search volume corresponding to the main beam and the sidelobes , respectively . this figure clearly shows that the sidelobes do not add much volume in the low mass range , but are very effective in finding large masses . since the observed velocity width is related to inclination @xmath101 and maximum rotational velocity @xmath102 as @xmath103 , the limiting depth is @xmath104 . the limiting depth is therefore weakly dependent on inclination , in the sense that low inclined galaxies can be detected to slightly larger distances . consequently , an survey would in principle preferentially select face - on galaxies . in practice , this effect is negligible , since the expectation for @xmath105 of a randomly oriented sample is close to unity : @xmath106 only the flux limited regime would be hampered by the inclination effect , but most galaxies in this regime are low mass , dwarf galaxies . the velocity spreads are small for these galaxies , and turbulent motions play an important part at establishing the profile width ( cf . , lo , sargent & young 1993 ) . a potential hazard in radio surveys is the influence of radio frequency interference ( rfi ) . rfi can in principle cause false negatives ( miss a significant signal ) in the sample if it affects the spectrum exactly at the frequency of a source , or when a narrow - lined source is mistaken for rfi . however , the driftscan method that has been used for the ahiss has proven to provide a very good stability for rfi signals ( see briggs et al . repetitive coverage of the same regions of sky makes the survey immune to rfi and unstable baselines . this has been demonstrated by the fact that all signals identified in the arecibo survey have been confirmed by the vla follow - up . the mass function ( himf ) is defined analogously to optical luminosity functions . the himf @xmath107 gives the total number of galaxies or intergalactic clouds per mpc@xmath108 in the mass interval @xmath109 centered on @xmath110 . here , we find it convenient in our analysis and figures to plot the himf as the number of galaxies or intergalactic clouds per decade in mass . in order to parameterize the shape of the himf , we adopt the conventional schechter ( 1976 ) function , @xmath111 with free parameters @xmath112 , the slope of the low - mass end , @xmath113 the characteristic mass that the defines the kink in the function and @xmath114 , a normalization factor . this function must be integrated over decade bins for comparison with the binned data in the figures . the classical method ( see e.g. , christensen 1975 , schechter 1976 ) for determining luminosity functions is based on the assumption that galaxies are distributed in a uniform manner . the luminosity function is determined by dividing the number of galaxies in a bin centered on @xmath115 by @xmath116 , the effective search volume for that particular @xmath115 . this method is easily applicable to the ahiss . the search volume can be evaluated with eq.[volume1.eq ] and eq.[volume2.eq ] , using a statistical relation between velocity width and mass . the advantages of this method are that it is nonparametric and that it is automatically normalized . the important disadvantage is that it assumes homogeneity and its use might lead to errors in @xmath117 if density fluctuations due to large scale structure occur on distance scales comparable to , or greater than the depth @xmath69 at which @xmath70 can be detected . a slightly modified form of the ` classical ' method is the @xmath118 method , first used by schmidt ( 1968 ) , which is also applicable to surveys . instead of calculating a mean survey volume for each mass bin , this method consists of summing the reciprocals of the volumes corresponding to the maximum distances to which the objects could be placed and still remain within the sample . the values of @xmath119 can be calculated directly using eq.[volume1.eq ] and eq.[volume2.eq ] , using now the measured velocity spread instead of the statistical value . the two procedures give similar results when the number of galaxies per bin is large . for the less densely populated bins the two procedures can give different results because @xmath119 of a particular galaxy can strongly deviate from the average @xmath87 of the mass bin it falls in . in section [ results.sect ] the himf is calculated using schmidt s method . like the classical method , schmidt s method is also vulnerable to errors caused by large scale structure . the possible effects of large scale structure are discussed in section [ lss.sect ] . [ himfhist.fig ] shows the principal results of this analysis . the lower panel shows the observed distribution of masses binned per half decade , with errorbars given by poisson statistics . this histogram shows that the survey has detected galaxies with masses in the range from @xmath120 to @xmath121 and therefore enables us to determine the himf over three orders of magnitudes in mass . the solid dots in the upper panel of fig . [ himfhist.fig ] show the himf determined by the @xmath122 method . the errorbars are given by poisson statistics . also drawn in this figure are analytical curves given by the schechter function of eq [ himf.eq ] . a satisfactory fit to the points is obtained with @xmath123 , @xmath124 and @xmath125 . mass functions with faint end slopes of 1.10 and 1.30 are shown to indicate the uncertainty in the value of @xmath112 . we note that in the present analysis the parameterization of the himf in the form of a schechter function is only used to enable comparison with other survey results and results based on the optically selected galaxy population . the @xmath122 method recovers the shape and amplitude of the himf simultaneously without using a schechter function ( or any other parameterization ) as an assumption about the intrinsic shape of the himf . the observational limits to the determination of the himf are also illustrated in fig . [ himfhist.fig ] by the thin line . this line represents the sensitivity function of the survey to objects of mass @xmath70 , defined as @xmath126 , where @xmath92 is the effective search volume . in the range @xmath127 this function defines an upper limit to the space density of intergalactic clouds without stars . it also shows that this survey is not capable of measuring the himf directly in the regions @xmath128 and @xmath129 if the extrapolation of the analytical form for @xmath117 holds . the sensitivity function does allow us to define upper limits in these ranges as indicated by arrows . the upper limit at @xmath130 is from the complementary survey with the arecibo telescope over the range 19,000 to 28,000 @xmath2 , only sensitive to high mass galaxies ( see sorar ( 1994 ) for details ) . the slice diagrams in fig [ slices.fig ] show that the survey strips sample a wide range of large scale structure , as the combined ra range extends nearly 2/3 of the way around the sky . on the other hand , the limited depth of the survey for small masses might cause the low mass end of the himf to be affected by local density fluctuations . therefore , a major concern in the determination of the mass function following schmidt s method , is whether homogeneity is a fair assumption . for instance , a local density enhancement would overestimate the number of low mass galaxies or clouds and would give rise to an overestimate of the faint end slope of the mass function . the biases in the shape of the mass function due to large scale structure can be avoided by making use of a maximum likelihood estimator ( cf . , efstathiou , ellis & peterson 1988 , saunders et al . if the spatial density distribution of galaxies and intergalactic clouds is given by @xmath131 , then the overall density @xmath132 and the mass function @xmath117 can be separated as @xmath133 . maximum likelihood estimators can then be used to solve either for @xmath117 without knowledge of @xmath132 or solve for @xmath132 without knowledge of @xmath117 . although the maximum likelihood method is a powerful procedure for determining optical luminosity functions ( marzke , geller & huchra 1994 , lin et al . 1996 , saunders et al . 1990 ) , we chose not to apply it to the ahiss . the most important problem is that we have to deal with small number statistics . especially the faint end slope of the mass function is defined by very few galaxies or clouds per bin . some experimentation with application of the algorithm to small samples such as ours showed that maximum likelihood methods can produce erratic results is these situations . another complication arises because maximum likelihood methods assume that the shape of the mass function is independent of the space density @xmath132 . in practice , high mass galaxies will have a higher statistical weight in the determination of @xmath132 since they are simply more numerous in our sample . it is questionable whether this space density defined by the high mass galaxies is a fair estimate for the galaxies at the faint end side of the mass function , in the region that is dominated by dwarf and lsb galaxies . although these galaxies are found to follow the same general large scale structures as the normal hsb galaxies , they preferentially avoid the highest density regions ( mo et al . 1994 , taylor 1996 ) . in principle , the spatial density @xmath134 could be separated into different functions for individual morphological types or different ranges in mass , but the limited number of galaxies in our sample does not allow this differentiation . in order to investigate the effect of large scale structure we have performed numerical experiments . these tests consisted of randomly placing artificial galaxies in a volume with weights given by the himf given in eq.[himf.eq ] . a range of himf parameters were investigated . the results are illustrated using the derived himf parameters from fig [ himfhist.fig ] : @xmath135 , @xmath136 and @xmath137 . the galaxies have random inclination , and rotational velocity related to mass as determined by br . galaxies were selected from these volumes in the same manner as the arecibo strip survey selects galaxies from the sky . [ mosaic.fig ] shows the results of three experiments , labeled as a , b and c. the presented results are each the average of 100 independent simulations . the top panels show the assumed normalized density fluctuations @xmath134 . the central panels show the redshift distribution of the simulated data as thick - lined histograms , thin solid lines show @xmath138 uncertainties . overlaid on these histograms is the measured redshift distribution of our survey sample , indicated by solid circles and dashed histograms . the bottom panels show the averaged himfs of the simulated catalogs as open squares , with @xmath138 errorbars computed as the standard deviations of the 100 simulations . the @xmath139 method has been used . in case a a homogeneous density distribution is assumed . it is not surprising to see that the input himf and the measured himf agree with high accuracy . the @xmath139 method should give reliable results in this situation . it is however apparent that the survey observations have several distance bins that are overdense or underdense compared to the uncertainty band of the ` uniform density ' simulation . the comparison implies that we may be observing underdensities at @xmath140 mpc and @xmath141 mpc and an overdensity at @xmath142 mpc . to test this hypothesis we constructed catalogs with a density fluctuation as indicated by the top panel in case b. the precise functional form of this fluctuation is not important , the condition is that is should produce consecutive under and overdensities at @xmath143 30 , 50 and 70 mpc . the central panel shows that the imposed density fluctuation reproduces the observed redshift distribution satisfactorily . it is surprising however that if the @xmath144 method is applied to these simulated data the resulting himf is still indistinguishable from the input himf . in other words , the @xmath139 method appears to be a robust method , and not very sensitive to the effects of large scale structure . this is even true for case c , where a more severe density fluctuation is imposed , and where the resulting himf is still in good agreement with the input himf . we can conclude from these simulations that although we see a hint of the effects of large scale structure in our data , the observed deviations from uniformity have no significant influence on our determination of the himf . the first surveys could only be used to set upper limits to the space density of intergalactic clouds without stars , and did not yield enough detections to allow the determination of the shape of the himf . shostak ( 1977 ) was to first to define these limits by means of blind driftscan emission and absorption searches in the 21 cm line with the nrao 91 m telescope . due to the relatively poor sensitivity ( 12 - 40 mjy ) and the small total effective search volume ( @xmath145 ) , these limits were not very strict . in the region @xmath146 the limits set by the arecibo strip survey are at least two orders of magnitude lower ( 99% confidence ) than those set by shostak . haynes & roberts ( 1979 ) , lo & sargent ( 1979 ) and fisher & tully ( 1981a ) search for invisible galaxies in groups of galaxies . haynes & roberts conclude from observations in the direction of the sculptor group that intergalactic clouds with @xmath147 do not exist . lo & sargent find four previously uncataloged lsb dwarf galaxies but their upper limits to the space density of unseen objects do not improve shostak s . fisher & tully used the nrao 91 m telescope to search for invisible galaxies in the m81 group . their null result allowed them to push the upper limits to the space density 0.6 dex lower those set by shostak . also krumm & brosch ( 1984 ) find no sources in their driftscan searches of void regions and are only able to define upper limits . these limits are not very strict since their survey is only sensitive in the redshift range above 5300 @xmath2 , and is consequently only capable of finding masses @xmath148 . a series of papers by kerr & henning ( 1987 ) and henning ( 1992 , 1995 ) describes blind surveys which consisted of observing a series of pointings along lines of constant declination in the zone of avoidance and at high galactic latitude . these surveys yielded 39 detections ( of which half were previously unknown ) and were the first to put serious constraints on the shape of the himf . for comparison , the himf calculated by henning ( 1995 ) is reproduced in fig . [ compa.fig ] as solid squares , together with the ahiss himf represented as a solid line . henning s points are significantly lower than our measured curve . according to henning s calculations objects with @xmath149 are deficient by a one order of magnitude compared to the ahiss himf . the presence of the local void in henning s survey could explain a part of this discrepancy , but even if the void is omitted the points are still beneath our himf . the most reasonable explanation for the discrepancy is that the sensitivity of henning s survey is not well understood , leading to a serious underestimate of the himf . the most recent determination of the himf is the one by schneider ( 1997 ) who used both the uncorrected arecibo data from sorar s ( 1994 ) survey and the data from spitzak s ( 1996 ) survey with the arecibo telescope . schneider assessed the completeness limits of the surveys with the use of @xmath150 tests . he calculated the average value of @xmath150 and scaled that to the correct value of 0.5 by increasing the limiting flux density @xmath151 . schneider noted the vulnerability of this method to the influence of large scale structure . despite this different approach , the resulting himf is in good agreement with our estimate , as can be seen in fig . [ compa.fig ] . only the low mass end deviates significantly . the statistics in the bins with @xmath152 are poor , and schneider could not rule out a slope as high as @xmath153 . the differences in the slopes can be partly explained by the fact that _ uncorrected _ masses are used in schneider s estimate of the himf from sorar s data . the effect that the flux is underestimated when the arecibo beam misses a galaxy by a certain angle @xmath44 was not taken into account . the vla masses that are used in our himf are on average 50% higher , leading to a shift of galaxies to higher mass bins . as an alternate approach , schneider also applied a maximum likelihood fit to all the signals brighter that @xmath154 . the resulting himf had a faint end slope @xmath155 . this result is also drawn in fig . [ compa.fig ] and is in very good agreement with our function over the whole range in mass . in addition to making comparisons with surveys that are equally blind as the ahiss , it is also interesting to compare our results with surveys that cover comparable volumes in underdense and overdense regions . it is noteworthy that the himf determined from hi observations in clusters ( mcmahon 1993 ) , as well as that determined in cosmic voids ( szomoru et al . 1996 ) are both well fit by a schechter function , with approximately the same shape as as our himf . the scaling @xmath114 , however , differs by a factor of ten or more between these different regions . since the ahiss crosses the galactic plane twice , and consequently a large fraction of the total survey volume lies in regions which are not complete in optical catalogs , it is difficult to calculate the exact void filling factor of our survey . if we limit the calculation to the regions where @xmath156 , we estimate that approximately 50% of the survey volume consists of regions where the average galaxy density is less than one third of the cosmic mean , and approximately 10% of the volume consists of structures where the galaxy density takes on more than three times the cosmic mean . the same percentages are obtained if random slices through the universe are taken . if voids would be omitted from the survey , the scaling @xmath114 of the himf would increase by a factor of two . the construction of an himfs from a sample of optically selected galaxies can either be done statistically , by studying optical luminosity functions and the dependence of mass on on optical luminosity , or directly by using 21 cm data of optically selected galaxies . the first method was used by rb , who reviewed the literature that describes luminosity functions and richness for individual morphological types . combining these data enabled them to construct an himf over three orders of magnitude in mass . their result is in good agreement with our estimate of the himf in the range of mass where our survey is sensitive . the second method is discussed by br who analyzed observations drawn from the catalogs of fisher & tully ( 1981b ) and hoffman et al . ( 1989 ) , and recently by solanes et al . ( 1996 ) who use 21 cm data of an optical magnitude - limited sample of galaxies in the direction of the pisces - perseus supercluster . both authors arrive at the same values for the normalization and @xmath113 , but their values of the faint end slope differ significantly . br find @xmath157 while solanes et al . find a declining slope . the solanes et al . result may not be relevant to the discussion here since their sample excluded all dwarf irregular galaxies and contained no galaxies with @xmath158 . the himf determined by br is reasonably consistent with with our estimate of the himf as shown in fig . [ compa.fig ] . the comparison between our result and the himfs based on optically selected galaxy samples provides a direct test of the existence of a new population of gas rich galaxies that has gone unnoticed by optical surveys . in a way , this comparison appraises the completeness of optically selected catalogs for gas rich galaxies . any contribution of uncataloged gas rich dwarf or gas rich lsb galaxies would yield an difference between himfs computed from optically selected and hi - selected galaxy catalogs . fig [ compa.fig ] shows that the himf derived from the results of the ahiss is in very good agreement with previous estimates based on optically selected galaxy catalogs . this implies that the optically selected samples that have been used to evaluate the himf are remarkably complete . there is no evidence for a large number of neutral gas rich objects that have escaped inclusion in these catalogs . to the extent that optically selected catalogs are incomplete for lsb galaxies , the excluded galaxies must be predominantly gas poor , consistent with the finding by sprayberry et al . ( 1997 ) that these excluded lsb galaxies are predominantly of low optical luminosity . several authors have speculated about the existence of a large class of gas rich dwarf galaxies ( e.g. , dekel & silk 1986 , tyson & scalo 1988 ) . these galaxies would yield a steep rise in the himf below @xmath159 . we find no observational support for the existence of this class of galaxies . based on the counting statistics in the mass bins @xmath160 to @xmath161 we can exclude a faint end slope of @xmath153 or steeper with a 99% confidence level . our knowledge of @xmath162 at high redshift is determined by the statistics of damped ly-@xmath112 systems , seen in absorption against background quasars ( e.g. , lanzetta , wolfe & turnshek 1995 , storrie - lombardi , irwin & mcmahon 1996 ) . the picture emerging from these studies is that @xmath162 reaches a maximum at @xmath163 and has been declining since then . at low redshifts different effects complicate the determination of @xmath162 . firstly , at redshifts @xmath164 , the ly-@xmath112 line is not redshifted to optical wavelengths and has to be observed from space ( lanzetta et al . 1995 , rao , turnshek & briggs 1995 ) . secondly , the evaluation of @xmath162 may depend on the selection effects in the sample of quasars that has been used . especially at these low redshifts , it is very difficult to compile a unbiased sample of quasars since qsos would need to be observed within the optical images of galaxies . at the one hand , gravitational lensing can bring faint quasars into the sample which should otherwise be below the detection limit ( smette , claeskens & surdej 1997 ) . but on the other hand , dust in ly-@xmath112 systems might obscure the background objects to a level where they are undetected ( fall & pei 1993 ) . the arecibo strip survey can be used to evaluate the gas density differently . the mentioned problems are circumvented since @xmath162 is not measured by using the ly-@xmath112 line , but the 21 cm line in emission . the space density of contained in objects of different masses is plotted in fig . [ hidens.fig ] . the solid line indicates an analytical expression of this function , determined from the product @xmath165 . the thin line represents the sensitivity limits and the arrows mark upper limits determined analogously to the upper limits in fig [ himfhist.fig ] . the integral mass density at the present epoch can be determined by taking the integral over the solid line in fig . [ hidens.fig ] . this yields : @xmath166 , where @xmath167 is the euler gamma function . using the best fit schechter parameters , we derive that @xmath168 or @xmath169 , with a statistical error of @xmath170 . a summation over all survey galaxies @xmath171 yields a slightly smaller value : @xmath172 since this calculation does not include the contribution to @xmath173 of the density function below @xmath100 or above @xmath174 . the cosmological mass density of at @xmath175 is @xmath176 . the total cosmological mass density of neutral gas at the present epoch is @xmath177 , assuming that the mass percentage of is 25% of the total gas density . this value agrees surprisingly well with earlier estimates by rb , who find @xmath178 by using optically selected galaxies . fall & pei ( 1993 ) arrive at approximately the same value of @xmath9 by simply computing the average @xmath179 of the huchtmeier & richter ( 1989 ) catalog and multiply that by the mean luminosity density in the local universe as estimated by efstathiou et al . the agreement between the cosmological mass density based on optically selected galaxies and that based on selected galaxies illustrates once more that there is not much neutral gas hidden in objects like lsb galaxies , dwarfs or intergalactic clouds that are missed by optical surveys . the contribution of dwarf galaxies to the content of the nearby universe is modest , galaxies with masses @xmath180 make up only @xmath181 of the integral density . the density function in fig . [ hidens.fig ] clearly illustrates that the integral mass density is dominated by high mass galaxies with masses in the range @xmath182 , which are @xmath183 galaxies . at @xmath184 the density function drops off sharply , indicating that malin 1 type galaxies make no significant contribution to @xmath9 . this sharp cutoff was already noted by bothun ( 1985 ) . much stronger upper limits to the contribution of galaxies with @xmath129 will be set in the near future by the results of the parkes multibeam survey ( staveley - smith et al . 1996 ) . the estimate of the integral gas density from the ahiss is a robust result . two effects cause a relatively low uncertainty in the determination of @xmath9 . firstly , the peak in the gas density function is conveniently caused by the galaxies that dominate the counting statistics . galaxies in the lower mass bins , where the poisson errors are large , contribute not much to the total density and therefore also not much to the uncertainty in @xmath9 . secondly , the effective search volume in the mass region that dominates the density function is mostly band width limited . uncertainties in @xmath70 and velocity width do not influence the search volume in this regime . the question arises which gas rich systems could be missed by the survey and could still contribute significantly to the integral gas mass density in the local universe . the only possible candidates that might escape detection are extremely low gas density systems with column densities below @xmath185 , the detection limit of the ahiss . it has been shown that the gas density of spiral galaxies is correlated with the optical surface brightness in such a way that lower optical surface brightness implies lower gas densities . de blok et al . ( 1996 ) observed a sample of lsb galaxies and showed that the neutral gas densities are generally a factor of two lower than those of late type high surface brightness galaxies . furthermore , van der hulst et al . ( 1993 ) has shown that the gas densities of lsb galaxies are often just below or equal to the critical density for star formation . if the relation between optical surface brightness and gas density extends to still lower surface brightnesses this implies that galaxies that most easily escape detection in optical surveys are also the ones that might be missed in 21 cm surveys . until recently , blind surveys were not able to reach column densities much lower than @xmath186 . these surveys would therefore still miss low density systems , and would not be capable to set strict constraints on the number density of lsb galaxies . however , the ahiss is sensitive to column densities @xmath187 at a @xmath4 level for gas filling the beam . using an extrapolation of the scaling of disney & banks ( 1997 ) , @xmath188 , where @xmath189 is the surface brightness averaged over the disk , and a typical value of @xmath190 for lsb galaxies , the ahiss would be capable of finding galaxies with @xmath191 . even if a ten times lower value of @xmath192 is used , and if the area of the galaxy were ten times smaller than the area covered by the arecibo beam , still galaxies as faint as @xmath193 would be detectable . the vla observations of the ahiss galaxies can be used to set lower limits to average column densities for the sample . due to the low spatial resolution of the observations ( @xmath194 ) , the measured column density is in many cases an average @xmath195 over the entire projected surface of the layer of the galaxy . if the galaxy is spatially unresolved , an upper limit to the extent @xmath196 of the layer can be determined , leading to a lower limit of the average column density @xmath197 . although many of the galaxies in the sample are unresolved , and many of the true values of @xmath195 may be higher , we find no values of @xmath195 below @xmath198 . hence , all galaxies in the sample have normal column densities , even though there are no observational selection effects against finding extreme low density systems . there is no indication of the existence of a group of extreme low column density galaxies that has been missed by previous surveys , simply because they were not capable of reaching the same low column densities as the ahiss . because we seem to be observing a lower limit to @xmath199 , much higher than our detection limit , it is very unlikely that galaxies with even lower column densities are missed by the ahiss . theoretical predictions of the ionization of gas layers by the extragalactic uv background ( corbelli & salpeter 1993 , maloney 1993 , charlton , salpeter & linder 1994 ) demonstrate physical mechanisms that can explain the possible non - existence of low column density neutral gas layers . these models predict photoionization by the extragalactic uv background of low column density regions , below @xmath200 . furthermore , models by quinn , katz & efstathiou ( 1996 ) show that the ionization only moderately suppresses the formation of galaxies with rotational speeds larger than 50 @xmath2 , but that it seriously affects the low density regions around these systems . the models are verified by very deep vla observations on one galaxy which appears to have a sharp truncation of the hi disk below a column density of @xmath201 ( van gorkom et al . further confirmation of ionization of low density hi comes from recent observations by bland - hawthorn et al . ( 1997 ) who have detected ionized gas beyond the truncated hi disk in ngc 253 . the calculation of the total masses in this paper is based on the assumption that the optical depth of the layer is close to zero . any possible influence of self absorption , which will cause an underestimation of the true mass , is ignored . in this paragraph we will make a rough estimate of the influence of self absorption on the determination of the cosmological mass density @xmath9 and the himf . the possible effect of self absorption could apply to all himfs compared in this paper ( see section [ comphi.sect ] , [ compopt.sect ] and fig . [ compa.fig ] ) , since none of these have addressed this possibility . the problem of self absorption for galaxies can be assessed statistically by plotting the 21 cm flux of different hubble types as a function of inclination @xmath101 to the line of sight . the line of sight through a inclined galaxy will be larger , generally causing a higher fraction of self absorption . haynes & giovanelli ( 1984 ) use data of a sample of 1500 galaxies with 21 cm fluxes measured with arecibo . they define a correction factor @xmath202 which is defined by the corrected flux divided by the measured flux and find a general trend : @xmath203 where @xmath204 is a constant dependent on morphological type . the values of @xmath204 are found to be 0.04 for sa and sab , 0.16 for sb and 0.14 for sbc and sc galaxies . no correlations are found for morphological types earlier than sa or later than sc , indicating that self absorption is negligible in these types . higher self absorption in types sb to sc can be explained by the fact that these galaxies generally have the highest surface densities ( cayatte et al . furthermore , rotation curves of early type spiral galaxies show the strongest deviations from solid body rotation implying large velocity spreads per line of sight , which further decreases self absorption . mean self absorption factors per morphological type can be obtained by averaging @xmath202 over a random distribution of inclinations . this yields @xmath205 the mean correction factors then become 1.04 for sa and sab , 1.19 for sb and 1.16 for sbc and sc galaxies . the cosmological mass density is dominated by high mass galaxies , which are statistically most likely to be late type spirals . the correction to @xmath9 due to self absorption will therefore probably not be more than the value of @xmath206 averaged over morphological types sb to sd . assuming that all types sb to sd contribute equally to @xmath9 , we derive that the mean value of @xmath207 is 1.10 . even in the most pessimistic approach , the correction factor can not be more than 1.19 . what will be the effect of self absorption on the shape of the himf ? using the same arguments as above , we conclude that the effect on the high mass range will be marginal . the normalization factor @xmath114 and the value which determines the kink , @xmath113 , will probably increase by no more than 10% . galaxies that determine the faint end slope @xmath112 are mostly dwarf and lsb galaxies . on the one hand , self absorption may be unimportant because the gas density in these galaxies is usually low ( van der hulst 1993 ) and the dust content is presumably low ( mcgaugh 1994 ) which implies scarcity of clumped gas ( haynes & giovanelli 1984 ) . on the other hand , the rotation curves of dwarf and lsb galaxies often show solid body rotation ( de blok et al . 1996 , swaters 1997 ) which causes a low velocity spread along a line of sight , leading to high fractions of self absorption . these two counteracting effects make the value of @xmath206 for low mass galaxies uncertain , but probably higher than that for high mass galaxies . after applying the correction factor to the low mass galaxies , some galaxies will shift to higher mass bins , eventually leading to a slightly shallower faint end slope of the mass function . the conclusion that the faint end slope of the mass function is shallow will therefore not be altered by the effects of self absorption . we have used the arecibo strip survey in combination with 21 cm follow - up observations with the vla to determine the mass function of gas rich galaxies in the local universe . the resulting himf is in good agreement with earlier estimates based on samples of optically selected galaxies . this implies that there is not a large population of gas rich dwarfs or low surface brightness galaxies , previously undetected by optical surveys . the observed faint end slope of the himf is flat ( @xmath208 ) leaving no room for a large class of gas rich dwarfs . the cosmological mass density of in the local universe is @xmath209 , also consistent with earlier estimates . the neutral gas content is dominated by high mass galaxies with @xmath182 . the observations indicate the existence of a lower limit to the average column density of 19.7 @xmath210 , consistent with theoretical predictions concerning the ionizing extragalactic uv background . bland - hawthorn , j. , freeman , k. c. , quinn , p.j . 1997 , apj , in press bothun , g. b. 1985 , 90 , 1982 bothun , g. b. , beers , t. , mould , j. , huchra , j. 1986 , apj , 308 , 510 briggs , f. h. 1990 , aj , 100 , 999 briggs , f. h. , & rao , s. 1993 , apj , 417 , 494 ( br ) briggs , f. h. , sorar , e. , kraan - korteweg , r. c. , & van driel , w. 1997 , pasa , 14 , 37 cayatte , v. , kotanyl , c. , balkowski , c. , & van gorkom , j. h. 1994 , aj , 107 , 1003 charlton , j. c. , salpeter , e. e. , & linder , s. m. 1994 , apj , 430 , l29 christensen , c. g. 1975 , aj , 80 , 282 corbelli , e. , salpeter , e. e. 1993 , apj , 419 , 104 dalcanton j. j. , spergel , d. n. , & summers , f. j. 1997 , apj , 482 , 659 de blok , w. j. g. , mcgaugh , s. s. , & van der hulst , j. m. 1996 , mnras , 283 , 18 dekel , a. , & silk , j. 1986 , apj , 303 , 39 disney , m. j. 1976 , nature , 263 , 573 disney m. j. , & banks , g. 1997 , pasa , 14 , 69 eder , j. a. , oemler , a. jr . , schombert , j. m. , & dekel , a. 1989 , apj , 340 , 29 efstathiou , g. , ellis , r. s. , & peterson , b. a. 1988 , mnras , 231 , 479 fall , s. m. , & pei , y. c. 1993 , apj , 402 , 479 fisher , j. r. , & tully , r. b. 1975 , a&a , 44 , 151 fisher , j. r. , & tully , r. b. 1981a , apjs , 47 , 139 fisher , j. r. , & tully , r. b. 1981b , apjl , 243 , l23 geller , m. j. , & huchra , j. p. 1989 , science , 246 , 897 giovanelli , r. , & haynes , m. p. 1989 , apjl , 346 , l5 giovanelli , r. , & haynes , m. p. 1985 , aj , 90 , 2445 haynes , m. p , & roberts , m. s. 1979 , apj , 227 , 767 haynes , m. p , & giovanelli , r. 1984 , aj , 89 , 758 henning , p. a. 1992 , apjs , 78 , 365 henning , p. a. 1995 , apj , 450 , 578 hoffman , g. l. , lewis , b. m. , helou , g. , salpeter , e. e. , & williams , b. m. 1989 , apjs , 69 , 65 hoffman , g. l. , lu , n. y. , & salpeter , e. e. 1992 , aj , 104 , 2086 huchtmeier , w. k. , & richter , o. -g . 1989 , a general catalog of observations of galaxies : the reference catalog ( new york : springer - verlag ) kerr , f. j. , & henning , p. a. 1987 , apj , 320 , l99 krumm , n. , & brosch , n. 1984 , aj , 89 , 1461 lanzetta , k. m. , wolfe , a. m. , & turnshek , a. m. 1995 , apj , 440 , 435 lin , h. , kirshner , r. p. , shectman , s. a. , landy , s.d . , oemler , a. , tucker , d. l. , & schechter , p. l. 1996 , 464 , 60 lo , k. y. , & sargent , w. l. w. 1979 , apj , 227 , 756 lo , k. y. , & sargent , w. l. w. , young , k. 1993 , aj , 106 , 507 maloney , p. 1993 , apj , 414 , 41 marzke , r. o. , geller , m. j. , & huchra , j. p. 1994 , aj , 108 , 437 mcgaugh , s. s. 1994 , apj , 426 , 135 mcgaugh , s. s. 1996 , mnras , 280 , 337 mcmahon , p. m. 1993 , ph.d thesis , columbia university mo , h. j. , mcgaugh , s. s. , & bothun , g. d. 1994 , mnras , 267 , 129 quinn , t. , katz , n. , & efstathiou , g. 1996 , mnras , 278 , l49 rao , s. , briggs , f. h. 1993 , apj , 419 , 515 ( rb ) rao , s. , turnshek , a. m. , & briggs , f. h. 1995 , apj , 449 , 488 salpeter , e. e. , & hoffmann , g. l. 1996 , apj , 465 , 595 saunders , w. , rowan - robinson , m. , lawrence , a. , efstathiou , g. , kaiser , n. , ellis , r. s. , & frenk , c. s. 1990 , mnras , 242 , 318 schechter , p. 1976 , apj , 203 , 297 schmidt , m. 1968 , apj , 151 , 393 schneider , s. e. 1989 , apj , 343 , 94 schneider , s. e. 1997 , pasa , 14 , 99 schombert , j. m. , bothun , g. d. , schneider , s. e. , & mcgaugh , s. s. 1992 , aj , 103 , 1107 shostak , g. s. 1977 , a&a , 54 , 919 smette , a. , claeskens , j. -f . , & surdej , j. 1997 , newa , 2 , 53 solanes , j. m. , giovanelli , r. , & haynes , m. 1996 , apj , 461 , 609 sorar , e. 1994 , ph.d . thesis , university of pittsburgh spitzak , j. g. 1996 , ph.d . thesis , university of massachusetts sprayberry , d. , impey , c. d. , irwin , m. j. , & bothun , g. d. 1997 , apj , 482 , 104 staveley - smith , l. , wilson , w. e. , bird , t. s. , disney , m. j. , ekers , r. d. , freeman , k. c. , haynes , r. f. , sinclair , m. w. , vaile , r. a. , webster , r. l. , & wright , a. e. 1996 , pasa , 13 , 243 storrie - lombardi , l. j. , irwin , m. j. , & mcmahon , r. g. 1996 , mnras , 282 , 1330 swaters , r. a. , 1997 , in preparation szomoru , a. , guhathakurta , p. , van gorkom , j. h. , knapen , j. h. , weinberg , d. h. , & fruchter , a. s. 1994 , aj , 108 , 491 szomoru , a. , van gorkom , j. h. , gregg , m. d. , & strauss , m. a. 1996 , aj , 111 , 2150 taylor , c. l. 1997 , apj , 480 , 524 thuan , t. x. , gott , j. r. , & schneider , s. e. 1987 , apjl , 315 , l93 thuan , t. x. , alimi , j - m . , gott , j. r. , & schneider , s. e. 1991 , apj , 370 , 25 tully , r. b. , & fisher , j. r. , 1977 , a&a , 54 , 661 tyson , n. d. , & scalo , j. m. 1988 , apj , 329 , 618 van der hulst , j. m. , skillman , e. d. , smith , t. r. , bothun , g. d. , mcgaugh , s. s. , & de blok , w. j. g. 1993 , aj , 106 , 548 van gorkom , j. h. 1993 , in the environment and evolution of galaxies , ed . j. m. shull and h. a. thronson ( kluwer academic publishers ) , 345 weinberg , d. h. , szomoru , a. , guhathakurta , p. , & van gorkom , j. h. 1991 , apj , 372 , l13

we constrain the distance of the gamma - ray burst ( grb ) prompt emission site from the explosion centre , , by determining the location of the electron s self absorption frequency in the grb prompt optical - to - x/-ray spectral energy distribution , assuming that the optical and the-ray emissions are among the same synchrotron radiation continuum of a group of hot electrons . all possible spectral regimes are considered in our analysis . the method has only two assumed parameters , namely , the bulk lorentz factor of the emitting source , and the magnetic field strength in the emission region ( with a weak dependence ) . we identify a small sample of 4 bursts that satisfy the following three criteria : ( 1 ) they all have simultaneous optical and-ray detections in multiple observational time intervals ; ( 2 ) they all show temporal correlations between the optical and-ray light curves ; and ( 3 ) the optical emission is consistent with belonging to the same spectral component as the-ray emission . for all the time intervals of these 4 bursts , it is inferred that cm . for a small fraction of the sample , the constraint can be pinned down to cm for . for a second sample of bursts with prompt optical non - detections , only upper limits on can be obtained . we find no inconsistency between the-constraints for this non - detection sample and those for the detection sample . gamma - rays : bursts - gamma - rays : theory - radiation mechanisms : non - thermal - radiative transfer

0906.2597

although the gamma - ray burst ( grb ) was discovered about 50 years ago first through its prompt @xmath1-ray emission , large uncertainties still remain in understanding the prompt emission site , namely , the distance of the emission site from the explosion centre @xmath0 , with controversial evidence . there are three possible sites discussed in the literature . one is the standard internal - shock site which depends on the fluctuation time scale @xmath7 seen in grb light curves ( e.g. , rees & mszros 1994 , see piran 2005 , mszros 2006 for reviews ) . it can have a large range of @xmath8 cm because @xmath7 and @xmath2 vary largely from burst to burst . the second is the photospheric radius at @xmath9 cm at which the prompt emission arises as a combination of the photosphere thermal emission and a comptonized component above it , the latter being induced by some energy dissipation process below and above the photosphere ( e.g. rees & mszros 2005 ; ryde et al . 2006 ; thompson et al . 2007 ) . the third one is a large radius ( @xmath10 cm ) as is supported by the swift xrt data ( lazzati & begelman 2005 ; lyutikov 2006 ; kumar et al . 2007 ) and fermi data of grb 080916c ( abdo et al . 2009 ; zhang & peer 2009 ) , possibly due to magnetic dissipation ( e.g. , lyutikov & blandford 2003 ) . the rapidly responding ability of a few grb - dedicated ground or space based optical telescopes , e.g. , rotse ( akerlof et al . 2003 ) , raptor ( vestrand et al . 2002 ) , tortora ( racusin et al . 2008 ) and the uvot ( roming et al . 2005 ) on aboard the swift satellite , has enabled the time - resolved detection of bright prompt optical emission before the @xmath1-rays die off , for about a dozen of grbs . five of these grbs , i.e. , 041219a ( vestrand et al . 2005 ) , 050820a ( vestrand et al . 2006 ) , 051111 ( yost et al . 2007a ) , 061121 ( page et al . 2007 ) and 080319b ( racusin et al . 2008 ) , show a temporal correlation between the strongly variable optical flux and the @xmath1-ray pulses , which suggests that the optical emission most likely shares the same dynamical process that is responsible for the highly variable @xmath1-ray emission . while the other four bursts have optical flux densities below or marginally consistent with the extrapolations from the low - energy power law of the @xmath1-ray spectra , the optical flux density in grb 080319b exceeds the @xmath1-ray extrapolation by 4 orders of magnitude ( racusin et al . 2008 ; kumar & panaitescu 2008 ) , suggesting that for this burst alone the optical emission has a spectral origin different from that of the @xmath1-rays . in this paper , for the four grbs - 041219a , 050820a , 051111 and 061121 - we assume that the prompt optical and the @xmath1-ray emissions are components belonging to the same synchrotron radiation continuum of a group of hot electrons . based on this assumption , the self - absorption frequency of the synchrotron electrons , @xmath11 , which causes a break in the long - wavelength part of the continuum , can be determined or constrained by studying the optical - to-@xmath1-ray spectral energy distribution ( sed ) to interpret the diversity in the prompt optical / @xmath1-ray temporal correspondence . ] . since @xmath11 is dependent on the properties of the prompt emission source , such as the distance of the emission site from the explosion centre @xmath0 , the bulk lorentz factor ( lf ) @xmath2 and the magnetic field @xmath3 of the source , from @xmath11 we can determine or make constraints on @xmath0 for these bursts , using information on @xmath2 and @xmath3 obtained in other ways . this is the main goal of this paper . since the prompt optical and @xmath1-ray components in grb 080319b are most likely of different spectral origins because of its peculiar sed shape , our approach is not applicable to this burst . on the other hand , for some other long grbs the rapid response of the dedicated rotse telescope has returned only upper limits of the optical flux density during the prompt phase ( yost et al . another goal of this paper is to get constraints on @xmath0 for these optically `` dark '' bursts and to study whether the prompt optical non - detection is caused by a heavier self - absorption due to a closer emission site to the explosion centre . in this paper , we first derive analytically @xmath11 in terms of @xmath0 , @xmath2 , @xmath3 and the emission properties in sec . the arguments that support our assumption of one synchrotron continuum component for both optical and @xmath1/x - ray are given in sec . we derive in sec . 4 the constraints on @xmath0 through @xmath11 explicitly , by determining the location of @xmath11 in the optical - to-@xmath1-ray sed and considering all possible spectral regimes . we apply this method to a prompt optical detection grb sample and a prompt optical non - detection sample which are described in sec . the results are presented in sec . 6 . finally the conclusion and discussions are given in sec . the grb high energy emission spectrum is characterized by a smoothly joint broken - power - law form ( band et al . 1993 ) . thus the relativistic electrons responsible for the grb prompt emission due to either synchrotron or synchrotron self - inverse compton ( ssc ) radiation are in a piece - wise two - power - law energy distribution : @xmath12 where @xmath13 is such defined that @xmath14 is the number density of electrons with energy in the interval of @xmath1 to ( @xmath15 ) , and @xmath16 is the minimum energy of these relativistic electrons ( for convenience we omit the factor @xmath17 in the electron energy @xmath18 throughout the text when electron energy is mentioned ) . note that although this distribution set - up is phenomenologically based on the two - power - law shape of the high energy radiation spectrum observed in grbs , it has specific physical meanings . within the shock acceleration scenario , newly accelerated electrons with a minimum energy @xmath19 and a power - law energy distribution are continuously injected . these electrons lose energy through radiative cooling , and the instantaneous electron spectrum steepens above a critical energy @xmath20 . all the electrons with energy larger than @xmath20 radiate away their energy within a time shorter than the dynamical time . when @xmath21 , our notation corresponds to @xmath22 and @xmath23 . when @xmath24 , the cooling causes a flatter power law between @xmath20 and @xmath19 , even though the newly accelerated electrons are injected in the energy range above @xmath19 . for this case one has @xmath25 and @xmath26 . in summary within the shock acceleration scenario one has @xmath27 and @xmath28 . more generally , one can also have a scenario that invokes continuous heating and cooling of electrons ( e.g. that envisaged in the reconnection models ) , and @xmath29 then reflects the intrinsic break in the steady state electron spectrum . in any case , our treatment is generic , which does not depend on the concrete particle acceleration mechanism and the origin of @xmath29 . the broken power - law electron energy spectrum naturally gives rise to a piece - wise power law photon spectrum as commonly observed : @xmath30 where @xmath31 is the observed flux density ( in units of mjy ) , @xmath32 and @xmath33 are the observed characteristic emission frequencies of the electrons with energy @xmath16 and @xmath29 , respectively , and @xmath33 is usually the peak frequency of the @xmath34 spectrum . the low - energy power law @xmath35 does not extend to low frequency indefinitely . without synchrotron self absorption , the spectral index below @xmath32 changes to 1/3 , regardless of whether @xmath21 or @xmath24 . below a certain frequency @xmath36 , the synchrotron self absorption starts to play a significant role at frequencies lower than @xmath11 the emitted photons are thermalized with the electrons . the self - absorption frequency @xmath11 is such defined that at this frequency the self - absorption optical depth @xmath37 . let us determine @xmath11 for an emitting grb ejecta moving with a lorentz factor ( lf ) @xmath2 at a distance @xmath0 from the center of the explosion . in the ejecta comoving frame ( hereafter the quantities measured in this frame are marked with the prime sign ) , @xmath38 can be determined by equating the un - absorbed source surface flux density , @xmath39 , at @xmath38 to a blackbody surface flux density with temperature @xmath40 in the rayleigh - jeans regime ( see appendix for the derivation ; also see sari & piran 1999 , li & song 2004 , mcmahon , kumar & piran 2006 ) : @xmath41 @xmath42 where @xmath43 , @xmath44 is the energy of electron whose characteristic emission frequency is @xmath38 , and @xmath45 is the boltzmann constant . the numerical factors @xmath46 and @xmath47 are functions of @xmath48 only whose values range from 1.2 to 4.5 and from 1.2 to 7.0 , respectively , for the range of observed @xmath48 values , but they have been neglected in previous works while we include them here . transforming the frequency to that measured in the observer s frame gives @xmath49 , where @xmath50 is the redshift of the grb host galaxy . measuring in the host comoving frame , the source has an isotropic luminosity of @xmath51 . this luminosity is also given by @xmath52 , where @xmath53 is the luminosity distance of the grb and @xmath54 is the observed peak flux density . thus we have @xmath55 $ ] . after applying these relations , ( 3 ) becomes @xmath56 after substituting the electron s energy @xmath1 using the following relation between @xmath1 and the photon frequency @xmath57 , @xmath58 the self - absorption frequency is calculated as : for synchrotron @xmath59 @xmath60 and , for ssc @xmath59 @xmath61 in the results above , the flux density , e.g. , @xmath54 , is in units of mjy , @xmath62 and the convention @xmath63 , e.g. , @xmath64 hz and @xmath65 gauss , is used for other quantities ; the same notations will be used in the rest of the paper . in the following , our discussion will be based on the synchrotron radiation only . but for the use of reference the expression of @xmath11 for ssc is also given here . in this paper , we make an assumption that for the two samples ( see sec . 5 for a description of the sample selection criteria ) studied , the optical and @xmath1/x - ray photons belong to a same synchrotron continuum spectrum generated by a same group of hot electrons . this assumption is based on the following three considerations . first , grb prompt @xmath1/x - ray emission is often interpreted as synchrotron radiation of a group of non - thermal electrons . if that is the case , the synchrotron spectrum _ must _ have a continuum extending to the low - frequency regime up to the optical band , presumably with a gentle slope of @xmath66 unless it has a self - absorption break . secondly , the optical flux density is expected to always lie below or near the extrapolation from the @xmath1-ray spectrum . this is generally consistent with observations in our sample ( see sec . 5 below and fig . 2 ) . finally , a temporal correlation between the optical flux variation and the @xmath1-ray lc is observed for the 4 grbs in the our sample , suggesting the two components likely have the same dynamical origin . this is the major supporting evidence for our assumption . we notice that there are other scenarios on prompt @xmath1/optical emission that have been discussed in the literature . these include the synchrotron + ssc model ( kumar & panaitescu 2008 ; racusin et al . 2008 ) , the models invoke different emission radii for optical and @xmath1-ray emissions ( li & waxman 2008 ; fan , zhang & wei 2009 ) , and the model invokes two shock regions at a same emission radius ( yu , wang & dai 2009 ) . these models are relevant to grb 080319b , which clearly requires a separate spectral component to interpret the prompt optical emission . for most other bursts studied in this paper , although these models are not ruled out , they are not demanded by the data . our simple one - component model is adequate to interpret these bursts , and we will hereafter adopt this one - component assumption . let us consider the synchrotron emission as the prompt emission mechanism . depending on the locations of @xmath32 and @xmath11 , the ratio of the optical flux density to the flux density at @xmath33 is given for four different spectral cases by @xmath67 @xmath68 fig . [ fig : spec_case ] illustrates these four spectral cases . there is also a variation of case ( iii ) ( let us call it case iii.5 ) : @xmath69 , for which the flux density ratio is @xmath70 . we will come back to this case later and show that it is almost exactly the same as case ( iii ) . in some grbs simultaneous observations of prompt x - ray and @xmath1-ray emissions are made . the x - ray spectrum corrected for the photo - absorption in the soft end always nicely matches the power law extrapolated from @xmath1-ray spectrum below @xmath33 , without a need to invoke a break ( e.g. cenko et al . 2006 ; romano et al . 2006 ; page et al . this requires that both @xmath11 and @xmath32 be smaller than @xmath71 0.3 kev , the lower end of the swift xrt band pass . we take this as a constraint in our analyses . we aim to constrain @xmath0 based on the spectral information , such as @xmath54 , @xmath72 and @xmath48 , using @xmath11 as a proxy . we know from sec . 2 that @xmath11 is expressed in terms of @xmath54 , @xmath32 and @xmath0 . so for each spectral case , we substitute the appropriate @xmath11 expression into the flux density ratio equation or into the constraint on @xmath11 implied by the definition of that spectral case ( eq . 8) , and then get the @xmath0 constraints . for case ( i ) , substituting @xmath11 and letting @xmath73 hz ( for r band ) , we have @xmath74 there is also a justification criterion due to the case definition @xmath75 , where @xmath11 is directly determined from the flux density ratio in eq . ( 8) by @xmath76 for case ( ii ) , the flux density ratio does not depend on @xmath11 and hence on @xmath0 . one justification criterion for this case is that the spectral slope from the optical to the x- or @xmath1-rays has to be consistent with @xmath48 , i.e. , @xmath77 . another criterion due to the case definition is @xmath78 . substituting with the expression of @xmath11 , the latter gives @xmath79 @xmath80 for case ( iii ) , when substituting the appropriate @xmath11 expression in eq . ( 6 ) into the flux density ratio relation , it gives the expression of @xmath0 in @xmath81 we find for case ( iii.5 ) that the @xmath0 expression obtained by substituting the appropriate @xmath11 expression into eq.(8 ) is exactly the same as eq . ( 12 ) except that @xmath46 is replaced with @xmath47 . we find the ratio @xmath82 lies in the range of ( 1 , 1.6 ) for @xmath83 ( see in appendix ) . according to the definition of case ( iii ) , @xmath84 . plugging this constraint of @xmath32 into eq . ( 12 ) and its counterpart equation for case ( iii.5 ) , it gives @xmath85 @xmath86 note that in writing this constraint we already combined the one for case ( iii ) with the one for case ( iii.5 ) . it is done by using @xmath46 in the lower limit and @xmath47 in the upper limit , such that the combined constraint is conservative . from now on we expand the case ( iii ) definition to be @xmath87 so that it includes case ( iii.5 ) . two last pieces of constraining information for case ( iii ) are from @xmath11 , i.e. , @xmath88 and @xmath89 @xmath90 , where we use the expressions for @xmath11 given in eq . ( 6 ) ; the first @xmath11 expression is for the situation of @xmath91 and the second for the situation of @xmath92 . the first constraint gives @xmath93 @xmath94 the @xmath11 expression in the second constraint , @xmath75 , contains @xmath32 which can be expressed in terms of @xmath72 and @xmath0 from the @xmath0-expression for this spectral case . after substituting for @xmath32 , the second constraint gives @xmath95 @xmath96^{\frac{1}{2/3 - 2\b_1 } } f_{\nu_{opt}}^{\frac{1}{2 } } \frac{d_{l,28}}{(1+z)}.\ ] ] here the constraint @xmath75 gives a lower limit of @xmath0 , contrary to what is inferred from the conventional relation between @xmath11 and @xmath0 . it is because in this subtle occasion @xmath11 depends not only on @xmath0 but also on @xmath32 , and @xmath32 is expressed in terms of the optical flux density and @xmath0 , hence the combined @xmath0-dependence of @xmath11 is positive . the final constraint for case ( iii ) should be the overlapping region among those three constraints obtained . for case ( iv ) , @xmath32 can be obtained from the flux density ratio relation by @xmath97^{\frac{1}{\b_1 - 1/3}},\ ] ] whose value will be used to justify the case definition @xmath84 . another constraint from the case definition is @xmath98 . substituting with the @xmath32 expression , this gives @xmath99 @xmath96^{\frac{1}{2/3 - 2\b_1 } } f_{\nu_{opt}}^{\frac{1}{2 } } \frac{d_{l,28}}{(1+z)}.\ ] ] notice that two contrary constraints , @xmath75 in case ( iii ) and @xmath78 in case ( iv ) , give exactly the same constraints on @xmath0 . this is because the @xmath11 expression in both cases contains @xmath0 and @xmath32 , but in case ( iii ) @xmath32 is a strong function of @xmath0 , @xmath100 , while in case ( iv ) @xmath32 is a function of the flux density ratio only . thus in case ( iii ) the @xmath0-dependence is reversed between two sides of the inequality relation @xmath75 . to summarize , the overall constraints on @xmath0 are : eq . ( 9 ) for case ( i ) , eq . ( 11 ) for case ( ii ) , eq . ( 13 - 15 ) for case ( iii ) and eq . ( 17 ) for case ( iv ) . in addition , when they are available , the calculated @xmath11 or @xmath32 must satisfy the case definitions . if the optical flux density has only an upper limit , the above @xmath0-constraints must be taken with a conservative point of view wherever @xmath72 is involved . let @xmath72 represent the measured upper limit . for case ( i ) , eq . ( 9 ) will give an upper limit for @xmath0 . for case ( ii ) , eq . ( 11 ) remains . for case ( iii ) , eq . ( 13 ) is left with only the upper limit of @xmath0 , eq . ( 14 ) remains and eq . ( 15 ) is useless . for case ( iv ) , eq . ( 16 ) gives a lower limit of @xmath32 which can be used to justify the case definition ; eq . ( 17 ) is useless . but we recall that eq . ( 17 ) is obtained by substituting the @xmath32 expression into the definition constraint @xmath98 . here , instead of using the @xmath32 expression , we plug in the upper boundary of @xmath32 : @xmath101 , then a new lower limit of @xmath0 is obtained for case ( iv ) : @xmath102 @xmath103 now we turn to the real grb data to which our method developed above can be applied . first we construct a small sample of grbs whose prompt optical emission is not only detected but is also variable and temporally correlated with the @xmath1-rays . excluding grb 080319b that requires a new spectral component for optical emission , we identify four of grbs in the sample , all belonging to the long - duration class . three of them show complex fluctuations in their prompt @xmath1-ray and optical lcs while the fourth is a single , smoothly peaked event , so we utilize multiple time intervals for each of the three . the emission properties of each time interval are listed in tab . 1 . grb 041219a is a very long ( @xmath104 500 s ) burst and has multiple peaks in @xmath1-ray lc . it has three optical detection intervals ; the first two are correlated with the first @xmath1-ray peak and the third with the second @xmath1-ray peak ( vestrand et al . 2005 ) . grb 050820a is a similar one except that it has denser optical temporal coverage . its optical lc is decomposed into two components : a smooth component with fast rise and power - law decay , and a strongly variable component superposed on it ( vestrand et al . the smooth component is well accounted for by the early afterglow due to the grb outflow interacting with the ambient medium . the variable component is found to correlate with the @xmath1-ray peaks , suggesting it has the same origin as the @xmath1-rays . in tab . 1 the optical emission properties for this burst are for the residual optical component after subtracting the smooth component . grb 051111 has a single fred ( fast rise and exponential decay ) peak in @xmath1-rays lasting @xmath105 90 s. the first optical observation starts at 30 s after the burst trigger when the @xmath1-ray lc began to decay ( yost et al . the prompt optical lc ( before the @xmath1-rays die off ) decays more steeply than that of the later optical afterglow emission . the prompt optical emission has an excess above the back extrapolation of the later optical afterglow component . it has a decay slope statistically compatible with that of the @xmath1-ray lc , and its flux density is also compatible with the spectral extrapolation of the @xmath1-rays ( yost et al . this is good evidence that the prompt optical excess has the same origin as the @xmath1-rays . we use the flux density of the excess - not the total - optical emission in the first optical observation time interval . grb 061121 has two separate @xmath1-ray peaks . the last peak was caught by xrt , uvot and rotse , and it appears in lcs in all bands ( page et al . we use the emission properties of two time intervals of the last peak , one during the rising phase , and the other just at the peak . we show the seds for all time intervals of the 4 grbs in fig . [ fig : obs ] using the data adapted from their original publications . two seds show almost no break between the optical and the @xmath1-rays ( corresponding to the theoretical spectrum case ii ) or a break very close to the optical . in all other cases , at least one break is needed between the optical and the @xmath1-rays bands . the break(s ) could be @xmath11 , @xmath32 , or both . besides this first data sample for optical detections , we also define a second data sample which is composed of those optically dark " grbs during the prompt phase . this sample is adopted from yost et al . ( 2007a ) , who reported the bursts whose prompt phase was observed by rotse but only upper limits on the optical flux were retrieved . each burst has either a single or multiple time intervals of rotse exposure during the prompt phase . for bursts with multiple optical time intervals , we use the interval which has the smallest measurement error in the @xmath1-ray flux density @xmath54 - usually the interval that has the brightest @xmath1-ray flux . the only exception is grb 061222a , for which three time intervals are used . this is because all the three intervals are located at the brightest part of the @xmath1-ray lc . they all have small errors in @xmath54 , and the instantaneous @xmath1-ray spectral index @xmath48 are available for all three intervals . the motivation of selecting this sample is the following : even if there is no direct detection of prompt optical emission , one can speculate the existence of a prompt optical emission component that tracking the @xmath1-ray lcs , which is the spectral extension of the prompt @xmath1-ray spectrum into the optical band . the flux level of this component must be fainter than the upper limit set by the rotse observations . we want to check whether inferred @xmath0 constraints of this sample is consistent with the sample with optical detection , and whether the non - detection of optical emission of this sample is due to stronger synchrotron self absorption associated with a smaller @xmath0 . we apply the constraints on @xmath0 derived in sec . [ sec : r - drv ] to the first sample with prompt optical detections . the results for all four broad - band spectral cases are listed in tab . 1 . for most bursts in the sample , case ( ii ) can be immediately ruled out because usually @xmath106 . case ( iv ) is also ruled out for some bursts because the calculated @xmath32 is @xmath107 0.3 kev . for grb 041219a , case ( i ) can be ruled out for its first time interval because the case definition is not satisfied by the calculated @xmath11 . actually this interval is consistent with case ( ii ) , i.e. , the optical intensity is consistent with the simple power - law extrapolation from the @xmath1-ray spectrum . we plot the permitted @xmath0-ranges for each observation time interval in the sample for all possible spectral cases as floating bars in fig . [ fig : hist ] . the observed optical to @xmath1-ray sed restricts @xmath11 from being much larger than @xmath108 . accordingly , the results in fig . [ fig : hist ] give a constraint on the emission site for most time intervals of this sample : @xmath109 a few @xmath110 cm . for two time intervals ( 041219a int . 3 and 050820a int . 3 ) in the sample , some spectral cases can be ruled out , thus the @xmath0-constraint can be pinned down to @xmath111 cm . similar results for the sample with only prompt optical upper limits are plotted in fig . 4 . in about half ( 6/13 ) of the sample a heavy self absorption , i.e. large @xmath11 , is needed to account for the optical deficit , corresponding to the spectral case ( i ) and ( iii ) , which implies a constraint of @xmath112 cm . for the remaining half ( 7/13 ) of the sample , a spectral break at @xmath32 which is below @xmath113 but is much larger than @xmath108 alone can give rise to the required deficit in optical , while @xmath11 can keep being smaller than @xmath108 , corresponding to case ( iv ) . thus for this half of the sample , we can provide no constraint on @xmath11 and hence on @xmath0 . comparing fig . 4 with fig . [ fig : hist ] , we find that there are always overlapping regions between the permitted @xmath0-ranges for the two samples . therefore we can _ not _ draw any statistically significant distinction between these two samples as regards the constraints on their emission sites . [ cols="<,<,<,<,^,<,<,<,^,^ " , ] + strictly speaking , the constraint on @xmath0 is dependent on the source lf @xmath2 and the magnetic field strength @xmath3 . independent determinations of @xmath2 and @xmath3 for each grb in our sample are not easy . so in this work we adopt the theoretically anticipated values in the standard internal - shock model : @xmath114 300 , @xmath115 g. in the following we will justify these adopted values based on the available information of the grbs in our sample . recently molinari et al . ( 2007 ) inferred @xmath116 for two grbs by directly observing the deceleration time of the grb outflow . this value is in agreement with what we adopt . in addition , here we present some attempts to estimate @xmath2 for each grb in our first sample ( with optical detections ) using three independent arguments , which suggests that the choice of @xmath117 is reasonable . \(1 ) the variability time scale argument . for the standard scenario in which the variability time scale is that of the central engine and that the ejecta form a conical jet with opening angle much larger than @xmath118 , the observed variability time scale should be at least the angular spreading time . this gives the constraint @xmath119^{1/2},\ ] ] where @xmath120 is the observed variability time scale , defined as the rising or decaying time scale of the pulses in the lc . from the data we find that @xmath120 is @xmath121 1 - 15 s for our sample . the @xmath2-constraint from this argument is rather weak compared with the other two constraints described below . \(2 ) the photon annihilation opacity argument . a grb releases a huge amount of energy ( @xmath122 erg isotropically ) in terms of high energy photons from a small volume , which are subject to the photon - photon annihilation ( e.g. baring & harding 1997 ; lithwick & sari 2001 ) . suppose @xmath123 is the maximum photon energy detected in the burst , the fact that a single power - law or a piece - wise power - law spectrum is detected for most grbs up to @xmath123 implies that the optical depth of photon - photon pair production @xmath124 for the photons of energy @xmath123 is smaller than unity . @xmath124 is related to the total number of photons and the size of the emission region , the latter of which can be expressed in terms of @xmath2 and @xmath120 within the internal shock model . thus this requirement can impose a lower limit on @xmath2 within the internal shock model . more generally the opacity argument can give a constraint in the @xmath125 space ( gupta & zhang 2008 ; murase & ioka 2008 ; zhang & peer 2009 ) . our following treatment applies to the internal shock model that is commonly discussed in the literature . the minimum photon energy at which the photons are able to annihilate with photons of energy @xmath123 is : @xmath126 the power - law form spectrum just below @xmath123 is described as in @xmath127 where @xmath128 is the detected photon energy , @xmath129 - in units of @xmath130s@xmath131@xmath132mev@xmath133 $ ] - is the number of photons detected per unit time per unit area per unit energy at @xmath128 , @xmath134 is equal to @xmath129 at @xmath135 mev , and @xmath136 is the photon index . we followed the formulae of lithwick & sari ( 2001 ) in their limit a case to calculate the lower limit of @xmath2 due to @xmath137 for our optical detection sample . the results , as well as the observational properties that are used , are summarized in tab . [ tab : opa ] . note that for all bursts in the sample , @xmath138 for reasonable values of @xmath2 ( e.g. , @xmath139 ) . thus our calculation has implicitly assumed that the power - law form spectrum detected below @xmath123 would actually extend well beyond @xmath123 and up to @xmath140 , which is usually below the detector s limited bandpass . more rigorous calculations require the knowledge of the spectral shape below peak energy of the spectrum ( gupta & zhang 2008 ) . however , the optical depth is much more sensitive to @xmath141 than the spectral indices , and the derived @xmath2 constraint is not significantly modified by performing the more rigorous treatment . the above treatment is adequate to serve our purpose . \(3 ) the deceleration time constraint . in the well established external shock model for grb afterglows ( e.g. , mszros & rees 1997 ; sari , piran & narayan 1998 ; chevalier & li 2000 ; see piran 2005 for a review ) , the onset of afterglow marks the time , as known as the deceleration time @xmath142 , when one half of the total kinetic energy of the grb outflow @xmath143 is deposited to the shocked circumburst medium . the afterglow light curve should rise before @xmath142 and decay in a power law with time after @xmath142 ( e.g. , molinari et al . 2007 ) . the deceleration time @xmath142 is determined by @xmath143 , outflow bulk lf @xmath2 and the density of the circumburst medium , either a constant density medium ( ism ) or a wind - like medium [ @xmath144 : @xmath145 where @xmath146 is the proton number density of the ism medium , and @xmath147 g cm@xmath131 is the wind medium density normalization parameter . the isotropic equivalent kinetic energy @xmath149 can be related to the isotropic energy release in @xmath1-ray radiation @xmath150 by an energy conversion efficiency factor @xmath151 . the afterglow observations ( x - rays and optical ) for our sample show either a power law decay starting from the earliest observation interval or , a long - lasting shallow decay followed by a normal power law decay . it suggests that the afterglow onsets should be earlier than the start of the single power law decay or the start of the shallow decay . one can use the first observation data in the decaying afterglow phase to constraint the deceleration time to be earlier than the observational epoch ( e.g. , zhang et al . 2006 ) . from the data , we find the deceleration time @xmath152 s for grb 041219a , and @xmath153 100 - 500 s for the other 3 grbs ( 050820a , 051111 , 061121 ) . assuming @xmath154 , @xmath155 @xmath156 and @xmath157 , we find : for ism , @xmath158 for 041219a , @xmath159 for the other 3 grbs ; for wind , @xmath160 for 041219a , @xmath161 for the other 3 . the results are summarized in tab . [ tab : dec ] . we have checked of the compliance with closure relationships predicted by the external shock models for these bursts during the afterglow phase . it turns out that two ( grb 050820a and 061121 ) out of the 4 grbs are consistent with and in favour of the ism environment scenario , while the other two are consistent with both scenarios and can not discriminate between them . to summarize , with the available data , one can only constrain but can not determine @xmath2 of the grbs in our sample . on the other hand , all three constraints derived from the data are consistent with @xmath162 adopted in our calculations . in particular , the @xmath2-constraint derived from the argument ( 3 ) , which is the most stringent one among the three arguments , indicate that the assumed value of @xmath162 is reasonable . the @xmath3 value in the emission region is a function of @xmath0 . there are two possible origins of the magnetic field in the emission region . the first component is the global magnetic field entrained by the ejecta from the central engine . let s assume @xmath163 g at the central engine , a typical value for a fast rotating magnetar or a fast rotating black hole accretion disk system - the two most plausible grb central engine candidates . the @xmath3 value drops as @xmath164 with @xmath0 up to the light cylinder , and then drops as @xmath165 thereafter ( goldreich & julian 1969 ) . given that the central rotating source has a radius of @xmath166 cm and a rotation period of @xmath167 ms , at a radius @xmath168 cm , the field that is carried within the outflow has a strength of @xmath169 g. the second @xmath3 component is a random field generated _ in - situ _ in the emission region , likely in a relativistic shock via the weibel instability ( medvedev & loeb 1999 ) . this random field also follows the same @xmath0-dependence and is of the same order as the engine - related @xmath3 component if @xmath170 - the ratio of the post - shock magnetic energy density to the total energy density - is not too small ( zhang & mszros 2002 ) . one can also briefly estimate the local field strength by relating the grb @xmath1-ray peak photon energy , typically @xmath171 mev , with the synchrotron characteristic frequency , @xmath172 , where @xmath1 is the typical energy of electrons . in the internal - shock model , @xmath173 , where @xmath174 is the ratio of the electron energy density over the total thermal energy density in the post - shock fluid , @xmath175 and @xmath176 is the electron energy spectral index . the parameter @xmath177 is the fractional energy gain of a proton passing the shock which depends only on the relative lf between the fluids downstream and upstream . for internal shocks , @xmath177 is _ not _ dependent on the shell bulk lf @xmath2 and is of the order of unity . for @xmath178 and @xmath179 , we have @xmath180 . thus the required @xmath3 value can be estimated as @xmath181 g. the three crude estimates are marginally consistent with each other . we have taken @xmath182 g as the typical value throughout the text . of course a large uncertainty exists due to our lack of understanding on the field properties , but it is reconciled by the very weak dependence of @xmath0 on @xmath3 ( 1/4 power ) . kumar & mcmahon ( 2008 ) developed a general method of modelling grb s @xmath1-ray emission properties . their method considers the synchrotron and the ssc emission , respectively , as the radiation mechanism , takes into account the radiative cooling of electrons , and uses observed emission properties ( such as the peak flux density and the pulse duration ) to search for the allowed space of the model parameters such as @xmath0 and @xmath2 . here we also apply their method for the synchrotron case to our optical detection sample using their code , and compare the results with ours . we add a new constraint into the module that controls the allowed model parameter space , which is that the optical flux density calculated from the model has to match the observed one within a @xmath183% range . to use this detailed modelling method , we have to specify which standard synchrotron spectral regime a grb emission interval is in . the spectral indices ( @xmath48 ) of the optical detection sample in tab . 1 have a variety of values around -1/2 , based on which we classify the sample into 4 categories and apply the detailed modelling method accordingly . ( 1 ) for those time intervals that are most probably consistent with the @xmath184 regime ( 051111 and 061121 int . 1 ) , this method gives @xmath185 cm . ( 2 ) for those possibly consistent with both @xmath184 and @xmath186 where @xmath187 ( 041219a int . 3 and 050820a int . 3 ) , the method in the @xmath184 regime gives @xmath5 cm , while in the @xmath186 regime it gives no allowed @xmath0-space but if we relax the @xmath72 constraint , it gives @xmath188 cm . ( 3 ) for those possibly consistent with both @xmath184 and @xmath186 where @xmath189 ( 041219a int . 2 , 050820a int . 1 and 2 ) , the method in the @xmath184 regime gives @xmath5 cm ; however , the code provided by the authors is not applicable when @xmath190 . ( 4 ) the last category are those inconsistent with @xmath184 but probably consistent with @xmath186 if @xmath189 ( 041219a int . 1 and 061121 int . 2 ) for which the code is not applicable . overall , we find that for the major part of the optical detection sample where the detailed modelling method ( kumar & mcmahon 2008 ) is applicable the allowed spaces for @xmath0 from this method are about @xmath191 cm . this is approximately consistent with the findings from our approach that @xmath0 is @xmath192 ( a few @xmath193 cm for most of the intervals in the sample , and @xmath194 cm for the remaining two intervals in the sample . based on the assumption that the prompt optical and @xmath1-ray emissions belong to the same synchrotron continuum of a group of hot electrons , we make constraints on the location of the prompt emission site for a sample of grbs whose prompt optical emission is detected to temporally tracking the @xmath1-ray lcs , by determining the location of @xmath11 in their sed . our analysis shows that for most of the intervals in this sample the distance of the prompt emission site from the explosion centre @xmath0 is @xmath192 ( a few @xmath195 cm , and for the remaining two intervals , the emission site is @xmath196 cm away from the explosion centre . the dependence of the distance constraint on the grb outflow lf @xmath2 is not negligible . on the other hand , various indirect observational constraints on @xmath2 point to @xmath197 ( e.g. molinari et al . 2007 ; and even @xmath198 for grb 080916c , abdo et al . 2009 ) . the derived observational constraints on @xmath2 for bursts of this sample ( sec . [ sec : gam - b ] ) are consistent with such an inference . in our paper , we take @xmath162 as a typical value . a higher @xmath2 would only make the above distance constraint even larger . our knowledge of the local magnetic field strength @xmath3 is less certain , although the @xmath0-dependence of @xmath3 is weak . several crude estimates of @xmath3 based on the synchrotron radiation mechanism for grb prompt emission suggest @xmath199 g at @xmath200 cm . this typical value has been adopted in our calculations . the @xmath0-constraint we obtained is inconsistent with the photospheric emission model in which the prompt emission arises at the photosphere radius of @xmath201 cm ( rees & mszros 2005 ; ryde et al . 2006 ; thompson et al . 2007 ) . this result alone can not discriminate between the fireball internal shock model and the magnetic outflow model . by comprehensive modeling the grb prompt @xmath1-rays and early x - rays , kumar et al . ( 2007 ) concluded a prompt emission site of @xmath202 cm , which is supported by their further general modeling of the @xmath1-ray emission properties ( kumar & mcmahon 2008 ) . a large @xmath0 is derived for grb 080916c through the pair opacity constraint ( abdo et al . 2009 ; zhang & peer 2009 ) , the inference of @xmath0 in abdo et al . ( 2009 ) is based specifically on the internal shock model while zhang & peer ( 2009 ) gave a more model - independent constraint on @xmath0 . ] and for grb 080319b through the synchrotron self - absorption constraint ( racusin et al . 2008 ) and the ssc scattering optical depth constraint ( kumar & panaitescu 2008 ) . in summary , a large prompt emission distance from the central engine seems to be supported by three independent approaches , respectively , i.e. in @xmath1/x - rays ( e.g. , lazzati & begelman 2005 ; lyutikov 2006 ; kumar et al . 2007 ) , in gev @xmath1-rays ( abdo et al . 2009 ; zhang & peer 2009 ) , and in the optical band ( this paper ) . we have also studied a sample of grbs with prompt optical non - detections . applying the same technique , we do _ not _ find any inconsistency between their @xmath0 constraints and those of the optical detection sample . this result is inherited from the findings by yost et al . ( 2007b ) that no distinction in distributions of @xmath48 and @xmath203 can be drawn between the optically dark grbs and the grbs with optical detections . however , this is only because the currently limited instrumental sensitivity prevents a distinction from being drawn . deeper observations in the future in the optical band would provide further information regarding whether the optical deficit is due to a heavier synchrotron self - absorption in these bursts . if multi - color photometry near the optical band exists for the same time interval during the prompt phase , it would provide the local spectral index near the optical band ( provided that the extinction correction is properly made ) . this would be helpful to identify the spectral case the data satisfy . for example , the spectral indices near optical differ by @xmath204 between the cases ( iii ) and ( iv ) , and by @xmath205 between the cases ( i ) and ( iii ) . unfortunately , this kind of observational information is unavailable for all the time intervals of the optical detection sample we have considered in tab . the hope is that future multi - band prompt optical detections may be able to break the spectral case degeneracy and to tighten the @xmath0-constraint . our method is based on the assumption that the optical emission is emitted from the same group of electrons that produce the @xmath1-rays in the same site via the same synchrotron radiation . this _ one component _ assumption has some supports ( see sec . 3 ) but certainly is not conclusive . there are three other scenarios that have been discussed in the literature ( mostly motivated by interpreting grb 080319b ) . the first one invokes two emission zones for optical and @xmath1-rays . for example , in the internal - shock - model based residual collision scenario proposed by li & waxman ( 2008 ) , the shells with high lf contrast in a grb outflow collide first and merge at smaller radii , producing the @xmath1/x - rays . later those merged shells with low lf contrasts would collide mildly at later times and larger radii , giving rise to optical emission . alternatively , if the outflow is neutron rich , the proton shells tend to collide at smaller radii to power @xmath1/x - rays , while the free neutrons only decay at large radii and the decay products would be collided by later injected faster proton shells and power optical emission at larger radii ( fan et al . 2009 ) . in both scenarios , it is expected that the observed optical pulse peak emission time is delayed by @xmath206 with respect to the @xmath1-ray pulse peak emission time , which may be in principle tested if the data quality is high . these models however do not naturally predict a smooth extension of @xmath1-ray spectrum to the optical band without distinct spectral features . although they can not be ruled out by the data in our sample , they are more complicated than our one - zone model . secondly , our analysis is not applicable in the following scenario ( yu et al . 2009 ) : a pair of shocks ( reverse shock and forward shock ) arise when two shells collide in the internal - shock model ; different populations of electrons are accelerated in each of the two shocks and the two populations have different typical electron energies and different shock - generated magnetic strengths . the characteristic synchrotron frequencies are different - the forward shock produces the optical emission , while the reverse shock produces the @xmath1-ray emission . these two emissions are two spectrally independent components but are temporally correlated because the heating of the two electron populations arises from the same dynamical process . although this model may interpret the peculiar sed shape of grb 080319b whose prompt optical flux density exceeds the extrapolation from the @xmath1-ray spectrum by 4 orders of magnitude , it is unclear whether it can work properly for the bursts in our optical detection sample . finally , the synchrotron + ssc scenario ( kumar & panaitescu 2008 ; racusin et al . 2008 ) has been proposed to interpret grb 080319b . for our sample , there is no need to introduce a second distinct spectral component . we consider only pure synchrotron radiation in deriving constraints on @xmath0 , although we have given the @xmath11 estimation for ssc radiation whose value is similar to @xmath11 for synchrotron . in principle , it is possible that the observed emission from optical to @xmath1-rays is dominated by ssc . if this is the case , our approach of constraining @xmath0 by calculating @xmath11 may give results somewhat different from the synchrotron case for the same sample of bursts . we did not carry out the analysis for the ssc case because the detailed shape of the ssc spectra is much more complicated than the synchrotron one . note that our approach assumes that the optical and the @xmath1-rays are from the same group of electrons due to the same radiation mechanism . in the ssc scenario this approach is applicable only if the ssc component dominates a large spectral band from the @xmath1-ray down to the optical and the synchrotron component has to lie well below the optical band . this is usually not expected in the ssc models ( e.g. piran , sari & zou 2009 ) . in any case , if ssc is involved in interpreting any part of the spectrum in our sample , then our analysis is no longer applicable . we thank pawan kumar for valuable comments and suggestions and for sharing his code for our self - consistency check . we also acknowledge the anonymous referee whose comments have greatly helped to improve the quality and presentation of the paper . r - fs would like to thank zhuo li for useful comments and discussion . this work is supported by nasa grants nng06gh62 g and nnx07aj66 g ( bz ) and by the nsf grant ast-0406878 . 99 abdo a. a. et al . , 2009 , science , 323 , 1688 akerlof c. w. et al . , 2003 , pasp , 115 , 132 band d. et al . , 1993 , apj , 413 , 281 baring m. g. , harding a. k. 1997 , apj , 491 , 663 blake c. h. et al . , 2005 , nature , 435 , 181 bykov a. m. , mszros p. , 1996 , apj , 461 , l37 cenko s. b. et al . , 2006 , apj , 652 , 490 chevalier r. , li z .- y . 2000 , apj , 536 , 195 daigne f. , mochkovitch r. , 1998 , mnras , 296 , 275 doi h. , takami k. , yamazaki r. , 2007 , apj , 659 , l95 eichler d. , waxman e. , 2005 , apj , 627 , 861 fan y .- z . , zhang b. , wei d .- 2009 , phys . d , 79 , 021301 goldreich p. , julian w. h. , 1969 , apj , 157 , 869 granot j. , piran t. , sari r. , 1999 , apj , 527 , 236 gupta n. , zhang b. , 2008 , mnras , 384 , l11 hill g. , prochaska j. x. , fox d. , schaefer b. , reed m. , 2005 , gcn circ . , 4255 kumar p. et al . , 2007 , mnras , 376 , l57 kumar p. , mcmahon e. , 2008 , mnras , 384 , 33 kumar p. , panaitescu a. , 2008 , mnras , 391 , l19 li z. , song l. m. , 2004 , apj , 608 , l17 li z. , waxman e. , 2008 , apj , 674 , l65 lithwick y. , sari r. , 2001 , apj , 555 , 540 lyutikov m. , blandford r. , 2003 , preprint ( astro - ph/0312347 ) lyutikov m. , 2006 , mnras , 369 , l5 lazzati d. , begelman m. c. , 2005 , apj , 641 , 972 mcmahon e. , kumar p. , piran t. , 2006 , mnras , 366 , 575 medvedev m. v. , loeb a. , 1999 , apj , 526 , 697 mszros p. 2006 , rep . prog . 69 , 2259 mszros p. & rees m. j. 1997 , apj , 476 , 232 molinari e. et al . , 2007 , a&a , 469 , l13 murase k. , ioka k. , 2008 , apj , 676 , 1123 page k. l. et al . , 2007 , apj , 663 , 1125 panaitescu a. , kumar , p. , 2000 , apj , 543 , 66 peer a. , waxman e. , 2004 , apj , 603 , l1 piran t. , 2005 , rev . phys . , 76 , 1143 piran t. , sari r. , zou y .- c . , 2009 , mnras , 393 , 1107 racusin j. l. et al . , 2008 , nature , 455 , 183 rees m. j. , mszros p. , 1994 , apj , 430 , l93 rees m. j. , mszros p. , 2005 , apj , 628 , 847 romano p. et al . , 2006 , a&a , 456 , 917 roming , p. w. a. 2005 , space sci . , 120 , 95 rybicki g. b. , lightman a. p. , 1979 , radiative processes in astrophysics . wiley - interscience press , new york . ryde f. , bjrnsson c .- i . , kaneko y. , mszros p. , preece r. , battelino m. , 2006 , apj , 652 , 1400 sari r. , piran t. , 1999 , apj , 520 , 641 sari r. , piran t. , narayan r. 1998 , apj , 797 , l17 thompson c. , mszros p. , rees m. j. , 2007 , apj , 666 , 1012 vestrand w. t. et al . , 2002 , proceedings of spie , 4845 , 126 vestrand w. t. et al . , 2005 , nature , 435 , 178 vestrand w. t. et al . , 2006 , nature , 442 , 172 yost s. a. et al . , 2007a , apj , 657 , 925 yost s. a. et al . , 2007b , apj , 669 , 1107 yu y. w. , wang x. y. , dai z. g. , 2009 , apj , 692 , 1662 zhang b. et al . , 2006 , apj , 642 , 354 zhang b. , mszros p. , 2002 , apj , 581 , 1236 zhang b. , peer a. , 2009 , apjl in press , preprint ( arxiv : 0904.2943 ) in this appendix , we provide a rigorous derivation of the blackbody equivalence equation ( eq . 3 ) that we use to calculate the self - absorption frequency ( eq . the derivation is carried in the grb ejecta comoving frame in which the relevant quantities are marked with the prime sign . the self - absorption frequency @xmath38 is defined as @xmath207 , where @xmath208 is the optical depth due to the self - absorption at @xmath209 . the optical depth @xmath210 deceases with the frequency , where @xmath211 is the self - absorption coefficient [ cm@xmath131hz@xmath131 ] and the integral is over the line - of - sight width of the emitting source . the integral can be calculated directly only when we have the exact information on the number density of the emitting particles and its distribution over the length , which is not easy . instead of directly calculating the integral @xmath213 , we turn to derive the emission coefficient @xmath214 [ erg s@xmath131@xmath156hz@xmath131sr@xmath131 ] and then express the integral of @xmath211 over width into the integral of @xmath214 over width , the latter is just the specific intensity at the source surface which is directly observable . the synchrotron radiation spectrum , or specific radiation power , of an electron with lf @xmath1 gyrating in a magnetic field @xmath3 with a pitch angle @xmath136 is @xmath215 where @xmath128 is the electron charge and @xmath216 is the characteristic photon frequency of the electron . the function @xmath217 has the asymptotic form @xmath218 where @xmath219 is the gamma function of argument 1/3 , and it reaches the maximum @xmath220 at @xmath221 . one integral property of the function @xmath222 is @xmath223 where @xmath224 is the gamma function of argument @xmath225 . we will use this property later . the self - absorption coefficient for any radiation mechanism is given ( rybicki & lightman 1979 ) by @xmath226},\ ] ] where @xmath227 is the single electron s specific radiation power , and @xmath14 is the number density of electrons with energy in the interval from @xmath1 to ( @xmath15 ) . in the case of grb , @xmath13 has a two - power - law form and was described in sec . 2 of the paper . we rewrite it here as @xmath228 where @xmath229 is the normalization constant . note that this distribution set - up is phenomenologically based on the two - power - law shape of the high energy radiation spectrum observed in grbs ; in the context of some specific particle acceleration scenario , e.g. , shock acceleration , where two characteristic electron energies , i.e. , the injection energy @xmath19 and the cooling energy @xmath20 , are involved , there will be @xmath27 and @xmath230 . thus @xmath231 / \partial \g = ( -p-2)n(\g)/\g^3 $ ] , where @xmath176 could be either @xmath232 or @xmath233 depending on the location of @xmath1 . for the synchrotron radiation ( eq . [ power_syn ] ) the self - absorption coefficient would be @xmath234,\ ] ] where @xmath235 . then we consider two different locations of @xmath236 : @xmath237 and @xmath238 , respectively . if @xmath237 , then @xmath222 falls in the @xmath239 asymptotic regime . one can transform the integral in eq . ( [ al_fx ] ) for the variable @xmath1 into the integral for the variable @xmath240 . notice the contribution from the second integral part in eq . ( [ al_fx ] ) is unimportant , as long as @xmath241 and @xmath242 . thus it gives @xmath243 if @xmath238 , then @xmath244 and @xmath245 . after the transformation of the variable @xmath1 into the variable @xmath240 , the first integral part in eq . ( [ al_fx ] ) is in effect integrating over the @xmath240-range from @xmath246 to @xmath247 , thus we can use eq . ( [ int_fx ] ) to calculate it . for the second integral part of eq . ( [ al_fx ] ) , @xmath248 , but its contribution is unimportant as long as @xmath241 and @xmath242 . therefore it gives @xmath249 to calculate @xmath38 from @xmath250 , one has to know the exact information about @xmath251 and its instantaneous distribution over the width of the emitting source along the line of sight , which are always subject to uncertainties . nevertheless , in their attempts to calculate @xmath38 , some authors have calculated @xmath251 by assuming all electrons swept up by the shock are accelerated to relativistic energies either in the blastwave model for grb afterglows ( granot , piran & sari 1999 ; panaitescu & kumar 2000 ; peer & waxman 2004 ) or in the internal - shock model for prompt emissions ( li & waxman 2008 ) . we warn that the reality in nature may be that _ not all _ but only a small fraction of the electrons encountered by the shock can be heated to relativistic energies and radiate , as was suggested by bykov & mszros ( 1996 ) and daigne & mochkovitch ( 1998 ) ( also see kumar & mcmahon ( 2008 ) for an idea of repeated acceleration of a group of electrons ) , and this will introduce the biggest uncertainty to @xmath251 hence to this `` conventional '' approach of calculating @xmath38 . in general , this conventional approach over - estimates the number of emitting ( and absorbing ) electrons and , hence , over - estimates @xmath11 . in the literature , it is usually suggested that @xmath11 is slightly below the x - ray band . according to our corrected calculation , @xmath11 is typically lower and can extend to close to the optical band in a wide parameter range . our new approach here is to express @xmath252 in terms of @xmath253 , both of which contain the term @xmath254 but the latter one is directly observable - it is just the specific intensity at the source surface . therefore the new approach can avoid the uncertainties associated with partial acceleration and inhomogeneity over the source radial width . the ratio of @xmath214 over @xmath211 , also called the source function , is @xmath257 which does not have dependence on @xmath251 . it shows that , for synchrotron radiation , the power - law index of the optical thick ( to the self absorption ) part of the emergent spectrum below @xmath258 is 2 , while the power - law index of the optical thick spectrum above @xmath258 is 5/2 . from the definition of the self - absorption frequency @xmath259 , we have @xmath260 . since @xmath261 does not depend on @xmath251 and its distribution over the source width , it can be taken out of the integral . thus we have @xmath262 , where @xmath39 is the specific flux at the source surface in the asymptotic optically thin regime at @xmath38 . rewriting the expression for @xmath263 ( eq . [ source_func ] ) at @xmath264 and using the photon frequency vs. electron energy relation @xmath265 for synchrotron radiation , we get @xmath266 where @xmath44 is the energy of the electron whose characteristic photon frequency is @xmath38 , and the correction factor @xmath267 if we assume a temperature @xmath268 then the last equation is @xmath269 where for practical uses the correction factor @xmath270 is changed to @xmath271 using the relation @xmath272 , and so @xmath273 in the samples presented in the main body of the paper , @xmath48 is among -1.4 to 0 , so the ranges for the correction factor are @xmath274 and @xmath275 . therefore eq . ( [ kt_f ] ) shows that , within a factor of a few , at @xmath38 the un - absorbed source surface flux density is equal to the flux density of the rayleigh - jeans part of the blackbody spectrum with a temperature corresponding to the lowest energy of those electrons that are barely affected by the self absorption . this equation is used to calculate @xmath38 in the main body of the paper where the correction factor @xmath271 is taken into account .

following prendergast we study the relativistically expanding electromagnetic fields generated by an axisymmetric explosion of magnetic energy in a small volume . the magnetic field expands uniformly either within a cone or in all directions and it is therefore accompanied by an electric field . in the highly conducting plasma the charges move to annul the electric field in the frame of the moving plasma . the solutions presented are analytical and semi - analytical . we find that the time - scale for the winding up of the initial magnetic field is crucial , as short time - scales lead to strong radiant fields . assuming a magnetic field of emerging from a magnetosphere of we end with a jet when confined by a pressure environment that falls more slowly than . the jet carries energy of , which is mostly due to differential rotation at the base . [ firstpage ] stars : magnetic fields , pulsars , magnetars ,-ray bursts , radio galaxies , quasars

0809.0253

observations of a wide variety of astronomical objects suggests the existence of magnetic fields in relativistic environments . there have been many studies of magnetic fields emanating from differentially rotating systems . some of them confine themselves to the non - relativistic regime in which the displacement currents can be neglected . even then few of them are analytic e.g. @xcite , @xcite , , @xcite , while most of them are computational e.g. @xcite , @xcite , @xcite . in the force - free case it was shown that the time evolution of the magnetic field arises solely from the time dependence of the boundary conditions so that the exact dynamical evolution can be calculated from the time dependent sequence of static models @xcite . those in turn can be derived from the energy principle . however , that simplification depends on both the force - free condition and the neglect of the displacement currents which is only valid when the velocities are much less than @xmath5 . relativistic problems are harder as that approximation is invalid , so most studies are purely computational e.g. @xcite , @xcite , @xcite , @xcite and @xcite or semi - analytical e.g. @xcite . despite this , @xcite was able to find an exact solution to the relativistic mhd problem of a point magnetic explosion . he derived a time dependent relativistic analogue of the grad - shafranov equation that governs axially symmetric force - free mhd by assuming that the radial coordinate and the time since the explosion only appear in the dimensionless combination @xmath6 . strictly speaking such equations are only valid when the length scale and @xmath7 the time scale of the region of the explosion are much smaller than @xmath8 and @xmath9 . the resulting prendergast equation is non - linear but becomes linear in a special case . it was this special case that prendergast studied in detail . however , he found that there are spherical nodes where the radial magnetic field is zero and these nodes occur before the highly relativistic regime @xmath10 is reached . this has the unfortunate consequence that magnetic field lines emanating from small @xmath8 turn back before they reach the extremely relativistic region where the displacement currents are very important . the field lines in that region are an appendage unattached to their origin . very similar effects are well known when one takes the analogous case , @xmath11 with @xmath12 constant , in non - relativistic mhd in spherical coordinates . when @xmath13 becomes large , the system gives way to oscillating solutions with a series of nodes . this difficulty occurs because @xmath14 has dimension @xmath15 and the field has to vary on this fixed length scale even at large r. at large distances there is too much current for unit field and the smaller scale is then reflected in the scale of oscillation . it is known that non - linear ansatzes for the non - relativistic grad - shafranov equation can avoid this , which we now recognise as a bad consequence of a mathematically simple linear approximation which is not generally justified in the physics of the problem . we shall therefore study the non - linear prendergast equation in all its glory ! we generalise the idea of force - free fields by considering a configuration where for each point there exists a lorentz frame in which the electric field is zero and the current is along the magnetic field . the force density @xmath16 in the frame fixed at the origin is then : @xmath17 where @xmath18 is the velocity of the moving frame . in this paper , we study such fields by solving the equation proposed by @xcite . however , before stepping to a solution , we extract as much information as possible about the fields that is independent of the detailed form of the solution . then we solve the equations semi - analytically as it impossible to achieve general analytical solutions . the solutions simplify in the non - relativistic limit and converge to those of @xcite . there are some other cases that are interesting and exactly soluble , namely the current - free magnetic dipole and the linear force - free field of prendergast . the introduction of currents in the system allows the existence of a toroidal component of the field at the cost of making analytical solutions much harder , however it is still possible to design analytical solutions for this structure . in accordance to prendergast s formalism we consider an electromagnetic field configuration for which there is at least one frame of reference at each point of space where the electric field vanishes and the magnetic field is parallel to the electric current . we denote these frames and the physical quantities appearing there by a prime . we take each of these frames to move uniformly with velocity @xmath6 . although each frame moves uniformly away from the origin and at each moment those frames further from the origin move faster , we are not using the expanding coordinates of cosmology . the frame of reference of the observer at the origin is the unprimed frame . the primed frames of reference move radially with respect to the unprimed frame under the scaled velocity @xmath19 @xmath20 then we may write for the primed frame of reference : @xmath21 @xmath22 the magnetic field in the unprimed frame of reference is @xmath23 where @xmath24 . the electric field in the unprimed frame of reference is @xmath25 the electric current in the unprimed frame of reference is @xmath26 by virtue of @xmath27 , @xmath28 , the self - similarity condition and ( 1 ) we re - express the fields in terms of the flux function @xmath29 and @xmath30 . @xmath31 is the magnetic flux through the cap of a sphere of radius @xmath8 subtending a semi - angle @xmath32 at the origin , and @xmath30 is related to the toroidal component @xmath33 and to the specific torque carried by field lines between @xmath29 and @xmath34 . remembering that @xmath6 and @xmath35 we have @xmath36,\end{aligned}\ ] ] @xmath37.\end{aligned}\ ] ] now we substitute the currents ( 7 ) and the electric and magnetic fields into @xmath38 and finally take into account charge conservation to get prendergast s equation for the flux function @xmath39 in prendergast s notation @xmath40 is @xmath41 and our @xmath29 and @xmath30 are @xmath42 times his . the toroidal part @xmath30 is related to the source function @xmath43 by @xmath44 our task is to determine the function @xmath29 by solving the partial differential equation ( 10 ) . this is not a straightforward task ; even under a numerical treatment of the problem , the solution depends on the form of the source function @xmath43 chosen . our strategy for solving equation ( 10 ) is + 1 . to study the non - relativistic limit so as to see how it connects to earlier work . + 2 . to study the elementary case when there are no currents in the body of the plasma . the resulting self - similar solutions of maxwell s equations become singular at r = ct if the fields at the origin are turned on suddenly . we demonstrate how these singularities disappear when the fields are turned on over a small finite time . find what form of source function @xmath45 allows solutions of ( 10 ) that are separable in @xmath46 and @xmath47 . the special form of @xmath45 required can not be made to vanish at @xmath48 . however it gives an exact solution in regions where p is above a certain threshold , @xmath49 , where @xmath50 . when @xmath29 is below that threshold the field joins smoothly to one with no currents . + 4 . the current - free solutions in this outer cocoon are not of product form but are found by analysis into suitable harmonics whose coefficients are computed to high accuracy . thus we end with a global solution that expands with the velocity of light within a narrow prescribed cone . it has high fields but no charges near @xmath51 . below that we enter the current free cocoon , soon followed by the basic relativistic solution of product form which is accompanied by charges and currents . in this section we focus on the properties of the field . it may seem more natural to achieve a solution for equation ( 10 ) and then investigate the details of the field . however it is possible to obtain some physical understanding by studying general properties of the field that do not depend on the details of the solution . we seek frames of reference where the electric field vanishes . such frames are particularly interesting as they determine the velocity of the field and allow us to trace the time evolution of the magnetic field lines @xcite . the setup of the problem gives an initial set of frames obeying this property , the uniformly expanding ones . indeed the motion of the field lines may be viewed as a simple expansion . a charge @xmath52 that is placed at distance @xmath53 from the origin at time @xmath54 with velocity @xmath55 , does not feel any force , as there is no electric field and the magnetic field in this frame has no effect on a particle at rest in it . the particle will remain on the same magnetic field line as those are determined by @xmath29 which in turn is a function of @xmath6 . however , as the fields have components in the all three directions the velocity field of uniform expansion is not perpendicular to the magnetic field . conventionally the velocity of field lines is chosen to be perpendicular to them . we examine other frames of reference where there is no electric field . in order to find such frames we use the following quadratic equation for the velocity field @xmath56 @xcite , that describes frames of reference where the magnetic and the electric fields are parallel to each other @xmath57 this velocity field is clearly perpendicular to @xmath58 . given that the electric and the magnetic field by construction satisfy @xmath59 , which is a relativistic field invariant , we conclude that in a frame where @xmath60 either the electric or the magnetic field vanishes . thus equation ( 12 ) may be used to define the velocity of the magnetic field . however the general velocity @xmath61 of a frame in which @xmath62 and @xmath63 are parallel allows boosts in the direction of the magnetic field and only determines the motion in the direction transverse to the magnetic field . the solution of the above quadratic equation gives the velocity we are interested in , however it is feasible to achieve a simpler formula for the motion of the field lines . equation ( 1 ) yields that @xmath64 . any velocity field of the form @xmath65 for arbitrary @xmath66 , gives a set of frames of reference where the electric field vanishes . the lorentz transformation of the electric field is given by @xcite @xmath67 substituting ( 13 ) into ( 14 ) and working the cross products we obtain that @xmath68 . therefore equation ( 13 ) gives the general velocities that can be attributed to the field lines ; we follow @xcite in using the expression for @xmath69 which gives a velocity perpendicular to the magnetic field . this is a solution to ( 12 ) provided @xmath70 . by virtue of equations ( 8) and ( 9 ) we find that the usual velocity field is @xmath71,\end{aligned}\ ] ] @xmath72 @xmath73 relations ( 15)-(17 ) give the details of the motion of the field lines . at the first instance we can make some reasonable assumptions for the flux function @xmath29 . a magnetic field consists of field lines rising from and returning to an imaginary spherical cap of radius @xmath53 and confined between angles @xmath74 and @xmath75 corresponding to @xmath76 and @xmath77 . in the case of a dipole field instead of a spherical cap we have the whole spherical surface , and @xmath78 . the flux function @xmath29 is zero at the boundaries and has a single maximum at @xmath79 where the field lines turn back . the toroidal component @xmath30 has to obey @xmath80 approaching the edges and normally the decrease has to be such so there is no singularity . the above assumption allows the determination of the signs of the derivatives appearing at relations ( 15)-(17 ) . therefore @xmath81 , as there can not be any flux generated as the field moves upwards ; @xmath82 for @xmath83 ; @xmath82 for @xmath84 and @xmath85 at @xmath79 . we conclude that the field lines viewed this way perform a complex motion that consists of a non - uniform radial expansion ; a meridional motion where the parts of the field lines that lie in @xmath86 move in the direction of @xmath87 , those in @xmath88 in the direction of @xmath89 ; and an azimuthal motion where the parts of the field lines lying in @xmath88 move in the direction of @xmath90 and the rest in the direction of @xmath91 . the magnitude of the velocity depends on the details of the solution and can not be determined by the previous arguments . the parameter @xmath47 , that is the expansion velocity scaled to the speed of light gives an estimate of the importance of the relativistic terms for the problem . in the limit of @xmath92 equation ( 10 ) becomes , @xmath93 equation ( 18 ) is similar to the grad - shafranov equation for axisymmetric magnetic fields @xcite . there are many solutions of this equation in astrophysical context , . amongst these solutions the most appropriate to our case is that of @xcite where the field is a self - similar quadruple . in our first paradigm we are going to apply this solution to a self - similar dipole . assuming a self - similar solution we set @xmath94 , where @xmath95 is a flux normalisation , formally the maximum flux at @xmath96 , so that the angular part of the flux @xmath97 , therefore equation ( 18 ) becomes @xmath98 in order to achieve self - similar solutions we need @xmath99 . with @xmath29 a product of a function of @xmath47 and a function of @xmath46 , @xmath45 can only be proportional to a power of @xmath47 if @xmath43 itself is a power of @xmath29 , since @xmath29 is proportional to @xmath100 the required power is given by @xmath101 a more detailed derivation is given in @xcite . setting @xmath102 , equation ( 19 ) reduces to @xmath103 to solve equation ( 21 ) in a sphere we apply the boundary conditions @xmath104 , and determine @xmath105 by the condition that @xmath106 has a single maximum @xmath107 . these conditions are sufficient to solve the equation for any @xmath108 . for @xmath109 the solution is @xmath110 and @xmath111 , for @xmath108 small the solution of ( 21 ) takes the form @xmath112 and in general for intermediate values it is solved numerically . therefore the fields take the following form in this non - relativistic limit @xmath113,\end{aligned}\ ] ] @xmath114.\end{aligned}\ ] ] the motion of the field lines can be found by equations ( 15)-(17 ) by applying the fields given at ( 22 ) and ( 23 ) . @xmath115.\end{aligned}\ ] ] this case is different to the grad - shafranov equation , as in the latter the field only depends on @xmath8 whereas here the field depends on @xmath116 . however , a snapshot of the configuration for any given @xmath117 satisfies the grad - shafranov equation . when there are no source terms on the right , equation ( 10 ) gives simple relativistic solutions with no toroidal magnetic field component . in this section we study this case ; we show that one solution reduces to a linearly increasing magnetic dipole , where the field is zero outside a sphere expanding at the speed of light . with @xmath118 equation ( 10 ) reduces to the simple form @xmath119 by using ogilvie s transformation @xmath120 equation ( 25 ) becomes @xmath121 we use the technique of separation of variables ; let @xmath122 and dash denotes differentiation with respect to @xmath123 , equation ( 26 ) becomes @xmath124 equation ( 27 ) for @xmath125 is the legendre differential equation , therefore the solution is a linear combination of @xmath126 , that has to be finite in the interval @xmath127 , therefore @xmath128 , as @xmath129 becomes infinite at @xmath130 . the angular part can be solved analytically using the following transformation @xcite , @xmath131 , therefore the second equality of ( 27 ) becomes : @xmath132 this is again the legendre differential equation , and the solution for the angular part of @xmath29 is @xmath133 . therefore the flux function of the field is @xmath134 let us consider @xmath109 , that corresponds to a field where a dipole of magnetic moment @xmath135 is initiated at @xmath136 . following the formalism of @xcite for the hertzian dipole modified for a magnetic dipole we can express the electric and the magnetic fields using the quantity @xmath137 which is a vector pointing at the axis of dipole @xmath138 and is a function of position and time ; dot denotes derivation with respect to @xmath9 @xmath139 @xmath140 it is shown in appendix a that for any function @xmath141 @xmath142 and the fields satisfy maxwell s equations . the fields are : @xmath143 @xmath144 therefore , we can evaluate the electric and the magnetic fields for a linearly increasing dipole that is switched on at @xmath136 and lies at @xmath145 , it corresponds to @xmath146 , where @xmath147 is the heaviside function which is unity for argument greater than zero and zero otherwise . we divide the space into three regions . the first one ( i ) is @xmath148 ; the second ( ii ) is the surface @xmath51 and the third ( iii ) is @xmath149 . in the first region there is an electromagnetic field because of the dipole ; in the the third region the field is zero , as the message of switching on the dipole has not arrived yet ; the second region is the horizon surface of discontinuity . therefore in region ( i ) the fields are : @xmath150 @xmath151 since the fields are related to derivatives of @xmath152 and the first derivative of @xmath152 is discontinuous at @xmath153 we expect singularities on the surface @xmath51 ( ii ) . indeed there are singular surface fields . these infinities we should have expected because in the analogous electric dipole case velocities are suddenly imposed on the charges to make the linearly growing dipole . thus initial accelerations are infinite so the power radiated should be singular . @xmath154 @xmath155 these fields are unacceptable . in region ( iii ) there are no fields at all . to avoid the singularities in ( ii ) we take into account the finite time needed to accelerate the dipole to a given growth rate . to do this we smooth by averaging the fields over a small spread @xmath156 of switch - on times . thus region ( i ) now extends @xmath157 , region ( ii ) spreads into a spherical shell @xmath158 and ( iii ) @xmath159 . @xmath160 becomes : @xmath161 this substitution increases the complexity of the formulae for the fields but removes the singularity at the horizon and allows us to study in greater detail the structure of the field near the horizon and how it depends on the time - scale for the dipole to establish a linear behaviour . the fields corresponding to @xmath162 are given by putting @xmath162 into ( 33 ) and ( 34 ) @xmath163 @xmath164 @xmath165 these are perfectly regular and acceptable . the higher order self - similar multipoles are given in appendix b. in the previous sections we have given analytical expressions consisting of magnetic fields that lie on the meridional planes and electric fields that are azimuthal ; such fields do not contain currents , or charge densities and each arises from a growing multipole at the origin . in this section we seek analytical solutions for electromagnetic fields in which the magnetic field has an azimuthal component , the electric has a meridional component and there are charge and current densities in the space . we need to solve ( 10 ) , prendergast s equation . we seek a solution that is a product of a function depending on @xmath47 and one depending on @xmath46 ; using the usual transformation @xmath166 equation ( 10 ) reduces to @xmath167+(1-\mu^{2})\frac{\partial^{2}p}{\partial \mu^{2}}=-\frac{1}{u^{2}-1}\beta \frac{d \beta}{d p}.\end{aligned}\ ] ] we multiply by @xmath168 and look for solutions in the form of @xmath169 @xmath170+(u^{2}-1)\frac{1-\mu^{2}}{f}\frac{d^{2}f}{d\mu^{2}}= \nonumber \\ = -\frac{\beta}{p}\frac{d\beta}{dp}.\end{aligned}\ ] ] we set : @xmath171,\\ m(\mu)=\frac{1-\mu^{2}}{f}\frac{d^{2}f}{d\mu^{2}},\\ \frac{\beta}{p}\frac{d\beta}{dp}=q(p),\end{aligned}\ ] ] so the equation takes the form : @xmath172 operating with @xmath173 we find @xmath174 which is of the form @xmath175 , with @xmath176 and @xmath177 . taking logs and operating with @xmath178 we find that ( i ) @xmath179 or alternatively ( ii ) @xmath180 which leads to prendergast s linear model @xmath181 . henceforward we consider only case ( i ) , @xmath182 is not constant since @xmath29 can not be a function of @xmath183 alone , so both @xmath184 and @xmath185 must equal the same constant @xmath186 . this is a generalisation of the @xmath108 used before in the current free case . thus without loss of generality @xmath187 and @xmath188 where @xmath189 is constant . finally from ( 77 ) @xmath190 , hence : @xmath191 , @xmath192 , @xmath193 so @xmath194 , our basic equation ( 44 ) now reads : @xmath195{\nonumber}\\ -(u^{2}-1)(\frac{1}{2}kl f^{2/l}+c)=-(\frac{1}{2 } kl p^{2/l}+c'),\end{aligned}\ ] ] but @xmath196=l(u^{2}-1)+l^{2}u^{2},\end{aligned}\ ] ] so equation ( 50 ) reduces to a linear equation in @xmath197 . comparing coefficients of @xmath197 @xmath198 and from the coefficients of @xmath199 in ( 50 ) @xmath200 evidently we must choose @xmath201 , so the first is satisfied and @xmath202 so the second is too . however in the non - relativistic approximation @xmath203 so then the second equation is all negligible compared with the first . from the definition of @xmath204 we have : @xmath205 so @xmath206 and by normalising @xmath207 , so @xmath208 @xmath209 which is the same equation as that studied by @xcite , who found it in non - relativistic mhd . we choose the solutions obeying boundary conditions that @xmath106 is zero at @xmath210 , ( @xmath76 ) and at any chosen outer boundary @xmath75 , ( @xmath211 ) . the constant @xmath212 is then determined in such a way that @xmath106 is one at its maximum . for small @xmath108 and @xmath213 the solution for @xmath106 is @xmath214,\end{aligned}\ ] ] where @xmath215 and @xmath183 is close to 1 and given by @xmath216 \nonumber\\ \simeq 1+\frac{1}{\nu}\ln2.\end{aligned}\ ] ] the value of @xmath217 associated with this solution is @xmath218 equation ( 57 ) gives solutions which are symmetric in @xmath219 around their maximum @xmath220 . this is the case for @xmath221 , when @xmath108 is small but not zero the maximum is slightly displaced . a more accurate formula taking into account this displacement is @xmath222\}\big),\end{aligned}\ ] ] where @xmath223 , @xmath224 , @xmath225 , @xmath226 and @xmath227 , for details on the derivation see @xcite ( appendix ) . the remaining equation comes from the definition of @xmath228 ( 47 ) , @xmath229 where @xmath230 is constant . if @xmath231 then @xmath232 has a minimum at @xmath233 . we choose @xmath230 so that this minimum is zero and take @xmath43 to be zero for @xmath234 so @xmath235-\big[(p / p_{0})^{2}-1\big]\big\}\\ 0 , p < p_{0 } \end{array } \end{aligned}\ ] ] had we chosen a lower value of @xmath230 then @xmath232 would have been negative for values of @xmath29 near @xmath49 . however @xmath232 must be positive . an unfortunate consequence of choosing the upper expression ( 62 ) for @xmath232 for all @xmath29 is that @xmath236 when @xmath48 . this entails an infinite @xmath33 near the axis which corresponds to a line current there . to avoid this in a continuous and differentiable way we choose the upper form ( 62 ) only where @xmath237 . elsewhere we choose @xmath238 , the current - free solution . we avoid the use of the linear form @xmath239 because those solutions have an infinity of oscillations in the sign of @xmath29 as they approach @xmath130 ( see prendergast 2005 ) . so they contain fields that are disconnected from the origin . thus , our solution of the relativistic prendergast equation is @xmath240 where @xmath106 is given by ( 57 ) or more accurately by ( 60 ) . in what follows we determine the fringing field in the cocoon region @xmath234 . we choose the boundary to be a cone with a total opening angle ( not semi - angle ) of less than @xmath241 . this is the case of greatest importance for applications to @xmath4-ray bursts , however we should point out that exactly similar methods work for any cone or even for the whole sphere . it is just that legendre polynomials or more generally legendre functions of @xmath46 replace the simpler bessel functions of the narrow cone cases . since our main applications will be the jets and @xmath4-ray bursts it makes sense to study solutions within conical boundaries with quite narrow cones @xmath242 . provided the total opening angle ( not the semi - angle ) of the cone is less than @xmath241 the actual opening angle does not affect the form of the solution whereas at larger angles it does . thus there is a real advantage in studying the small angle case which allows us to replace high order legendre functions of angles by bessel functions of order 1 . the @xmath243 limit comes from our replacement of @xmath244 by 2 . the fractional error in neglecting @xmath245 as compared with 2 is @xmath246 which is less than @xmath247 for @xmath248 in the region where @xmath234 , @xmath232 is zero so the solutions of ( 43 ) that we need , have no currents , however they are solutions with a very awkward boundary on @xmath249 . in place of trying to fit a current - free solution on that boundary to our known solution within the region with @xmath234 we adopt the current - free solutions within the whole cone as our complete set . we expand the known right hand side of equation ( 43 ) in terms of our bessel functions of angle within the cone . we then get a known function of @xmath123 as a source term on the right hand side of each bessel component of the equation ( 43 ) . we solve for the radial functions of @xmath123 which are the coefficients of the different bessel components , each with its own known source term . under the assumption of narrow opening angle and in the absence of currents the angular part of ( 43 ) becomes @xmath250 setting @xmath251 the above equation becomes : @xmath252=0,\end{aligned}\ ] ] so @xmath253 so far , we have only defined j up to an arbitrary additive constant which is eliminated in @xmath254 so we absorb the constant into @xmath255 and get : @xmath256 the solution is @xmath257 , where @xmath258 . if we want a solution regular at @xmath210 we must omit the @xmath259 term so @xmath260 , and @xmath261 where @xmath262 . in order that @xmath254 should vanish at @xmath75 we need @xmath263 . the solutions for @xmath264 are @xmath265 where @xmath266 , @xmath267 , @xmath268 etc p 409 of a & s , and @xmath269 is the s - th zero of @xmath270 for @xmath271 , called @xmath272 in a & s. now any function @xmath273 , @xmath274 can be expanded as a series . @xmath275 the orthogonality relation is a & s 11.4.5 @xmath276^{2}\delta_{sn},\end{aligned}\ ] ] to find @xmath277 we multiply by @xmath278 and integrate : @xmath279d\big(\frac{\theta}{\theta}\big)= \nonumber \\ = \frac{1}{2}[j_{1}'(j_{n})]^{2}f_{n}.\end{aligned}\ ] ] so the @xmath280 corresponding to any @xmath182 can by found by integration and a knowledge of the @xmath281 and @xmath282 found in the a & s p 409 . now write : @xmath283 and @xmath284 then multiply ( 43 ) by @xmath285 and integrate from @xmath74 to @xmath75 ; on division by @xmath286^{2}$ ] we get , since @xmath287 satisfies ( 67 ) ; @xmath288-\frac{j_{n}^{2}}{\theta^{2}}p_{n}(u)=k_{n}(u).\end{aligned}\ ] ] the above equation is legendre s with a known right hand side @xmath289 . this may be solved using the boundary conditions by variation of parameters . we write each @xmath290 in the form @xmath291 where @xmath292 and @xmath293 , @xmath294 are legendre functions satisfying @xmath295+\nu(\nu+1)p_{n}=0,\end{aligned}\ ] ] and @xmath296 likewise . we choose @xmath297 and @xmath298 to satisfy @xmath299 thus @xmath300 substituting ( 78 ) into ( 74 ) the terms with @xmath301 and @xmath302 undifferentiated vanish since @xmath293 and @xmath294 satisfy legendre s equation so the only terms that survive are @xmath303=k_{n}(u ) . \end{aligned}\ ] ] we solve ( 77 ) and ( 79 ) using the wronskian @xmath304 . but the wronskian is @xmath305 see a & s 8.1.9 . , so @xmath306 similarly , @xmath307 where @xmath308 is the value of @xmath123 at the surface where the field expands at some given time , and @xmath309 is the value of @xmath123 at highest point of the cocoon @xmath310 . @xmath311 can be determined by analysing @xmath312 into @xmath313 and by ( 75 ) @xmath314 one must choose the end points of the integrations to ensure the boundary conditions are satisfied . indeed formula ( 72 ) provides accurate results when the first few terms are evaluated and we are not very close to the extreme relativistic limit . in the paradigm solved for @xmath315 the deviation between the reconstructed solution and the solution satisfying ( 56 ) , which holds inside the lobe where @xmath316 , is @xmath317 at @xmath318 and @xmath319 at @xmath320 , @xmath321 at @xmath322 and @xmath323 at @xmath324 by taking into account the first fifteen terms ( figures 1 - 3 ) . as @xmath123 comes closer to unity the deviation between the reconstructed solution and the self - similar one increases . this deviation is systematic as the reconstructed solution always gives a flux function that is greater than the self - similar one . this is because the self - similar solution is proportional to @xmath325 and becomes @xmath74 at @xmath130 , whereas the legendre p functions converge to 1 at the top . very close to the top of the lobe we reconstruct the solution by analysing @xmath326 into bessel functions , then by virtue of ( 76 ) and setting @xmath327 because of ( 81 ) we can determine @xmath328 . then from ( 82 ) it is clear that for @xmath329 @xmath330 . thus we have a complete description of the field everywhere . = 8.7 cm = 8.7 cm = 8.7 cm in this section we evaluate physical quantities such as the twist , the pressure and the energy corresponding to the mathematical solutions we describe above . we use quantities appropriate for @xmath4-ray bursts of the long - soft type . we study two cases ; one with semi - opening angle of @xmath331 which is the maximum value for the assumption of narrow cones to hold and one with semi - opening angle of @xmath332 which is closer to what is expected for @xmath4-ray bursts . we find that in the latter case the fields are more twisted so the energy of the initial purely poloidal field is amplified by a greater factor . the form of the field defines the distribution of the twist . for the reasons discussed in the section 3.2.1 we use the the solution for the flux function given by ( 63 ) for @xmath316 . we have taken as a given that the magnetic field is contained within a cone . although a detailed hydrodynamic study is needed in order to find the shape of the magnetic cavity ; a pressure environment that depends in @xmath8 like @xmath333 leads to the conical jets . inspection of ( 62 ) shows that toroidal torque @xmath43 increases with @xmath29 when @xmath316 but it is zero for @xmath334 . however our solution ( 63 ) shows that the lines of force for large @xmath29 do not go to large radii where the twist is . thus the twist of the lines is greatest for some @xmath335 greater than @xmath49 and less than @xmath95 . when @xmath108 is large @xmath336 is close to zero , but as @xmath108 decreases it shifts closer to @xmath95 . the maximum twist increases as @xmath108 decreases ( figures 4 and 5 ) . smaller @xmath108 corresponds to fields that decrease slower with @xmath8 and at the same time they demand greater twist . = 8.7 cm = 8.7 cm = 8.7 cm the lines of force of the magnetic field are given : @xmath337 in the region where @xmath234 there is no twist in the field , as @xmath118 , thus @xmath338 . in the region where @xmath316 the field can be evaluated by virtue of ( 63 ) and ( 8) and by using the second equality of ( 83 ) the total twist @xmath339 for a field line determined by @xmath29 is given by : @xmath340 where @xmath341 and @xmath342 are the polar angles of the points where the field line rises and sinks at the base . we then express @xmath123 as a function of @xmath32 for some given @xmath29 and we integrate . this gives the detailed shape of the field lines ( figure 7 ) . most of the twist is concentrated near the top of each field line . from integral ( 84 ) for @xmath341 and @xmath342 corresponding to the footpoints of the field lines we evaluate the total twist a field line carries . the maximum twist depends on the opening angle of the jet . the overall behaviour of @xmath343 can be approximated by the analytical formula for the total twist in the non - relativistic case @xcite @xmath344 ( figure 7 ) or @xmath345 for the case of narrow cones we are interested in this study . thus the smaller the opening angle the more twisted the field is , giving evidence for the relation between collimation and the number of the turns performed . although the total twist is inversely proportional to the semi - opening angle , the constant of proportionality is somewhat smaller than what was found from the non - relativistic analytical formula . the total twist for a given field line does not depend on time , indeed they merely expand with the expanding frame . nevertheless the field lines rotate as they expand relative to the stationary frame . the azimuthal velocity component given in equation ( 17 ) describes that rotation of the field lines and is a fraction of the speed of light , its maximum value is @xmath346 . as we have a non - stationary solution of maxwell s equations , problems at the light cylinder can not occur , and even a radius for the light cylinder is not defined . rotation has been observed in non - relativistic jets emerging from t - tauri stars @xcite . = 8.7 cm we can easily find the pressure and the energy carried by a given electromagnetic field since we know the actual fields . as the energy density of an electromagnetic field is proportional to the pressure we are going to treat them together . the energy density is given by @xmath347 from equations ( 8) and ( 9 ) and the flux function @xmath29 from equation ( 63 ) we can evaluate the pressure at some given time @xmath117 ( figures 8 and 9 ) . the energy is mainly concentrated near the base of the jet and near the top of the jet . it is expected that the energy is strong near the base as the fields decrease with distance from the centre . however unlike other cases studied where the fields extend to infinity , there is a very strong field near the top . this is because of the turn over of the field lines , which are confined inside the light sphere , so they have to close . thus , they have a very strong @xmath348 component . = 8.7 cm = 8.7 cm as we approach the light horizon where the expansion velocity is equal to the speed of light the energy strongly depends on the smoothing of the fields . if we considered an infinitely fast turning on of the magnetic field then it would carry infinite energy , however this is unphysical as it takes finite time to switch on . that time scale determines the energy of the initial pulse ; it is inversely proportional to the time - scale of switching on the dipole . the physical system we have in mind is the following : as the core of a massive star collapses to form a compact object or a black hole it spins much faster and winds up the magnetic field where it connects to the rest of the star or to any disc that may form . this leads to a dramatic increase in the energy by two orders of magnitude in the case of a narrow jet , as shown in the next section . all the twist is imposed in this stage . the concentrated energy of the field leads to an explosion . the magnetic field expands in the star and a jet forms . the scales chosen for this study are @xmath349 for the size of the jet . this is the typical size of the jet related to the prompt emission of a @xmath4-ray burst . @xmath350 for the smoothing scale that is a few times the length scale of the schwarzschild radius of a black hole of @xmath351 . we consider a magnetic field of order @xmath352 at @xmath353 powering the jet . this magnetic field comes from the central compact object , a discussion on the origin of strong magnetic fields in neutron stars can be found in @xcite and references therein . the strongest magnetic fields estimated for magnetars are @xmath354 . we assume that such a field originates at @xmath355 and then expands outwards , as the surrounding pressure near the centre drops relatively slowly . a constant pressure environment leads to a field which is confined within a cylinder without decreasing at all , we take an intermediate case for which the field between @xmath355 and @xmath356 drops by two orders of magnitude . above this radius and below @xmath357 we apply the self - similar solutions inside a narrow cone . these give a total energy of order @xmath3 in the jet from which @xmath358 is due to the switching on of the field concentrated in a very thin layer just below the light sphere . the width of this layer is @xmath359 and the fields contained are of order @xmath360 whereas the field at the main body of the jet at @xmath361 is of order of @xmath362 . the energies quoted above and on the tables are at @xmath363 after the initial explosion in the centre of the star , when the top of the jet reaches the surface of the star . @xmath364 @xmath365 the total energy depends on @xmath108 ( figures 10 and 11 ) . it can be seen ( tables 1 and 2 ; figures 4 and 5 ) that @xmath108 parametrises the twist . we prefer the use of the twist as parameter as it describes a physical process rather than @xmath108 which is just an index . thus a decrease in @xmath108 can be viewed as an increase on the differential rotation at the base . when differential rotation is imposed at the base there is a toroidal component in the field which increases the energy of the configuration , and in addition to that the poloidal part of the field expands to reach a force free state . this dual process leads to an increase in the total energy . the factor by which the total energy is amplified depends on the maximum twist the configuration can tolerate and thus the opening angle which determines the maximum twist . narrower cones get a greater amplification on their total energy ; in the limiting case which is equivalent to an infinite cylinder , the twist has no constraint , thus the energy can be infinitely amplified . = 8.7 cm = 8.7 cm as the whole structure expands there is a flow of energy in the jet . the energy contained in an element of the the jet from angle @xmath32 to @xmath366 and in velocity space between @xmath47 and @xmath367 is @xmath368(1+v^{2})\big)d\theta dv.\end{aligned}\ ] ] this energy is inversely proportional to time . the poynting vector for the fields is @xmath369 substituting from ( 8) and ( 9 ) @xmath370 the @xmath32 component of the poynting vector is zero at the boundaries of the jet and on the axis , so it is consistent with the demand that the jet is constrained within a cone and there is no flow of energy out of the sides of the cone . we now apply the poynting theorem for the expanding volume @xmath371 as opposed to the usual form for the poynting theorem , there is an extra term in the right hand side because of the expansion of the volume . the term in the left hand side is the integral of equation ( 86 ) . the terms in the right hand side show in which direction the energy is transfered in @xmath47 space . the integrals should be performed on the surfaces of the inner and the outer boundary of the expanding volume . nevertheless , by integrating only on one surface we can find whether the energy flow is towards larger or smaller @xmath47 . @xmath372\frac{1-v^{2}}{2}-\frac{1}{2}\big(\frac{\partial p}{\partial \theta}\big)^{2}\big ) \frac{d \theta}{\sin \theta}.\end{aligned}\ ] ] the integral of equation ( 90 ) is negative when integrated for the solutions of @xmath29 and @xmath30 found in section 3.2 . thus the energy flows towards smaller @xmath47 . nevertheless the energy flow relative to fixed axes is outwards everywhere . this discrepancy occurs because the energy flow as described by the poynting vector is slower than the velocity due to the uniform expansion . the region at the top has to be treated separately . near the top @xmath373 and it is only the component due to @xmath374 that contributes to the energy . the energy in the top is compressed in a thin shell ( region ii ) . the width of this shell is @xmath355 . thus near the top the flux function drops from @xmath375 to @xmath74 within a distance @xmath355 as described in appendix b for the smoothing of the fields in the case of self - similar multipoles . @xmath375 is approximated with good accuracy by @xmath376 , where @xmath49 is defined for equation ( 62 ) and @xmath377 is the solution to equation ( 56 ) , note that the exact value of @xmath375 is not given by equation ( 63 ) but it is a sum of multiples of bessel and legendre functions . after we smooth the field we find that the energy contained in region ii @xmath378 thus the energy of the shell ( region ii ) is constant . as the pressure of the magnetic field is proportional to @xmath379 for the case of the self - similar solutions in the non - relativistic region the field shall demand a special pressure environment so that it is contained within a cone . let us assume that the pressure of the surrounding medium is described by a power - law @xmath380 , thus as the field is differentially rotated it will expand primarily up the axis and if the index of the pressure of the medium @xmath381 is greater that @xmath382 then the jet will not have a constant opening angle but will expand sideways . near the top the case is rather different , as there is a particularly strong ram pressure . this is because the head of the jet encounters the gas of the medium in a velocity @xmath383 that is close to the speed of light , let this velocity correspond to some lorentz factor @xmath4 . as there is only a @xmath348 component there , in the comoving frame the magnetic field is : @xmath384 this magnetic field encounters an environment of density @xmath385 where @xmath386 is the density of the medium in its rest frame . the @xmath4 factor is because the coordinate in the direction of expansion is shrunk by a factor of @xmath4 . balance of momenta gives a stagnation pressure in the comoving frame @xmath387 setting the stagnation pressure equal to the magnetic pressure of the top field we can determine the @xmath4 that the magnetic field reaches as it expands . this depends on the magnetic field and the density the jet encounters ; as @xmath348 and @xmath386 are known we can easily find the value of @xmath4 . for densities around @xmath388 a lorentz factor of @xmath389 is found , whereas for densities smaller than @xmath390 lorentz factors greater than 10 are found , and for the case of an hii region surrounding a massive star the jet becomes extremely relativistic with @xmath391 . however , near the centre of the star the density greatly exceeds the value of @xmath392 , thus the jet pushes its way through the material at non - relativistic speeds . however it is possible that a funnel forms by an other mechanism . in the case of @xmath4-ray bursts there is a black hole formed in the centre of a rotating star thus , the material on the axis will fall into the black - hole . then there will be an underdense region on the axis allowing the expansion of the jet , see e.g. @xcite . relativistic jets were first found emanating from radio galaxies and quasars , then more spectacularly still from @xmath4-ray burst sources and microquasars . related but non - relativistic jets arise from young stars with accretion discs and generate herbig - haro objects . we consider that all these phenomena come from the magnetic energy generated when a magnetic field is wound up by the differential rotation of its footpoints . the opening angles of many of these jets is of order @xmath393 , giving semi angles of @xmath394 or so . it is believed that massive wolf - rayet stars in rapid rotation generate the long - soft @xmath4-ray bursts when their cores collapse eventually to make black holes and the associated supernovae @xcite . then the differential rotation is between the collapsing core of the star and the enclosing envelope . it has also been suggested that the outcome of the collapse is a strongly magnetised compact object @xcite releasing energy in magnetic field . this magnetic energy is generated by differential rotation deep within the star . while the strong magnetic field expands to balance the external gas pressure the density of the external envelope initially prevents relativistic expansion along the rotation axis . under expansion at constant flux the field in a magnetic bubble or cavity varies , with length scale , @xmath15 , as @xmath395 and the magnetic pressure as @xmath396 . thus while it is within the body of the star the magnetic cavity is collimated when the ambient pressure falls less rapidly than @xmath2 subtending smaller angles at the star s centre as @xmath8 increases . such configurations of non - relativistic jets were described in @xcite and references therein and have been applied to @xmath4-ray bursts by @xcite . in the outer parts of the star the pressure falls more rapidly than @xmath2 and the head of the magnetic cavity accelerates until it breaks through the stellar surface and expands relativistically . numerical studies of ultrarelativistic mhd jets have been done by @xcite , where it was found that high collimation angles can be achieved by a configuration where the toroidal component of the magnetic field dominates . when the central object is a rotating black hole then the kerr geometry was taken into account . in the case we study the main source of confinement is ram pressure on the low density stellar wind . in this paper we have solved analytically the equation proposed by prendergast for relativistically expanding axisymmetric self - similar force - free fields . this equation is a relativistically moving extension of the grad - shafranov equation for static force - free fields with axial symmetry . to excellent accuracy our solution is given everywhere by equations ( 62 ) and ( 63 ) with the @xmath397 function given by equation ( 57 ) or more accurately ( 60 ) . the resulting electromagnetic fields are given by equations ( 8) and ( 9 ) . all solutions of the prendergast equation need to be averaged over a small time @xmath398 that corresponds to the light crossing time across the magnetic explosion . his equation assumes a point explosion . exact solutions of prendergast s equation give singular fields where the expansion speed is that of light but we have shown how averaging over @xmath398 yields genuine solutions of maxwell s equations without singularities . nevertheless this @xmath398 may be quite short and the resulting fields close to @xmath51 are strong , carrying around @xmath247 of the system s energy . this may be related to @xmath4-ray burst precursors . however , the fields at the top require further modification as the ram pressure of their relativistic motion is retarded by the interstellar medium . considerations as those of @xcite are clearly important there , whatever drives the shock waves and it is there that the eponymous @xmath4-rays themselves are generated . while this paper has given a basic mechanism that produces highly relativistic motion there is a big gap between that and the phenomenology of @xmath4-ray bursts . on the way to understanding the solutions of the fully relativistic prendergast equation we were led to give purely poloidal solutions of maxwell s equations for time - dependent multipoles , interesting in their own right . we have also seen how in the non - relativistic limit our problem is related to the solutions found in @xcite and @xcite . finally we have seen how small opening angles of the cone correspond to greater coiling of the magnetic field and thus to mush greater concentration of energy into the jet . setting some values for the magnetic field quoted from the stronger magnetars we find that we can power jets up to the energies of @xmath4-ray bursts , for which the isotropic energy is between @xmath399 to @xmath400 erg @xcite . in the case studied we found that the energy inside the jet is of order @xmath401 , where the isotropic energy is two orders of magnitude greater . the very strong magnetic field found at the top of the magnetic configuration can oppose the strong ram pressure because of expansion and swipe material out of the way for the jet to expand . this feature may be related to the precursor observed in some @xmath4-ray bursts e.g. @xcite , as it is the first to break out and carries a small fraction of the energy of the @xmath4-ray burst . jets of narrow opening angles can be more twisted than those of wide opening angles , giving evidence that collimation increases with twist . finally a strong stagnation pressure is found at the head of the jet . it prevents the jet from being relativistic near the base but as it expands outwards and encounters underdense material it becomes extremely relativistic as is indeed observed in @xmath4-ray bursts . we thank uden sherpa for his organisation of the himalayan camp where the partial solution ( 63 ) was obtained and profs gough and pringle for mathematical discussion of some of the finer points of the averaging of the singular fields in region ( ii ) . we also thank drs eldridge and tout for the discussion of @xmath4-ray burst progenitors . we thank the referee for demanding that we clarify the physics of rotation and energy flow in this solution . j.j . 1994 , , 288 , 1012a a.r . , lucek s.g . 1995 , 277 , 1327b d. , bacciotti f. , ray t. p. , eislffel j. , woitas j. 2007 , 663 , c j. 1995 , 446 , 67c j .- p . , hawley j.f . , krolik j.h . & hirose s. 2005 , 620 , 878d r.c . , thompson c. , 1992 , , 392l , 9d c. , ouyed r. 2004 , 608 , 378f , c. 1997 , 319 , 1025f c.f . , mckinney j.c . & tth g. , 2003 , 589 , 444 g h. , rubin h. 1958 , mhd equilibrium in an axisymmetric toroid . proceedings of the 2nd un conf . on the peaceful uses of atomic energy , 31 , geneva : iaea p.190 . k.n . , 2008 , 385 , 875 g a. , fryer c.l . , woosley s.e . , langer n. & hartmann d.h . 2003 , 591 , 288h j.d . 1975 , classical electrodynamics , wiley21 s.s . 2002 , 336 , 759k c. , 1998 , , 393 , 235k l.d . , lifshitz e.m . 1975 , the classical theory of fields , elsevier j.m . , wilson j.r . , 1970 , 161 , 541l z .- y . , chiueh t. & begelman m. 1992 , 394 , 459l r.v.e . 1976 , , 262 , 649l d. 1978 , , 17 , 185l d. , boily c. 1994 , , 267 , 146l d. 2006 , , 369 , 1167l l. 1999 , cosmic magnetism , oxford t. , inoue h. , nishimura j. et al . , 1991 , 350 , 592 m r. , pudritz r.e . & stone j.m . 1997 , 385 , 409o t. 2005 , , 76 , 1143p k.h . 2005 , , 359 , 725p e.r . 1984 , solar magnetohydrodynamics , reidel , dordrecht m.j . & mszros p 2005 , 628 , 847r c. & tsinganos k. 1994 , 287 , 893s v.d . 1966 , plasma equilibrium in a magnetic field , reviews of plasma physics , vol . 2 , new york : consultants bureau , p. 103 , lynden - bell d. 2007 , , 378 , 409s a. , 1952 , electrodynamics , new york h. 2008 , , 983 , 391s p.a . 1994 , plasma physics , cambridge university press a. , mckinney j.c . & narayan r. 2008 , 388 , 551 t d.a . & macfadyen a.i , 2006 , 647 , 1192u r. , low b.c . 1992 , 391 , 353w following the formalism of @xcite we express the electric and the magnetic fields in the following way , dot denotes differentiation with respect to time , @xmath139 @xmath402 the fields must satisfy maxwell s equations in vacuum . by construction they satisfy @xmath403 @xmath404 the other two equations will let us determine function @xmath137 . substituting into @xmath405 we take @xmath406 and by substituting in the equation @xmath407 we have : @xmath408 given that the fields that have physical meaning are expressed in forms of derivatives of @xmath137 we can choose it to satisfy the wave equation without overconstraining the fields . the proof that follows uses a gauge transformation of @xmath137 . @xmath409 for @xmath410 . indeed : @xmath411 @xmath412 integrating equations ( a5 ) and ( a6 ) , we take : @xmath413 where @xmath414 and @xmath415 . using the gauge transformation that leaves the fields unaffected as we have shown above , we can substitute in favour of @xmath416 , and equation ( a10 ) becomes : @xmath417 note that the divergence of the rhs vanishes . thus by choosing @xmath418 we can make the rhs of ( a11 ) vanish and therefore , equations ( a5 ) and ( a6 ) reduce to the wave equation . the next task is to solve the wave equation . in axial symmetry with @xmath137 along @xmath138 everywhere , the wave equation becomes : @xmath419\pi=0,\end{aligned}\ ] ] when @xmath420 , the magnitude of @xmath137 , has spherical symmetry , the solution to the above equation describes a wave emerging from the centre and expanding outwards . it is @xmath142 where @xmath141 is any function . it is possible to extend the idea of the hertzian dipole to describe a multipole . these multipoles have electric and magnetic fields confined between co - axial cones . it is possible to isolate and apply the solutions only within the central cone or between any two cones ignoring the rest of the field as there is no interaction between them . in order to find such solutions , the vector potential @xmath137 will be a function of @xmath8 , @xmath117 and @xmath32 . although it is possible to find general solutions for the wave equation in spherical coordinates , we can by demanding self - similarity , design a set of multipole solutions that coincide with the prendergast system . where @xmath147 is heaviside s step function and @xmath46 is @xmath428 . for @xmath429 , zero satisfies equation ( b2 ) . in order to find the equivalence of these solutions to the prendergast fields in empty space we will check if the fields defined in ( a1 ) and ( a2 ) obey the conditions set in expressions ( 1 ) and ( 6 ) indeed ( b5 ) reduces to ( b2 ) after the differentiations are done ; therefore , a field that obeys the self - similar wave equation , it will obey ( 1 ) and ( 6 ) as well . thus the hertzian multipole is equivalent to the multipole fields that obey the empty space prendergast equation . to check that this solution ( b4 ) holds not merely for @xmath425 and @xmath429 but also through @xmath130 we divide equation ( b4 ) by the regular function @xmath431 and integrate over the small region @xmath432 . after integrating by parts the left hand side gives : @xmath433(u-1)\big\}^{1+\epsilon}_{1-\epsilon}{\nonumber}\\ -\big[(u^{2}-1)p_{l}'(u)h\big]^{1+\epsilon}_{1-\epsilon } & & { \nonumber}\\ -l(l+1)\int^{1+\epsilon}_{1-\epsilon}(u-1)p_{l}'(u)hdu\big)p_{l}(\mu ) & & \end{aligned}\ ] ] where @xmath147 stands for @xmath434 so it is @xmath435 when @xmath436 and zero when @xmath437 and its derivatives are zero at both places . the first term reduces to @xmath438 and the second to @xmath439 and the third reduces to the same integral without the @xmath147 but with the lower limit replaced by @xmath435 . it is clearly @xmath440 . thus the whole expression is @xmath441 and there is no @xmath442 function discontinuity in the equation at @xmath130 . it is interesting to remark that had we divided equation ( b4 ) by @xmath443 before integrating we have found the result which equals @xmath445 , so the equation when divided by @xmath443 does have a @xmath442 function discontinuity between the two sides at @xmath130 ! it is only the extra factor @xmath446 that saves the situation above . the derivation of equation ( b2 ) from maxwell s equations does not involve any multiplication by @xmath447 . having demonstrated that our solution is a solution everywhere we recall that our starting equations are independent of the zero point of time so an explosion that starts at @xmath449 will be described by setting @xmath450 rather than @xmath451 . also , since our equations are the linear maxwell equations , we can superpose solutions to obtain , if @xmath420 satisfies maxwell s equations @xmath453 , will too . the physics of this is that the explosion takes a finite time of order @xmath156 and occurs in a finite volume of order @xmath454 so we need to average the fields generated by a point explosion over radial distances @xmath455 . where we used the fact that @xmath462 satisfies legendre s equation . now when the step in the heaviside function in equation ( b8 ) is at later times than those involved in the integral in ( b9 ) we replace h by unity . so provided @xmath463 we perform the integral in ( b9 ) and find however when @xmath465 the heaviside function cuts into the range of integration in ( b9 ) . if we multiply equation ( b10 ) by @xmath434 we then find that the range of integration has its upper limit changed from @xmath466 to @xmath467 . to evaluate @xmath453 we need the three integrals as a further check we took @xmath109 and showed that this expression for @xmath453 does indeed obey equation ( a13 ) which itself follows directly from maxwell s equations . thus our procedure of finding singular self - similar solutions and then averaging out the singularities is indeed giving non - singular solutions of maxwell s equations even in the region over which the singularity is smoothed out . of course in the new region ( iii ) @xmath475 there is no signal and @xmath453 is zero . remembering that @xmath476 we see that @xmath453 is given by @xmath477p_{l}'(u_{0})p_{l}(\mu ) & u_{0}-1 > \frac{c\tau_{0}}{2r}\\ \{[(u_{0}^{2}-1)i_{0}+\frac{1}{2}l(l+1)i_{2}]p_{l}'(u_{0})+ & \\ + l(l+1)i_{1}p_{l}(u_{0})\}p_{l}(\mu ) & |u_{0}-1|\leq \frac{c\tau_{0}}{2r } \\ 0 & u_{0}-1<-\frac{c\tau_{0}}{2r}. \end{array } \end{aligned}\ ] ] it is easy to see that @xmath453 is continuous as @xmath478 at the edges of region ( ii ) , so @xmath479 and @xmath480 go from 1 and @xmath481 at the inner edge of region ( ii ) , to zero at the outer edge and @xmath482 is zero at both edges . such a continuity is actually inevitable since even the unsmoothed @xmath483 is continuous since @xmath484 is continuous but furthermore the first derivative of @xmath420 includes a heaviside function discontinuity . the result of smoothing in @xmath117 ensures that the @xmath485 is continuous but @xmath486 may have a heaviside function corresponding to finite sudden changes in field strength at the edges of region ( ii ) . even such jumps would be avoided had we used the gentler function @xmath487 with @xmath488 in the range @xmath489 to @xmath490 in place of the simpler function @xmath491 used above . the more complicated function has zero first derivatives at both ends of the range and would lead to continuous and differentiable field strengths everywhere but we felt that the much more complicated expression for the @xmath492 added unnecessary complexity . we notice that when @xmath156 is small , the smoothing produced a negligible modification of order @xmath493 in region ( i ) . in region ( ii ) the changes are significant in that the @xmath442 function fields are removed and spread out over a region @xmath494 around @xmath51 . the @xmath482 term is especially significant here , despite its falling to zero at the inner and outer edges of this region .

the aim of this work is to identify the mechanisms responsible for very large rates and other peculiarities observed in low - energy positron annihilation on molecules . the two mechanisms considered are : ( i ) direct annihilation of the incoming positron with one of the molecular electrons . this mechanism dominates for atoms and small molecules . i show that its contribution to the annihilation rate can be related to the positron elastic scattering cross section . this mechanism is characterized by strong energy dependence of at small positron energies and high values ( up to ) for room temperature positrons , if a low - lying virtual level or a weakly bound state exists for the positron . ( ii ) resonant annihilation , which takes place when the positron undergoes resonant capture into a vibrationally excited quasibound state of the positron - molecule complex . this mechanism dominates for larger molecules capable of forming bound states with the positron . for this mechanism averaged over some energy interval , e.g. , due to thermal positron energy distribution , is proportional to the level density of the positron - molecule complex , which is basically determined by the spectrum of molecular vibrational states populated in the positron capture . the resonant mechanism can produce very large annihilation rates corresponding to . it is highly sensitive to molecular structure and shows a characteristic behaviour of at small positron energies . the theory is used to analyse calculated and measured for a number of atoms and molecules .

physics9912023

the aim of this work is to develop the framework for the description of low - energy positron annihilation on molecules , and to analyse its two main mechanisms : direct and resonant annihilation . there are a number of remarkable phenomena associated with this process : very large annihilation rates @xcite , high sensitivity of the rates to small changes in the molecular structure @xcite , large ionization - fragmentation cross sections for organic molecules at sub - ps - threshold positron energies @xcite , and rapid increase of the fragmentation and annihilation rates towards small positron energies @xcite . in spite of decades of study , there is no consistent physical picture or even general understanding of these processes , and there have been very few calculations @xcite , which leaves too much room for speculations @xcite . my main objective is to consider real mechanisms of positron annihilation on molecules , describe their characteristic features , make estimates of the corresponding annihilation rates , and formulate the terms in which positron - molecule annihilation should be described and analysed . in recent work @xcite , iwata _ _ describe new experiments to study positron annihilation on molecules . some of these experiments test specific features of the annihilation processes described in the present paper . though some aspects of the experimental work are discussed here , futher details and comparison with theory and various models of positron annihilation can be found in ref . @xcite . the annihilation rate @xmath5 for positrons in a molecular or atomic gas is usually expressed in terms of a dimensionless parameter @xmath0 : @xmath6 where @xmath7 is the classical radius of the electron , @xmath8 is the non - relativistic spin - averaged rate of electron - positron annihilation into two @xmath9 quanta , and @xmath10 is the number density of molecules @xcite . equation ( [ eq : zeff ] ) implies that @xmath0 is the effective number of target electrons contributing to the annihilation process . in terms of the annihilation cross section @xmath11 the rate is @xmath12 , so by comparison with eq . ( [ eq : zeff ] ) , we have @xmath13 where @xmath14 is the positron velocity . accordingly , the spin - averaged cross section of annihilation of a non - relativistic positron on a single electron corresponds to @xmath15 , see e.g. @xcite . if the annihilation occurs during binary positron - molecule collisions , as in the experiments of the san diego group @xcite who use a positron trap and work at low gas densities , the parameter @xmath0 is independent of the density . it characterizes the annihilation of a positron on a single molecule . one could expect that @xmath0 is comparable to the number of electrons @xmath16 in an atom or molecule . moreover , low - energy positrons do not penetrate deep into the atom , and annihilate most probably with the valence electrons only . however , even for hydrogen @xmath17 at low energies @xcite . this is a manifestation of correlation effects . the most important of them is polarization of the atom by the positron and , as a result , an attractive @xmath18 positron - atom potential , @xmath19 being the atomic dipole polarizability . an additional short - range contribution to the positron - atom attraction comes from virtual ps formation , i.e. , hopping , or rather , tunneling of an electron between the atomic ion and the positron . the electron density on the positron is also enhanced due to the coulomb attraction between them . these effects make atomic @xmath0 large , e.g. , @xmath20 for room temperature positrons on xe @xcite . even compared with this large number , annihilation rates for low - energy ( room temperature ) positrons on polyatomic molecules are huge . they increase very rapidly with the molecular size , and depend strongly on the chemical composition of the molecules , see fig . [ fig : zeffmol ] . this has been known for quite a while , after early measurements for ccl@xmath21 , @xmath22 @xcite , butane , @xmath23 @xcite , and @xmath0 ranging between @xmath24 and @xmath25 for large alkanes c@xmath26h@xmath27 , @xmath28416 @xcite ( see also @xcite ) . the largest @xmath0 values measured so far are @xmath29 for antracene c@xmath30h@xmath31 @xcite and @xmath32 for sebacic acid dimethyl ester c@xmath33h@xmath34o@xmath21 @xcite . thus , while @xmath0 up to five orders of magnitude greater than @xmath16 have been observed , the physical processes responsible for these anomalously large annihilation rates have not been really understood . in other words , if the observed @xmath0 are parametrically large , compared to the number of available electrons , then what are the parameters that determine large annihilation rates for positrons on molecules ? in this work i consider two basic mechanisms of positron - molecule annihilation . the first mechanism is _ direct annihilation _ of the incoming positron with one of the molecular electrons . the contribution of this mechanism to the annihilation rate is proportional to the number of valence electrons available for annihilation . it can be enhanced by the positron - molecule interaction which distorts the positron wave . in particular , the positron density in the vicinity of the molecule increases greatly if a low - lying virtual state ( @xmath35 ) or a weakly bound level ( @xmath36 ) exists for the @xmath37-wave positron . in this case @xmath38 for small positron energies @xmath39 @xcite . this type of enhancement is responsible for large @xmath0 values observed in heavier noble gas atoms , where successively lower virtual levels exist for the positron ( @xmath40 , 90.1 and 401 for ar , kr and xe , respectively @xcite ) . this understanding is confirmed by the temperature dependences of the annihilation rates measured for the noble gases in @xcite . note that for room - temperature positrons , @xmath41 , even for @xmath42 the size of the enhancement due to virtual / weakly bound states is limited . the second mechanism is _ resonant annihilation_. by this i mean a two - stage process . the positron is first captured into a feshbach - type resonance , where positron attachment is accompanied by excitation of some molecular degrees of freedom . such process is well known for electrons @xcite . the positron in the quasi bound state then annihilates with a molecular electron . enhancement of annihilation due to a single resonance was considered theoretically in @xcite . the possibility of forming such resonances by excitation of the vibrational degrees of freedom of molecules was proposed by surko _ @xcite to explain high annihilation rates and their strong dependence on the molecular size observed for alkanes . it was also considered in relation to the problem of fragmentation of molecules by positron annihilation @xcite . however , its contribution to the annihilation have never been properly evaluated . to make this mechanism work for low - energy positrons one must assumed that positrons can form bound states with large neutral molecules , i.e. , the positron affinity of the molecule is positive , @xmath43 @xcite . the capture is then possible if the energy of the incoming positron is in resonance with the vibrationally excited state of the positron - molecule complex @xcite . the density of the vibrational excitation spectrum of this complex can be high , even if the excitation energy supplied by positron binding , @xmath44 , is only few tenths of an ev ( it is reasonable to assume that the presence of the positron does not change the vibrational spectrum of the molecule by too much ) . for positrons with thermal maxwellian energy distribution the contribution of the resonant annihilation mechanism averaged over a number of resonances @xmath45 is observed . the magnitude of @xmath45 is determined by three parameters of the positron - molecule resonant states : their annihilation width @xmath46 , the autodetachment width @xmath47 , which also determines the probability of positron capture , and the level density @xmath48 of the positron - molecule resonant states populated in positron capture . the magnitude of @xmath46 for positron - molecule bound states is comparable to the spin - averaged annihilation width of the ps atom ( @xmath49 s ) . note that @xmath46 does not increase with the size of the molecule , because the increase in the number of electrons is accompanied by thinning of the positron density in the ( quasi)bound positron - molecule state . it turns out ( see sec . [ sec : anmech ] ) that for @xmath50 the magnitude of @xmath45 is simply proportional to @xmath48 . this density increases rapidly with the size of the molecule , @xmath51 , where @xmath52 is the number of vibrational modes , @xmath53 is the effective number of vibrational quanta excited in positron capture , and @xmath54 is a typical molecular vibrational frequency . thus , the resonant annihilation mechanism can explain the rapid increase of @xmath0 with the size of the molecule shown in fig . [ fig : zeffmol ] . moreover , my estimates show that for thermal positrons @xmath45 up to @xmath55 could be observed . a necessary condition for the resonant annihilation to occur is the existence of positron - molecule bound states . until recently there was almost no positive information about the possibility of positron binding to neutral atomic species . the experimental results and their interpretation by surko _ @xcite could be viewed as the strongest , albeit indirect , evidence of positron binding to large molecules . this situation has changed now . many - body theory calculations of dzuba _ et al . _ @xcite indicated strongly that positrons can be bound by mg , zn , cd , and hg and , possibly , many other atoms . recently the variational calculations of ryzhikh and mitroy proved rigorously that positrons form bound states with li atoms , and demonstrated that bound states also exist for na , be , mg , zn , cu and ag @xcite . molecules are much larger potential wells for the positron , and it seems natural that many of them should be capable of binding positrons . ideas about different mechanisms in positron - molecule annihilation have been discussed earlier in a number of theoretical @xcite and experimental @xcite works . however , there is a need to re - examine this question using a unified approach to the annihilation mechanisms , and define clearly the physical variables which determine the observed annihilation rates . the latter is especially important for the present work which aims to provide understanding of a whole variety of phenomena , including the origins of the high values of @xmath0 for molecules and their dependence on the chemical composition and positron energy . in this section a derivation of the positron annihilation rate within a standard scattering theory formalism is presented . i show how to estimate the contributions of the direct and resonant mechanisms , and examine specific features of these mechanisms . the effective number of electrons @xmath0 related to the annihilation rate through eq . ( [ eq : zeff ] ) is determined by the positron density on the electrons @xmath56 where @xmath16 is the number of target electrons , @xmath57 and @xmath58 are the coordinates of the electrons and positron , respectively , and @xmath59 is the total wave function of the system . it describes scattering of the positron with initial momentum @xmath60 from the atomic or molecular target in the ground state @xmath61 , and is normalized as @xmath62 where @xmath63 is the radius of the target ( atomic units are used throughout ) . note that for molecular targets @xmath64 and @xmath61 should , strictly speaking , depend on the nuclear coordinates as well . let us first assume that the electron - positron degrees of freedom are completely decoupled from the nuclear motion . the scattering wave function is then determined by the positron interaction with the charge distribution of the ground - state target and electron - positron correlation interaction ( polarization of the target , virtual ps formation , etc . ) . let us denote the corresponding wave function @xmath65 . at positron energies of a few electron volts the molecule can be excited electronically , and the positron may find itself trapped in electronically excited feshbach resonance states . this may result in rapid resonant energy dependence of the @xmath65 wave function . however , at small sub - ev or room - temperature positron energies electron excitations can not be produced , and @xmath65 behaves smoothly . on the other hand , if the positron affinity of a molecule is positive , the system ` molecule@xmath66positron ' is capable of forming a stable `` positronic ion '' , whose lifetime is only limited by positron annihilation . this system will also have a number of excited bound states @xmath67 corresponding to vibrational excitations of the positron - molecule complex . their typical energies are of the order of 0.1 ev and smaller , as determined by the vibrational spectrum of the molecule . if we now turn on the coupling @xmath68 between the electron - positron and nuclear degrees of freedom the total scattering wave function will be given by @xmath69 the first term on the right - hand side describes direct , or _ @xcite , scattering of the positron by the ground - state molecule . the second term describes positron capture into bound positron - molecule states . equation ( [ eq : totpsi ] ) has the appearance of a standard perturbation - theory formula . the energy of the system is @xmath70 , where @xmath71 is the target ground state energy . the energies of the positron - molecule ( quasi)bound states @xmath67 in the denominator are complex , @xmath72 , because these states are , in fact , unstable against positron annihilation with one of the target electrons , and against positron emission , a process inverse to positron capture . therefore , the total width of state @xmath73 is the sum of the annihilation and emission ( or capture ) widths : @xmath74 @xcite . these states manifest as resonances in positron - molecule scattering . they may not give a sizeable contribution to the scattering cross section , but , as i show below , they can contribute a lot to the positron - molecule annihilation rate . the contribution of a particular resonant state @xmath73 to the wave function is proportional to the corresponding capture amplitude @xmath75 , which also determines the capture width @xmath76 where the latter formula is valid for the positron @xmath37 wave which dominates at low positron energies ( see below ) . if the positron interaction with vibrations can not be described by perturbations eqs . ( [ eq : totpsi ] ) and ( [ eq : gamc ] ) remain valid , provided we replace the first - order capture amplitudes @xmath75 with their non - perturbative values . the annihilation width of the positron - molecule state @xmath67 is a product of the spin - averaged electron - positron annihilation cross section @xmath77 , the positron velocity @xmath14 , and the density factor , @xmath78 where @xmath79 is the average positron density on the target electrons in the @xmath80th bound state . for the ground state positronium @xmath81 . one can use this value to estimate the annihilation width of the positron - molecule complex . the presence of many electrons in a large molecule does not lead to an increase of the width , because the positron is spread over a larger volume due to the normalization condition @xmath82 therefore , using the ps estimate of the density one obtains @xmath83 a.u . @xmath84 @xmath85ev , which corresponds to the annihilation lifetime @xmath86 s. to calculate @xmath0 wave function ( [ eq : totpsi ] ) is substituted into eq . ( [ eq : zeffpsi ] ) , which yields @xmath87 where @xmath88 is the capture amplitude introduced as @xmath89 [ cf . ( [ eq : gamc ] ) ] . the terms on the right - hand side correspond to the contributions of direct annihilation , resonant annihilation ( i.e. , annihilation of the positron captured into the positron - molecule quasibound state ) , and the interference between the two . the direct annihilation term in eq . ( [ eq : zeff1 ] ) @xmath90 is a smooth function of the positron energy . let us estimate its magnitude and find its energy dependence at small positron energies . when the positron is outside the atomic system , @xmath91 , the wave function @xmath92 contains contributions of the incoming and scattered positron waves @xmath93 ~,\ ] ] where @xmath94 is the scattering amplitude . due to positron repulsion from the atomic nuclei the low - energy positron does not penetrate deep inside the atomic system . accordingly , the positron annihilates mostly with the outer valence electrons , where the electron and positron densities overlap . this takes place `` on the surface '' of the atomic system , and eq . ( [ eq : out ] ) essentially determines the amplitude of finding the positron there . of course , due to short - range electron - positron correlations the true wave function at small distances can not be factorized similarly to eq . ( [ eq : out ] ) . the coulomb interaction between the positron and electron increases the probability of finding both at the same point in space , as required by the @xmath95-function in eq . ( [ eq : zeffpsi ] ) . this effect enhances the annihilation rate @xcite . however , since small distances and relatively large interactions are involved , these correlations do not depend on the momentum of the incoming positron at low energies . on the other hand , to participate in the annihilation event the positron must first approach the target , and this is described by eq . ( [ eq : out ] ) . unlike the short - range correlation effects , the scattering amplitude can be very sensitive to the positron energy . this effect is fully accounted for by eq . ( [ eq : out ] ) , and i use it to evaluate the energy dependence and magnitude of @xmath96 . after substitution of expression ( [ eq : out ] ) into eq . ( [ eq : zdir ] ) one obtains @xmath97 \left[e^{-i{\bf kr}}+f^*(\omega ) \frac{e^{-ikr}}{r } \right ] r^2drd\omega ~,\ ] ] where @xmath98 is the electron density in the ground state of the system . the electron density drops quickly outside the atom , and the positron density decreases rapidly inside the atom . therefore the integration in eq . ( [ eq : dir1 ] ) should be taken over a relatively thin shell of thickness @xmath99 enclosing the atomic system . let us approximate the integration domain by a spherical shell of radius @xmath100 , where @xmath63 is the typical distance between the positron and the target during the annihilation , comparable to the size of the atom or molecule . for small positron momenta , @xmath101 , eq . ( [ eq : dir1 ] ) then yields @xmath102 where @xmath103 is the electron density in the annihilation range ( which can be enhanced due to short - range electron - positron correlations ) , @xmath104 is the elastic cross section , @xmath105 , and @xmath106 is the spherically symmetric part of the scattering amplitude , @xmath107 . for positron interaction with an atom the latter is simply equal to the @xmath37-wave scattering amplitude . its real part is expressed in terms of the @xmath37 wave phase shift @xmath108 as @xmath109 . the @xmath37 wave gives a dominant contribution to the cross section @xmath104 at low projectile energies @xcite . for @xmath110 it is determined by the scattering length @xmath111 , @xmath112 , as @xmath113 in this limit . a similar description is also valid for positron scattering from a molecule at small momenta . note that the relation between @xmath96 and elastic scattering given by eq . ( [ eq : zdir1 ] ) could also be derived by matching the true many - body wave function of the positron - target system at low energy ( @xmath114 ) with the asymptotic form ( [ eq : out ] ) . in this case @xmath63 will be the matching radius , and the factor before the brackets will remain a free atomic - sized parameter . however , even in the form ( [ eq : zdir1 ] ) the electron density @xmath103 and the overlap @xmath99 are effective parameters , and the accurate value of the pre - factor can only be found by comparison with numerical calculations ( see sec . [ subsec : illdir ] ) . nevertheless , eq . ( [ eq : zdir1 ] ) is very useful for the analysis of direct annihilation . the three terms in brackets are due to the incoming positron plane wave , the scattered wave , and the interference term , respectively , cf . ( [ eq : out ] ) and ( [ eq : dir1 ] ) . even if the cross section @xmath104 is zero or very small , as in the case of a ramsauer - townsend minimum , the annihilation rate @xmath96 is nonzero . its magnitude is determined by the effective annihilation radius @xmath63 , electron density @xmath103 and @xmath99 , which gives @xmath11510 , since the quantities involved have `` normal '' , atomic - size values . equation ( [ eq : zdir1 ] ) shows that the annihilation rate for slow positrons is greatly enhanced if the scattering cross section is large . this occurs when the scattering length is large , because the positron - target interaction supports a low - lying virtual @xmath37 level ( @xmath116 ) or a weakly bound @xmath37 state ( @xmath117 ) @xcite . their energies , @xmath118 , respectively , must be much smaller than typical atomic energies , @xmath119 . for @xmath120 the scattering cross section at low energies is much greater than the geometrical size of the target . this effect leads to strong enhancement of @xmath96 @xcite . theoretically , this gives a possibility of infinitely large cross sections and annihilation rates at zero positron energy , if @xmath121 . however , for nonzero momenta the @xmath37 wave cross section does not exceed the unitarity limit @xmath122 ( for the @xmath37 wave ) . this fact puts a bound on the enhancement of @xmath96 . for example , for thermal positrons with @xmath123 at room temperature ( @xmath124 a.u . ) we obtain @xmath125 from eq . ( [ eq : zdir1 ] ) . consequently , much higher values of @xmath0 can not be produced by the direct annihilation mechanism . a more detailed discussion of this point and illustrations of the validity of eq . ( [ eq : zdir1 ] ) are presented in sec . [ subsec : illdir ] . unlike the direct annihilation term , the interference and the resonant terms on the right - hand side of eq . ( [ eq : zeff1 ] ) are rapidly varying functions of energy . the energy scale of this variation is given by the mean spacing @xmath126 between the resonances . if the resonances are due to vibrational excitations of a single mode of the positron - molecule complex then @xmath127 , with @xmath128 ev for a typical vibrational frequency . in a complex molecule the positron attachment energy is sufficient for excitation of several modes , and @xmath126 can be much smaller . to describe the annihilation rates observed in experiments with non - monochromatic , e.g. , thermal , positrons , one needs to average the interference and resonance terms over an energy interval @xmath129 which contains many resonances : @xmath130\end{aligned}\ ] ] upon averaging the first , interference term vanishes . in the second , resonance term the diagonal items in the sum ( @xmath131 ) dominate . averaging is then reduced to the integral over the breit - wigner resonant profiles . the number of resonances within @xmath129 is @xmath132 . therefore , the total annihilation rate is the sum of the direct and resonant contributions , @xmath133 with the resonant contribution given by @xmath134}\right\rangle ~,\ ] ] where the angular brackets stand for averaging over the resonances , and @xmath135 substituted for the total width . below i will show that the resonant term in eq . ( [ eq : result ] ) can be much greater than the direct one , and very high @xmath0 values can be achieved . it is easy to see that the resonant contribution could also be derived from standard resonant scattering theory developed originally to describe neutron scattering via compound nucleus resonances ( @xcite , ch . the maximal @xmath37-wave capture cross section is given by @xmath136 . the true capture cross section is smaller than @xmath137 , because the capture takes place only when the positron energy matches the energy of the resonance . for positrons with finite energy spread ( e.g. , thermal ones ) , the capture cross section is then @xmath138 , where @xmath126 is the mean energy spacing between the resonances . more accurately , @xmath139 @xcite . if we are concerned with the annihilation process , the capture cross section must be multiplied by the probability of annihilation , @xmath140 , which gives the energy - averaged resonance annihilation cross section @xmath141 where averaging over resonances similar to that in eq . ( [ eq : zres ] ) is assumed . by comparison with eqs . ( [ eq : siganzeff ] ) and ( [ eq : gama ] ) , the resonant contribution to @xmath0 , eq . ( [ eq : zres ] ) , is recovered . the way eq . ( [ eq : zres ] ) has been derived implies that the positrons are captured in the @xmath37 wave . otherwise , an additional factor of @xmath142 , where @xmath143 is the positron orbital momentum , appears in the formula @xcite . at low positron energies the capture widths behave as @xmath144 for resonances formed by positron capture with the orbital momentum @xmath143 @xcite ( @xmath145 is the typical radius of the target ) . so , the @xmath37 wave capture indeed dominates in the resonant annihilation of slow positrons . at higher energies contributions of several lowest partial waves should be added in @xmath45 . let us estimate the rate of resonant annihilation and compare it with the maximal direct contribution @xmath125 for room - temperature positrons . the typical annihilation widths for positron - molecule ( quasi)bound states are very small , @xmath146 ( see sec . [ subsec : gen ] ) . if one assumes that the positron capture width is much greater , @xmath147 the total width @xmath148 cancels the capture width in eq . ( [ eq : zres ] ) , and the resonant contribution is given by @xmath149 in the last equality i use the fact that electron - positron degrees of freedom are almost unaffected by the vibrational motion of the nuclei . hence , for a given molecule the positron density on the target electrons @xmath150 is the same for different vibrational resonances . i have also introduced the density of resonances @xmath151 , where @xmath44 is the vibrational excitation energy due to positron - molecule binding . equation ( [ eq : resres ] ) shows that for @xmath152 @xmath85ev the contribution of the resonant mechanism is _ independent _ of the capture width , and is determined by the density of positron - molecule resonant states populated by positron capture . suppose that only a single mode with @xmath153 ev is excited . equation ( [ eq : resres ] ) then yields @xmath154 , if i use the estimates @xmath155 , and @xmath156 for room - temperature positrons . the resonance spacing @xmath126 can not be smaller than the widths of the resonances , which are limited by the annihilation width @xmath46 . thus , one can obtain an upper estimate of the resonant annihilation rate from eq . ( [ eq : zres ] ) by putting @xmath157 a.u . , and @xmath158 , which gives the maximal possible capture cross section @xmath137 . these estimates yield @xmath159 at room temperature ( cf . @xmath160 for c@xmath33h@xmath340@xmath21 @xcite ) . this theoretical maximum of @xmath45 corresponds to the unitarity limit of the @xmath37 wave capture cross section . however , this estimate of @xmath0 is not trivial . the resonance mechanism shows that such large cross sections can be achieved for the annihilation process , in spite of the fact that it is suppressed by the relativistic factor @xmath161 , in atomic units [ see eq . ( [ eq : siganzeff ] ) ] . equation ( [ eq : resres ] ) predicts unusual low - energy threshold behaviour @xmath162 ( the latter for thermal positrons ) . in a standard situation the cross section of an inelastic process involving a slow projectile in the initial state behaves as @xmath163 . this dependence is characteristic of the @xmath37 wave scattering , which dominates at low projectile energies , and is valid in the absence of long - range forces between the target and the projectile . it is known as the `` @xmath164 '' law , and its examples are numerous : from the @xmath165 nuclear reaction to dissociative electron attachment to molecules , where it is observed below 1 mev @xcite . therefore , one would expect the positron annihilation cross section to behave as @xmath166 . accordingly , @xmath0 , which is proportional the annihilation rate , is expected to be constant at low positron energies . the anomalous threshold dependence of eq . ( [ eq : resres ] ) clearly contradicts this general statement . this `` puzzle '' is easily resolved if we recall condition ( [ eq:2gam ] ) that has lead to eq . ( [ eq : resres ] ) . for very low positron momenta the @xmath37-wave capture width behaves as @xmath167 , so that ( [ eq:2gam ] ) is clearly violated , and the resonant contribution in eq . ( [ eq : result ] ) becomes constant as @xmath110 . however , at higher positron energies the @xmath168 behaviour of @xmath0 may be observed . this dependence corresponds to the @xmath169 drop of the cross section which is reported in some electron attachment experiments ( see , e.g. , @xcite ) . the fact that positron - molecule resonances give a large contribution to the annihilation rate , as compared to the direct annihilation , does not mean that they also contribute much to the elastic scattering cross section . in analogy with eq . ( [ eq : sigan ] ) , the resonant contribution to the elastic scattering is given by @xmath170 and for @xmath171 it is much smaller than the direct , or potential , scattering cross section . if low - energy positron scattering is dominated by the presence of a virtual or weakly bound state at @xmath172 , the corresponding cross section has the form ( for scattering by a short - range potential @xcite ) @xmath173 where @xmath174 . according to eq . ( [ eq : zdir1 ] ) a similar maximum should appear in the momentum dependence of the annihilation rate . its magnitude at @xmath175 can be arbitrarily large if @xmath176 ( @xmath177 ) , which corresponds to a level at zero energy . however , for nonzero momenta the maximal cross section is finite , @xmath178 , which corresponds to the unitarity limit for the @xmath37-wave cross section . real atomic and molecular targets have nonzero electric dipole polarizabilities @xmath19 , which give rise to the long - range polarization potential @xmath179 for the positron . its effect is taken into account by the modified effective - range formula for the @xmath37-wave phase shift @xcite , @xmath180^{-1}~ , \\ \sigma _ { \rm el}=\frac{4\pi a^2 } { \left [ 1-(\pi \alpha k/3a)-(4\alpha k^2/3 ) \ln \left ( \frac{c}{4}\sqrt{\alpha } k \right ) \right]^2+a^2k^2 } ~ , \label{eq : mersig}\end{aligned}\ ] ] the latter formula being valid when the scattering length is large and the @xmath37-wave scattering dominates at small @xmath181 . in equations ( [ eq : mer ] ) and ( [ eq : mersig ] ) @xmath182 is a dimensionless positive constant . note that for @xmath183 , eq . ( [ eq : csvirt ] ) is immediately recovered . the polarization potential modifies the behaviour of the cross section at low energies . for example , it leads to a more rapid decrease of the cross section for @xmath116 , @xmath184 $ ] . however , this does not change the estimates of the maximal values of @xmath0 that could be produced in direct annihilation . to illustrate the relation between direct annihilation and elastic scattering , and the enhancement of both due to the presence of a low - lying virtual level , let us compare the behaviour of @xmath0 and @xmath104 for ar and kr . the results shown in fig . [ fig : zandcs ] were obtained within the polarized - orbital method @xcite , which takes into account the polarization of the target by the positron . these calculations yield large negative values of the scattering length for ar , kr and xe ( see table [ tab : ar ] ) , indicating the presence of positron - atom virtual levels formed due to strong positron - atom attraction . the increase of @xmath185 correlates with the increase of the dipole polarizability in these atoms . similar values of @xmath111 have been obtained in the many - body theory calculations of dzuba _ et al . figure [ fig : zandcs ] shows that both @xmath104 and @xmath0 are enhanced at low momenta due to the presence of the virtual @xmath37 levels . this effect is stronger for kr , which has a greater absolute value of the positron scattering length . as illustrated by fig . [ fig : zandcs]a for kr , eq . ( [ eq : mersig ] ) provides a good description of the cross section at small @xmath181 . the visible difference between @xmath0 and @xmath104 in fig . [ fig : zandcs ] is due to the background given by the energy - independent term @xmath186 in eq . ( [ eq : zdir1 ] ) . figure [ fig : zdir ] provides a direct comparison between @xmath0 and the right - hand side of eq . ( [ eq : zdir1 ] ) , and shows that this relation is valid at low positron energies . the comparison is based on the polarized - orbital method results for the noble - gas atoms @xcite , and the values of @xmath0 and @xmath104 obtained for the ethylene molecule ( c@xmath187h@xmath21 ) by the schwinger multichannel method @xcite . in this comparison i have considered @xmath63 and the pre - factor @xmath188 in eq . ( [ eq : zdir1 ] ) as fitting parameters . their values are listed in table [ tab : ar ] together with the values of @xmath111 obtained in those calculations . note that the theoretical results used to produce this plot are not necessarily `` exact '' or accurate ( although , experimental data confirm that they are reasonable @xcite ) . it follows from the derivation that eq . ( [ eq : zdir1 ] ) holds for any calculation , as long as the same wave function is used in the scattering and annihilation calculations @xcite . in agreement with the estimates made in sec . [ subsec : dir ] , fig . [ fig : zdir ] shows that direct annihilation is indeed strongly enhanced by the presence of low - lying virtual levels . nevertheless , even for targets with very large scattering lengths , such as xe or c@xmath187h@xmath21 , the annihilation rates do not exceed @xmath189 for room - temperature positron momenta ( 0.05 a.u . ) . direct annihilation is the only annihilation mechanism for atoms and molecules which do not form bound states with positrons . it will also dominate for small molecules which do form a weakly bound state with the positron , but whose vibrational frequencies are high . in this case the energy @xmath190 is simply insufficient for the excitation of the resonant quasibound states at low impact positron energies @xmath4 . for large molecules the difference between the resonant and direct mechanisms is probably most obvious when one compares the experimental values of @xmath0 for alkanes and perfluorinated alkanes shown in fig . [ fig : zeffmol ] . the large annihilation rates of the alkane molecules with more than two carbon atoms can not be explained by direct annihilation . they also display a very rapid increase with the size of the molecule , which is typical of resonant annihilation . on the other hand , the @xmath0 values of the perfluorinated alkanes remain comparatively small , in spite of their softer vibrational spectra . thus , one is lead to conclude that the resonant mechanism is switched off for them . the latter is explained by the very weak attraction between the positron and fluorine atoms @xcite , insufficient to provide positron - molecule binding . let us examine the effect of fluorination on @xmath0 for the lightest molecule of the series , methane . the experimental data at room temperature are : @xmath191 , 715 , 411 , 127 , and 38 , for ch@xmath21 , ch@xmath192f , ch@xmath187f@xmath187 , chf@xmath192 , and cf@xmath21 , respectively ( data from @xcite normalized to the given value for methane ) . these values are small enough to be accounted for by the direct mechanism . within its framework the increase and subsequent drop of @xmath0 could be explained by the existence of a loosely bound state for the positron on methane , which turns into a virtual level as the number of substitute fluorine atoms increases @xcite . in terms of @xmath193 parameter this would mean that @xmath193 is small and positive for ch@xmath21 , and then goes through zero , and becomes negative upon fluorination . accordingly , both the cross section and the annihilation rate peak for the molecule with the smallest absolute value of @xmath193 , namely ch@xmath192f . this picture is considered in ref . @xcite in more detail using the zero - range potential model for positron - molecule interaction . besides having a larger value of @xmath0 , the molecule with a smaller @xmath194 ( i.e. , larger @xmath185 ) should have a more rapid dependence of the annihilation rate on the positron energy , cf . figure [ fig : zdir ] . if the experiment is done with thermal positrons this should manifest in a stronger temperature dependence of the maxwellian average of @xmath195 @xmath196 on the positron temperature @xmath197 . the overbar is usually omitted , as it is clear from the context whether one is dealing with @xmath195 at a specific positron momentum , or with a thermal average @xmath198 . the temperature dependences of the annihilation rates for methane and fluoromethane measured in ref . @xcite are shown in fig . [ fig : ch4ch3f ] . also shown are low - temperature theoretical fits obtained using eqs . ( [ eq : zdir1 ] ) , ( [ eq : mersig ] ) and ( [ eq : zt ] ) . their parameters are given in the caption . the dipole polarizability of ch@xmath192f @xmath199 a.u . is close to that of methane , @xmath200 a.u . , and i use the latter for both molecules . the constant @xmath182 appears in eqs . ( [ eq : mer ] ) and ( [ eq : mersig ] ) under the logarithm , and the result is not very sensitive to it , so @xmath201 has been chosen . the value of the characteristic radius @xmath202 a.u . is similar to those for noble gas atoms and ethylene ( table [ tab : ar ] ) , and the pre - factor @xmath203 is between those for noble gas atoms and c@xmath187h@xmath21 . of course , the number of independent parameters ( @xmath111 , @xmath182 , @xmath63 and @xmath204 ) is too large to enable their unique determination from the experimental data . however , the fits clearly demonstrate that very different @xmath205 curves can be obtained _ only _ due to different @xmath193 values ( @xmath206 and 0.01 , for ch@xmath21 and ch@xmath192f , respectively . these values imply that both molecules have bound states with the positron . the binding energy for ch@xmath21 is @xmath207 a.u.@xmath208 ev , and the binding energy corresponding to @xmath209 is just 1 mev . there is a large uncertainty in the latter value , because measurements performed at and above room temperature , @xmath210 ev , are not really sensitive to such small @xmath193 . this can be seen , e.g. , from eq . ( [ eq : csvirt ] ) , which becomes @xmath193-independent for @xmath211 . zero - range model calculations presented in ref . @xcite show that the last three members of the fluoromethane sequence have negative @xmath193 , corresponding to virtual levels with increasing energies . this causes the decrease of their @xmath0 values . as seen in fig . [ fig : ch4ch3f ] , equation ( [ eq : zdir1 ] ) for the direct annihilation combined with the modified effective range formula ( [ eq : mersig ] ) works well in the low - energy part of the graph . however , the data for methane clearly show an abrupt departure from this law at higher @xmath197 , and the formation of some kind of a plateau in @xmath205 . in principle , one could think that this is due to contributions of higher partial waves , not included in @xmath104 , eq . ( [ eq : mersig ] ) . however , their contribution has been included via the @xmath63 term of eq . ( [ eq : zdir1 ] ) . also , the contributions of higher partial waves to @xmath0 emerge as @xmath212 , which is a manifestation of the wigner threshold law @xcite . for thermally averaged rates this corresponds to @xmath213 . thus , it can not be responsible for this sudden feature . on the other hand , if the methane molecule forms a bound state with the positron the system can also have vibrationally excited positron - molecule resonant states . the positron bound state on ch@xmath21 must belong to the @xmath214 symmetry type of the molecule . since the positron @xmath37 wave dominates at low energies , its capture into the @xmath214 state can result in the excitation of @xmath214 vibrational modes of the molecule . the frequency of this mode for methane is @xmath215 @xmath216 ev . assuming that the positron binding does not change this frequency much , the lowest vibrationally excited positron - molecule resonance will occur at @xmath217 ev . it is easy to estimate the contribution of a single narrow vibrational resonance located at positron energy @xmath218 to the thermally averaged @xmath0 @xcite , @xmath219 the latter formula valid for @xmath220 , which implies that the resonance has a capture width greater than 1 @xmath85ev . figure [ fig : ch4ch3f ] shows the effect of the lowest vibrational @xmath214 resonance at @xmath221 ev on @xmath0 for methane ( chain curve ) . its onset is indeed quite rapid , due to the exponent in eq . ( [ eq : singres ] ) , which makes @xmath222 very small for @xmath223 . to fit the experimental data the density @xmath224 is chosen to be 25% of @xmath225 . one could expect that for a weakly bound state ( @xmath226 ev ) , where the positron spends most of its time outside the molecule , its density on the electrons is reduced below that of ps ( binding energy 6.8 ev ) @xcite . equation ( [ eq : resres ] ) derived in sec . [ subsec : res ] shows that the annihilation rate due to positron capture into resonances is determined by the level density of these quasibound vibrationally excited states of the positron - molecule complex . this density depends on the excitation energy available , as defined by the positron kinetic energy and positron affinity , @xmath44 , and also on the structure of the molecular vibrational spectrum . suppose that the molecule possesses a particular symmetry , which is true for most of the molecules where positron annihilation has been studied so far @xcite . the electronic ground state wave function of the molecule is usually nondegenerate and invariant under all symmetry transformations . let us call this symmetry type @xmath227 . depending on the actual symmetry of the molecule this can be @xmath214 , @xmath228 , or @xmath229 . if the positron can be bound by such molecule , the electron - positron part of the wave function of the positron - molecule complex will also be fully symmetric , i.e. , of the @xmath227 symmetry type . consider now the capture of a continuous spectrum positron into the bound positron - molecule state . at low positron energies this process is dominated by the incident positron @xmath37 wave , higher partial waves being suppressed as @xmath230 , compared to the @xmath37 wave [ cf . ( [ eq : gamcth ] ) ] . as a result , the electron - positron part of the wave function of the initial ( molecule and the @xmath37-wave positron ) and final ( bound positron - molecule complex ) states of the capture process are characterized by the same full molecular symmetry @xmath227 . this imposes a selection rule on the nuclear vibrations which can be excited during the capture process . they must also belong to the @xmath227 symmetry type . therefore , the selection rule limits the spectrum of possible vibrationally excited resonances which could in principle be formed . it allows arbitrary excitations and combinations of the @xmath227 modes . it also allows overtones and combinations of the other symmetry types , provided such excitations contain the @xmath227 symmetry type , i.e. , the ( symmetric ) product of the symmetry types involved contains @xmath227 among its irreducible representations @xcite . this does not mean that all such vibrations will contribute to the density factor @xmath48 in eq . ( [ eq : resres ] ) for @xmath0 . some of them may have extremely weak coupling to the electron - positron degrees of freedom , with capture widths much smaller than 1 @xmath85ev . in this case they will be effectively decoupled from the positron capture channel , and hence , will not contribute to @xmath0 . of course , this can only be found out by doing detailed calculations for specific molecules . nevertheless , it is instructive to compare eq . ( [ eq : resres ] ) with experimental data . this comparison enables one to extract the effective mean spacing @xmath126 between the positron - molecule resonances . for experiments with thermal positrons eq . ( [ eq : resres ] ) must be averaged over the maxwellian positron momenta distribution , @xmath231 let us use the ps value , @xmath232 , to estimate the electron - positron density , and apply eq . ( [ eq : zrest ] ) to simple symmetric molecules with @xmath233 , where resonant annihilation must be the dominant mechanism . the effective spacings @xmath234 ( in @xmath235 ) obtained from the experimental @xmath0 values measured with room - temperature positrons @xcite are listed in table [ tab : vibfreq ] . they are compared with the low frequency vibrational modes of the @xmath227 symmetry type of these molecules taken from ref . @xcite . as discussed above , vibrations of the @xmath227 symmetry type also occur in overtones and combinations of other modes . however , their frequencies scale with the size and chemical composition of the molecule in a way similar to the @xmath227 modes , and the @xmath227 mode frequencies listed in the table are representative of the lower vibrational modes on the whole . for molecules with moderate @xmath0 at the top of the table , such as ccl@xmath21 , the effective resonance spacing @xmath126 is comparable to the frequencies of single modes . with the increase of the size of the molecule ( alkanes ) , or masses of the constituents ( e.g. , cbr@xmath21 ) , the vibrational modes are softened , and the number of low - frequency modes increases . at the same time one can expect that the positron binding energy increases for these molecules . these effects , and especially the increase of the number of modes , facilitate multimode excitations , whose density is much greater that the level density of the individual modes . accordingly , we see that @xmath126 becomes much smaller that the frequencies of the individual modes at the bottom of the table . in the simplest model this effect can be estimated as follows . suppose the vibrational modes in question are characterized by some typical frequency @xmath54 , and the molecule has @xmath52 such modes . suppose , the positron binding energy is @xmath236 , where @xmath10 is the number of vibrational quanta excited due to positron binding . if we neglect the small kinetic energy of the positron , @xmath237 , the total number of various vibrational excitations at energy @xmath238 is given by @xmath239 $ ] ( number of ways to distribute @xmath10 vibrational quanta among @xmath52 modes ) . for large molecules @xmath240 remains finite , whereas @xmath52 increases linearly with the size of the molecule , the total number of vibrational modes being @xmath241 , where @xmath242 is the number of atoms . therefore , the number of vibrational excitations available , and the density of the resonant vibrational spectrum , increase as @xmath243 . such rapid increase is indeed observed for alkanes and aromatic hydrocarbons , see fig . [ fig : zeffmol ] . the effective number of vibrational modes excited in the capture process , @xmath244 and 8.2 , respectively , is compatible with the positron binding energy of few tens of an electron volt . for example , if i use the lowest @xmath228 mode frequency of hexane ( table [ tab : vibfreq ] ) , the positron affinity is @xmath245 ev . this number looks reasonable , compared with positron binding energies on single atoms , e.g. , @xmath246 , 0.15 , and 0.38 , for be , cu and mg , respectively @xcite . apart from the rapid growth , @xmath0 for alkanes shows clear signs of saturation , when the number of carbon atoms becomes greater than 8 or 10 . apparently , this takes place well before the unitarity limit derived in sec . [ subsec : res ] is reached . this behaviour can be understood if we recall that eq . ( [ eq : resres ] ) is valid only when the capture width @xmath47 is greater than the annihilation width @xmath46 . with the increase of the number of vibrational modes their coupling to the electron - positron degrees of freedom decreases . this coupling is represented by @xmath47 , and for small capture widths , @xmath247 , @xmath45 from eq . ( [ eq : zres ] ) is estimated as @xmath248 where eq . ( [ eq : gama ] ) is used together with @xmath249 , in atomic units . the decrease of @xmath47 is a simple consequence of sum rules , because the total strength of positron coupling is distributed among larger number of possible vibrational excitations . in this regime @xmath47 is proportional to @xmath126 , and the increase of @xmath45 related to the increase of the density of vibrational excitation spectrum stops . the relation @xmath250 which characterizes this regime is well known in neutron capture into compound resonances @xcite . it takes place in complex atomic spectra , e.g. , in rare - earths , where the oscillator strengths are distributed among very large numbers of transitions @xcite . it also emerges in the unimolecular reaction treatment of dissociative electron attachment @xcite , where it is responsible for very large lifetimes ( i.e. , small state widths ) of transient molecular anions . let us now look at the energy dependence of the resonant annihilation rate . at very small positron energies @xmath45 must be constant ( see discussion at the end of sec . [ subsec : res ] ) . however , as soon as the @xmath37-wave capture width becomes greater that 1 @xmath85ev , the corresponding annihilation rate shows a @xmath251 dependence on positron energy , as predicted by eq . ( [ eq : resres ] ) . for a thermally averaged rate this is described by eq . ( [ eq : zrest ] ) . figure [ fig : c4h10 ] presents a comparison between the @xmath252 law and the experimental temperature dependence of @xmath0 for c@xmath21h@xmath31 @xcite . this molecule has @xmath253 . within the present theoretical framework this large value must be due to the resonant annihilation process . the theory and experiment agree well at low temperatures . one may notice that the measured @xmath0 show a slightly steeper rise towards small @xmath197 . however , the difference is not large , both in relative and absolute terms . it could be explained by a direct contribution @xmath96 in eq . ( [ eq : result ] ) , which peaks sharply at small energies , if the positron - molecule scattering length is large ( see sec . [ subsec : illdir ] ) . in spite of the dominance of the resonant contribution , @xmath254 for butane , the addition of @xmath125 at small positron energies would still be noticeable . a more pronounced feature of the experimental data , which is not accounted for by eq . ( [ eq : zrest ] ) , is the plateau observed at higher temperatures , @xmath255 ev , where @xmath0 goes well above the @xmath252 curve . to find its possible origins let us first take a closer look at eq . ( [ eq : zrest ] ) and its predecessor , eq . ( [ eq : resres ] ) . for small impact positron energies @xmath4 the vibrational excitation energy is given by @xmath237 . accordingly , the resonance density @xmath48 in eq . ( [ eq : resres ] ) , and the mean spacing @xmath126 in eq . ( [ eq : zrest ] ) are approximately constant . as the positron energy , or temperature , increase , the resonance density factor should also increase , since @xmath48 is a strong function of the excitation energy for multimode vibrational spectra . therefore , the decrease of @xmath45 should be slower than @xmath168 , or @xmath252 . moreover , the density factor may even produce a rise in the energy dependence of @xmath45 . besides this , contributions of higher positron partial waves which emerge as @xmath197 , @xmath256 , etc . , at small @xmath197 , may also contribute to @xmath45 in the plateau region . it might even seem that these effects could lead to a rapid increase of @xmath45 with positron energy . however , there is an effect that suppresses the increase of resonant annihilation . throughout the paper i have assumed that the positron - molecule resonances have only two decay channels , annihilation and detachment , the latter being the reverse of positron capture . when the positron energy rises above the threshold of molecular vibrational excitations , the resonances can also decay into the ` positron @xmath66 vibrationally excited molecule ' channels . in this situation the total width of a resonance will be given by @xmath257 , where @xmath258 is the decay width due to positron detachment accompanied by the vibrational excitation of the molecule . this leads to a modification of eq . ( [ eq : resres ] ) , which now reads @xmath259 this equation shows that as soon as the positron energy exceeds another inelastic vibrational - excitation threshold , the factor in brackets drops , thereby reducing the resonant annihilation contribution . such downward step - like structures at vibrational thresholds are well known in dissociative electron attachment experiments ( see , e.g. , refs . when the positron energy is well above the lowest inelastic vibrational threshold the `` elastic '' width @xmath47 will become much smaller than the `` inelastic '' width @xmath260 , due to a large number of open inelastic vibrational - excitation scattering channels , and due to a kinematic increase of @xmath260 above the respective thresholds . this will strongly suppress the resonant annihilation contribution ( [ eq : zgv ] ) with respect to that of eq . ( [ eq : resres ] ) at larger positron energies . one may speculate that it is precisely the increase of @xmath258 that counteracts the rise of @xmath48 , and prevents rapid growth of @xmath261 with positron energies . it may also be true that a similar mechanisms is behind the dramatic drop of the dissociative attachment cross sections for projectile energies above few lower vibrationally inelastic thresholds @xcite . in this work i have considered two possible mechanisms of low - energy positron annihilation in binary collisions with molecules . the first mechanisms is direct annihilation . it describes positron annihilation with atoms and small molecules , as well as molecules which do not form bound states with the positron . the annihilation rate due to this mechanism has been related to the positron elastic scattering properties . in particular , it is enhanced when the positron has a low - lying virtual @xmath37-type level or a weakly bound state at @xmath172 . for zero - energy positrons the direct annihilation rate is inversely proportional to @xmath262 . small @xmath193 , together with the dipole polarizability of the target , also determine the rapid energy dependence of @xmath0 at small positron energies . estimates show that for room - temperature positrons @xmath0 of up to @xmath1 can be produced due the virtual / weakly bound state enhancement . the second mechanism is resonant annihilation . it is operational when the positron forms temporary bound states with the molecule . as a necessary condition , the positron affinity of the molecule must be positive . the positron capture is a resonant process , whereby the energy of the positron is transferred into vibrational excitations of the positron - molecule complex . the contribution of this mechanisms to the annihilation rate is proportional to the level density of the positron - molecule resonances @xmath263 . these resonances are characterized by the capture width @xmath47 and annihilation width @xmath264ev . for @xmath265 its contribution is independent of @xmath47 , and is basically determined by the density @xmath263 . the resonant mechanism can give very large annihilation rates ( up to @xmath55 ) . through its dependence on the vibrational excitation spectrum of the positron - molecule complex , this mechanism shows high sensitivity to the chemical composition of the target , and the size of the molecule . both are essential features of the experimental data @xcite . the difference between the two mechanisms is illustrated most clearly by comparison of the annihilation rates of alkanes and perfluoroalkanes . for example , c@xmath266h@xmath30 has @xmath267 , whereas for c@xmath266f@xmath30 , @xmath0 is only 630 . the present theory attributes this huge difference to the fact that perfluorocarbons do not form bound states with the positrons , and hence , the resonant annihilation is switched off for them . on the other hand , this mechanism is behind the the high @xmath0 values of alkanes . the experimental group at san diego has performed a number of measurements on protonated and deuterated molecules to test the sensitivity of @xmath0 to the molecular vibrational modes @xcite . for example , their data for benzene show that a replacement of a single hydrogen atom with deuterium changes the annihilation rate from @xmath268 for c@xmath266h@xmath266 to @xmath269 for c@xmath266h@xmath270d . on the other hand , the data on fully protonated vs fully deuterated alkanes shows very little difference between the two cases . such behaviour is natural for smaller alkanes , e.g. , methane , where direct annihilation is the dominant mechanism . however , observed for large alkanes , it can not be readily interpreted by means of eq . ( [ eq : resres ] ) or alike . it is possible that the vibrational excitations are dominated by low - lying c@xmath271c modes which are weakly affected by deuteration . on the other hand , deuteration may also influence positron coupling to the molecular vibrations , which will most likely lead to a reduction of @xmath47 in eq . ( [ eq : zres ] ) . if the system is in the regime where @xmath272 , this effect may offset the decrease of the vibrational spacings . in spite of these difficulties , which could only be resolved by doing calculations for specific molecules , the present theory offers a consistent description of positron - molecule annihilation in real terms , through some well defined parameters which characterize the system . it clearly identifies the two basic mechanisms of positron annihilation and discusses their specific features . it also shows that studies of positron annihilation on molecules may give a unique insight into the physics of molecular reactions which go through formation of vibrationally excited intermediate states . such processes are very likely to be responsible for large dissociative electron attachment cross sections observed for molecules such as sf@xmath266 . they are also of key importance for the whole class of chemical reactions , namely , for unimolecular reactions ( see , e.g. , @xcite ) . this work was strongly stimulated by the vast experimental data of the san diego group , and i very much appreciate numerous discussions with its members , especially c. surko and k. iwata . i am thankful to my colleagues at the university of new south wales , v. flambaum , a. gribakina , m. kuchiev , and o. sushkov for their encouragement and useful discussions . my thanks also go to s. buckman for the information on vibrational excitations and dissociative attachment . support of my work by the australian research council is gratefully acknowledged . e. p. da silva , j. s. e. germane , and m. a. p. lima , phys . 77 * , 1028 ( 1996 ) . the @xmath228 symmetry ( `` @xmath37 wave '' ) dominates in both @xmath104 and @xmath0 at low positron energies , and i use @xmath273 to extract the @xmath37 wave phase shift and amplitude @xmath106 necessary for implementation of eq . ( [ eq : zdir1 ] ) . in this paper the word ` molecules ' is often used interchangeably with ` atoms ' , or ` atoms or molecules ' . however , there are specific phenomena , e.g. , low - energy positron capture ( sec . [ subsec : res ] ) , which involve vibrational degrees of freedom , and hence , apply to molecules only . the energy of positrons at room temperature , @xmath274 mev , is too small to excite any electronic degrees of freedom . the formation of ps is also impossible if we consider molecules with ionization potentials greater than the ps binding energy of 6.8 ev . g. g. ryzhikh and j. mitroy , phys . lett . * 79 * , 4124 ( 1997 ) . b * 31 * , l401 ( 1998 ) ; * 31 * , 4459 ( 1998 ) ; * 31 * , 5013 ( 1998 ) ; * 32 * , 1375 ( 1999 ) ; g. g. ryzhikh , j. mitroy , and k. varga , j. phys . b * 31 * , 3965 ( 1998 ) . at low positron energy annihilation and emission are the only decay channels of the resonances . at higher energies positron emission accompanied by the vibrational ( and then electronic ) excitation of the molecule becomes possible , see sec . [ subsec : illres ] . r. p. mceachran , d. l. morgan , a. g. ryman , and a. d. stauffer , j. phys . b * 10 * , 663 ( 1977 ) ; * 11 * , 951 ( 1978 ) ; r. p. mceachran , a. g. ryman , and a. d. stauffer , j. phys . b * 11 * , 551 ( 1978 ) ; * 12 * , 1031 ( 1979 ) ; r. p. mceachran , a. d. stauffer , and l. e. m. campbell , _ ibid _ * 13 * , 1281 ( 1980 ) . the most accurate theoretical data on positron scattering and annihilation are available for hydrogen . to avoid cluttering the plot the corresponding results are not presented in fig . [ fig : zdir ] . for hydrogen eq . ( [ eq : zdir1 ] ) provides a good fit of @xmath96 [ j. w. humberston , adv . at . phys . * 15 * , 101 ( 1979 ) ] with @xmath275 a.u . , @xmath276 a.u . , when variational scattering phase shifts are used to obtain @xmath104 and @xmath106 [ a. k. bhatia _ et al . _ , rev . a * 3 * , 1328 ( 1971 ) ; * 9 * , 219 ( 1974 ) ; d. register and r. t. poe , phys . a * 51 * , 431 ( 1975 ) ] . experimental data for a variety of molecules show that a non - zero dipole moment of the molecule does not have any direct effect on @xmath0 , ref . for example , @xmath277 , 1090 , and 1600 , for h@xmath1870 , no@xmath187 , and nh@xmath192 , whereas their dipole moments are 1.85 , 0.32 , and 1.47 debye , respectively . there also seems to be no correlation between @xmath0 values of ch@xmath192f , ch@xmath187f@xmath187 , chf@xmath192 , quoted in the text , and their respective dipole moments , 1.85 , 1.97 , and 1.65 debye . the contribution of a particular resonance to @xmath0 is given by the breit - wigner formula , cf . terms with @xmath131 in eq . ( [ eq : zeff1 ] ) . for the purpose of thermal averaging , the contribution of the @xmath73th resonance can be approximated by the @xmath95-function , @xmath278 , without any loss of accuracy . numerical calculations of positron - atom bound states show that for atoms with ionization potentials greater than 6.8 ev and positron affinities in the range @xmath279@xmath280 a.u . ( zn , be , cu , and ag ) the annihilation rates are 2030% that of ps , see g. ryzhikh and j. mitroy , j. phys . b * 31 * , 5013 ( 1998 ) ; j. mitroy and g. ryzhikh , j. phys . b * 32 * , 1375 ( 1999 ) .

we investigate the predictability problem in dynamical systems with many degrees of freedom and a wide spectrum of temporal scales . in particular , we study the case of turbulence at high reynolds numbers by introducing a finite - size lyapunov exponent which measures the growth rate of finite - size perturbations . for sufficiently small perturbations this quantity coincides with the usual lyapunov exponent . when the perturbation is still small compared to large - scale fluctuations , but large compared to fluctuations at the smallest dynamically active scales , the finite - size lyapunov exponent is inversely proportional to the square of the perturbation size . our results are supported by numerical experiments on shell models . we find that intermittency corrections do not change the scaling law of predictability . we also discuss the relation between finite - size lyapunov exponent and information entropy .

chao-dyn9606014

the ability to predict has been the single most important qualifier of what constitutes scientific knowledge , all since the successes of babylonian and greek astronomy . indeed , the famous statement of laplace that an intelligent being with complete knowledge of the present and of the laws of nature will know the future for all time , assumes that the future is completely predicated by the past , and that perfect prediction would in principle be possible . in more mathematical terms one can say that in the physical sciences , whether in the classical or the quantum regime , one believes that nature is ultimately described by differential equations , and if one knows them and how to solve them , one knows all there is to know about the world@xcite . laplacian determinism is always conditioned by the fact that in the real world initial conditions can never be known to arbitrary accuracy . more recent is the general appreciation of the fact that in the presence of deterministic chaos , predictability is even more severely limited , because small errors typically grow exponentially in time@xcite . most sufficiently complex systems in the world display chaos . therefore , most sufficiently complex systems can only be predicted for a finite time . however , there may be some aspects of a system that are stable , while others vary . to take a familiar example , weather prediction is possible , typically for about ten days on temperate latitudes , but how the wind blows on the corner of the street is in practice unpredictable from one moment to the next@xcite . the words predictability and prediction are rather empty by themselves : one has to ask for predictability of what feature against what perturbation , in particular , against a perturbation of what size . in complex and spatially extended systems one can typically talk about large - scale features and small - scale features . the predictability of a small - scale feature against a small - scale perturbation is typically shorter than the predictability of a large - scale feature against a large - scale perturbation . certainly this is not always true , and we will consider a counter example , @xmath1 turbulence , below . but it seems to be a very common situation . that successful large - scale prediction essentially can mean just assuming that things do not change , is not _ per se _ an argument against prediction in the large . also in weather prediction the assumption that the weather tomorrow will be like today is a fairly good one . in this work , a further development of results presented in the a recent brief report @xcite , we will introduce a quantity which measures predictability in the large , and apply it to hydrodynamic turbulence . before we proceed to the definitions , let us first recall some facts about predictability in the small , i.e. the effects of dynamical chaos . a system is said to be chaotic if small i.e. infinitesimal perturbations grow exponentially in time . if the initial perturbation is of size @xmath2 , and the accepted error tolerance is @xmath3 , still small , then a rough estimate gives that the predictability time is @xmath4 where @xmath5 is the leading lyapunov exponent@xcite . already within the framework of infinitesimal perturbations there are important modifications to ( [ eq : lyapunov - predict])@xcite . in fact , in typical chaotic systems , ( [ eq : lyapunov - predict ] ) is not quite true@xcite . the exponent @xmath5 is a global quantity which measures the _ average _ exponential rate of separation of nearby trajectories , and fluctuations of the local exponential grow should be taken into account @xcite , but these effects are not what concerns us here . in this paper we shall address the problem of predictability in systems with many characteristic times , e.g. the case of fully developed turbulence where a hierarchy of different eddy - turnover times do exist , or when the threshold @xmath2 is not small . in these cases the predictability time @xmath6 is determined by the details of the nonlinear mechanism responsible for the growth of the error @xcite . in particular , @xmath6 may have no relation with the maximum lyapunov exponent governed by the linearized equations for the infinitesimal error . in general , in this case the predictability time strongly depends on the details of the system @xcite . according to oseledec theorem , the leading eigenvalue of the linearized equations of motion is @xmath7 , except on a set of points of measure zero . the sub - leading eigenvalues have the form @xmath8 , and , taken together , the leading and sub - leading lyapunov exponents measure the growth rate of @xmath9-dimensional volumes spanned by @xmath9 infinitesimal vectors , where @xmath9 can range from one to the dimensionality of the space where the motion takes place . in dynamical systems , in addition to lyapunov exponents , an important dynamical characterization is given by the kolmogorov - sinai entropy , which measures the bandwidth necessary to observe a system over time , so that it could later be faithfully reproduced from the observations@xcite . arguing heuristically , new observations are necessary if an error grows in time , and the necessary rate of accumulation of information is the growth rate of the error , and therefore one expects pesin theorem , @xmath10 to hold true for a large class of systems . it was realized by shannon that with finite error tolerance , the relevant quantity is the bandwidth necessary to observe a system such that it could later by reproduced within this error , not to arbitrarily high accuracy . this quantity was called by `` source entropy with respect to a fidelity criterion '' by shannon@xcite , and `` @xmath11-entropy '' by kolmogorov@xcite , who found analytic formulae , valid for gaussian variables and gaussian stationary processes . recently the concept of @xmath11-entropy was taken up by gaspard and wang , who computed it from experimental data in thermal turbulence@xcite , and for a very large variety of model problems in stochastic processes and statistical physics@xcite . we especially recommend their lucid and remarkably complete review@xcite . however , the @xmath11-entropy does not say all one wants to know about predictability in the large . just as the lyapunov exponent is often more relevant , and more easily computable , than the kolmogorov - sinai entropy , so the predictability time with respect to a finite perturbation should be determined by a quantity analogous to the lyapunov exponent , and not by the @xmath11-entropy . the natural starting point in looking for such a quantity is the time it takes for a perturbation to grow from an initial size @xmath2 to a tolerance @xmath3 . we call this the @xmath12 predictability time and denote it by @xmath13 . generally speaking , the predictability time will fluctuate . the natural definition of the finite - size lyapunov exponent is therefore an average of some function of the predictability time , such that if both @xmath2 and @xmath3 are in the infinitesimal range , we will recover the usual lyapunov exponent , and an obvious choice is then @xmath14 in appendix [ app : a ] we discuss other possible definitions for the finite - size lyapunov exponent and the relation with the @xmath11-entropy . in contrast to infinitesimal perturbations , for finite perturbations the threshold @xmath3 is typically not to be taken much larger than the perturbation @xmath2 . what is interesting , and what makes finite - size lyapunov exponents different from lyapunov exponents for infinitesimal perturbations , is the dependence on @xmath2 . this paper is organized as follows : in section [ s : multifractal ] we recall the multifractal approach to turbulence , and lorenz approach to the predictability problem within the kolmogorov theory . we show that there are no multifractal corrections to the results of lorenz , but that the scaling range for the finite - size lyapunov exponents is shorter . in section [ s : shell - model ] we describe numerical experiments on predictability in shell models for @xmath0 and @xmath1 fully developed turbulence . in section [ s : edqnm ] we present the results from the eddy damped quasi normal markovian ( edqnm ) approximation for the shell model , and compare them with the results of section [ s : shell - model ] . in section [ s : conclusion ] we summarize our results and present conclusions . in appendix [ app : a ] we discuss alternative ways of defining the finite - size lyapunov exponent . appendix [ s : kolmogorov ] contains a derivation of the kolmogorov results of @xmath11-entropy for gaussian processes and gaussian random fields . in appendix [ s : spacetime ] we apply the results of appendix [ s : kolmogorov ] to space - time gaussian fields with spectra as in kolmogorov 1941 theory of @xmath0 turbulence , and to fictitious @xmath15-dimensional fields that describe shell models . we show that the finite - size lyapunov exponent is the the same for any dimensionality , but the @xmath11-entropy largely depends on the dimensionality - dependent density of degrees of freedom . the results of sections [ s : multifractal ] and [ s : shell - model ] were also presented in less generality in @xcite . our understanding of high reynolds - number turbulence is still mainly based on the fundamental contribution of kolmogorov in 1941@xcite . we will here just discuss on a phenomenological level the predictions of the kolmogorov theory and of its multifractal generalizations @xcite . turbulence is a statistically stationary state of matter on macroscopic scales maintained by external forces . one considers only effects that are captured in a hydrodynamic level of description , that is the time evolution is supposed to be completely described by the macroscopic navier - stokes equations@xcite @xmath16 from the typical length scale @xmath17 , the typical fluctuations of velocity on that scale @xmath18 , and the viscosity @xmath19 , we can form the reynolds number , @xmath20 which characterizes the flow . the multifractal model @xcite consists in assuming that at scales much less than @xmath17 , however sufficiently large that the action of viscosity is weak , the velocity differences assume a scaling form @xmath21 different values of @xmath22 are assumed to occur according to a probability distribution , which also takes a scaling form @xmath23 \right\ } \sim \left(\frac{l}{l}\right)^{3-d(h ) } dh . \label{eq : probability - ansatz}\ ] ] the function @xmath24 is the fractal dimension of the subset with scaling exponent @xmath22 . the moments of the velocity differences on length scale @xmath25 can be computed as @xmath26 and , for small-@xmath25 ( i.e. in the inertial range ) , the integral in ( [ eq : moment - q ] ) can be evaluated by saddle point method : @xmath27 . \label{eq : zeta - q}\end{aligned}\ ] ] the model is physically reasonable for a large set of possible choices of the function @xmath24 , but not entirely arbitrary . by normalization , the value of @xmath24 must always be less or equal than three , and the maximum must be obtained for some @xmath22 . that is @xmath28 the function @xmath24 can have support only at positive @xmath22 , because a negative value of @xmath22 implies that velocity fluctuations in a local inertial frame of size @xmath25 increase without limit as @xmath25 tends to zero . the navier - stokes equations are derived under the assumption that all velocities are much smaller than the velocity of sound , and this condition would then no longer hold @xcite . furthermore , an exact result of kolmogorov assures that @xmath29 , so that @xmath30 where equality is obtained for at least one value of @xmath22@xcite . the inequality ( [ eq : zeta-3-inequality ] ) is the analogous for turbulence of the inequality @xmath31 for multifractal measures @xcite . in terms of the multifractal model , the kolmogorov theory is formulated by supposing that the function @xmath24 has support only at a single point . from ( [ eq : zeta-0-inequality ] ) and ( [ eq : zeta-3-inequality ] ) it then follows that this point must be @xmath32 , and that @xmath33 . it further follows the kolmogorov law @xmath34 energy dissipation per unit mass and time , @xmath11 , has dimension @xmath35 . we could therefore also write the right - hand side in ( [ eq : kolmogorov-41 ] ) in the more familiar form @xmath36 from ( [ eq : kolmogorov-41 ] ) it follows by balancing in ( [ eq : navier - stokesa ] ) , that viscous forces become comparable with inertial forces at the kolmogorov scale @xmath37 which marks the lower end of the inertial range : @xmath38 if there exists more than one value of @xmath22 then each @xmath22 selects a different damping scale @xmath39 . by using ( [ eq : scaling - ansatz ] ) and balancing , one gets @xcite @xmath40 lorenz investigated the predictability problem within the framework of the kolmogorov theory@xcite . let us assume that we have a disturbance on scale @xmath25 , and that it grows at a characteristic rate given by the turn - over time at this scale : @xmath41 we can turn around ( [ eq : turn - over - time ] ) and say that after a time @xmath42 a disturbance will have grown large on all scales smaller than @xmath43 the size of the disturbance will then be @xmath44 , since all the smaller scales contribute relatively little . if we call the difference between two fields @xmath2 , we can rewrite ( [ eq : turn - over - time ] ) as the predictability time with respect to a perturbation of size @xmath2 : @xmath45 in other words , the predictability time of a perturbation of size @xmath2 grows as @xmath46 in lorenz scenario . the finite - size lyapunov exponent thus decreases with the error threshold as @xmath47 . finally , we can insert @xmath2 in ( [ eq : decorrelated - scale ] ) , and find how the error grows with time : @xmath48 the upshot of these simple estimates is that finite error growth and predictability in high reynolds number turbulence are characterized by _ algebraic _ laws , very different form the _ exponential _ laws characteristic of infinitesimal perturbations in chaotic dynamical systems . we now turn to possible consequences of a spectrum of @xmath22 s to the predictability problem , in direct analogy with the lorenz theory . we start by assuming the inverse of ( [ eq : probability - ansatz ] ) that is @xmath49 \right\ } \sim \left(\frac{\delta}{v}\right)^{\frac{3-d(h)}{h}}dh , \label{eq : probability - ansatz - inverse}\ ] ] where we have identified @xmath50 with @xmath2 , and used ( [ eq : scaling - ansatz ] ) to relate @xmath25 and @xmath2 . the length @xmath25 is now assumed a fluctuating quantity , and has the interpretation of a scale such that two fields are uncorrelated on all smaller scales , given that the distance of the two configurations is @xmath2 . the finite - size lyapunov exponent is , according to ( [ eq : predict ] ) , proportional to the expectation value of the inverse predictability time : @xmath51 in the small error limit we thus expect the finite - size lyapunov exponent to scale as a power of the error size : @xmath52 . \label{eq : chi}\end{aligned}\ ] ] the exponent @xmath53 is always equal to the lorenz value of @xmath54 . this follows easily from ( [ eq : zeta-3-inequality ] ) , which can be rewritten in the form @xmath55 where the equality holds for the exponent @xmath56 which dominates the third order structure function . as far as we know this result is new . one could therefore conclude that the exponent @xmath53 for the scaling of the finite - size lyapunov exponent with error threshold is a new invariant of the multifractal approach to turbulence , and that the law @xmath57 can be easily observed in numerical experiments . this is not quite simple , due to the influence of the fluctuating cut - off ( [ eq : frisch - vergassola - scale ] ) . the smallest fluctuation in a field scaling with exponent @xmath58 is @xmath59 which inversely determines the smallest value of @xmath22 contributing to a fluctuation of size @xmath2 . a modified version of ( [ eq : multifractal - predict ] ) therefore reads @xmath60 the integral is dominated by @xmath56 as long as @xmath2 is much larger than @xmath61 . for smaller @xmath2 values , the integral is dominated by the lower end - point in ( [ eq : multifractal - predict - intermediate ] ) , which leads to an intermediate dissipative range , in the sense of frisch and vergassola@xcite . as a consequence , we have @xmath62 with @xmath63 where @xmath58 and @xmath2 are related via ( [ eq : fluctuation - scale ] ) . from ( [ eq : zeta-3-inequality ] ) follows that @xmath56 is less or equal to @xmath64 in the multifractal approach , and therefore the bottom of the scaling range of the finite - size lyapunov exponent is larger than the corresponding error size in the kolmogorov theory . the scaling range for the finite - size lyapunov exponent thus is generally shorter in a multifractal model . simplified dynamical models of fluid turbulence with relatively few degrees of freedom , collectively referred to as _ shell models _ , have been studied since the seventies . these models by construction typically include a richardson cascade of energy from large to small scales . some of the models are dynamically stable , with a fixed point which reproduces the kolmogorov law for the energy spectrum , @xmath65 . an historical overview , with references to much of the early work , can be found in a recent monograph @xcite . more interestingly , other models are dynamically unstable , with chaotic motion taking place on a strange attractor where the kolmogorov @xmath66 law holds to good accuracy , but not exactly . one of the simplest example is the family of models introduced by gledzer@xcite , and yamada and ohkitani @xcite , now commonly called the goy models@xcite . they have recently been the subject of several investigations @xcite . an in - depth description of this work can also be found in @xcite . the goy models are defined as follows : fourier space is divided into @xmath67 shells , labeled by the wave - vector modulus @xmath68 , where @xmath69 is a constant . the velocity difference over a length scale @xmath70 are represented , each by one complex variable @xmath71 , which obey the following system of coupled ordinary differential equations : @xmath72 where @xmath73 is the strength of the external force , acting on large scales , and @xmath19 the viscosity . for any values of the three coefficients @xmath74 , @xmath75 and @xmath76 , phase space volume in preserved in the force - free inviscid limit . the restricted number of degrees of freedom is both the main advantage and disadvantage of shell models . it is an advantage , because it allows simulations at much lower viscosity and for much longer time than in the full navier - stokes equations . but it is also a severe departure . all spatial structure of the field is ignored . one of the coefficients in ( [ eq : sh3 ] ) , say @xmath74 , can be scaled to one and the condition of energy conservation fixes one more , such that , in terms of one parameter @xmath11 , @xmath75 is equal to @xmath77 and @xmath76 to @xmath78 . with @xmath11 greater than one the goy equations conserve one more positive definite quantity besides energy , i.e. an analogous situation to @xmath1 hydrodynamics . the dynamical behavior of the goy models in this range is rather far removed from @xmath1 turbulence as was shown in recent papers@xcite . we include below a study of predictability in such shell models , but just as a simplified model to demonstrate one possible scaling behavior of the finite - size lyapunov exponent . with parameter @xmath11 in the range between and zero and one , the goy models also preserve another invariant , but which is not positive definite . in the following we will look at the goy model with the standard choice of @xmath11 equal to @xmath79 , i.e. @xmath80 the second invariant then has physical dimension of helicity@xcite . presumably that is the reason why this particular model has turned out to be so close to numerical and experimental data on navier - stokes turbulence in @xmath0 @xcite . before we turn to the numerical experiments , let us summarize some salient features of the system defined by ( [ eq : sh1])-([eq : goy_standard ] ) . energy is pumped into the system by the force , which acts only on shells with low values of @xmath81 , and is removed at high shells by viscosity . from @xmath69 , @xmath19 and the typical fluctuations of velocity on large scales , @xmath18 , we can form a reynolds number @xmath82 . we consider the situation where @xmath83 is large , such that there is a wide range in @xmath81 where the external and viscous forces are both negligible compared to the inertial forces . in this inertial range , we have @xmath84 , where the exponent @xmath85 is close to @xmath86@xcite . an estimate of the smallest excited scale is in analogy with the kolmogorov scale @xmath87 . the slowest dynamical scale is the time - scale of the the shells containing most energy , about @xmath88 , and the fastest is @xmath89 , or , about @xmath90 . from the fastest time - scale follows by dimensional analysis in the kolmogorov theory that the leading lyapunov exponent should grow with reynolds number as @xmath91 , a prediction due to ruelle@xcite . in the multifractal picture there are corrections to this estimate and that the leading lyapunov exponent of the goy shell model scales as @xmath92 where @xmath93.\ ] ] it is indeed numerically observed to scale as @xmath94 , in good agreement with a computation of @xmath95 staring from a function @xmath24 , obtained by a parametric fit of measured values of the scaling exponents @xmath96 in experiments@xcite . the mean square fluctuations at the kolmogorov scale are @xmath97 . if we compute the distance between two shell variable configurations as @xmath98 an error smaller than @xmath99 is relatively small all over the inertial range . it can be taken to be infinitesimal , and its growth rate will be the fastest linear growth rate . if , however , the error is larger than @xmath99 , it could be larger than the typical size of the fluctuations at the kolmogorov scale . such an error would have to be concentrated on larger scales , since otherwise we have that for some @xmath81 @xmath71 and/or @xmath100 is much larger than the typical size . in other words , a physical perturbation larger than @xmath99 can not be obtained by a random perturbation of that size uniformly distributed over all the shells . in the appendix [ app : a ] we discuss some possible definitions for the finite lyapunov - exponents . in what follows we adopt the following procedure : after a long integration time to let the system relax towards the statistically stationary state , we introduce a very small error . this is done by generating a new shell variable configuration @xmath101 differing from @xmath71 by a small fraction of @xmath102 . another possible approach would be , that if we want an initial error of size @xmath11 , we determine a shell @xmath103 such that @xmath104 is about @xmath11 , and we concentrate the perturbation on shells above @xmath103 . we then iterate @xmath71 and @xmath101 ( perturbed system ) for again a long time , such that the error has grown to a threshold , which is still small compared to @xmath105 . we thus have two realizations of configurations in the statistically stationary state , which only differ by a small error , which we call @xmath106 . further we define a series of thresholds @xmath107 , and we measure the times it takes for the error to grow from @xmath106 to @xmath108 , and so on . for brevity we will call these times error doubling times , even if @xmath109 can be different from two . the threshold rate @xmath109 should not be taken too large , because then the error has to grow through several different scales before reaching the next threshold . on the other hand , the rate @xmath109 can not be too close to one , so a sensible threshold rate is on the order of two . the most convenient choice of @xmath109 clearly depends on the how the fluctuations in the shell variable depend on @xmath81 , and in what way we measure the error . for our model ( [ eq : sh3 ] ) with error measured by ( [ eq : error_def ] ) , the range of error sizes in the inertial range is not large , scaling as @xmath110 . for 35 shells , which is the largest systems we simulate , we can take @xmath83 equal to @xmath111 , which gives an error range of about @xmath112 . with error threshold rate equal to two , that would give about @xmath113 data points , with some points lost on both ends due to boundary effects . for practical reasons we therefore take @xmath109 equal to @xmath114 . when we have performed @xmath115 error doubling experiments , we can form an estimate of the expectation value of some quantity @xmath116 : @xmath117 this is not the same as taking a time - average , since different error doubling experiments may take different times . indeed , we have @xmath118 a particular case of the above relation concerns the mean error doubling times themselves . let @xmath119 be the time it takes for an error to grow from threshold @xmath120 to @xmath121 . then @xmath122 where we have used the definition of ( [ eq : predict ] ) . the finite - size lyapunov exponents , @xmath123 , can be compared with shell turn - over times as follows : we first select a shell @xmath124 such that @xmath125 is about @xmath2 , and then estimate @xmath126 as @xmath127 , which scales as @xmath47 . this argument for typical error growth times is the same as lorenz argument for @xmath0 turbulence@xcite , discussed above in section [ s : multifractal ] . in fig.[f : doubling - times ] we compare error doubling times and shell turn - over times , as a function of size of the perturbation and of the typical fluctuations in the corresponding shell . below the kolmogorov scale , the turn - over times increase : we are here in the dissipation range , where the shell amplitudes decrease quickly . on the other hand , the doubling times tend to a constant as the error threshold is small . we are here in the infinitesimal range , and the constant is approximatively the inverse of the lyapunov exponent . at the kolmogorov scale , there is rather large discrepancy between the lyapunov exponent and the turn - over time . this observation , that the lyapunov exponent obeys a scaling law with a sizeable numerical pre - factor , has been made before@xcite , but without a plausible explanation . we here find nice agreement with our prediction that the inertial range for the finite - size lyapunov exponent is shorter than the spectral inertial range , because the first is limited from below by the scaling exponent @xmath56 , as in equation ( [ eq : fluctuation - scale ] ) , while the second is limited from below by the scaling exponent @xmath128@xcite . in fig . [ f : re - scaling ] we compare the error doubling times for different reynolds numbers . for small thresholds the doubling times scale as the lyapunov exponent , i.e. as @xmath129 . we also observe that the bend away from the infinitesimal growth rate occurs at smaller error scales for larger reynolds numbers . this suggests that a simple scaling ansatz can be sought in the following form : times and errors are scaled with the turn - over time and the typical scale of fluctuations at the kolmogorov scale , that is by @xmath129 and @xmath110 , respectively . in fig . [ f : re - data - collapse ] we show such re - scaled data . the data collapse is reasonable . to improve the data collapse , taking into account multifractal corrections , we made a scaling based on multiscaling @xcite , i.e. of the form @xmath130 where @xmath131 , @xmath132 are parameter to be fixed , and @xmath133 is the scaling function . according the argument at the end of section [ s : multifractal ] , we have @xmath134 for large @xmath135 , while @xmath133 is constant for small values of @xmath135 . in the intermediate regime @xmath133 has a nontrivial form which depends on the shape of @xmath24 , as follows from eqs . ( [ eq : multi - verg])-([eq : chi1 ] ) . the result is shown in fig . [ fig : multiscaling ] . the data collapse is clearly improved . we conclude this section discussing the case of @xmath1 turbulence . two dimensional euler equation has the peculiar property of an infinite number of invariants . two of them are retained in a finite fourier discretization , the energy and the average square vorticity , or enstrophy . as we previously discussed , the second conserved quantity in the goy shell model depends on the choice of the parameter @xmath11 . with the choice @xmath136 , leading to @xmath137 equations ( [ eq : sh1 ] ) conserve in the unforced and inviscid limit , in addition to the energy , the enstrophy here defined as @xmath138 despite the fact that the two - dimensional shell model has superficially the same physical justification of its three - dimensional corresponding model , it has been demonstrated that it has little to do with turbulence @xcite . moreover , all the numerical simulations of two - dimensional navier - stokes equation at sufficiently high reynolds number have demonstrated the dynamical relevance of coherent structures which emerge spontaneously from the turbulent flow . the predictability problem , which is more relevant for geophysical flows in this case than in @xmath0 turbulence , is also ruled by coherent vortex motion in the physical space , rather than modes dynamics in fourier space @xcite . with this limitations , the study of the predictability problem , as addressed in the present paper , in two - dimensional shell model is nevertheless interesting because of the different scaling behavior with respect to the @xmath0 situation . dimensional analysis @xcite shows that in the enstrophy cascade one expects constant i.e. independent on the scale turn - over times . hence an argument similar to that of section [ s : multifractal ] shows that @xmath139 where @xmath5 is the largest lyapunov exponent . the predictability time for two - dimensional shell model is thus determined by a single value @xmath140 , where @xmath141 is the enstrophy flux toward small scales , up to a perturbation of the order of the large scale velocity field , where saturation effects are dominant . figure [ fig:2d ] shows the finite - size lyapunov exponent @xmath142 for a simulation with @xmath143 shells . the forcing term is now @xmath144 and @xmath145 . because of the inverse energy cascade we introduce in the equation ( [ eq : sh1 ] ) an artificial large - scale dissipation @xmath146 ( @xmath147 ) in the first shells @xmath148 @xcite . the velocity field shows a pseudo - cascade power law , see @xcite , @xmath149 in a wide range @xmath150 . the mean enstrophy flux in this range is @xmath151 which is in agreement with the dimensional evaluation of the eddy turnover time @xmath152 . the data plotted in figure [ fig:2d ] are obtained by using the method described in appendix [ app : a ] for computing the size dependent lyapunov exponent . the same results , not reported , can be obtained from the doubling time algorithm . we stress once more that , in the light of the results discussed in @xcite , two - dimensional shell model is not a good model for two - dimensional turbulence . as a consequence a discussion of the predictability problem for two - dimensional turbulence requires the direct study of navier - stokes equations @xcite . preliminary results show that this scenario remains , nevertheless , valid in the direct cascade @xcite . in this section we describe the results obtained from the eddy damped quasi - normal markovian approximation ( edqnm ) for the shell model . the basic idea of closure approximations is quite simple : write down the reynolds hierarchy for moments of the shell variables and truncate the chain to the lowest sensible order . the important point is that in the closure approximation intermittent effects are washed out , so we can directly test if the relevant mechanism is due to the existence of many characteristic times . we do not report the derivation of the edqnm equations for the shell model . the interested reader can find it in ref . @xcite . we consider two independent realizations of the shell model field , @xmath71 and @xmath153 , with the same energy spectrum @xmath154 , and both evolving according the shell model equations ( [ eq : sh1])-([eq : goy_standard ] ) . the distance between the two fields can be obtained from , cfr . ( [ eq : error_def ] ) , the energy difference at shell @xmath81 : @xmath155 where @xmath156 , and @xmath157 denotes the real part . . from the definition it follows @xmath158^{1/2}.\ ] ] the evolution equations of @xmath159 and @xmath160 in the edqnm approximation read : @xmath161 \\ & \,+ 2\,\epsilon\,\delta_{n,4 } , \end{array}\ ] ] and @xmath162 \\ & \,+ 2\,\epsilon\,\delta_{n,4 } , \end{array}\ ] ] where @xmath163\,t } } { \displaystyle \nu ( k_n^2 + k_{n+1}^2 + k_{n+2}^2 ) + \mu_n + \mu_{n+1 } + \mu_{n+2 } } , \label{eq : theta}\ ] ] and @xmath164 we have one free parameter , the dimensionless constant @xmath95 . it should be adjusted such that the spectrum is as similar as possible to the spectrum obtained in simulations of the full equation . the energy spectrum of the shell model in the edqnm approximation must therefore obey @xmath165 in the inertial range . the undetermined function @xmath166 is the kolmogorov constant . on the other hand it has become clear in several independent investigations that intermittency corrections exist in shell models . the energy spectrum is therefore in reality more closely described by @xmath167 , where the exponent @xmath85 has been estimated to be @xmath168 @xcite . the function @xmath169 that gives the prefactor to the power law in the inertial range should not depend on viscosity , but depends on the forcing through @xmath11 , the mean dissipation of energy per unit time , or , equivalently , the mean energy input into the system from the force . in a really large inertial range the two power - laws are not good approximations to one other . the best that can be done is to demand that the spectra agree as closely as possible at the upper end of the inertial range . a reasonable agreement is obtained for @xmath170 , leading to @xmath171 which is the value observed both in simulations of the shell model and in experiments @xcite . the procedure described in the previous section to compute the scale dependent lyapunov exponent for the shell model can be adopted here for the closure equations . in practice after a long iteration time , to have a well stabilized energy spectrum @xmath159 , we take a small initial distance @xmath172 and perform the doubling experiment similar to those of the previous section iterating equations ( [ eq : wnedqnm ] ) . in fig . [ fig : edqnmfig1 ] we show @xmath173 as a function of the rescaled distance @xmath174 , for different @xmath175 . the other parameters are @xmath176 shells , @xmath177 , integration step @xmath178 and @xmath179 . from fig . [ fig : edqnmfig1 ] we see that the closure approximation leads to the same scenario observed for the shell model , confirming that this is due to the existence of many characteristic scales . we note that the slope of the curve for @xmath180 is the lorenz value @xmath54 since in the edqnm approximation there are not intermittent corrections . we note that in this case , since there is not intermittency , the effective inertial range roughly coincides with the inertial range . in fig . [ fig : edqnmfig2 ] we compare @xmath181 and the inverse of the turnover time @xmath182 as a function of the distance @xmath183 . in the figure we used @xmath184 . we have introduced a generalization @xmath185 of the leading lyapunov exponent to finite perturbations of size @xmath2 . unlike the lyapunov exponent , @xmath185 contains direct information on the predictability time for an extended chaotic system and it is particularly useful in presence of several characteristic times . moreover , it is computationally no more expensive than the standard lyapunov exponent . in the limit of infinitesimal perturbation @xmath2 , the finite - size lyapunov exponent gives the leading lyapunov exponent . the way in which this limit is reached depends on the details of the particular system and gives informations about the characteristic time scales . in the case of @xmath0 fully developed turbulence we have found an universal scaling law @xmath186 where the value @xmath54 of the exponent is an invariant of the multifractal approach . the scaling law is confirmed by extensive numerical simulations on shell model for turbulence at very high reynolds numbers , but we warn that it would be very difficult to observe such a scaling in experimental data because of the reduction of the scaling range in presence of intermittency . we have also shown and discussed the relationship between our approach and the @xmath11-entropy results in literature . in conclusion , we think that it would be useful to compute the finite - scale lyapunov exponent in other complex dynamical systems , as a new tool for investigating the presence of characteristic times and the predictability properties . this work was supported by the swedish natural science research council through grant s - fo-1778 - 302 ( e.a . ) , by infn ( _ iniziativa specifica meccanica statistica fi11 _ ) and by cnr ( progetto coordinato _ predicibilit in turbolenza geofisica _ ) . thanks dipartimento di fisica , universit di roma `` la sapienza '' for hospitality . g. b. thanks the `` istituto di cosmogeofisica del cnr '' , torino , for hospitality . here we discuss some alternative ways of computing the finite - size lyapunov exponent beyond the definition ( [ eq : predict ] ) . the first method is a modification of the standard technique @xcite . we integrate two trajectories @xmath187 and @xmath188 with initial euclidean distance @xmath189 until their separation becomes larger , at a given time @xmath190 than @xmath191 where @xmath109 is a given constant coefficient . the perturbed trajectory @xmath192 is then rescaled at the original distance @xmath2 , keeping the direction @xmath193 constant . the possible problem with this definition is that it assumes that the statistically stationary state of the system is homogeneous with respect to perturbations of finite size . one may plausibly argue that the structure of the attractor in phase space on which the motion takes place may be fractal , and not at all equally dense at all distances from a given point . the procedure outlined would then not necessarily sample in a faithful way the motion on the attractor . in practice , as it is showed by the numerical experiments , we do not find any difference with the numerical method presented in the main text , so the effect must be quite small . for the usual lyapunov exponent the problem does not exist since we only use small finite perturbations as an approximation to infinitesimal perturbations . the finite - size lyapunov exponent at scale @xmath2 is obtained by averaging the divergence rate @xmath194 along the unperturbed trajectory @xmath187 . the average @xmath195 is over many error - doubling experiments . in the case of infinitesimal error @xmath2 , and not too large factor @xmath109 , this definition leads to the maximal lyapunov exponent @xmath5 . for maps , the above methods have to be slightly modified since the distance between two states is not continuous in time . from eq . ( [ eq : finite - lyapunov ] ) readily follows @xmath196 where @xmath190 is the minimum time such that the distance between two realizations is larger or equal to @xmath197 and @xmath198 is the distance at that time . definition ( [ eq : alterna1 ] ) is valid for any value of @xmath109 , which however is generally assumed not too large to avoid the interference of different scales . in particular , one could think of removing the threshold condition used for defining @xmath190 and simply compute the average error growth rate at every time step . in other words , at every time step @xmath199 in the integration , the perturbed trajectory @xmath188 is rescaled to the original distance @xmath2 , keeping the direction @xmath200 constant . the finite - size lyapunov exponent is given by the average of the one - step exponential divergence : @xmath201 which is equivalent to the above definition ( [ eq : alterna1 ] ) . this method is indeed the one used for computing @xmath185 in the case of @xmath1 shell model . the one - step method ( [ eq : app3n ] ) has the advantage that it can be easily generalized to compute the sub - leading finite - size lyapunov exponent following the standard orthonormalization method @xcite . one introduces @xmath202 perturbed trajectories @xmath203 each at distance @xmath2 from @xmath204 and such that @xmath205 are orthogonal each to the others . at every time step , any difference @xmath205 is rescaled at the original value and orthogonalized , while the corresponding finite size lyapunov exponent is accumulated according to ( [ eq : app3n ] ) . here we have again the problem of the implicitly assumed homogeneity of the attractor , but also a problem of isotropy when we re - orthogonalize the perturbations . we note that this could be a more serious problem , which will not be discussed here any further . for systems with only one positive lyapunov exponent the size dependent lyapunov exponent @xmath185 , definition ( [ eq : app2 ] ) , or ( [ eq : predict ] ) , coincides with the @xmath11-entropy widely described below in appendix [ s : kolmogorov ] and by gaspard and wang @xcite . we consider now few examples . consider a chaotic deterministic map perturbed by a gaussian term : @xmath206 where @xmath207 is a stochastic variable with gaussian distribution of zero mean and unit variance . the @xmath207 for different times @xmath81 are independent . when @xmath208 the map @xmath133 is chaotic with positive lyapunov exponent @xmath209 . a simple computation shows that @xmath210 for @xmath211 the noise term is negligible , so that @xmath212 . in the opposite limit , @xmath213 , the first term in ( [ eq : app4 ] ) can be neglected so that @xmath214 . thus in one iteration of the map the distance between the two trajectories grows to @xmath215 , larger than the tolerance . from ( [ eq : app2 ] ) we have @xmath216 . these are the same result obtained for the @xmath11-entropy , see sect . 3.5 of ref . @xcite . consider the gaussian process described by the langevin equation @xmath217 where @xmath218 and @xmath37 is a white noise with zero mean and correlation @xmath219 . the formal solution of ( [ eq : app5 ] ) reads @xmath220 so that the distance between two different process behaves in time as @xmath221 this implies that @xmath222 , and thus the predictability time @xmath223 behaves as @xmath224 so that @xmath225 , as found for the @xmath11-entropy , see below or sect . 3.6.2 of @xcite . similar computations can be done for the yaglom noise , see sect . 3.6.3 of @xcite . the one - dimensional map @xmath226 presents deterministic diffusion . for example for @xmath227 the diffusion coefficient is @xmath228 . the size dependent lyapunov exponent for this map can be computed numerically using the definition ( [ eq : app2 ] ) . from fig . [ fig : det - diff ] we see that @xmath229 for small @xmath2 , while @xmath230 for large values of @xmath2 . the @xmath11-entropy shows the same behavior , see fig . 25.b of ref . @xcite . note that in ref . @xcite the @xmath11-entropy is measured in unit of digit / iteration , so there is a multiplicative factor @xmath231 with @xmath185 of fig . [ fig : det - diff ] . the purpose of this appendix is to derive systematically on a physical level of rigor the results of kolmogorov on the @xmath11-entropy of gaussian variables and gaussian processes and random fields . these results were stated without proof by kolmogorov @xcite , and in the review by tikhomirov @xcite . the published proofs we are aware of are either not easily accessible , or carried out in such generality that the simple underlying idea may be lost to the less mathematically inclined reader . hence the interest of including a simple derivation here . let us just for completeness refer to comparatively recent paper @xcite . the general setting is that of two random variables @xmath232 and @xmath37 , taking values in spaces @xmath233 and @xmath234 . the values in two realizations of @xmath232 and @xmath37 are denoted @xmath135 and @xmath235 , respectively , where @xmath135 lies in @xmath233 and @xmath235 lies in @xmath234 . when @xmath233 and @xmath234 are continuous the probability distributions of @xmath232 and @xmath37 will be denoted @xmath236 and @xmath237 . except for the trivial case , the variables @xmath232 and @xmath37 are not independent , but characterized by the joint probability distribution @xmath238 . the single - variable probabilities are determined from the joint distribution as @xmath239 the conditional probabilities are @xmath240 we suppose in this subsection that the spaces @xmath233 and @xmath234 are discrete , and consist of finitely many points , @xmath241 and @xmath242 , and associated probabilities @xmath243 and @xmath244 . the _ entropy _ of the random variable @xmath232 is @xmath245 if we want to code a message of @xmath115 letters , which are given by @xmath115 consecutive independent realizations of @xmath232 , so taking values in @xmath246 , this can be done , in the limit when @xmath115 is large , by using @xmath247 bits per letter @xcite . the pair @xmath232 and @xmath37 specify both a source signal and the output after sending the signal through a channel of transmission . the joint probability distribution @xmath248 can be decomposed either to the pair @xmath232 ( an input ) and @xmath249 ( an output , given the input ) , or to the pair @xmath37 and @xmath250 . in the discrete case these second conditional probabilities are denoted @xmath251 , where @xmath252 ranges from one to @xmath81 , and @xmath253 ranges from one to @xmath254 . the _ equivocation _ is @xmath255 shannon @xcite considered a _ experiment where we send an error - correcting message parallel to the transmission of @xmath232 to @xmath37 , and asked for the number of bits in the error - correcting message needed to transmit @xmath115 letters in the original message virtually without error . in the limit of large @xmath115 , the answer is @xmath256 bits per letter . suppose further that we have some source of information which we can recode into letters from @xmath232 , then transmit to @xmath37 , and then observe . the rate of transmission of information is not the source entropy , since we must correct for the equivocation , but the _ mutual information _ : @xmath257 if we introduce the discrete joint probabilities @xmath258 and rearrange terms , we see that ( [ eq : mutual - information ] ) can be rewritten as in a more symmetrical way , namely : @xmath259 a channel of transmission of information can be considered as a collection of pairs @xmath232 and @xmath37 , where the @xmath232 s are the possible inputs , and the @xmath37 s the corresponding outputs . the _ capacity _ of the channel is the maximum rate of information transfer : @xmath260 where the maximization is performed over the collection of pairs @xmath232 and @xmath37 that describe the channel . the fundamental shannon theorem says that a source signal can be transmitted over a channel , if the source entropy is less than the channel capacity . suppose finally that we have an input @xmath232 and we wish that the output @xmath37 contains only some partial information about @xmath232 , saying that some fidelity criterion is fulfilled . we consider all channels consisting of the fixed source @xmath232 and different outputs @xmath37 . the least rate of information transfer needed to specify @xmath232 in this way is _ source entropy with respect to the fidelity criterion _ : @xmath261 where the minimization is performed over all @xmath37 which satisfy the criterion with @xmath232 . in practice many fidelity criteria can be written @xmath262 where @xmath263 is some suitable function and @xmath11 is a measure of the fidelity . in this case the source entropy is naturally considered as a function of @xmath11 , and we have the kolmogorov @xmath11-entropy @xmath264 suppose we have a random variable with a continuous distribution , that is @xmath265 with continuous density @xmath266 . then ( [ eq : entropy ] ) is infinite . as stressed by kolmogorov , the mutual information , ( [ eq : mutual - information ] ) , the capacity , ( [ eq : capacity ] ) , and , in particular , the rate of information production with respect to a fidelity criterion , ( [ eq : source - entropy - shannon],[eq : source - entropy - kolmogorov ] ) , are all well - defined also in the continuous case . in other words , although a real number observed with infinite accuracy contains an infinite amount of information , a real number observed with finite accuracy contains only a finite amount of information . we can see this in a simple heuristic way as follows : if we introduce a discretization of the space @xmath233 into boxes with diameter @xmath2 , we have a new random variable that we can call @xmath267 with entropy @xmath268 where @xmath9 is the dimension of @xmath233 . similarly , from the conditional variable @xmath250 we have a new random variable @xmath269 with equivocation @xmath270 rearranging terms as in ( [ eq : mutual - information - symmetrized ] ) , the mutual information between @xmath267 and @xmath37 is thus @xmath271 as the discretization tends to zero the mutual information tends to a finite value , provided that the joint probability @xmath238 is not singular with respect to the product @xmath272 , a result due to yaglom and gelfand@xcite . we can give a meaning to the most random continuous distribution with respect some given constraints by maximizing the entropy of a discretized continuous variable , as in ( [ eq : entropy - discretized ] ) , subject to these constraints , and then letting the discretization tend to zero . the most random distribution with given first and second moments is thus , not surprisingly , a gaussian distribution with the same first and second moments . the entropy of a discretization of a @xmath9-dimensional real gaussian random variable @xmath232 with correlation matrix @xmath273 is @xmath274 all these results on the gaussian distributions are due to shannon@xcite . the spaces @xmath233 and @xmath234 are now identical . we consider first a one - dimensional gaussian random variable , and then a finite - dimensional variable and show that the kolmogorov formula holds in these cases . in the next subsection , we then observe that the fourier components of gaussian processes and gaussian random fields are independent gaussian random variables , and so , taking the the formal infinite - dimensional limit , we find the desired result . the fidelity criterion will be the mean square deviation : @xmath275 let us first estimate the answer by dimensional arguments . if the finite - dimensional normal variable @xmath232 is decomposed in its principal components , a fluctuation in direction @xmath252 will be of typical size @xmath276 . to estimate the fluctuation in direction @xmath252 with an accuracy @xmath11 we need about @xmath277 bits . the answer should therefore be of the form @xmath278 the result of the analysis will be that @xmath116 is one and @xmath279 is zero . let @xmath232 be a gaussian one - dimensional random variable with variance @xmath280 . the random variable @xmath250 is defined on the support of @xmath281 . we suppose it to have mean @xmath282 and variance @xmath283 . we have the obvious inequality @xmath284 and the equality @xmath285 on the other hand , the equivocation , ( [ eq : equivocation - discretized ] ) , with fixed outcome @xmath235 of @xmath37 and given mean @xmath282 and variance @xmath283 is maximized if the variable is gaussian with variance @xmath283 . therefore , the mutual information between @xmath232 and @xmath37 is bounded from below by @xmath286 we can therefore pose a constrained minimization problem over the auxiliary positive function @xmath283 . strictly speaking , @xmath287 is only bounded from below by the minimization , but , as we will see , the solution can be realized in terms of variables @xmath232 and @xmath37 with desired properties . either one or the other of the two constraints will be satisfied at the minimum in ( [ eq : constrained - minimization ] ) . a variation with lagrange multipliers gives that @xmath283 must be constant , either equal to @xmath289 or to @xmath280 . we therefore have @xmath290.\end{aligned}\ ] ] the derivation also shows clearly that the solution can be realized as follows : if @xmath291 is less than @xmath11 , then @xmath250 is gaussian with mean zero and variance @xmath280 , and @xmath37 a delta function with support at the origin . if , on the other hand , @xmath291 is greater than @xmath11 , then @xmath250 is gaussian with mean @xmath282 equal to @xmath235 and variance @xmath289 , and @xmath37 is gaussian with mean zero and variance @xmath292 . by the semigroup property of gaussian kernels follows in both cases that @xmath232 is gaussian with mean zero and variance @xmath280 . the generalization to higher - dimensional case is as follows . the variable @xmath232 is gaussian with second moments @xmath293 , with principal components in a diagonal basis @xmath294 . at a point @xmath295 in the support of @xmath281 we consider the variable @xmath250 with first moments @xmath296 and second moments @xmath297 . we have , in analogy with the one - dimensional case , @xmath298 and @xmath299 the mutual information between @xmath232 and @xmath37 is bounded by @xmath300 some of the inequalities ( [ eq : sigma - equations ] ) may hold as equalities in the solution , some not . it does however follow from the diagonal structure of ( [ eq : sigma - equations ] ) and ( [ eq : epsilon - equation ] ) and the relation @xmath301 that when ( [ eq : mutual - information - gaussian - manyd ] ) is minimized , @xmath302 must be constant in @xmath235 and diagonal in @xmath252 and @xmath253 . let the diagonal elements of @xmath95 be @xmath303 . we then have the following simpler discrete constrained minimization problem : @xmath304 the minimum can be found by starting with all @xmath305 s very small and increasing them proportionally to the gradient , that is , at the same rate . when one of the @xmath305 s hits the constraint ( [ eq : sigma - constraints ] ) we keep it constant from thereon and increase the others . the constraint ( [ eq : epsilon - constraint ] ) will eventually be fulfilled when the @xmath305 s that still change are equal to @xmath306 , and there we stop . the solution is thus implicitly given in terms of @xmath306 , which can be interpreted as a cut - off threshold for modes where normal fluctuations are small : @xmath307 , \qquad \epsilon^2 = \sum_i \min\,\left[\sigma_i^2 , \theta^2\right ] \label{eq : kolmogorov - formula}\ ] ] the solution can be realized by taking @xmath37 a random variable with independent components , diagonal in the same basis as @xmath232 , and letting the @xmath252th component of @xmath232 only depend on the @xmath252th component of @xmath37 . the variables @xmath308 and @xmath309 are then constructed as in the one - dimensional case , with the only difference that the parameter @xmath306 substitutes for the error @xmath11 . let us now consider a scalar gaussian random field in @xmath310-dimensional space . a particular example ( @xmath311 ) is gaussian processes . we begin by the approximation that the field is periodic with period @xmath17 in all directions . fourier components of gaussian random fields are independently distributed gaussian random variables . therefore ( [ eq : kolmogorov - formula ] ) can be rewritten , using the volume element @xmath312 , equal to @xmath313 : @xmath314\ , \delta k , \\ \epsilon^2 \left(\frac{2\pi}{l}\right)^d & = & \sum_{\bbox{k } } \min\ , \left[\sigma_{\bbox{k}}^2 , \theta^2 \right]\ , \delta k.\end{aligned}\ ] ] in the limit when @xmath17 tends to infinity the right hand sides turn into integrals , and the left hand sides to entropy and mean - square distance per unit volume . therefore we have @xmath315\ , d^d k , \label{eq : kolmogorov - entropy } \\ \epsilon^2 & = & \left(\frac{1}{2\pi}\right)^d \int \min\ , \left[e(\bbox{k } ) , \theta^2\right]\ , d^d k. \label{eq : kolmogorov - error}\end{aligned}\ ] ] which , for @xmath310 equal to one , is the kolmogorov result for the @xmath11-entropy per unit time of a gaussian random process . we note that @xmath306 is still a cut - off to eliminate modes such that the energy of the fluctuations in these modes is less than @xmath316 . in fields with a power - law spectrum , such as turbulence , @xmath306 can simply be substituted for a wave - number cut - off @xmath317 . the mean square fidelity criterion is convenient for analytical computations , but it is not the only one possible . shannon lists also the maximum distance , another choice would be the mean absolute distance . more interestingly , there may be cases where the most appropriate fidelity criteria is not simply proportional to the distance in the mean or in maximum value . if the signal @xmath232 is a lvy - process it will have a non - negligible probability of changing by a large amount over a short interval of time@xcite . in many applications one would then like an approximating signal @xmath37 to capture well the large jumps , but one would be less interested in a very precise approximation when @xmath232 changes comparatively little@xcite . a reasonable fidelity criterion would then be that there is only a small probability that the distance between @xmath232 and @xmath37 is large . as far as we know no investigations have been performed of the entropy of such sources with respect to such fidelity criteria . of course , for a lvy process a mean square error function is not possible , since the second moment is infinite . it would be possible to use mean absolute error , as was done implicitly by gaspard and wang in @xcite , but the relevance of such computation would have to be motivated in each case by the application at hand in this appendix we want to discuss and compare the finite - size lyapunov exponent and the @xmath11-entropy for a turbulent flow and for the shell model . we assume that the flow and the shell model are both gaussian as in appendix [ s : kolmogorov ] , and we assume that the spectrum follows the kolmogorov theory , as described in section [ s : multifractal ] . non gaussian effects and intermittency correction to the spectrum are not taken into account . at a given error threshold @xmath2 , we have seen that the finite - size lyapunov exponent scales as @xmath186 , and this holds as well for the shell model as for @xmath0 turbulence . let us now consider the @xmath11-entropy , and first give a simple dimensional estimate . it will then be seen that the expressions ( [ eq : kolmogorov - entropy ] ) ( [ eq : kolmogorov - error ] ) just reproduce this result . the dimensional estimate starts from the time scales in the kolmogorov theory , @xmath318 . the distance between two fields that only differ in wave numbers greater than @xmath202 is @xmath319 , that is , it does not depend on the dimensionality of space , but only of the form of the spectrum @xmath320 . on the other hand , the number of degrees of freedom at wave numbers less than or equal to @xmath202 is proportional to @xmath321 . the system must be observed at a rate @xmath322 to capture all the motion in wave numbers less than @xmath202 . the amount of information per unit time and space needed to describe the system up to an error tolerance @xmath2 is thus @xmath323 . translating this into a functional dependence on @xmath2 we have @xmath324 which , for @xmath325 , leads to @xmath326 . as far as we can see there is an inconsistency between this result , also derived in @xcite , and the result stated in @xcite , which we believe is a misprint resulting from using an argument that really applies to the finite - size lyapunov exponent and not the @xmath11-entropy . the very different scaling laws @xmath327 and @xmath230 is also a further motivation why introducing the quantity finite - size lyapunov exponent . this straight - forward dimensional analysis can easily be generalized to a generic stationary gaussian processes with spectral density ( see section 6.2.6 of @xcite ) , @xmath328 the function @xmath169 is supposed to vanish for argument much larger and much smaller than unity , and its integral is on the order of unity . the energy spectrum is therefore @xmath329 we will here check that the previous dimensional estimates can also be obtained from the kolmogorov formulae ( [ eq : kolmogorov - entropy ] ) and ( [ eq : kolmogorov - error ] ) , that we write in this case as follows : @xmath330\ , d^d k\ , d\omega , \label{eq : space - time - entropy } \\ \delta^2 & = & \left(\frac{1}{2\pi}\right)^{d+1 } \int \min\ , \left[\phi({\bf k},\omega ) , \theta^2\right]\ , d^d k\ , d\omega . \label{eq : space - time - error}\end{aligned}\ ] ] the first observation is that the integral over @xmath331 in ( [ eq : space - time - error ] ) gives a contribution of order @xmath332 if @xmath333 is less than @xmath316 , but otherwise a contribution of order @xmath334 . the error is thus determined by a cut - off wave number @xmath317 such that @xmath335 the second observation is that the integral over @xmath331 in ( [ eq : space - time - entropy ] ) gives a vanishing contribution is @xmath333 is less than @xmath316 , that is , for wave numbers larger than the cut - off @xmath317 . for smaller wave numbers the integration over @xmath331 gives a contribution of the order of @xmath336 , up to a logarithmic correction that we do nt take into account . the @xmath11-entropy with cut - off wave number @xmath317 is thus @xmath337 combining ( [ eq : space - time - error ] ) and ( [ eq : space - time - entropy ] ) we find @xmath338 which is equation ( 6.14 ) of @xcite . by inserting the values @xmath325 , @xmath339 and @xmath340 of the kolmogorov theory we reproduce the result ( [ eq : gaspard - wang - scaling ] ) .

employing realistic and consistent atmosphere boundary conditions , we have generated evolutionary models for brown dwarfs and very - low - mass stars ( vlms ) for different atmospheric metallicities ( [ fe / h ] ) , with and without clouds . we find that the spread in radius at a given mass and age can be as large as% to% , with higher - metallicity , higher - cloud - thickness atmospheres resulting quite naturally in larger radii . for each 0.1 dex increase in [ fe / h ] , radii increase by% to.5% , depending upon age and mass . we also find that , while for smaller masses and older ages brown dwarf radii decrease with increasing helium fraction ( ) ( as expected ) , for more massive brown dwarfs and a wide range of ages they increase with helium fraction . the increase in radius in going from to can be as large as.025 (.5% ) . furthermore , we find that for vlms an increase in atmospheric metallicity from 0.0 to 0.5 dex , increases radii by% , and from -0.5 to 0.5 dex by% . therefore , we suggest that opacity due to higher metallicity might naturally account for the apparent radius anomalies in some eclipsing vlm systems . ten to twenty - five percent variations in radius exceed errors stemming from uncertainities in the equation of state alone . this serves to emphasize that transit and eclipse measurements of brown dwarf radii constrain numerous effects collectively , importantly including the atmosphere and condensate cloud models , and not just the equation of state . at all times , one is testing a multi - parameter theory , and not a universal radiusmass relation .

1102.3922

for the last fifteen years , the evolutionary , spectral , and atmospheric theory of brown dwarfs has evolved in tandem with the explosive growth in the number of known examples . that number now stands near 1000 . this maturing theory has been used to interpret spectroscopic and photometric observations of such substellar - mass objects ( smos ) and , thereby , to derive effective temperatures ( ) , gravities ( @xmath5 ) , bolometric luminosities , compositions , masses ( @xmath6 ) , and ages ( @xmath7 ) . the focus of study has been the new spectroscopic types , l and t , that have emerged to accomodate objects with from @xmath0500 k to @xmath02200 k. both chemical equilibrium and non - equilibrium models have found roles and gradual improvements in molecular opacity and thermochemical databases suggest that objects with molecular atmospheres may someday be studied with the same precision as are stars with simpler atomic atmospheres , the latter albeit after almost a century of development . however , most brown dwarfs are found as singletons , and those in binaries have only rarely been used to derive dynamical masses with which to provide independent checks on spectroscopically - derived masses ( konopacky et al . 2010 ; liu , dupuy , & leggett 2010 ) . too often , however , such comparisons are cast as fundamental tests of core assumptions in the models . little mention is made of the equal importance of atmospheric opacities , metallicities , cloud models , and ages in determining whether theory fits data . in fact , these last - mentioned quantities are crucial in determining the physical properites of a given brown dwarf . in addition to obtaining independent masses , measuring the radii of brown dwarfs can be an equally powerful means of testing the associated physical theory ( burgasser , burrows , & kirkpatrick 2006 ) . however , here too one is testing a multi - parameter theory , and not a universal radius ( @xmath8)@xmath4mass relation . the latter exists only for cold , very old , objects of a given elemental composition . hot , evolving brown dwarfs with a priori unknown atmospheric and interior compositions ( and opacities ) can span a ( wide ) range ( perhaps 25% ) of radii for a given mass and age . the brown dwarf s atmospheric opacities determine the rate with which heat escapes the convective core , and at which the core entropy evolves , and it is the core entropy which sets the radius , for a given mass and internal composition . gas - phase atmospheric opacities can be augmented by silicate and iron cloud opacities . the presence of clouds , for which there is as yet no rigorous or credible model , complicates the theory and interpretation of the measured physical properties of brown dwarfs . the thicker the atmosphere , the slower the radius shrinks and the larger is the radius at a given mass and age . since the gas - phase opacities depend upon composition ( atmospheric metallicity ) and extant cloud models are still crude , without additional constraints on these factors there is an inherent ambiguity in the comparison between measured and theoretical radii that should temper any conclusions drawn . what is more , rapid rotation and ( perhaps ) magnetic activity ( chabrier , gallardo , & baraffe 2007 ; morales et al . 2009 ) can further complicate the interpretation of an observed radius and what one can deduce from differences between measured and model radii . for a given mass and age there is a physical , realistic _ range _ of possible radii that varies with atmospheric metallicity and cloud optical depth , both of which would need to be measured to properly falsify theoretical models . for simplicity , evolutionary models have frequently been provided by theorists for only a restricted set of parameters ( e.g. , at solar metallicity ) and for a single ad hoc prescription for clouds ( or with no clouds ) ( burrows et al . 1993,1997,2001,2006 ; chabrier & baraffe 1997 ; baraffe et al . 1998,2003 ; saumon & marley 2008 ) . however , these models have been applied quite generally , with an assumption of universality that is not warranted . as a consequence , any deviation between these theoretical models and physical measurements , be they of radius or mass , must first account for the effects of different atmospheres before declaring anything about , for example , the underlying equation of state or fundamental assumptions of the modeling exercise . moreover , without a good constraint on the age of the brown dwarf ( perhaps from a good estimate of the primary s age and the assumption of coevality ) , any radius measurement is of correspondingly limited utility if the goal is to test structural theory beyond its zeroth - order " aspects . recently , numerous exoplanet transit surveys have collectively discovered and characterized more than one hundred transiting companions to nearby stars . these surveys include _ kepler _ ( koch et al . 2010 ; borucki et al . 2010 ) , _ corot _ ( baglin et al . 2006 ) , wasp ( pollacco et al . 2004 ) , ogle ( udalski et al . 2002 ; konacki et al . 2003 ) , and hat ( bakos et al . 2002 ) . importantly , byproducts of such surveys , and their radial - velocity follow - up , are radii and masses for almost all of them , enabling the type of structural study to which we alluded above . most of the transiting companions have masses near or below that of jupiter and most of them are very close to their primaries . such proximity introduces the need to incorporate the effects of irradiation ( and perhaps tidal heating ) in any interpretative study of the radius@xmath4mass relation of objects in this mass range . however , a small subset of the transiting companions found are quite massive ( above @xmath015 ) and the radii of such smos should not be affected by stellar irradiation ( burrows et al . 2007 ) . therefore , data for these more massive objects can be used to constrain brown dwarf theory . of most interest for such a study are _ kepler s _ lhs 6343 ( johnson et al . 2010 ) , corot-15b ( bouchy et al . 2011 ) , corot-3b ( deleuil et al . 2008 ) , and wasp-30b ( anderson et al . 2011 ) . since they reside in a physical region that is seamlessly connected to that of brown dwarfs , the very - low - mass stars ( vlms ) ogle - tr-122b ( pont et al . 2005 ) and ogle - tr-123b ( pont et al . 2006 ) , though of stellar mass , test the same basic theories . in this paper , we calculate a new suite of evolutionary models with new sets of atmospheric boundary conditions for various atmospheric metallicities , with and without clouds ( i.e. , clear ) . with these models , we are able to demonstrate and quantify the dependence of the radius of a brown dwarf of a given mass and age on the variety of different atmospheric characteristics expected . brown dwarfs should vary in metallicity and cloud cover . higher - metallicity atmospheres cool more slowly and , therefore , result in larger brown dwarf radii , all else being equal . cloudy atmospheres similarly retard radius shrinkage and lead to larger radii at a given mass and age . though we do not yet know how to incorporate this possibility , higher - metallicity atmospheres may also have thicker cloud decks , amplifying the effect of increasing metallicity . be that as it may , one should expect a range of realistic brown dwarf radii at a given mass and age . our paper quantifies this range to highlight this important fact , while also exploring the dependence on the helium mass fraction ( @xmath1 ) . with the models we generate , we fit various of the recently measured transiting brown dwarf radii . we have chosen for this study the brown dwarfs lhs 6343c ( johnson et al . 2010 ) , corot-15b ( bouchy et al . 2011 ) , corot-3b ( deleuil et al.2008 ) , and wasp-30b ( anderson et al . 2011 ) , and the vlms ogle - tr-122b ( pont et al . 2005 ) and ogle - tr-123b ( pont et al . 2006 ) , but the generic nature of the expected spread in radius at a given mass and age is the true focus of this investigation . while documenting representative theoretical differences in radius evolutions for different metallicities and helium fractions ( with and without clouds ) , we at the same time demonstrate that radius@xmath4mass measurements do not constrain solely the equation of state . the atmospheric opacities ( and , less so , the helium fraction and overall internal compositions ) are important factors in the interpretation of any radius measurement . moreover , the significant remaining uncertainties in the ages of all these systems , and the difficulties and potential systematic inaccuracies in the measurements themselves , partially undermine any attempt to arrive at definitive conclusions concerning the accuracy or usefulness of any theoretical evolutionary model . nevertheless , the magnitude of the expected theoretical spread in radii for brown dwarfs at a given mass and age should be more robust , and , as we demonstrate , is certainly not trivial . in section [ data ] , we summarize the data for the transiting objects upon which we focus in this paper . then , in section [ method ] we review the techniques we have employed to calculate the evolutionary models , using an atmosphere / spectral code to obtain the boundary conditions for our henyey evolutionary code . in section [ general ] , we discuss the general trends and behaviors of the full model suite , after which in section [ individual ] we determine individual object characteristics from comparisons with the collection of new models . finally , in section [ conclusion ] , we review the overarching conclusions concerning the radius@xmath4mass relation that have emerged from this study . one of the most interesting objects in the observed set is lhs 6343c ( johnson et al . 2010 ) , discovered using _ kepler_. it has a measured mass of @xmath9 , a measured radius of @xmath10 , and orbits one of a pair of m dwarfs ( lhs 6343a , with an inferred mass of @xmath00.37 m@xmath11 ) with a period of 12.71 days . from the minimal chromospheric activity , the authors suggest an age greater than 1 gyr ( gigayear ) , and prefer a range of @xmath01 gyr to @xmath05 gyr . anderson et al . ( 2011 ) quote a metallicity ( [ fe / h ] ) for lhs 6343a of @xmath12 , but johnson et al . ( 2010 ) provide a value of @xmath13 . the triple " nature of this system complicated the extraction of physical parameters . we quote inferred metallicities for the primary of lhs 6343c ( and for the other primaries discussed below ) under the assumption that the primary and secondary metallicities of coeval objects formed in the same context are likely to be similar and that the primary s [ fe / h ] can serve as a guide when fitting secondary characteristics . corot-15b ( bouchy et al . 2011 ) has a measured mass of @xmath14 , a measured radius of @xmath15 , and orbits its f7v star primary with a period of 3.06 days . bouchy et al . ( 2011 ) suggest an age for corot-15b of @xmath01.14@xmath43.35 gyr and a metallicity of @xmath0@xmath16 , basically solar . deleuil et al . ( 2008 ) obtain a mass for corot-3b of @xmath17 and a radius of @xmath18 . these authors prefer an age and metallicity for the f3v primary corot-3 of @xmath02 gyr and @xmath19 , respectively , the latter consistent with solar . corot-3b s orbital period is 4.26 days . wasp-30b ( anderson et al . 2011 ) has a measured mass of @xmath20 and a measured radius of @xmath21 . anderson et al . ( 2011 ) quote a metallicity for wasp-30 of @xmath22 , also basically solar , and prefer an age of @xmath23 gyr . wasp-30b is orbiting its f8v primary with a period of 4.16 days . the very - low - mass stars ( vlms ) ogle - tr-122b and ogle - tr-123b have measured masses of @xmath24 ( pont et al . 2005 ) and @xmath25 ( pont et al . 2006 ) , respectively . their measured radii are @xmath26 and @xmath27 . pont et al . ( 2005,2006 ) suggest that the ages of these vlms are probably less than @xmath28 gyrs . the metallicity of ogle - tr-122 is estimated to be @xmath29 , and , hence , can be considered unconstrained . ogle - tr-122b is orbiting a solar - like star with a period of 7.3 days and ogle - tr-123b is orbiting an f star with a period of 1.8 days . both are at large distances ( @xmath01000 and @xmath01600 parsecs ) and , hence , their primaries are difficult to characterize accurately . an interesting eclipsing system is 2mass j0535 ( stassun et al . 2006 ) , for which , curiously , the least massive of the pair has the higher measured . however , though its masses are both in the brown dwarf regime ( @xmath30 m@xmath11 and @xmath31 m@xmath11 ) and the radii of these two objects are measured to be @xmath32 and @xmath33 , respectively , the estimated system age is @xmath01 megayear ( myr ) . at such young ages , one needs a theory for their formation in the context of the protostellar disk . extant evolutionary models for isolated spherical objects that assumed the initial conditions were lost on timescales of tens of myrs are not suitable for comparisons at myr ages . our spectral and atmosphere methods have previously been described in burrows et al . ( 1993,1997,2003,2006 ) , but we repeat the salient points here for completeness . given the surface gravity ( @xmath5 ) , effective temperature ( @xmath34 ) and cloud properties , the planar code ` cooltlusty ` solves for the atmosphere and spectrum of an object using the hybrid technique of complete linearization and accelerated lambda iteration ( hubeny 1988 ; hubeny & lanz 1995 ) , adopting the solution to the ( gray ) milne problem as a first guess . convection is treated using mixing length theory , where the mixing length @xmath35 in units of the pressure scale height ( @xmath36 ) is set to 1.0 . iteration proceeds until adjacent layers of the 1d atmosphere are in radiative equilibrium . we have computed cloud - free evolutionary tracks for @xmath37 , @xmath36 and @xmath38 and find that within the ranges of effective temperature ( @xmath393000 k ) and surface gravity ( @xmath405.5 ) considered in this paper , the models generated are essentially identical . the main gas - phase species used in these theoretical atmospheres are h@xmath41 , he , h@xmath41o , co , ch@xmath42 , n@xmath41 , nh@xmath43 , feh , crh , na i , and k i and our spectral model employs 30000 frequency points from @xmath00.35 to 300 microns . the opacities are taken from sharp & burrows ( 2007 ) . to generate molecular abundances in chemical equilibrium at a given atmospheric metallicity , we employed the thermochemistry found in sharp & burrows ( 2007 ) and followed nearly 500 species ( with over 150 condensates ) containing 27 elements . for these abundance calculations , the atmospheric element fractions for a given metallicity were taken from allende - prieto , lambert , & asplund ( 2001 ) , asplund et al . ( 2005 ) , and allende - prieto & lambert ( 2005 ) , which replaced the anders & grevesse ( 1989 ) compositions we used previously . refractory silicates condense out at temperatures below @xmath02300 k and above @xmath01700 k. in our chemical code , these species are rained out ( burrows & sharp 1999 ) to determine the resulting gas - phase abundances . however , the clouds of the condensed refractories have a meteorology , physical extent , and particle - size distribution that are as yet unconstrained ( ackerman & marley 2001 ; helling et al . 2001,2004 ) . therefore , to incorporate clouds into the spectral and atmosphere models , we introduce a flexible parametrization that allows us to vary cloud opacities and effects . for this study , we assume that the numerous cloud species are mimicked by one representative species ( in this case forsterite , mg@xmath41sio@xmath42 ) that extends from a base at 2300 k to the intersection point of the atmospheric temperature - pressure profile with the forsterite clausius - clapeyron " condensation line . the baseline distribution of the cloud particle density is assumed to follow the gas - phase pressure profile , but we multiply this baseline reference distribution by a shape function , @xmath44 , to determine the actual model cloud distribution in pressure space . @xmath44 is never greater than one and can be made to cut off sharply at the top and bottom of the cloud . specifically , we define @xmath44 as : @xmath45 where @xmath46 is the pressure at the cloud base and @xmath47 . here , @xmath48 is the pressure at the intersection of the atmospheric @xmath49 profile and the forsterite condensation curve . the indices @xmath50 and @xmath51 define the rapidity with which the clouds are cut off on their upper and lower boundaries . the larger they are , the sharper the cutoff . when @xmath52 the cloud has a flat portion in its middle between @xmath48 and @xmath46 . hence , the parameters @xmath50 , @xmath51 , and @xmath53 define the cloud spatial structure . we chose @xmath02300 k as our cloud base since it is where the most refractory species ( calcium - aluminum silicates ) condense . we chose the forsterite condensation line to cap this region because forsterite has one of the lowest condensation temperatures ( for a given pressure ) . hence , the physical region where refractory clouds reside is more extensive than where forsterite alone would reside . to attempt a multi - species meteorological model , incorporating the optical properties of all the expected refractories and the role of convection in grain growth , and to pretend to model the various particle sizes and shapes and the interaction between the various grain species ( mantling , collisions , etc . ) , would introduce extra complexity wholely out of proportion with the sparse dataset available for these very - remotely - sensed objects . note that if only forsterite clouds obtained , the clausius - clapeyron " condensation line would be the natural cloud base ( and not the @xmath02300 k line ) , but that convection and turbulence would still likely extend the cloud upward to lower pressures . the characteristic scale of this is often assumed to scale with the pressure scale height . this is the motivation for introducing the parameter @xmath50 , thereby capturing this physical extension above the forsterite condensation line . for the cloud models in this study , we set @xmath54 and @xmath55 and use a modal particle size @xmath56 @xmath57 m ( sudarksy et al . 2000 ) , where the size distribution is given by : @xmath58}\ , .\ ] ] we use mie scattering theory with a table of complex indices of refraction as a function of wavelength to calculate the absorption and scattering opacities for the grains in the clouds . in summary , with this general cloud model we place clouds where they may be likely to reside , over the collective condensation region , and then vary the modal particle size of forsterite grains to create a model set with a wide range of possible cloud optical thicknesses . to mimic the transformation of a young , cloudy object evolving into an old , cloud - free object , and the l - to - t transition , we linearly combine cloudy ( @xmath59 ) and cloudfree ( @xmath60 ) spectra using the phenomenological prescription : @xmath61 thus , @xmath62 is the transitional temperature at and below which clouds in the l dwarf atmosphere begin to both disperse and reside below the photosphere , while the power law index @xmath63 describes the speed of the transition . effectively , our hybrid models have five parameters : @xmath64 , @xmath65 , @xmath66 , @xmath62 and @xmath63 . we are guided by empirical measurements of the color - magnitude diagram to choose @xmath67 k and @xmath68 . although our approach is intrinsically different from @xmath69 parametrization of saumon & marley ( 2008 ) , the resulting cloud structures and temperature profiles are similar across the l - to - t transition . an evolutionary henyey code is used to establish the mapping between @xmath70 and @xmath71 ( mass - age ) pairs . with the equation of state ( eos ) for hydrogen / helium mixtures of ( saumon , chabrier , & van horn 1995 ) and specified values of the deuterium and helium fractions , the code solves the equations of stellar structure . we emphasize that the hydrogen / helium eos of saumon , chabrier , & van horn ( 1995 ) employs the volume addition law for such mixtures and does not have a rigorous means to incorporate heavy elements in a consistent fashion . hence , though the core and atmosphere will certainly have the same [ fe / h ] , we do not include heavies in our eos . currently , there does not exist a reliable eos for more complicated mixtures . if the matter were an ideal gas , then one could easily incorporate any range of elemental abundance mixtures by calculating the appropriate mean molecular weight . however , the material in smos experiences significant coulomb interaction effects . indeed , at the high densities and low entropies encountered in their cores , the coulomb effects are important . at higher densities , degeneracy effects predominate , and the eos , depending as it does on the electron fraction per baryon , again simplifies . however , with brown dwarfs we are often in the intermediate regime where coulomb and degeneracy effects compete , an awkward regime to treat well . at times , theorists have employed altered helium fractions to treat heavies , boosting @xmath1 by an appropriate " amount and then using the h / he eos with the enhanced @xmath1 ( spiegel , burrows , & milsom 2011 ; guillot et al . 2006 ) . at solar metallicity , [ fe / h ] = 0.0142 and the effect of neglecting heavies in the eos , while not large , is still interesting however , the reader should be aware that the results we quote do not automatically incorporate heavies in the core eos and that the stopgap of using a suitably " enhanced @xmath1 to gauge their effects remains an option . if this be the case , then a @xmath72solar metallicity would be equivalent to a @xmath00.04 augmentation in @xmath1 , comparable to the range of @xmath1 we explore in this paper . the boundary condition between the atmosphere and the convective core is the entropy at the radiative - convective transition layer of the planar atmosphere , which is an output of ` cooltlusty ` . by merging the evolutionary and spectral computations in this manner , we ensure that our spectral - evolutionary calculations are self - consistent . for this investigation , we calculate models from 1 to 0.15 m@xmath11 ( @xmath0157 ) and focus on three atmospheric metallicities : [ fe / h]=@xmath730.5 and 0.0 ( solar ) . hence , our models are 1 ) clear with the three different metallicities and 2 ) cloudy hybrid with [ fe / h]=0.0 and 0.5 . these five models , along with the heritage model from burrows et al . ( 1997 ) , are used to fit the measured radii of the brown dwarfs lhs 6343c , corot-3b , corot-15b , & wasp-30b and the very - low - mass ( vlm ) stars ogle-122b and ogle-123b and to arrive at the various general conclusions . in this section , we discuss the qualitative behavior of both the radius evolution with time and the radius@xmath4mass isochrones as a function of the presence or absence of clouds , the atmospheric metallicity , and the helium fraction . the importance of including clouds in evolutionary models was previously emphasized by saumon & marley ( 2008 ) . figure [ fig : evolution ] depicts the temporal evolution of the radii of theoretical brown dwarfs at four different masses ( @xmath74 , @xmath75 , @xmath76 , and @xmath77 m@xmath11 , equivalent to @xmath058 , @xmath063 , @xmath068 , and @xmath073 ) for models with cloudy atmospheres at [ fe / h ] = 0.0 and 0.5 and for cloud - free atmospheres at [ fe / h ] = 0.0 . the cloud model incorporates the cloud prescription described in section [ method ] . since the physics of cloud particles and structure are not yet understood , this cloud model should be viewed merely as representative of the effects of clouds . however , it is neither extreme in its properties , nor unrealistic , and introduces the extra degree of atmospheric opacity expected of such clouds . as fig . [ fig : evolution ] indicates , before @xmath01 gyr brown dwarf radii evolve very quickly , but later their shrinkage moderates significantly . importantly , however , this figure shows the radius hierarchy that emerges from this study . models with higher atmospheric opacity , due either to higher metallicity or to the presence of thick clouds , have larger radii at a given mass and age than those with lower opacity . the latter would be associated with lower metallicity and/or the absence of clouds . importantly , the natural spread in theoretical radii at a given mass and age can be significantly larger than the quoted formal error bars of the measured radii . to demonstrate this , we have superposed onto this figure the data for the three brown dwarfs in the mass range @xmath060 to @xmath070 for which we have useful transit radii and masses ( lhs 6343c , wasp-30b , and corot-15b ) . we caution that the formal error bars in measured radii may be trumped by systematic errors in the observational analysis and that the true error in the data may be larger than published . in turn , there are , no doubt , errors " in the theoretical models that spring from uncertainties in the gas - phase opacities , uncertainties in the the equation of state , elemental abundances patterns for a given metallicity that are different from what is assumed , and the many uncertainties in designing accurate cloud models . nevertheless , as fig . [ fig : evolution ] demonstrates , for reasonable stellar/(brown dwarf ) metallicities expected in the solar neighborhood , the radius spread at a given mass and age can be as large as @xmath00.2 . this exceeds any reasonable error stemming from uncertainities in the equation of state alone and serves to emphasize that measurements of brown dwarf radii constrain a collection of effects , importantly including the atmosphere and condensate cloud models . without an independent measure of the atmospheric metallicity and constraints on the surface clouds , and a good age constraint , a measurement of a brown dwarf radius may be less useful than some might expect for testing the hydrogen / helium equation of state . one way to understand the effect of enhanced atmospheric opacity on the radius of an smo is by way of the consequent alteration in atmospheric temperature / pressure profiles ( for a given t@xmath78 and gravity ) with increasing atmospheric metallicity or upon the introduction of clouds . figure [ tpcloud ] depicts three sets of atmospheric thermal profiles at t@xmath78 = 1300 k and five gravities from @xmath79 = 3.5 to @xmath79 = 5.5 . the three sets are cloud - free ( red ) , cloudy at solar metallicity ( green ) , and cloudy at 3@xmath80solar metallicity ( blue ) . as is clear from the figure , when comparing cloudy with cloud - free models we notice that at a given temperature , the pressure is everywhere lower for the cloudy model . this is also the case when comparing cloudy / solar with the cloudy/3@xmath80solar models . this indicates that the entropy at depth , the core entropy , is correspondingly higher for the atmospheres with enhanced opacity due either to increased metallicity or the presence of clouds . since at a given mass higher entropy translates into larger radii , fig . [ tpcloud ] neatly demonstrates the opacity / radius connection and our general thesis . figure [ burrows97 ] portrays isochrones from 0.5 to 5.0 gyr in radius@xmath4mass space from 1 to 120 for the heritage " model published by burrows et al . superposed are the corresponding data for the six objects of this study . there is a slight peak near @xmath04 m@xmath81 . at larger masses ( in the brown dwarf regime " ) , radii decrease with increasing mass , monotonically shrinking as they age . depending upon age , for masses around @xmath060 to @xmath070 , the radius then begins to increase with mass . this is a manifestation of the onset of significant thermonuclear burning and the transition to a star . hence , there is a minimum radius for a given model and age . from 1 gyr to 5 gyr that minimum radius ranges from @xmath01.1 to @xmath00.75 for the five - model suite studied in this paper and from @xmath00.9 to 0.77 for the burrows et al . ( 1997 ) models . as fig . [ burrows97 ] shows , ogle-122b and ogle-123b are clearly stars , though a comparison with these heritage models requires ogle-123b to be quite young . pont et al ( 2005,2006 ) suggest that the ogle objects might indeed be younger than 0.5 gyrs . the quoted age range of lhs 6343 is between one and five billion years ( johnson et al . 2010 ) and fig . [ burrows97 ] implies that the burrows et al . ( 1997 ) model set could fit this object . anderson et al . ( 2011 ) suspect that wasp-30b is younger than 1 gyr , but given the inherent ambiguities in such an estimate fig . [ burrows97 ] indicates that the burrows et al . ( 1997 ) models might provide an adequate fit . the age of corot-3 is not well constrained and its radius error bars are large , making a comparison between theory and measurement an easy , though rather useless , exercise . however , with a suggested age range of @xmath01.14@xmath43.35 gyr ( bouchy et al . 2011 ) , corot-15b might be the only object among the six we are highlighting in this paper for which a solution using the burrows et al . ( 1997 ) models might be problematic . figure [ rvsm_clear_cloudy ] compares radius@xmath4mass trajectories for cloudy - atmosphere models and clear - atmosphere models at four different ages ( 0.5 , 1.0 , 3.0 , and 5.0 gyr ) . the clear models have atmospheric metallicities ( [ fe / h ] ) of -0.5 , 0.0 , and + 0.5 , while the cloudy models have atmospheric metallicities of 0.0 and + 0.5 . as fig . [ rvsm_clear_cloudy ] demonstrates , increasing the metallicity increases the radius , with or without clouds . we note that atmospheres without clouds , or a cloud effect , are not expected for metallicities of solar and above . nevertheless , for clear models at early ages , the spread in radius for brown dwarfs with masses between @xmath030 and @xmath070 and for the metallicity range of this study can be as much as @xmath00.15 . for the same clear model set at late ages , that same spread narrows between @xmath055 and @xmath070 to @xmath00.1 . for the cloud models , at early ages the radii can range by @xmath00.15 in going from [ fe / h]=0.0 to [ fe / h]=0.5 , but that range tightens at the later ages to @xmath00.05 for the more massive brown dwarfs . however , the spread in radius from clear models at [ fe / h ] = -0.5 to cloudy models at [ fe / h ] = 0.5 can be as large as @xmath00.25 at early ages , and as large as @xmath00.1 at late ages . this implies that each increment in [ fe / h ] of 0.1 can translate into a radius change of @xmath01@xmath42.5 % , depending upon age and mass . in the stellar realm above @xmath090 , atmospheric temperatures are above the condensation temperatures of iron and refractory silicates . as a consequence , and as we see in fig . [ rvsm_clear_cloudy ] , the radii of stars depend only weakly on the presence or absence of clouds . however , their radii are still modest functions of atmospheric metallicity . increasing the metallicity of a vlm star from 0.0 to 0.5 increases the radius by @xmath04% and from -0.5 to 0.5 by @xmath010% . the latter percentage is slightly above what is oft - quoted as the discrepancy in radius between measurement and theory in the vlm regime . though we do not in this investigation discuss models in the @xmath00.2 to 0.25 band , the implication of the systematic behavior of fig . [ rvsm_clear_cloudy ] in the stellar realm is that opacity due to higher metallicity can naturally account for this apparent radius anomaly in , for instance , the cm dra ( morales et al . 2009 ) and koi-126 ( carter et al . 2011 ) systems . neither rotation nor magnetic activity ( chabrier , gallardo , & baraffe 2007 ) need be the only ( nor the dominant ) explanation . summarizing , the effect of atmospheric metallicity on the radii of brown dwarfs and vlms is straightforward and natural , and the magnitude of the potential spread in radii for reasonable ranges in metallicity in the solar neighborhood is comparable to published apparent discrepancies . figure [ helium ] recapitulates some of the information on fig . [ rvsm_clear_cloudy ] , but highlights the helium fraction dependence of the radius@xmath4mass relation and is for only solar atmospheric metallicity . shown are trajectories for @xmath82 ( blue / aqua ) and @xmath83 ( red / magenta ) , where @xmath1 is the helium mass fraction . at zero temperature , the higher the mean molecular weight and the lower the electron - to - baryon ratio ( both consequences of higher @xmath1 ) , the smaller the brown dwarf . at later ages and lower masses , fig . [ helium ] clearly demonstrates this tendency , and this is the widely - expected result . however , for the higher temperatures and entropies which persist for more massive brown dwarfs , the opposite is true @xmath4 a higher helium fraction leads to a larger radius . the magnitude of the effect is slight , but can amount to as much as @xmath00.025 . as fig . [ helium ] shows , this slight reversal " continues into the stellar realm . we note that the burrows et al . ( 1997 ) models ( also shown on the figure for comparison ) were calculated for @xmath82 , but that a value of @xmath3 is generally preferred ( saumon & marley 2008 ) . this slightly counterintuitive result at larger masses is due to the enhanced core hydrogen burning rate for the larger @xmath1 , despite the slightly smaller hydrogen fraction . in hydrostatic equilibrium , the fact that at higher mean molecular weight the adiabat is higher results in a higher core temperature for an isentropic core . hence , larger @xmath1 ( which translates into a higher mean molecular weight ) results in higher central temperatures , which generate higher thermonuclear powers . this also results in higher core entropies for a given convective mass and age , and , hence , larger outer radii . when the nuclear luminosity is very small , the larger @xmath1 results in more compact configurations , as is expected from the behavior of the eos , and smaller outer radii . however , for larger masses , the core temperatures eventually become significant , even if the object is not destined to settle onto the main sequence . the condition for the latter is that the core thermonuclear power equal the atmospheric luminosity @xmath4 the first mass for which this condition is satisfied defines the main sequence edge . however , even for lower masses , thermonuclear power can retard the decay of the core entropy , which for a given mass determines the outer radius . that the specific heat decreases with increasing @xmath1 is a slightly mitigating effect , but this is trumped by the enhancement in nuclear burning rate . the same phemonenon was identified in spiegel , burrows , & milsom ( 2011 ) in the context of deuterium burning . figure [ ratio ] portrays the ratio of the nuclear luminosity to the surface photon luminosity for representative masses from 0.05 to 0.08 m@xmath11 and @xmath1 = \{0.25 , 0.28}. when this ratio is equal to one we are on the main sequence . as fig . [ ratio ] demonstrates , even for masses near 0.05 m@xmath11 and far below the main sequence , core burning is both non - trivial and larger for larger @xmath1 . the higher nuclear luminosity for higher @xmath1 also helps explain why vlm radii are larger for larger @xmath1 ( fig . [ helium ] ) . we end this section by reiterating that our core eos does not automatically incorporate the slight effect of metallicity , but that one can approximately account for heavy elements in the eos by augmenting the helium fraction accordingly . this suggests , given the findings summarized in fig . [ helium ] , that properly including them in the eos would further ( though only slightly ) increase the radii of vlms and of the more massive brown dwarfs younger than @xmath05 gyrs , emphasizing yet again the potential role of metallicity in influencing smo radii . the six objects ( lhs 6343c , wasp-30b , corot-15b , corot-3b , ogle - tr-122b , and ogle - tr-123b ) that we investigate in this paper have measured transit radii and radial - velocity masses . the errors in the measured radii quoted in the discovery papers are usually purely statistical , yet systematic errors may actually dominate . therefore , the reader should keep this in mind when we derive approximate physical properties for these objects from a comparison with our theoretical models . there may in fact be more interpretive latitude than we have allowed ourselves . given this caveat , we now proceed to discuss each individual brown dwarf or vlm . on fig . [ lhs6343 ] , we have plotted four panels in radius@xmath4mass space from 40 to 80 , each for a different model set , with and without clouds and for [ fe / h ] equal to either 0.0 or 0.5 . four isochorones at 0.5 , 1.0 , 3.0 , and 5.0 gyr are given on each panel . the data from johnson et al . ( 2010 ) for lhs 6343c ( @xmath84 ; @xmath85 ) are superposed , where the errors are 1-@xmath86 . johnson et al . ( 2010 ) suggest an age range from 1 to 5 gyrs . lhs 6343c is seen to be a small brown dwarf , and , therefore , fig . [ lhs6343 ] suggests that older ages , clearer models , and lower metallicities might be preferred . however , a range of combinations can be shown to fit . for the cloudy model , the older solar - metallicity models may be indicated ( bottom right panel ) . anderson et al . ( 2011 ) suggest a metallicty of @xmath12 , but johnson et al . ( 2010 ) quote @xmath13 . if we were to take the johnson et al . number and note that low gas - phase metallicity may suggest thinner clouds , we get a consistent set of characteristics in the lower - right - hand panel with an age near 5 gyrs . clear models at the higher metallicity fit for an age near @xmath04@xmath45 gyrs . clear models at the lower metallitiy fit for ages greater than @xmath02 gyrs ( bottom left panel ) . however , the high - metallicity , cloudy model ( top right panel ) does not fit unless the age is greater than @xmath07 gyrs . curiously , the burrows et al . ( 1997 ) models ( see fig . [ burrows97 ] ) fit well for a suggested age of @xmath02 gyrs . figure [ wasp30 ] is the same as fig . [ lhs6343 ] , but for wasp-30b and with the [ fe / h ] = 0.5 and [ fe / h ] = 0.0 models switched ( top @xmath87 bottom ) . superposed on all panels is the data point for wasp-30b at @xmath20 and @xmath21 ( anderson et al . 2011 ) . the metallicity of wasp-30 is quoted to be [ fe / h ] = @xmath22 , consistent with solar . as fig . [ wasp30 ] indicates , a variety of models and age@xmath4metallicity combinations fit the wasp-30b data . clear models with [ fe / h ] = 0.0 fit well for @xmath01@xmath42 gyrs . clear models with [ fe / h ] = 0.5 fit well for ages from @xmath02 to @xmath03 gyrs . cloudy models with [ fe / h ] = 0.0 fit well for ages of @xmath88 gyrs and our cloudy model with [ fe / h ] = 0.5 still fits near ages of @xmath05 gyrs . cloudy models with @xmath82 ( not shown ) fit at slightly younger ages . as fig . [ burrows97 ] suggests , the heritage models from burrows et al . ( 1997 ) fit wasp-30b for an age near @xmath01 gyr . therefore , as figs . [ wasp30 ] and [ burrows97 ] together demonstrate , the wasp-30b radius is rather easily fit within its suggested mass , age , and metallicity constraints ( soft as they are ) . we superpose the data from bouchy et al . ( 2011 ) for this brown dwarf ( @xmath89 ; @xmath15 ) on fig . [ corot15 ] , which is in the same format as fig . [ wasp30 ] , but has a slightly different range for the radius and mass axes . corot-15b has the widest error bars for a measured radius among the set of brown dwarfs upon which we focus in this study . bouchy et al . ( 2011 ) estimate a metallicity of @xmath16 ( consistent with solar ) and an age for its primary in the range @xmath01.14@xmath43.35 gyr . [ corot15 ] suggests , our solar - metallicity models fit in the lower age range to within @xmath01-@xmath86 to @xmath01.5-@xmath86 , with the best fit for the [ fe / h ] = 0.0 cloudy model . however , our [ fe / h ] = 0.5 models fit the suggested age range better , with the clear [ fe / h ] = 0.5 models fitting an age of @xmath01 gyr to within @xmath01-@xmath86 and the cloudy [ fe / h ] = 0.5 models fitting anywhere in the suggested age range . cloudy models with [ fe / h ] = 0.0 and clear models with [ fe / h ] = 0.5 fit corot-15b almost equally well . figure [ burrows97 ] indicates that the burrows et al . ( 1997 ) solar - metallicity models would fit only for very young ages less than @xmath00.5 gyrs . if the highish " value of the radius survives , a high - opacity atmosphere model , either due to high metallicity or thick clouds or both , may be indicated for corot-15b . however , the range of possible models available to fit this object remains comfortably broad , particularly for ages younger than @xmath02 gyrs . figure [ corot3 ] is similar to figs . [ lhs6343 ] , [ wasp30 ] , and [ corot15 ] , but is constructed for corot-3b and has a lower mass range between 5 and 40 . deleuil et al . ( 2008 ) measure the mass of corot-3b to be @xmath17 and its radius to be @xmath18 . corot-3 has an estimated metallicity of @xmath19 and , hence , the models with solar metallicity may best describe the data . note that due to the large uncertainty in its radius , many models , both clear and cloudy with a variety of age@xmath4metallicity pairs , can fit corot-3b comfortably . deleuil et al.(2008 ) quote an age near @xmath02 gyr . fortunately or unfortunately , due to the large radius error bars , all the model sets plotted ( cloudy and clear , [ fe / h ] = 0.0 and 0.5 ) fit the corot-3b data reasonably well . higher metallicity / cloudy models favor older ages and low metallicity / clear models favor younger ages , but ages between 1 and 5 gyr are consistent with the quoted mass / radius data for the putative estimated age and metallicity . in addition , as fig . [ burrows97 ] demonstrates , the burrows et al . ( 1997 ) models fit for any age younger than @xmath06 gyrs . figure [ ogle ] compares the measured radius@xmath4mass points for the vlm stars ogle - tr-122b and ogle - tr-123b with cloudy and clear models at ages of from 0.5 to 5.0 gyrs and for the [ fe / h ] = 0.0 and 0.5 . pont et al . ( 2005,2006 ) suggest that the ages of these vlms are probably less than @xmath28 gyrs . if this is the case , as fig . [ ogle ] suggests , most of our models provide acceptable fits , with the best fits for ogle - tr-123b given by either the cloudy or the clear models with [ fe / h ] = 0.5 . higher metallicities would fit ogle - tr-123b even better , but all models fit ogle - tr-122b . the large radius error bars , as well as the ambiguity in the age of the systems , limit the strength of any general conclusions . moreover , given the difficulty of these measurements , one should not discount the possibility that these radii might be revised in the future . not shown on fig . [ ogle ] are models for [ fe / h ] = -0.5 ( see fig . [ rvsm_clear_cloudy ] ) , which predict slightly smaller radii at a given mass and age . nevertheless , whatever the age of these ogle stars , and in this stellar mass realm , the set of different models ( all reasonable in their character and inputs ) we have calculated allow a spread in expected radii of @xmath010% . in this paper , we have generated a collection of evolutionary models for brown dwarfs and very - low - mass stars for different atmospheric metallicities , with and without clouds . these models employ realistic atmosphere boundary conditions that allow us to consistently predict , given a detailed opacity model , the evolution of the object s radius . we have sought to demonstrate that with transit or eclipse radius measurements one is testing a multi - parameter theory , and not a universal radius@xmath4mass relation . the spread in radius at a given mass can be as large as @xmath00.1 to @xmath00.25 ( or @xmath010% to @xmath025% ) , with higher - metallicity , higher - cloud - thickness atmospheres resulting quite naturally in larger radii , all else being equal . for each 0.1 dex increase in atmospheric [ fe / h ] , the radius is expected to increase by @xmath01% to @xmath02.5% , depending upon age and mass . therefore , in order to constrain the hydrogen - helium equation of state one must control for the metallicity and cloud model . if the goal is to test structural theory and the viability of a suite of theoretical models , absent measurements of , for example , the metallicity , and a good constraint on the age , any radius measurement is of correspondingly limited utility . conversely , one should expect a range of radii for the natural range of metallicities and possible cloud properties expected for substellar objects and vlm stars in the solar neighborhood . in addition , we have calculated the effect of helium fraction on brown dwarf and vlm radii and find that , while for smaller masses and older ages radius decreases with increasing helium fraction ( as expected ) , for more massive brown dwarfs and a wide range of ages it increases with helium fraction . this runs counter to common lore , which expects that higher mean molecular weights universally result in smaller radii . we find that the increase in radius in going from @xmath2 to @xmath3 can be as large as @xmath00.025 ( @xmath02.5% ) . furthermore , we suggest that properly including the trace of heavy elements in the core eos should further augment this effect . we do not suggest that the cloud model we constructed for this investigation is definitive , nor uniquely applicable . rather , we engineered a cloud model , and its corresponding opacities and optical depths , to be representative of the generic effect of the clouds we know reside in brown dwarf atmospheres . we have sought merely to demonstrate that the presence or absence of clouds has an important effect on the radius of a brown dwarf . we note that many other cloud models can be constructed which may prove in the long run to be more viable . nevertheless , the qualitative effect of clouds that we have highlighted is robust . similarly , and more straightforwardly , atmospheric metallicity has a direct and clear effect on brown dwarf radii that needs to be accounted for in any interpretation of radius measurements . ten to twenty - five percent variations in radius exceed any reasonable error stemming from uncertainities in the equation of state alone and serve to emphasize that measurements of brown dwarf radii constrain a collection of effects , importantly including the atmosphere and condensate cloud models . without an independent measure of the atmospheric metallicity , constraints on the surface clouds , and a good age estimate , a measurement of a brown dwarf radius may be more difficult to interpret than previously thought . increasing the atmospheric metallicity of the vlm stars we studied in this paper from 0.0 to 0.5 increases their radii by @xmath04% . if we increase their atmospheric metallicitiy from -0.5 to 0.5 , their radii increase by @xmath010% . the latter percentage is slightly above what is often quoted as the discrepancy in radius between measurement and theory in the vlm regime ( morales et al . 2009 ; carter et al . 2011 ) . though we do not in this paper discuss models in the @xmath00.2 to 0.25 band , the clear implication of the systematic behavior we have derived in the stellar realm is that opacity due to higher metallicity might naturally account for the apparent radius anomalies in some eclipsing vlm systems . be that as it may , the effect of metallicity and clouds on the radii of brown dwarfs and vlms is straightforward and natural . therefore , we suggest it is necessary to incorporate these extra degrees of freedom into any interpretation of brown dwarf and vlm radius measurements and into any attempts usefully to constrain their equation of state . we acknowledge useful conversations with dave spiegel and nikku madhudsudhan , and ivan hubeny for his general support of the cooltlusty code . we also acknowledge support in part under nasa atp grant nnx07ag80 g , hst grants hst - go-12181.04-a and hst - go-12314.03-a , and jpl / spitzer agreements 1417122 , 1348668 , 1371432 , and 1377197 . baglin , a. , auvergne , m. , boisnard , l. , lam - trong , t. , barge , p. , catala , c. , deleuil , m. , michel , e. , & weiss , w. 2006 , 36th cospar scientific assembly , 36 , 3749 . bakos , g.a . , lzr , j. , papp , i. , sri , p. , & green , e.m . 2002 , , 114 , 974 johnson , j.a . , apps , k. , gazak , j.z . , crepp , j. , crossfield , i.j . , howard , a.w . , marcy , g.w . , morton , t.d . , chubak , c. , and isaacson , h. 2010 , submitted to ( arxiv:1008.4141 ) koch , d.g . 2010 , , 713 , l79 konacki , m. , torres , g. , jha , s. , & sasselov , d. 2003 , nature , 421 , 507 morales , j.c . 2009 , , 691 , 1400 pollacco , d. et al . 2004 , , 385 , 1576 pont , f. , melo , c.h.f . , bouchy , f. , udry , s. , queloz , d. , mayor , m. , & santos , n.c . 2005 , , 433 , l21 pont , f. , moutou , c. , bouchy , f. , behrend , r. , mayor , m. , udry , s. , queloz , d. , santos , n. , & melo , c. 2006 , , 447 , 1035 saumon , d. , chabrier , g. , & van horn , h. 1995 , , 99 , 713 = 0.0 [ dashed ] and 0.5 [ solid ] , shown in red ) and 2 ) clear atmospheres ( at [ fe / h ] = 0.0 [ dashed ] , shown in black ) . ( see section [ method ] for a discussion of the cloud prescription . ) the four masses shown are @xmath74 , @xmath75 , @xmath76 , and @xmath77 solar mass ( or approximately @xmath90 , @xmath91 , @xmath92 , and @xmath93 jupiter masses ) . the helium mass fraction ( @xmath1 ) for each model is here set to @xmath94 , i.e. , roughly the solar value . this figure shows that the radius of a brown dwarf is an increasing function of metallicity and is larger for models with clouds , all else being equal . three of the brown dwarfs which we highlight in this paper ( lhs 6343c , wasp-30b , and corot-15b ) have masses within this range and are included ( in blue ) on the figure , with putative observational error bars . the formal errors in the age are the most uncertain . as this figure indicates , before @xmath01 gyr , the radius evolves quickly , but afterwards significantly decelerates its shrinkage . ] , @xmath95 , @xmath96 , and @xmath97 gyr . notice the slight peak near @xmath04 m@xmath81 , the decrease in radius with increasing mass in the brown dwarf regime " at greater masses , and the monotonic radius shrinkage with increasing age . the curves start to rise near the main - sequence edge ( near @xmath075 m@xmath81 ) and ogle-122b and ogle-123b are clearly stars . six objects ( four brown dwarfs , two vlms ) are shown in blue ( with putative mass and age error bars ) : lhs 6343c ( johnson et al . 2010 ) , wasp-30b ( anderson et al . 2011 ) , corot-3b ( deleuil et al . 2008 ) , corot-15b ( bouchy et al . 2011 ) , ogle - tr-122b ( pont et al . 2005 ) , and ogle - tr-123b ( pont et al . 2006 ) . the quoted age of lhs 6343 is between one and five billion years . wasp-30b is suspected to be younger ( see fig.[fig : evolution ] ) . pont et al ( 2005,2006 ) suggest that the ogle objects might be younger than 0.5 gyrs , and , hence , their measured radii might be consistent with these models . the age of corot-3 is not well constrained . therefore , with the suggested age range of @xmath01.14@xmath43.35 gyr for the star corot-15 ( bouchy et al . 2011 ) , corot-15b is the only object in this set for which a solution using these heritage models may be problematic . see the text for a discussion . ] . the heritage models from burrows et al . ( 1997 ) ( in green ) are included for comparison . cloudy - atmosphere models are shown in red and clear - atmosphere models are shown in black . the clear - atmosphere models have metallicities ( [ fe / h ] ) of -0.5 , 0.0 , and + 0.5 , while the cloudy - atmosphere models have metallicities of 0.0 and + 0.5 . a general trend is that the presence of clouds retards shrinkage , as does higher metallicity . the model with [ fe / h ] = 0.5 and a cloudy atmosphere generally has the largest radii and the variation in radii among the models shown can be 0.1@xmath40.25 r@xmath81 at a given mass and age . ] , but highlighting the differences between models with two different helium fractions , @xmath2 ( blue / aqua ) and @xmath3 ( magenta ) . the gas - phase metallicity is solar for all models shown . as a reference , the model from burrows et al . ( 1997 ) is also plotted ( in green ) . the solid lines represent models with clouds , while the dashed lines are those with clear atmospheres . the solid ( cloudy ) models are generally larger than the corresponding dashed ( clear ) models . interestingly , the magenta ( @xmath3 ) models are larger than the blue / aqua ( @xmath2 ) models in the mass range above @xmath055 - 60 at later ages ( 3.0 and 5.0 gyr ) and above @xmath035 - 40 at earlier ages ( 0.5 and 1.0 gyr ) . this is counter to common lore , which suggests that planets with a higher molecular weight and lower electron fraction should be smaller . this is true only for cold planets . note that the burrows et al . ( 1997 ) models were calculated for @xmath2 ( saumon & marley 2008 ) . see text for a discussion . ] = 0.5 ( top two panels ) and [ fe.h ] = 0.0 ( bottom two panels ) . all the models shown are for a helium fraction , @xmath1 , of 0.28 . on each panel are shown isochrones at 0.5 , 1.0 , 3.0 , and 5.0 gyrs . superposed on all panels is the data point for lhs 6343c at @xmath89 and @xmath98 ( johnson et al . 2010 ) . the metallicity of lhs 6343c s primary , lhs 6343a , is suggested by anderson et al . ( 2011 ) to be [ fe / h ] = @xmath12 , i.e. super - solar , but johnson et al . ( 2010 ) quote a value of @xmath13 . clear models for [ fe / h ] = 0.0 ( solar ) and [ fe / h ] = 0.5 metallicities fit for ages greater than @xmath02 gyrs , but the cloudy model with @xmath83 and [ fe / h ] = 0.0 also fits for late ages . models with @xmath82 fit slightly better , but given the systematic observational uncertainties , nothing substantive can be said about @xmath1 for lhs 6343c . the new model with clouds and super - solar metallicity does not fit for ages less than @xmath07 gyr . as fig.[burrows97 ] demonstrates , the burrows et al . ( 1997 ) models fit well for the suggested age of @xmath02 gyrs . see the text for a discussion of the implications of all these model comparisons . ] , but for wasp-30b and with the [ fe / h ] = 0.5 models at the bottom and the [ fe / h ] = 0.0 models at the top . ( note that a slightly different range of radii on the ordinate is used . ) as in fig . [ lhs6343 ] , all the models shown are for a helium fraction of 0.28 and isochrones at 0.5 , 1.0 , 3.0 , and 5.0 gyrs are plotted . superposed on all panels is the data point for wasp-30b at @xmath20 and @xmath21 ( anderson et al . 2011 ) . the metallicity of wasp-30 is quoted to be [ fe / h ] = @xmath22 , basically solar . a variety of models and age@xmath4metallicity combinations fit the wasp-30b data . clear models with [ fe / h ] = 0.0 fit well for @xmath01@xmath42 gyrs . clear models with [ fe / h ] = 0.5 fit well for ages from @xmath02 to @xmath03 gyrs . cloudy models with [ fe / h ] = 0.0 fit well for ages of @xmath88 gyrs and our cloudy model with [ fe / h ] = 0.5 still fits near ages of @xmath05 gyrs . cloudy models with @xmath82 fit at slightly younger ages . as fig . [ burrows97 ] suggests , the heritage models from burrows et al . ( 1997 ) fit wasp-30b for an age near @xmath01 gyr . see the text for a discussion of these fits and conclusions concerning wasp-30b . ] , but for corot-15b and with slightly different ranges for the radius and mass axes . the mass of corot-15b is measured to be @xmath89 and its radius is measured to be @xmath15 ( bouchy et al . 2011 ) . corot-15 has an estimated metallicity of @xmath16 , again basically solar . bouchy et al . ( 2011 ) suggest that its parent star has an age in the range @xmath01.14@xmath43.35 gyr . our solar - metallicity ( [ fe / h ] = 0.0 ) models can fit the lower age range to @xmath01-@xmath86 to @xmath01.5-@xmath86 , with the best fit for the [ fe / h ] = 0.0 cloudy model . however , our [ fe / h ] = 0.5 models fit the suggested age range better , with the clear [ fe / h ] = 0.5 models fitting an age of @xmath01 gyr within @xmath01-@xmath86 and the cloudy [ fe / h ] = 0.5 models fitting anywhere in the suggested age range . cloudy models with [ fe / h ] = 0.0 and clear models with [ fe / h ] = 0.5 fit corot-15b almost equally well . figure [ burrows97 ] suggests that the burrows et al . ( 1997 ) solar - metallicity models would fit only for very young ages less than @xmath00.5 gyrs . see the text for details . ] , [ wasp30 ] , and [ corot15 ] , but for corot-3b and for a lower brown - dwarf mass range between 5 and 40 . the clear models are on the left panels and the cloudy models are on the right panels . the top models are for [ fe / h ] = 0.0 and the bottom models are for [ fe / h ] = 0.5 . the measured mass of corot-3b is @xmath17 and its measured radius is @xmath18 ( deleuil et al . see text for a discussion . ] = 0.0 ( top panels ) and 0.5 ( bottom panels ) and for clear ( left panels ) and cloudy ( right panels ) models . the data for ogle - tr-122b and ogle - tr-123b are taken from pont et al . ( 2005 ) and pont et al . ( 2006 ) , respectively . since these objects have measured masses of @xmath24 ( ogle - tr-122b ) and @xmath25 ( ogle - tr-123b ) , the plots are for a mass range from 60 to 120 . the measured radii are @xmath26 and @xmath27 for ogle - tr-122b and ogle - tr-123b , respectively , and the plotted radius range is 0.8 to 1.6 . see text for a discussion of the issues involved . ]

we show that the unit tangent bundle of and a real cohomology admit riemannian metrics with positive sectional curvature almost everywhere . these are the only examples so far with positive curvature almost everywhere that are not also known to admit positive curvature .

math9910187

-0.05 in a manifold is said to have quasi - positive curvature if the curvature is nonnegative everywhere and positive at a point . in analogy with aubin s theorem for manifolds with quasi - positive ricci curvature one can use the ricci flow to show that any manifold with quasi - positive scalar curvature or curvature operator can be deformed to have positive curvature . by contrast no such result is known for sectional curvature . in fact , we do not know whether manifolds with quasi - positive sectional curvature can be deformed to ones with positive curvature almost everywhere , nor is it known whether manifolds with positive sectional curvature almost everywhere can be deformed to have positive curvature . what is more , there are only two examples ( @xcite and @xcite ) of manifolds with quasi - positive curvature that are not also known to admit positive sectional curvature . the example in @xcite is a fake quaternionic flag manifold of dimension @xmath2 . it is not known if it has positive curvature almost everywhere . the example in @xcite is on one of milnor s exotic @xmath3-spheres . it was asserted without proof , in @xcite , that the gromoll meyer metric has positive sectional curvature almost everywhere , but this assertion was disproven by mandell , a student of gromoll ( @xcite , cf also @xcite ) . thus there is no known example of a manifold with positive sectional curvature at almost every point that is not also known to admit positive curvature . we will rectify this situation here by proving the following theorem . [ unit tangent bundle ] the unit tangent bundle of @xmath0 admits a metric with positive sectional curvature at almost every point with the following properties . * the connected component of the identity of the isometry group is isomorphic to @xmath4 and contains a free @xmath5subaction . * the set of points where there are @xmath6 sectional curvatures contains totally geodesic flat @xmath7tori and is the union of two copies of @xmath8 that intersect along a @xmath9 . * remark*in the course of our proof we will also obtain a precise description of the set of @xmath6curvature planes in the grassmannian . this set is not extremely complicated , but the authors have not thought of a description that is succinct enough to include in the introduction . * remark*in the sequel to this paper , @xcite , the second author shows that the metric on the gromoll meyer sphere can be perturbed to one that has positive sectional curvature almost everywhere . in contrast to @xcite the metric we construct here is a perturbation of a metric that has zero curvatures at every point . by taking a circle subgroup of the free @xmath5action in theorem [ unit tangent bundle](1 ) we get the following . [ 6-dim quot ] there is a manifold @xmath10 with the homology of @xmath1 but not the cohomology ring of @xmath1 , that admits a metric with positive sectional curvature almost everywhere . there are also flat totally geodesic @xmath7tori in @xmath10 so as a corollary of lemma 4.1 in @xcite ( cf also proposition 3 in @xcite ) we have the following . [ no 1st order perturbation ] there are no perturbations of our metrics on the unit tangent bundle and @xmath10 whose sectional curvature is positive to first order . that is , there is no smooth family of metrics @xmath11 with @xmath12 the metric in theorem [ unit tangent bundle ] or corollary [ 6-dim quot ] so that @xmath13 for all planes @xmath14 that satisfy @xmath15 . * remark * although the unit tangent bundle of @xmath16 is a homogeneous space ( see below ) , the metric of theorem [ unit tangent bundle ] is obviously inhomogeneous . what is more if the unit tangent bundle of @xmath16 admits a metric with positive sectional curvature , then it must be for some inhomogenous metric see @xcite . the space in corollary [ 6-dim quot ] is a biquotient of @xmath17 . it follows from ( * ? ? ? * theorem 6 ) that @xmath10 does not have the homotopy type of a homogeneous space . before outlining the construction of our metric we recall that the @xmath5bundles over @xmath0 are classified by @xmath18 as follows ( @xcite , @xcite ) . the bundle that corresponds to @xmath19 is obtained by gluing two copies of @xmath20 together via the diffeomorphism @xmath21 given by @xmath22 where we have identified @xmath23 with @xmath24 and @xmath5 with @xmath25 . we will call the bundle obtained from @xmath26 `` the bundle of type @xmath27 '' , and we will denote it by @xmath28 . translating theorem 9.5 on page 99 of @xcite into our classification scheme ( [ gluing map ] ) shows that the unit tangent bundle is of type @xmath29 . we will show it is also the quotient of the @xmath5action on @xmath17 given by @xmath30 ( it was shown in @xcite that this quotient is also the total space of the bundle of type @xmath31 , so we will call it @xmath32 . ) the quotient of the biinvariant metric via @xmath33 is a normal homogeneous space with nonnegative , but not positive sectional curvature . to get the metric of theorem [ unit tangent bundle ] we use the method described in @xcite to perturb the biinvariant of @xmath17 using the commuting @xmath5actions @xmath34 we call the new metric on @xmath17 , @xmath35 , and will observe in proposition [ s^3 times s^3 times s^3 times s^3 ] that @xmath33 is by isometries with respect to @xmath35 . our metric on the unit tangent bundle is the one induced by the riemannian submersion @xmath36 . in section 1 we review some generalities of cheeger s method . in section 2 we study the symmetries of @xmath32 . in section 3 we analyze the infinitesimal geometry of the riemannian submersion @xmath37 , given by projection onto the first column . this will allow us to compute the curvature tensor of the metric , @xmath38 , obtained by perturbing the biinvariant metric via @xmath39 and @xmath40 . in section 4 we compute the @xmath41tensor of the hopf fibration @xmath42 , because it is the key to the geometry of @xmath38 . in section 5 we specify the zero curvatures of @xmath38 , and in section 6 we describe the horizontal space of @xmath43 with respect to @xmath38 and hence ( via results from section 1 ) with respect to @xmath44 . in section 7 we specify the zero curvatures of @xmath32 , first with respect to @xmath38 and then with respect to @xmath44 , proving theorem [ unit tangent bundle ] . in section 8 we establish the various topological assertions that we made above , that @xmath32 is the total space of the unit tangent bundle and that while the cohomology modules of @xmath45 are the same as @xmath1 s the ring structure is different . using these computations we will conclude that @xmath32 and @xmath45 do not have the homotopy type of any known example of a manifold of positive curvature . we assume that the reader has a working knowledge of oneill s `` fundamental equations of a submersion '' @xcite and the second author s description of the tangent bundle of @xmath17 , @xcite , we will adopt results and notation from both papers , in most cases without further notice . it will also be important for the reader to keep the definitions of the five @xmath5 actions @xmath46 , @xmath47 , @xmath48 , @xmath40 and @xmath33 straight . to assist we point out that the letters @xmath49 , @xmath50 , @xmath51 and @xmath52 stand for `` up '' , `` down '' , `` left '' and `` right '' and are meant to indicate the row or column of @xmath17 that is acted upon . * acknowledgments * we are grateful to claude lebrun , wolfgang ziller , and the referee for several thoughtful and constructive criticisms of the first draft of this paper . the first author is supported in part by the nsf . support from national science foundation grant dms-9803258 is gratefully acknowledged by the second author . in @xcite a general method for perturbing the metric on a manifold , @xmath53 , of nonnegative sectional curvature is proposed . various special cases of this method were first studied in @xcite and @xcite . if @xmath54 is a compact group of isometries of @xmath53 , then we let @xmath54 act on @xmath55 by @xmath56 if we endow @xmath54 with a biinvariant metric and @xmath57 with the product metric , then the quotient of ( [ skew action ] ) is a new metric of nonnegative sectional curvature on @xmath53 . it was observed in @xcite , that we may expect the new metric to have fewer @xmath6 curvatures and symmetries than the original metric . in this section we will describe the effect of certain cheeger perturbations on the curvature tensor of @xmath53 . most of the results are special cases of results of @xcite , we have included them because they can be described fairly succinctly and are central to all of our subsequent computations . the quotient map for the action ( [ skew action ] ) is @xmath58 the vertical space for @xmath59 at @xmath60 is @xmath61 where the @xmath62 in the first factor stands for the value at @xmath63 of the killing field on @xmath54 given by the circle action @xmath64 and the @xmath65 in the second factor is the value of the killing field @xmath66 on @xmath53 at @xmath67 . until further notice all cheeger perturbations under consideration will have the property : @xmath68 where @xmath69 is the lie algebra of @xmath54 . notice that @xmath46 , @xmath47 , @xmath40 and @xmath48 have this property . in this case the horizontal space for @xmath70 is the direct sum @xmath71 where @xmath72 and @xmath73 and @xmath74 is the space that is normal to the orbit of @xmath54 . the image of @xmath75 under @xmath76 is @xmath77 it follows that the effect of cheeger s perturbation is to keep @xmath74 perpendicular to the orbits of @xmath54 , to keep the metric restricted to @xmath74 unchanged and to multiply the length of the vector @xmath78 by the factor @xmath79 if @xmath80 is a fixed biinvariant metric on @xmath54 and @xmath81 is a positive real number , then we let @xmath82 denote the metric we obtain on @xmath53 via the riemannian submersion @xmath83 when the metric on the @xmath54factor in @xmath57 is @xmath84 . when @xmath85 , @xmath80 will always be the unit metric . as pointed out in ( [ orbit length change ] ) , @xmath82 converges to the original metric , @xmath86 , as @xmath87 . to emphasize this point we will also denote @xmath86 by @xmath88 . let @xmath89 denote the tangent distribution in @xmath53 to the orbits of @xmath54 . then any plane @xmath14 tangent to @xmath53 can be written as @xmath90 where @xmath91 and @xmath92 . we let @xmath93 denote the plane in @xmath94 that is horizontal with respect to @xmath70 and satisfies @xmath95 where @xmath96 denotes the projection onto the second factor . if @xmath97 , then @xmath98 has the analogous relationship to @xmath99 as @xmath93 has to @xmath14 . as pointed out in @xcite we have the following result . [ cheeger s curvature condition ] let @xmath100 and @xmath101 denote the lengths of the killing fields on @xmath54 and @xmath53 corresponding to @xmath102 via the procedures described in ( [ right field on g ] ) and ( [ left field on m ] ) . let @xmath103 and @xmath104 have the analogous meaning with respect to @xmath105 . * if the curvature of @xmath14 is positive with respect to @xmath88 , then the curvature of + @xmath106 + is positive with respect @xmath82 . * the curvature of @xmath107 is positive with respect to @xmath82 if the @xmath41tensor , @xmath108 , of @xmath70 is nonzero on @xmath93 . * if @xmath109 , then the curvature of @xmath107 is positive if the projection of @xmath14 onto @xmath89 is nondegenerate . * if the curvature of @xmath93 is @xmath6 and @xmath108 vanishes on @xmath93 , then the curvature of @xmath107 is @xmath6 . \(2 ) and ( 4 ) are corollaries of oneill s horizontal curvature equation . to prove ( 1 ) notice that the curvature of @xmath107 is positive if the curvature of its horizontal lift , @xmath93 , is positive . the curvature of @xmath93 is positive if its image , @xmath14 , under @xmath110 has positive curvature , proving ( 1 ) . let @xmath111 be the projection onto the first factor . the curvature of @xmath93 is also positive if its image under @xmath112 is positively curved . if @xmath113 , then this is the case , provided the image of @xmath114 is nondegenerate , proving ( 3 ) . using ( [ orbit length change ] ) we get the following . [ new horizontal space ] let @xmath115 be an action that is by isometries with respect to both @xmath88 and @xmath82 . let @xmath116 denote the distribution of vectors that are perpendicular to the orbits of @xmath117 . @xmath14 is in @xmath116 with respect to @xmath88 if and only if @xmath107 is in @xmath116 with respect to @xmath82 . just combine our description of @xmath82 with the observation that if we square the expression in ( [ orbit length change ] ) we get the reciprocal of the quantity in ( [ perturbed p ] ) . ultimately we will be studying cheeger perturbations via commuting group actions , @xmath118 , that individually have property ( [ simple cheeger perturbations ] ) . generalizing our formulas to this situation is a simple matter once we observe the following result . [ commuting cheeger perturbations ] let @xmath119 act isometrically on @xmath120 . fix biinvariant metrics @xmath121 and @xmath122 , on @xmath123 and @xmath124 . let @xmath125 and @xmath126 be the metrics obtained by doing cheeger perturbations of @xmath127 with @xmath123 and @xmath124 respectively . * @xmath124 acts by isometries on @xmath128 and @xmath123 acts by isometries on @xmath129 . * let @xmath130 denote the metric obtained by doing the cheeger perturbation with @xmath124 on @xmath128 and let @xmath131 denote the metric obtained by doing the cheeger perturbation with @xmath123 on @xmath129 . then @xmath132 in fact @xmath130 coincides with the metric obtained by doing a single cheeger perturbation of @xmath127 with @xmath119 . * @xmath133 is horizontal for @xmath134 with respect to @xmath135 if and only if @xmath136 is horizontal for @xmath137 with respect to @xmath138 and @xmath139 is horizontal for @xmath140 with respect to @xmath141 . the proof of ( 1 ) is a routine exercise in the definitions which we leave to the reader . part ( 1 ) gives us a commutative diagram @xmath142 of riemannian submersions from which ( 2 ) readily follows . it follows from the diagram that if @xmath143 is horizontal for @xmath144 with respect to @xmath135 , then @xmath145 is horizontal for @xmath146 with respect to @xmath147 and @xmath148 is horizontal for @xmath149 with respect to @xmath150 . this proves the `` only if '' part of ( 3 ) . the `` if '' part of ( 3 ) follows from the `` only if '' part and the observation that @xmath151 via a dimension counting argument . rather than changing the metric of @xmath32 directly with a cheeger perturbation , we will change the metric on @xmath17 and then mod out by @xmath33 . the constraint to this approach is that the cheeger perturbations that we use can not destroy the fact that @xmath33 is by isometries . fortunately it was observed in @xcite that if the metric on the @xmath54 factor in @xmath57 is biinvariant , then @xmath54 acts by isometries with respect to @xmath82 . therefore we have the following result . [ s^3 times s^3 times s^3 times s^3 ] let @xmath44 denote a metric obtained from the biinvariant metric on @xmath17 via cheeger s method using the @xmath152action , @xmath153 . then @xmath153 is by isometries with respect to @xmath44 . in particular , @xmath33 is by isometries with respect to @xmath44 . we include a proof of proposition [ s^3 times s^3 times s^3 times s^3 ] even though it follows from an assertion on page 624 of @xcite . we do this to establish notation that will be used in the sequel , and because the assertion in @xcite was not proven . throughout the paper we will call the tangent spaces to the orbits of @xmath48 and @xmath40 , @xmath154 and [ v_1 , v_2 , h - page ] @xmath155 . the orthogonal complement of @xmath156 with respect to the biinvariant metric will be called @xmath157 . according to proposition 2.1 in @xcite , @xmath17 is diffeomorphic to the pull back of the hopf fibration @xmath158 via @xmath159 , where @xmath160 is the antipodal map and @xmath158 is the hopf fibration that is given by right multiplication by @xmath5 . moreover , the metric induced on the pull back by the product of two unit @xmath161 s is biinvariant . through out the paper our computations will be based on perturbations of the biinvariant metric , @xmath162 , induced by @xmath163 , where @xmath164 is the sphere of radius @xmath165 . observe that if @xmath65 is a killing field on @xmath5 whose length is @xmath166 with respect to the unit metric , then the corresponding killing field on @xmath17 with respect to either @xmath48 or @xmath40 has length @xmath165 with respect to @xmath162 . it follows from this that the quantity ( [ orbit length change ] ) is constant when we do a cheeger perturbation on @xmath162 via either @xmath48 or @xmath40 . thus the effect of these cheeger perturbations is to scale @xmath154 and @xmath155 , and to preserve the splitting @xmath167 and @xmath168 . the amount of the scaling is @xmath169 and converges to @xmath166 as the scale , @xmath81 , on the @xmath5factor in @xmath170 converges to @xmath171 and converges to @xmath6 as @xmath172 . we will call the resulting scales on @xmath154 and @xmath155 , @xmath173 and @xmath174 , and call the resulting metric @xmath38 . with this convention the biinvariant metric @xmath162 is @xmath175 . it follows that @xmath38 is the restriction to @xmath17 of the product metric @xmath176 where @xmath177 denotes the metric obtained from @xmath178 by scaling the fibers of @xmath179 by @xmath180 . since @xmath48 and @xmath40 are by symmetries of @xmath179 in each column , they are by isometries on @xmath176 and hence also on @xmath181 . let @xmath182 denote the metric obtained from @xmath162 via the cheeger perturbation with @xmath183 when the metric on the @xmath184factor in @xmath185 is @xmath186 . an argument similar to the one above , using rows instead of columns , shows that @xmath187 is by isometries on @xmath188 . since @xmath187 commutes with @xmath189 , it follows that @xmath187 acts by isometries on @xmath181 . doing a cheeger perturbation with @xmath187 on @xmath190 produces a metric @xmath191 which can also be thought of as obtained from @xmath17 via a single cheeger perturbation with @xmath192 . since @xmath189 acts by isometries on @xmath181 and commutes with @xmath187 , @xmath189 acts by isometries with respect to @xmath191 . but @xmath191 can also be obtained by first perturbing with @xmath187 and then perturbing with @xmath189 . so repeating the argument of the proceeding paragraph shows that @xmath187 acts by isometries with respect to @xmath191 . it follows that @xmath193 is a riemannian submersion with respect to both @xmath38 and @xmath191 . we will abuse notation and call the induced metrics on @xmath32 , @xmath38 and @xmath191 . since @xmath189 commutes with @xmath33 and is by isometries on @xmath194 , it is by isometries on @xmath195 . however on the level of @xmath32 it has a kernel that at least contains @xmath196 . to see this just observe that the action of @xmath197 on @xmath17 via @xmath189 is the same as the action of @xmath198 via @xmath33 . it turns out that the kernel is exactly @xmath196 , and that @xmath189 induces the @xmath4action whose existence was asserted in theorem [ unit tangent bundle](1 ) . [ a^l times a^r on e_2,0 ] * @xmath48 and @xmath40 act freely on @xmath32 . * @xmath199 is diffeomorphic to @xmath0 and the quotient map @xmath200 is the bundle of type @xmath31 . * @xmath40 acts by symmetries of @xmath201 . the induced map on @xmath0 is the join of the standard @xmath196ineffective @xmath5action on @xmath202 with the trivial action on @xmath203 . * the kernel of the action @xmath189 on @xmath32 is generated by @xmath204 , so @xmath189 induces an effective @xmath4action on @xmath32 . it is easy to see that @xmath205 and @xmath206 are free on @xmath17 and hence that @xmath48 and @xmath40 are free on @xmath32 , proving ( 1 ) . it was observed in @xcite that @xmath207 is diffeomorphic to @xmath0 and in @xcite that @xmath201 is the bundle of type @xmath31 . combining these facts proves ( 2 ) . part ( 3 ) is a special case of proposition 5.5 in @xcite . it follows from ( 3 ) that if @xmath208 is in the kernel of @xmath189 , then @xmath209 . on the other hand , @xmath48 is the principal @xmath5action for @xmath210 . combining these facts we see that the kernel has order @xmath211 , and we observed above that @xmath197 is in the kernel . the @xmath212 action on @xmath17 , @xmath213 , that is given by @xmath214 commutes with @xmath33 and so is by symmetries of @xmath43 . since it also commutes with @xmath189 it acts by isometries on @xmath215 . it is also acts by symmetries of @xmath201 according to proposition 5.5 in @xcite . we shall see , however , that it is not by isometries on @xmath216 . ( note that it commutes with neither @xmath46 nor @xmath47 . this is one of the central ideas behind the curvature computations in the proof of theorem [ unit tangent bundle ] . ) we may nevertheless use @xmath217 to find the @xmath6curvatures of @xmath215 . ( we will then use proposition [ cheeger s curvature condition ] to see that most of them are not @xmath6 with respect to @xmath218 . ) a special case of proposition 5.7 in @xcite is the following . [ isometric reduction ] every point in @xmath32 has a point in its orbit under @xmath219 that can be represented in @xmath17 by a point of the form @xmath220 with @xmath221 , re(\alpha ) = 0 $ ] and @xmath222 . we will call points of the form ( [ modulo isometries ] ) representative points . * notational convention * we have seen that @xmath40 , @xmath48 and @xmath223 all induce actions on @xmath32 , and that @xmath40 and @xmath223 even induce actions on @xmath0 . to simplify the exposition we will make no notational distinction between these actions and their induced actions . thus @xmath40 stands for an action on @xmath17 , @xmath32 , or @xmath0 . the space that is acted on will be clear from the context . we will study the curvature of ( @xmath17 , @xmath38 ) by analyzing the geometry of the riemannian submersion @xmath224 given by projection onto the first column . the moral of our story is that the geometry of @xmath225 resembles the geometry of @xmath226 very closely . @xmath225 is a principal bundle with a `` connection metric '' . that is the metric is of the form @xmath227 where @xmath228 is a principal bundle , @xmath229 is a biinvariant metric on @xmath54 , @xmath230 is an arbitrary metric on @xmath231 , and @xmath232 is a connection map . in the case of @xmath225 , @xmath233 and the action is given by @xmath40 . it will be important for us to understand how the infinitesimal geometry of @xmath225 changes as @xmath234 and @xmath235 change . combining 2.12.3 of @xcite with a rescaling argument we get the following . [ nash ] let @xmath236 be a principal bundle with a connection metric @xmath237 . let @xmath238 , @xmath239 and @xmath240 denote the covariant derivative , @xmath41tensor and curvature tensor of @xmath241 . if the paremeter @xmath242 is omitted from the superscript of one of the objects @xmath238 , @xmath243 , @xmath240 , or @xmath244 , then implicity that parameter has value @xmath166 . if @xmath245 , @xmath246 and @xmath247 are horizontal fields and @xmath248 , @xmath249 and @xmath250 are vertical fields , then : * the fibers of @xmath251 are totally geodesic . * @xmath252 , @xmath253 , @xmath254 , @xmath255 , @xmath256 , * @xmath257 @xmath258 , * @xmath259 , * @xmath260 , * @xmath261 , and * @xmath262 . the next step is to determine the @xmath41 and @xmath263 tensors of @xmath225 . the vertical space of @xmath225 is @xmath264 where @xmath265 is the vertical space for the hopf fibration @xmath226 that is given by left quaternionic multiplication . we will also denote vectors in @xmath266 by @xmath267 , and we will often abuse notation and write just @xmath268 for @xmath269 . * tensor * given two vector fields , @xmath270 , @xmath271 , with values in @xmath266 we compute @xmath272 since @xmath273 is totally geodesic . therefore @xmath274 * the a tensor * @xmath275 splits as the direct sum @xmath276 where @xmath154 and @xmath157 are as defined on page . so any vector field with values in @xmath275 can be written uniquely as @xmath277 where @xmath278 takes values in @xmath157 and @xmath279 takes values in @xmath154 . given two such vector fields @xmath280 and @xmath281 we compute @xmath282^v \ ) = \\ ( \ 0 , \ \nabla^{1 } _ { dp_{2}^{2 } ( z_1 ) } dp_{2}^2 ( z_2 ) \ ) , \nonumber \end{aligned}\ ] ] where @xmath283 denotes the projection of @xmath17 onto its last factor . the second equality is due to the fact that a vector in @xmath155 is @xmath6 in the first entry and a vector in @xmath154 is @xmath6 in the last entry . the upshot of ( [ t - tensor of p_1 ] ) and ( [ a - tensor of p_k , k-1 ] ) is that the @xmath263 and @xmath41 tensors of @xmath225 are essentially the `` @xmath263 and @xmath41 tensors of @xmath179 in the last factor '' . the only difference is that @xmath284 has a @xmath285dimensional kernel , @xmath154 , and @xmath286 has kernel @xmath287 . this principal also holds for the vertizontal @xmath41tensor . if @xmath288 is a vector field with values in @xmath155 and @xmath289 is a vector field with values in @xmath290 , then @xmath291 the only components of the curvature tensor of @xmath17 which are not mentioned in ( [ nash ] ) are those of the form @xmath292 and @xmath293 . according to formulas @xmath294 and @xmath295 in @xcite these are @xmath296 and @xmath297 ( we use the opposite sign convention for the curvature tensor than oneill . ) if we choose @xmath298 and @xmath299 to be basic horizontal fields the first equation simplifies to @xmath300 where the superscript @xmath301 denotes the component in @xmath157 . it appears in the formula because the orthogonal complement of @xmath157 in the horizontal space for @xmath225 , @xmath154 , is the kernel of @xmath41 . similarly ( [ mixed curvature ] ) simplifies to @xmath302 according to oneill , the curvatures with exactly three horizontal terms are @xmath303 through out the rest of this section we will let the superscript @xmath304 denote the component in @xmath154 . by ( [ a - tensor of p_k , k-1 ] ) and the fact that the analogous equation holds for the hopf fibration we have @xmath305 of course @xmath306 since @xmath154 is the kernel of @xmath41 . using this again we get @xmath307 where the last equality is because @xmath308 , since the orbits of @xmath48 are totally geodesic . also @xmath309 keeping in mind that @xmath157 is the horizontal space for @xmath310 and that @xmath154 is in the vertical space for @xmath310 we can rewrite the right hand side of ( [ bad 3a prelim ] ) and get @xmath311 using ( [ bad 3a ] ) and the antisymmetry of the curvature tensor we get @xmath312 combining the first bianchi identity with ( [ bad 3a ] ) and ( [ bad 3b ] ) we find that @xmath313 it turns out that the sum of the two terms on the right is always @xmath6 and hence that @xmath314 this can be seen by direct computation combining proposition [ a - tensor of h ] ( below ) with section 6 of @xcite , but there are many details to check . a less direct , but much quicker proof begins by using the second equation in [ nash](2 ) to observe that the right hand side of ( [ 3 horiz , kernal 3prelim ] ) is @xmath315 times the corresponding curvature quantity for the biinvariant metric . therefore it suffices to prove ( [ 3 horiz , kernal 3prelim - b ] ) for the biinvariant metric . to do this observe that distributions @xmath154 and @xmath155 have the form @xmath316 @xmath317 where @xmath318 denotes the differential of left translation by @xmath319 . thus @xmath320 = 0 $ ] , where @xmath321 and @xmath322 denote the extensions of @xmath323 and @xmath248 to left invariant fields . therefore @xmath324,\ ; [ e_{3}^v , \sigma ] \rangle = 0 , \end{aligned}\ ] ] where @xmath325 and @xmath326 denote the extensions of @xmath327 and @xmath328 to left invariant fields , proving ( [ 3 horiz , kernal 3prelim - b ] ) . combining ( [ e^v_1 , e_2^v , e^3 ] ) , ( [ e^h_1 , e_2^v , e^3_v ] ) , and ( [ 3 horiz , kernal 3prelim - b ] ) gives us @xmath329 a review of the results of this section shows that the only difference between the infinitesimal geometries of @xmath179 and @xmath225 is the fact that for @xmath225 the terms ( [ bad 3a ] ) and ( [ bad 3b ] ) are not always @xmath6 but for @xmath179 the curvatures terms with exactly three horizontal vectors are always @xmath6 . this is the root cause of many of the @xmath6 curvatures in @xmath181 . in section 3 we showed that the @xmath41tensor of @xmath225 is essentially the @xmath41tensor of the hopf fibration , @xmath179 , in the last factor . @xmath330 certainly ought to be very well known , but the authors are not aware of any computation of it in the literature . so for completeness we prove the following . [ a - tensor of h ] let @xmath331 be a point in @xmath332 . * the vertical space of @xmath179 at @xmath331 , @xmath333 , is @xmath334 * if @xmath278 is in the horizontal space , @xmath335 , for @xmath336 at @xmath331 , then @xmath337 a proof of this for the case of the complex hopf fibration can be found in @xcite . the proof for the quaternionic case is nearly identical . \(1 ) is an immediate consequence of the definition @xmath179 . to prove ( 2 ) we let a superscript @xmath301 denote the horizontal component with respect to @xmath179 and we compute @xmath338 where for the last equality we have used the fact that @xmath339 is invariant under right quaternionic multiplication . because the metric @xmath38 is a simpler metric than @xmath340 , we study its @xmath6 curvatures in this section . where all statements about curvatures of @xmath17 are understood to be with respect to @xmath38 , for some @xmath341 . a plane @xmath14 tangent to @xmath17 has the form @xmath342 where @xmath343 and @xmath344 . [ generic planes ] * the curvature of @xmath14 is positive if the projection of @xmath14 on to one of @xmath157 , @xmath345 , or @xmath346 is two dimensional . * if all three projections in ( 1 ) are degenerate , then the curvature of @xmath14 is @xmath6 , if its projection onto @xmath157 is @xmath6 . * if all three projections in ( 1 ) are degenerate and the projection of @xmath14 onto @xmath157 is one dimensional , then we may assume that @xmath347 and @xmath348 . the curvature of @xmath14 is positive unless + @xmath349 * if all of the hypotheses of ( 3 ) hold and ( [ bad inner product2 ] ) holds , then the curvature of @xmath14 is @xmath6 . * remark * since @xmath350 is vertical for @xmath351 and @xmath352 , ( [ bad inner product2 ] ) can be thought of as @xmath353 . since the fibers of @xmath351 are totally geodesic this version of ( 3 ) is a corollary of [ nash](5 ) . we give an alternative proof below . if the projection of @xmath14 onto @xmath157 , @xmath154 and @xmath155 is degenerate , then by replacing the second vector with the appropriate linear combination of the vectors in ( [ generic form of p ] ) we may assume that @xmath14 has the form @xmath354 where @xmath355 and @xmath356 are multiples of each other . by replacing the first vector in ( [ reduced p ] ) by the appropriate vector in @xmath14 we can further assume that @xmath357 , so @xmath14 has the form @xmath358 where @xmath359 and @xmath360 are multiples of each other . it follows that @xmath361 where we have used the results of section 3 to conclude that many terms are @xmath6 . \(3 ) and ( 4 ) follow from ( [ vertizontal zeros ] ) . * notation * since the @xmath41tensors in ( [ vertizontal zeros ] ) are essentially the @xmath41tensors of @xmath179 in the first and second factor we will shorten the notation and set @xmath362 and we will denote the inner product in ( [ vertizontal zeros ] ) by @xmath363 the idea behind this notation is that we are taking a certain type of inner product between the @xmath41tensors of @xmath179 in the first and second factors . the subscript @xmath364 is meant to remind us of the role that the antipodal map of @xmath0 plays in determining this inner product . * proof of ( 1 ) * if the projection of @xmath14 onto either @xmath154 or @xmath155 is nondegenerate , then according to [ cheeger s curvature condition](3 ) the curvature of @xmath14 is positive . so we may assume that the projection of @xmath14 onto @xmath157 is nondegenerate and the projections onto both @xmath154 and @xmath155 are degenerate . since the projection of @xmath14 onto @xmath155 is degenerate we may replace the first vector with another vector in @xmath14 that satisfies @xmath365 since the projection onto @xmath154 is also degenerate we may replace our vectors by appropriate linear combinations to get that either @xmath366 if @xmath367 , @xmath14 is spanned by @xmath368 it follows that @xmath369 where @xmath370 denotes the component of @xmath371 that is perpendicular to @xmath278 , and we have relied heavily on results of the previous two sections to conclude that many of the components of this curvature are @xmath6 . using ( [ 3 horiz , kernal 3 ] ) and the symmetries of the curvature tensor we see that the fourth and the sixth terms on the right are @xmath6 . the last two terms are also @xmath6 . one way to see this is to combine ( [ fundamental equation ] ) with the fact that @xmath359 is in the kernel of @xmath372 . use ( [ bad 3a ] ) to evaluate the other two terms to get @xmath373 for @xmath374 let @xmath375 and @xmath376 where @xmath377 denotes the unit norm on @xmath161 . combining [ nash](5 ) , ( [ vertical a - tensor of p_k , k-1 ] ) and ( [ a - tensor of h ] ) and meticulously chasing through the definitions we get that @xmath378 from ( [ evaluating middle terms ] ) we see that sum of the last three terms in ( [ prelim v^1 = 0 ] ) is always nonnegative so @xmath379 where the last inequality is due to our hypothesis that the projection of @xmath14 onto @xmath157 is nondegenerate . it remains to consider the case @xmath380 in ( [ v^ = 0 or w^1 = 0 ] ) . if so , our plane is spanned by @xmath381 arguing as in ( [ prelim v^1 = 0 ] ) and ( [ final story v^1 = 0 ] ) we can show that @xmath382 @xmath383 @xmath384 @xmath385 if @xmath386 where @xmath387 and @xmath388 , then @xmath389 @xmath390 because the @xmath263tensor of @xmath225 is @xmath6 and @xmath154 is the kernel of the @xmath41tensor of @xmath225 . from this point forward we will use the notation from section 6 in @xcite for specific tangent vectors to @xmath17 . recall that @xmath392 $ ] is one of the coordinates of our representative points ( see proposition [ modulo isometries ] ) . [ h_p_m,-1 ] for @xmath393 the horizontal space of @xmath394 with respect to @xmath38 is spanned by @xmath395 * notation * we will call the seven vectors in ( [ h_p_m,-1 ] ) @xmath396 @xmath397 , @xmath398 , @xmath399 and @xmath400 respectively . we will call the set consisting of the first four @xmath401 , and the set consisting of the last three @xmath402 . the proof is just straight forward computations of inner products . many of the required computations were done in proposition 6.5 of @xcite . a corollary of [ generic planes](1 ) is the following . [ generic planes in e_2,0 ] let @xmath14 be a plane in @xmath32 whose @xmath242coordinate is in @xmath403 . if the horizontal lift of @xmath14 with respect to @xmath38 has a nondegenerate projection onto either @xmath404 or @xmath405 , then @xmath14 is positively curved with respect to @xmath38 . we will combine theorem [ generic planes](1 ) with the following observations . * any plane in @xmath406 has a nondegenerate projection ( with respect to the splitting @xmath407 ) onto @xmath157 . * @xmath408 . it follows from ( a ) and ( a@xmath409 ) that if the projection of @xmath14 onto @xmath406 is nondegenerate , then so is the projection onto @xmath157 ( with respect to the splitting @xmath407 ) . * any plane in @xmath410 has a nondegenerate projection ( with respect to the splitting @xmath407 ) onto @xmath154 . * @xmath411 . it follows from ( b ) and ( b@xmath409 ) that if the projection of @xmath14 onto @xmath410 is nondegenerate , then so is the projection of @xmath14 onto @xmath154 ( with respect to the splitting @xmath407 ) . [ h_p_m,-1 , t= pi/4 ] for @xmath412 the horizontal space of @xmath394 with respect to @xmath38 is spanned by @xmath413 again the proof is just straight forward computations of inner products . in fact the computations that proved ( [ h_p_m,-1 ] ) will suffice for all of the vectors except @xmath414 and @xmath415 . these are just multiples of the limits of @xmath416 and @xmath417 as @xmath418 . * notational convention * let @xmath419 be a riemannian submersion . to simplify the exposition we will make no notational distinction between a vector that is tangent to @xmath231 and its horizontal lift to @xmath53 . * remark * since the seven vectors listed in ( [ h_p_m,-1 , t= pi/4 ] ) are limits of ( function multiples of ) the seven vectors listed in ( [ h_p_m,-1 ] ) we will call them @xmath420 @xmath397 , @xmath398 , @xmath399 and @xmath400 . notice that via @xmath421 , @xmath422 and @xmath423 project to the unit normals of @xmath424 , where @xmath424 is the @xmath7sphere in @xmath0 that is fixed by the action induced on @xmath0 by @xmath40 via @xmath425 ( cf the notational remark in section 6 of @xcite ) . because of this their definitions do not seem to be very canonical when @xmath426 . the approach we have taken is to define them at representative points with @xmath426 as extensions of their definitions at representative points when @xmath242 is generic . we will extend these definitions ( as needed ) to all points with @xmath412 by letting the isometry group act . also notice that the vectors @xmath427 are in @xmath428 when @xmath412 . notice furthermore that at @xmath412 @xmath429 for any @xmath430 and any @xmath431 . here and in the sequel @xmath432 stands for @xmath433 . it follows from ( [ generic planes in e_2,0 ] ) that any @xmath6curvature planes for @xmath435 have the form @xmath436 where @xmath437 , @xmath438 and @xmath393 . about such planes we will prove the following . [ 0 s wrt g_nu_1 , nu_2 ] * for @xmath439 the @xmath6curvature planes of @xmath215 satisfy ( [ difficult planes ] ) and also , + @xmath440 + where @xmath441 is a convex combination of @xmath442 and @xmath443 is the same convex combination of @xmath444 . * when @xmath412 the planes of the form ( [ extra 0 s when t = pi/4 ] ) have @xmath6 curvature in @xmath17 with respect to @xmath38 , and , in addition , given any convex combination @xmath278 of @xmath445 and any convex combination @xmath279 of @xmath446 , there is a unique convex combination @xmath447 of @xmath448 , so that + @xmath449 + for all @xmath450 . + there are no other zero curvatures when @xmath451 . to prove ( 1 ) split @xmath371 into @xmath452 where @xmath453 and @xmath454 . to be more concrete suppose that @xmath455 is a multiple of @xmath456 . suppose @xmath279 has the form @xmath457 for some purely imaginary unit quaternion @xmath458 . notice that @xmath459 and @xmath460 if @xmath461 . combining [ generic planes](3),(4 ) with ( [ y inner products ] ) and ( [ perp y inner products ] ) we get that the curvature of @xmath14 is positive except possibly if @xmath462 . on the other hand , the projection of @xmath14 onto @xmath155 is nondegenerate unless @xmath463 or @xmath464 so @xmath14 is positively curved unless @xmath465 or @xmath279 is a multiple of the same convex combination of @xmath444 as @xmath455 is of @xmath442 . to complete the proof of ( 1 ) it remains to consider the case when @xmath466 in which it is easy to see , using [ generic planes](3),(4 ) that @xmath14 is positively curved unless @xmath467 . we saw in ( [ h_p_m,-1 , t= pi/4 ] ) that at points where @xmath412 , @xmath468 using ( [ basis at t= pi/4b ] ) it is easy to see that the only planes in @xmath469 whose projections onto both @xmath154 and @xmath155 are degenerate are those of the form ( [ extra 0 s when t = pi/4 ] ) . it follows that the only other possible @xmath6 s are of the type described in [ generic planes](3),(4 ) . combining this with straightforward computations of @xmath41tensors completes the proof of ( 2 ) , and even supplies explicit solutions for @xmath470 in terms of @xmath279 and @xmath278 . for example when @xmath471 and @xmath472 , @xmath473 , and when @xmath474 and @xmath472 , @xmath475 . * remark * we have shown that the planes in ( [ 0 s wrt g_nu_1 , nu_2 ] ) have @xmath6 curvature in @xmath476 @xmath477 . they also have zero curvature in @xmath478 . to verify this we have to evaluate the @xmath41tensor of @xmath391 on these planes . it turns out that it is @xmath6 . we will not verify this since most of these planes become positively curved with respect to @xmath479 . now we study the zero curvatures of @xmath480 . to do this we adopt the following convention . * notational convention * let @xmath481 denote the horizontal space of @xmath391 with respect to @xmath38 and @xmath482 the horizontal space of @xmath391 with respect to @xmath44 . let @xmath483 be the quotient map given in ( [ skew action ] ) for @xmath187 . according to ( [ new horizontal space ] ) @xmath14 is in @xmath481 if and only if @xmath484 is in @xmath485 . to keep the notation simpler we will think of this correspondence as a parameterization of @xmath486 by @xmath481 and we will denote vectors and planes in @xmath487 by the corresponding vectors and planes in @xmath488 . we will do this without any further mention or change in notation . for example @xmath489 now stands for @xmath490 . with this convention it follows from [ cheeger s curvature condition](1 ) that the zero curvatures of @xmath44 are a subset of the zero curvatures of @xmath38 . we will show that most of the planes in ( [ 0 s wrt g_nu_1 , nu_2 ] ) have nondegenerate projections onto the orbits of @xmath46 and hence are positively curved with respect to @xmath218 according to [ cheeger s curvature condition](3 ) . the root cause of this is that @xmath223 is not by isometries with respect to @xmath44 . this seems extremely plausible because @xmath223 commutes with neither @xmath46 nor @xmath47 . a rigorous proof that it does not act by isometries follows from our curvature computations . in any case we no longer know that @xmath223 is by isometries , so ( [ modulo isometries ] ) are no longer our representative points . the representative points are now @xmath491 for all @xmath392 $ ] and for all @xmath492 . the interval for @xmath493 is @xmath494 because @xmath223 is @xmath196ineffective . using this we will show the following . [ proj onto eta ] for any @xmath393 the planes of the form @xmath495 have positive curvature with respect to @xmath44 unless @xmath496 or @xmath497 . here @xmath498 and @xmath443 stand for ( the same ) convex combinations of @xmath499 and @xmath500 respectively . the main idea is that most planes are eliminated because their projection onto the orbits of @xmath46 are nondegenerate . combining [ cheeger s curvature condition](3 ) with ( [ commuting cheeger perturbations ] ) we see that it is actually enough to check that a projection is degenerate with respect to @xmath162 . the tangent space to the orbit of @xmath46 is spanned by the @xmath162orthogonal basis @xmath501where @xmath502 and @xmath503 are purely imaginary , unit quaternions that satisfy @xmath504 . before we can find the projections we also need a formula for @xmath505 along the orbits of @xmath223 . at points of the form ( [ theta , t representatives ] ) @xmath506 is @xmath507to find the projections we compute @xmath508according to ( [ h_p_m,-1 ] ) we should compute the inner products of @xmath509 and @xmath510 with the vectors in ( [ basis for to_a^u ] ) with respect to @xmath511 . these are the same as the inner products of @xmath512 and @xmath513 with the vectors in ( [ basis for to_a^u ] ) with respect to @xmath514 , where @xmath515 . we will compute the latter inner products since the notation is simpler . we will only do this explicitly for @xmath512 , since the computations for @xmath513 are the same modulo obvious changes in notation . @xmath516 where `` @xmath517 '' stands for a nonzero , but irrelevant term . combining the previous 2 equations we get @xmath518 @xmath519 combining ( [ < ( 0 , vartheta ) , u_gamma_1 > ] ) and ( [ < u_gamma_1 , ( vartheta_1 , 0 ) > ] ) we get @xmath520 @xmath521 @xmath522combining ( [ < u_gamma_2 , eta_1 > ] ) and ( [ < u_gamma_2 , vartheeta_1 > ] ) we get @xmath523 @xmath524 combining ( [ < u_gamma_2 , vartheeta_1 > ] ) and ( [ < u_gamma_2 , ( vartheta_1 , 0 ) > ] ) we see that @xmath525 combining ( [ < ( - vartheta , vartheta ) , u_gamma_1 > ] ) , ( [ < u_gamma_2 , eta_1 ^ 2,0 > ] ) , and ( [ < ( -vartheta_1 , vartheta_1 ) , u_gamma_2 ] ) we see that the projection of @xmath14 onto @xmath526 is nondegenerate unless @xmath527 or @xmath528 , proving the proposition . * remark * one might hope to show that the planes from ( [ proj onto eta ] ) with @xmath529 , and @xmath530 are positively curved by studying the projections onto @xmath531 . these are also nondegenerate for most values of @xmath493 . unfortunately they are degenerate precisely when @xmath529 , and @xmath530 . [ x - y , planes ] let @xmath441 and @xmath443 be as in ( [ proj onto eta ] ) . for @xmath393 if @xmath532 where @xmath533 , then the curvature of @xmath14 is positive with respect to @xmath44 unless @xmath534 or @xmath535 . along the orbit of @xmath536 the formula for @xmath489 is @xmath537 so @xmath538 similar computations show that @xmath539 combining ( [ < ( - vartheta , vartheta ) , u_gamma_1 > ] ) , ( [ < u_alpha , x^2,-1 > ] ) , and ( [ projections of x and y onto u_gamma_i and d_gamma_i ] ) we see that the projection of @xmath14 onto @xmath526 is nondegenerate unless @xmath540 or @xmath527 , proving the proposition . * remark * one might hope that there really is positive curvature on these planes when @xmath541 and @xmath530 because perhaps the projection of these planes onto @xmath531 is nondegenerate . while this is true for most values of @xmath493 , it is false precisely when @xmath541 and @xmath530 . finally we study the zero curvatures of @xmath542 at points when @xmath412 . at such points the action induced by @xmath223 on @xmath0 via @xmath543 fixes @xmath424 , so the orbits of @xmath223 are vertical with respect to @xmath201 when @xmath426 . combining this with the fact that @xmath48 acts transitively on the fibers of @xmath201 , we see that to study the curvature of these points we may assume that @xmath544 , in other words we are back to representative points of the form ( [ modulo isometries ] ) with @xmath545 . at such points we get the following as a corollary of ( [ < ( 0 , vartheta ) , u_gamma_1 > ] ) , ( [ < u_gamma_1 , ( vartheta_1 , 0 ) > ] ) , ( [ < u_gamma_2 , vartheeta_1 > ] ) and ( [ < u_gamma_2 , ( vartheta_1 , 0 ) > ] ) . [ t = pi/4 ] when @xmath412 and @xmath546 the planes of the form ( [ extra 0 s when t = pi/4 ] ) have positive curvature except if they have the form @xmath547 where @xmath548 is any convex combination of @xmath549 . to determine which of the planes in ( [ x - y zeros at pi/4 ] ) still have @xmath6curvature we prove the following . [ stable x - y 0 s at pi/4 ] at points with @xmath426 and @xmath550 , the only planes of the form ( [ x - y zeros at pi/4 ] ) that have @xmath6curvature with respect to @xmath44 are those of the form @xmath551 and @xmath552 where @xmath548 stands for any convex combination of @xmath549 and @xmath553 is any real number . the point is that if @xmath278 is any other convex combination of @xmath445 , then the projection of the plane in ( [ x - y zeros at pi/4 ] ) onto @xmath526 is nondegenerate . to see this we will need to compute the projection of @xmath554 onto @xmath526 . although we will only need to know its values for @xmath555 , we will do the computation for arbitrary @xmath556 , since it may be of interest to some readers . along the orbit of @xmath536 the formula for @xmath554 is @xmath557so @xmath558 similar computations show that @xmath559 combining ( [ < u_alpha , y > ] ) and ( [ projection of y onto u_gamma_i ] ) with ( [ < ( 0 , vartheta ) , u_gamma_1 > ] ) , ( [ < u_gamma_1 , ( vartheta_1 , 0 ) > ] ) , ( [ < u_gamma_2 , vartheeta_1 > ] ) , ( [ < u_gamma_2 , ( vartheta_1 , 0 ) > ] ) , ( [ < u_alpha , x^2,-1 > ] ) and ( [ projections of x and y onto u_gamma_i and d_gamma_i ] ) we see that if @xmath14 is any plane of the form of ( [ x - y zeros at pi/4 ] ) and not of the form of either ( [ stable x o s at pi/4 ] ) or ( [ stable y o s at pi/4 ] ) , then the projection of @xmath14 onto @xmath526 is nondegenerate and hence @xmath14 is positively curved . * remark * as was the case with our previous results one could study projections onto @xmath531 to prove ( [ t = pi/4 ] ) and ( [ stable x - y 0 s at pi/4 ] ) . unfortunately these projections are degenerate on the same planes as the projections onto @xmath526 and hence do not give us additional positive curvature . the curvature computations for the proof of theorem [ unit tangent bundle ] are completed with the following . [ really are 0 s ] * the planes in ( [ generic eta - vartheta 0 s ] ) , ( [ generic x , vartheta 0 s ] ) , ( [ special vartheta 0 s ] ) , ( [ stable x o s at pi/4 ] ) and ( [ stable y o s at pi/4 ] ) all have @xmath6curvature in @xmath560 . * the plane spanned by @xmath489 and @xmath561 in @xmath32 is tangent to a totally geodesic flat @xmath7torus . * let @xmath247 be the pointwise zero locus in @xmath562 , then @xmath247 is the union of two copies of @xmath8 that intersect along a copy of @xmath9 . to prove ( 1 ) we must check that the @xmath41tensors of @xmath563 and @xmath391 vanish on these planes . these computations are rather long , but are quite straight forward so we leave them to the reader . recall @xcite that @xmath564 denotes the circle in @xmath0 that is fixed by the action induced from @xmath40 via @xmath201 . it is easy to see that @xmath489 and @xmath561 span a @xmath7torus , @xmath263 , in @xmath17 whose tangent plane is horizontal with respect to @xmath391 at every point . since @xmath263 is horizontal , @xmath565 is a covering map . the order of the covering is at least two because @xmath198 leaves @xmath263 invariant under @xmath33 . we show next that the order is exactly two . let @xmath566 and @xmath567 be the geodesic circles that are generated by @xmath489 and @xmath561 . then with respect to the join decomposition @xmath568 , @xmath569 is a radial circle and for @xmath570 , @xmath571 is an intrinsic geodesic in a copy of @xmath202 . using these observations we see that a point in @xmath572 has only two preimages under @xmath573 from which it follows that @xmath574 is a two fold covering . the fields @xmath489 and @xmath575 are both parallel along @xmath263 . straightforward computation shows that the results of transporting them from a point @xmath576 to @xmath577 agrees with their images under the differential of the map induced on @xmath263 by @xmath198 via @xmath33 . therefore @xmath574 is a 2fold orientation preserving cover , and @xmath578 is a torus rather than a klein bottle . it follows from ( 1 ) that @xmath578 is flat and computations similar to those in the proof of ( 1 ) show that @xmath578 is totally geodesic . combining part ( 1 ) with ( [ 0 s wrt g_nu_1 , nu_2 ] ) , ( [ proj onto eta ] ) and ( [ x - y , planes ] ) , we see that @xmath247 consists of the points with @xmath579 or @xmath530 or @xmath412 . using the join structure @xmath580 we see that the union of the points with @xmath546 , @xmath581 and @xmath426 is a @xmath285sphere in @xmath0 . ( keep in mind that @xmath223 is @xmath196ineffective on @xmath0 . ) for similar reasons the union of the points with @xmath582 , @xmath583 and @xmath426 is another @xmath285sphere . these two three spheres intersect along the common @xmath7sphere @xmath424 and their inverse images via @xmath201 are of course diffeomorphic to @xmath8 . finally notice that every point in this inverse image has zero curvatures because @xmath201 is the quotient map of @xmath40 and @xmath40 acts by isometries . the example , @xmath10 , of corollary [ 6-dim quot ] is @xmath584 where @xmath458 is any purely imaginary , unit quaternion , and @xmath585 is the circle subaction of @xmath48 that is generated by @xmath458 . when @xmath586 the torus @xmath587 is normal to the orbit of @xmath585 . some further curvature computations show that its image in @xmath10 is totally geodesic and flat . [ e_2,0 is the unit tangent bunlde ] @xmath32 is diffeomorphic to the total space of the unit tangent bundle of @xmath0 . in fact , the riemannian submersion @xmath588 that is given by @xmath589 is bundle isomorphic to the unit tangent bundle of @xmath0 , where @xmath590 is the hopf fibration given by left multiplication . translating theorem 9.5 on page 99 of @xcite into our classification scheme ( [ gluing map ] ) shows that the unit tangent bundle is the bundle of type @xmath29 . we will show via direct computations ( similar to those in @xcite ) that @xmath591 is also the bundle of type @xmath29 . as in @xcite we define @xmath592 by @xmath593 and we define explicit bundle charts @xmath594 by @xmath595 and @xmath596 @xmath597 and @xmath598 are embeddings onto the open sets @xmath599 and @xmath600 in fact their inverses are given by @xmath601 and @xmath602 thus @xmath603 so @xmath604 is the bundle of type @xmath29 and hence is the unit tangent bundle of @xmath0 . suppose @xmath619 is a set of generators for @xmath620 , and @xmath621 generates @xmath622 . then using the fact that the map @xmath623 given by @xmath624 has degree @xmath65 we can see that @xmath625 ( to evaluate the first map represent @xmath626 by @xmath627 . ) from this point a routine algebraic computation shows that @xmath630 it follows from ( [ matrix of phi_3 ] ) that @xmath628 is injective if @xmath608 , and hence we get using ( [ meyer - vietoris ] ) that @xmath631 completing the proof of ( 1 ) . the known examples of simply connected @xmath3manifolds with positive curvature are given in @xcite , @xcite and @xcite . with the exception of @xmath161 and the example , @xmath154 , of berger none of them are @xmath7connected . since @xmath632 , @xmath32 is not a homotopy sphere . it is not homotopy equivalent @xmath154 since , according to proposition 40.1 in @xcite , @xmath154 is not a @xmath633cohomology sphere . * @xmath10 has the same integral cohomology modules as @xmath1 but not the same integral cohomology algebra . * @xmath10 does not have the homotopy type of any known example of a manifold of positive curvature . from the long exact homotopy sequence of the fibration @xmath634 we see that @xmath635 and @xmath636 . thus @xmath637 and @xmath638 . to compute @xmath639 and the cup products we appeal to the gysin sequence @xmath640 of the fiber bundle @xmath634 with integer coefficients . we already know that @xmath641 . moreover , it follows from the universal coefficient theorem that @xmath642 since @xmath643 has no torsion and @xmath644 is a torsion group . thus the gysin sequence shows that @xmath645 and hence that @xmath646 has the same integral cohomology groups as @xmath647 to see that @xmath646 does not have the cohomology ring of @xmath648 we look at the gysin sequence again : @xmath649 we therefore have @xmath650 this means that @xmath651 maps the generator @xmath652 for @xmath653 to twice a generator @xmath654 for @xmath655 if we let @xmath656 we therefore have @xmath657 therefore , it suffices to show that @xmath658 in order to show that the cohomology ring of @xmath646 is not that of @xmath647 we have that @xmath659 is an isomorphism . thus @xmath660 must also be twice a generator for @xmath661 therefore , if @xmath662 we would have that @xmath663 generates @xmath661 this , however , is impossible . besides @xmath664 and @xmath648 , there are two examples given in @xcite and @xcite of simply connected @xmath665manifolds with positive curvature . neither of these has the cohomology modules of @xmath648 so @xmath646 does not have the homotopy type of any known example .

we describe a possible implementation of the nanomechanical quantum superposition generation and detection scheme described in the preceding , companion paper [ armour a d and blencowe m p 2008 _ new . j. phys . _ * xx * xxx ] . the implementation is based on the circuit quantum electrodynamics ( qed ) set - up , with the addition of a mechanical degree of freedom formed out of a suspended , doubly - clamped segment of the superconducting loop of a dc squid located directly opposite the centre conductor of a coplanar waveguide ( cpw ) . the relative merits of two squid based qubit realizations are addressed , in particular a capacitively coupled charge qubit and inductively coupled flux qubit . it is found that both realizations are equally promising , with comparable qubit - mechanical resonator mode as well as qubit - microwave resonator mode coupling strengths .

0804.2179

in reference @xcite , we described a scheme for generating and detecting superposition states of a nanomechanical resonator . the scheme employs a qubit with dispersive coupling to the mechanical resonator . by preparing the qubit in a superposition of energy eigenstates and the mechanical resonator in a displaced thermal state , the latter subsequently evolves into a superposition state , the existence of which can be inferred through appropriate qubit state control and measurement . in the present paper , we describe a possible implementation of this scheme ( see figure [ schemefig ] ) that is based on the demonstrated circuit quantum electrodynamics ( qed ) set - up @xcite . control and readout of the qubit is achieved by pumping and probing the appropriate mode of a microwave coplanar waveguide ( cpw ) resonator that couples electromagnetically to the qubit . we shall consider two types of superconducting qubit : a capacitively coupled charge qubit , otherwise known as a cooper pair box ( cpb ) @xcite , and an inductively coupled flux qubit @xcite . as we shall see , both types of qubit appear equally promising for implementation , having comparable strength couplings to the cpw resonator and to the nanomechanical resonator . one motivation for adopting the circuit qed set - up is the adequate demonstrated qubit coherence times for our purpose @xcite . we suppose that a segment of the qubit flux loop directly opposite the centre conductor of the cpw is freely suspended , forming a doubly - clamped beam mechanical resonator . related schemes can be found in references @xcite . furthermore , in addition to the usual ghz microwave qubit control / readout line that is capacitively coupled to the cpw , we require a separate , low frequency ( i.e. , tens of mhz and below ) centre conductor bias line , in order to achieve the necessary strong couplings between the mechanical resonator and both qubit and cpw mode , as well as to drive the mechanical resonator on resonance , producing the displaced thermal state . in the case of the capacitively coupled charge qubit , the required centre conductor voltage bias can be introduced via a series inductor , otherwise known as a ` bias tee ' that is microwave engineered to reflect the separate , much higher frequency ghz qubit control signal @xcite . in the case of the inductively coupled flux qubit , the required current bias can be introduced via two superconducting low pass filters at each end of the centre conductor @xcite . the high frequency thermal johnson noise entering the bias line can be significantly suppressed with additional low - pass filtering at each temperature stage of the set - up , while at the same time allowing sufficiently large dc voltage or current biases for the required mechanical resonator - qubit coupling strengths @xcite . on the other hand , the high frequency thermal noise entering the microwave control line can be reduced by attenuating the signal @xcite ; only low drive power signals are required to control the qubit @xcite . sections 2 and 3 of this paper derive the respective charge and flux qubit hamiltonians with coupled , single microwave and mechanical modes . both qubit hamiltonians have a common origin in the cpw - coupled - dc squid circuit of figure [ schemefig ] . section 2 discusses the classical dynamics , while section 3 derives the quantum hamiltonians of the charge and flux qubit systems , equations ( [ concisechargehamiltonianeq ] ) and ( [ concisefluxhamiltonianeq ] ) respectively . while there is nothing particularly new in terms of theory in these two sections , they serve the purpose of mapping the scheme of reference @xcite onto the circuit qed set - up with a unified treatment of the charge and flux qubit implementations , facilitating a direct comparison between them . readers not interested in the derivations may proceed directly to section 4 , where the relative merits of the qubit hamiltonians for actual device implementation are addressed . section 5 gives a brief conclusion . appendix a gives a derivation of the single mode approximation to the cpw superconducting phase wave equation , while appendix b derives the qubit langevin equations . in this section , we analyze the classical equations of motion for the phase field coordinate @xmath0 of a length @xmath1 coplanar waveguide ( cpw ) with longitudinal coordinate @xmath2 and gauge invariant phases @xmath3 and @xmath4 across each of the two josephson junctions ( jj ) of the dc squid . these equations are the common starting point for the subsequent derivations of both the microwave resonator mode - flux qubit and microwave resonator mode - charge qubit hamiltonians . related analyses can be found in the review references @xcite . the cpw and squid are coupled via a mechanical , position - dependent mutual inductance @xmath5 and capacitance @xmath6 at @xmath7 , arising from the adjacent , parallel centre conductor and suspended squid loop segments . the position dependence and hence mechanical degree of freedom will be suppressed for the time being , to be introduced in section [ quantumequations ] once the qubit hamiltonians have been obtained . figure [ schemefig ] details the closed cpw - squid circuit . not shown are the capacitively - coupled microwave signal pump - probe line , as well as the low frequency bias line . again , we will ignore these external lines , focusing first on the closed cpw - squid dynamics . the external bias dependences will be introduced later below once the relevant classical cpw - coupled squid lagrangians have been derived [ see equations([lagrangiancapeq ] ) and ( [ lagrangianindeq ] ) ] . for @xmath8 and @xmath9 , the cpw phase field coordinate obeys the wave equation @xmath10 where @xmath11 is the phase velocity and @xmath12 and @xmath13 are the inductance and capacitance per unit length of the cpw , respectively . at the cpw boundaries @xmath14 , the current @xmath15 vanishes , since as mentioned above we are in the first instance considering the cpw - squid system to be closed . we assume small cpw - squid couplings , i.e. , @xmath16 and @xmath17 . the equations of motion for the coupled squid can be conveniently derived by applying kirchhoff s laws for the voltages and currents at @xmath18 and in the squid loop . referring to figure [ schemefig ] , current conservation gives @xmath19 the voltage relations are @xmath20 and @xmath21 where @xmath22 , @xmath23 is the charge on the capacitance @xmath6 and @xmath24 is the squid loop self inductance . current conservation at the jj nodes gives @xmath25 where @xmath26 is the jj critical current and @xmath27 the jj capacitance ( assumed identical for each junction ) . the jj phases are related to the net flux @xmath28 threading the squid loop as @xmath29 where the net flux is @xmath30 with @xmath31 the externally applied flux . it is more convenient to work with the squid phase variables @xmath32 , in which case equation ( [ currentjjreleq ] ) becomes @xmath33 and @xmath34 , \label{gammaminuseq}\ ] ] where we have used equations ( [ phasefluxeq ] ) and ( [ fluxeq ] ) , @xmath35 is the jj plasma frequency and @xmath36 . subtracting and adding the voltage relations ( [ voltagerel1eq ] ) and ( [ voltagerel2eq ] ) in order to express them in terms of variables @xmath37 , we obtain @xmath38 , \label{voltdiffeq}\ ] ] where we have again used equations ( [ phasefluxeq ] ) and ( [ fluxeq ] ) , and @xmath39 where we have taken time derivatives to replace the charges with currents and we have also used the current conservation relation ( [ currentconseq ] ) . equations ( [ gammapluseq ] ) , ( [ gammaminuseq ] ) , ( [ voltdiffeq ] ) , ( [ voltsumeq ] ) along with the wave equation ( [ waveeq ] ) completely specify the classical dynamics of the coupled cpw - dc squid system . we shall restrict ourselves to cpw mode solutions that reflect the symmetry of the circuit . in particular , we consider two types of solution : 1 ) voltage antinode and current node at @xmath7 : @xmath40 , @xmath41 ; 2 ) voltage node and current antinode at @xmath7 : @xmath42 , @xmath43 . in the former case , the cpw - squid coupling is predominantly capacitive and results in the microwave resonator mode - charge qubit hamiltonian , while in the latter case the coupling is inductive and results in the microwave resonator mode - flux qubit hamiltonian . the following two subsections deal with each case in turn . for a voltage antinode - current node solution , we have from equation ( [ voltdiffeq ] ) that @xmath44 must be a constant . from equation ( [ gammaminuseq ] ) , we then require that @xmath45 and @xmath46 , where @xmath47 is an arbitrary integer . thus , we see that in order to have a non trivial solution for the remaining @xmath48 variable , the external flux is constrained to be an integer multiple of the flux quantum . however , if @xmath49 , i.e. , the squid loop self inductance @xmath24 is small , then to leading , zeroth order in @xmath50 a voltage antinode solution is allowed provided @xmath51 and the external flux need not be constrained . assuming therefore a small loop area squid with @xmath49 in the cpw resonator voltage antinode case and neglecting the inductance dependent terms in equation ( [ voltsumeq ] ) , we obtain @xmath52 using equation ( [ voltsum2eq ] ) , equation ( [ gammapluseq ] ) for @xmath48 thus becomes @xmath53 where @xmath54 and we have set @xmath55 , since observable quantities do not depend on @xmath56 . equations ( [ voltsum2eq ] ) and ( [ gammaplus2eq ] ) along with the wave equation ( [ waveeq ] ) completely specify the voltage antinode restricted classical dynamics . the qubit control and readout protocols @xcite require driving the cpw close to one of its resonant modes . in the voltage antinode case , this is usually the second fundamental mode @xcite . the subsequent analysis of the cpw - squid dynamics is greatly simplified if we can first effectively replace the cpw with a single harmonic oscillator at the relevant microwave mode frequency . in [ singlemode ] , we derive these approximate , effective single oscillator mode equations and also give the conditions under which these equations are valid . in the voltage antinode second fundamental mode case , we obtain for the approximate coupled single oscillator - squid equations of motion : @xmath57 and @xmath58=c_m \frac{\phi_0}{2\pi}\ddot{\gamma}_+ , \label{phiosceq}\ ] ] where @xmath59 is the oscillator coordinate giving the cpw phase field amplitude at the squid location @xmath7 and where we have neglected an order @xmath16 shift in the cpw second fundamental mode frequency @xmath60 . equations of motion ( [ gammaplus3eq ] ) and ( [ phiosceq ] ) follow from the lagrangian @xmath61\cr & & -c_m \frac{\phi_0}{2\pi}\dot{\gamma}_+\left(\frac{\phi_0}{2\pi}\dot{\phi}+v(t)\right ) , \label{lagrangiancapeq}\end{aligned}\ ] ] where we have included a cpw voltage bias @xmath62 , assumed to be slowly varying in time as compared with the microwave mode frequency @xmath63 . as we shall see in section [ comparison ] , this additional voltage bias is necessary in order to tune the qubit s operating point , couple strongly the qubit with the mechanical oscillator , and also drive the mechanical oscillator on resonance @xcite . using the lagrangian ( [ lagrangiancapeq ] ) to construct the associated hamiltonian , we have for the generalized momenta : @xmath64 and @xmath65 expressing the velocities in terms of the momenta and coordinate variables and substituting into the definition for the hamiltonian , @xmath66 , we obtain after some algebra @xmath67 where we have have dropped overall constant ( i.e. , variable independent ) terms . from the definition ( [ pgammaeq ] ) for the generalized momentum @xmath68 and recalling that @xmath69 , we see that @xmath70 , \label{neq}\ ] ] where @xmath71 is the number of excess cooper pairs on the island represented by the directly connected nodes between the @xmath6 and @xmath27 capacitances ( see figure [ schemefig ] ) and @xmath72 is the polarization charge induced by the slowly varying voltage bias on the @xmath6 electrode , expressed in units of cooper pair number . expression ( [ neq ] ) then suggests an alternative formulation of the hamiltonian in terms of the cooper pair number @xmath71 : @xmath73 ^ 2-\frac{i_c\phi_0}{\pi}\cos(\pi\phi_{\mathrm{ext}}/\phi_0)\cos\gamma_+\cr & & + \frac{2}{c_w l}\left(\frac{2\pi}{\phi_0}\right)^2 \frac{p_{\phi}^2}{2}+\frac{1}{2 } c_w l \left(\frac{\phi_0}{2\pi}\right)^2 \left(\frac{2\pi v}{l}\right)^2\frac{\phi^2}{2}\cr & & -4e\frac{c_m}{c_w l c_{\sigma}}[n - n_m ( t)]\frac{2\pi}{\phi_0}p_{\phi } \label{hamiltoniancap2eq}\end{aligned}\ ] ] and we recognize in the top line of equation ( [ hamiltoniancap2eq ] ) the usual cooper pair box ( cpb ) hamiltonian @xcite . for a voltage node - current antinode solution , we see from equations ( [ voltsumeq ] ) and ( [ gammapluseq ] ) that @xmath74 , where @xmath47 is an arbitrary integer . thus , equation ( [ gammaminuseq ] ) for @xmath44 becomes @xmath75 where the mutual capacitance @xmath6 drops out and we have set @xmath76 . equation ( [ voltdiffeq ] ) also simplifies to @xmath77 where we have assumed that @xmath78 , i.e. , changes slowly compared with the dynamical timescales of the squid and cpw . equations ( [ gammaminus2eq ] ) and ( [ voltdiff2eq ] ) along with wave equation ( [ waveeq ] ) completely specify the voltage node restricted classical dynamics . the approximate , first fundamental ( i.e. , lowest frequency ) microwave mode - coupled squid equations that follow from equations ( [ gammaminus2eq ] ) and ( [ voltdiff2eq ] ) are ( see [ singlemode ] ) : @xmath79=\frac{m_m}{l } \frac{\phi_0}{2\pi l_w}k_0\tan(k_0 l/2)\phi \label{gammaminus3eq}\ ] ] and @xmath80=\frac{m_m}{l } \frac{\phi_0}{2\pi l_w}k_0 \gamma_- , \label{phiosc2eq}\ ] ] where from equation ( [ phipeq ] ) in [ singlemode ] the term @xmath81 appearing in the @xmath44 equation of motion ( [ gammaminus3eq ] ) is just the cpw current at the squid location @xmath7 . the wavenumber @xmath82 is the fundamental solution to the equation @xmath83 neglecting order @xmath84 corrections , we have approximately for the fundamental wavenumber and mode frequency : @xmath85 and @xmath86 , respectively . equations of motion ( [ gammaminus3eq ] ) and ( [ phiosc2eq ] ) follow from the lagrangian @xmath87\right)^2\right]\cr & & + \frac{1}{4}c_w l\left(\frac{\phi_0}{2\pi}\right)^2\tan^2(k_0l/2)\left[\frac{\dot{\phi}^2}{2}-\left(\frac{\pi v}{l}\right)^2\frac{\phi^2}{2}\right]\cr & & + \frac{m_m}{l}\frac{\phi_0}{2\pi}\gamma_- \left[\frac{\phi_0}{2\pi l_w}k_0\tan(k_0 l/2)\phi\right ] , \label{lagrangianindeq}\end{aligned}\ ] ] where we have included a cpw current bias @xmath88 , assumed to be slowly varying in time as compared with the fundamental microwave mode frequency @xmath63 . as we shall see in section [ comparison ] , this additional current bias is necessary in order couple strongly the qubit with the mechanical oscillator and also drive the mechanical oscillator on resonance @xcite . the hamiltonian associated with lagrangian ( [ lagrangianindeq ] ) is @xmath89\right)^2\right]\cr & & + \frac{4}{c_w l}\tan^{-2}(k_0l/2)\left(\frac{2\pi}{\phi_0}\right)^2\frac{p_{\phi}^2}{2 } + \frac{1}{4}c_w l \left(\frac{\phi_0}{2\pi}\right)^2\tan^2(k_0l/2 ) \left(\frac{\pi v}{l}\right)^2\frac{\phi^2}{2}\cr & & -\frac{m_m}{l}\frac{\phi_0}{2\pi}\gamma_- \left[\frac{\phi_0}{2\pi l_w}k_0\tan(k_0 l/2)\phi\right ] . \label{hamiltonianindeq}\end{aligned}\ ] ] the goal of this section is to derive the quantum langevin equations for the coupled cpw microwave oscillator - charge(flux ) qubit - mechanical oscillator system , taking into account the effects of damping and noise on the three subsystems . we shall begin by deriving the closed system quantum hamiltonians , treating the ( voltage antinode ) capacitively - coupled charge qubit and ( voltage node ) inductively - coupled flux qubit systems separately as we have done in the previous section . it is most convenient to work in terms of the cpw microwave oscillator raising and lowering operators : @xmath90 where , from equation ( [ hamiltoniancap2eq ] ) , @xmath91 and @xmath60 . the hamiltonian ( [ hamiltoniancap2eq ] ) then becomes @xmath92 ^ 2-\frac{i_c\phi_0}{\pi}\cos(\pi\phi_{\mathrm{ext}}/\phi_0)\cos\gamma_+ + \hbar\omega_w a_w^+ a_w\cr & & -\frac{c_m}{c_{\sigma}}\sqrt { \hbar\omega_w\frac{(2e)^2}{c_w l } } [ n - n_m(t ) ] i(a_w^+ -a_w ) , \label{hamiltoniancap3eq}\end{aligned}\ ] ] where for notational convenience we have dropped the hats and also the ` - ' superscript on the lowering operator . examining the interaction hamiltonian , we see that the coupling constant in units of @xmath93 depends essentially on the ratio of the cpw single cooper pair charging energy to the cpw mode single photon energy @xcite . note from equation ( [ neq ] ) that the standard poisson bracket relation @xmath94 gives @xmath95 for the variables @xmath48 and @xmath71 . thus , when we quantize using the correspondence principle between poisson brackets and commutators , i.e. , @xmath96=c$ ] ( for some constant @xmath97 ) , we therefore have that @xmath98=-i$ ] . we now truncate the squid hilbert space down to the lowest energy , two - dimensional hilbert space , obtaining the charge qubit hamiltonian . working in the representation for which the number operator @xmath99 is diagonal , the commutation relation @xmath98=-i$ ] gives @xmath100 . for the number operator , we have @xmath101 , where we have truncated to the two - dimensional hilbert space spanned by eigenkets @xmath102 , @xmath103 of the number operator ; in order to be within this subspace , we require @xmath62 to be tuned such that @xmath104 , where recall @xmath72 . the truncated charging energy term can similarly be expressed as @xmath105 ^ 2=(n - n_m)^2 |n\rangle \langle n| + ( n+1-n_m)^2 |n+1\rangle\langle n+1| ) = [ \delta n(t)]^2 + \frac{1}{4 } + \delta n(t ) \sigma_z$ ] , where @xmath106 . writing @xmath107 , we have @xmath108 . substituting the above truncated approximations into equation ( [ hamiltoniancap3eq ] ) , we obtain for the truncated hamiltonian : @xmath109 ^ 2 , \label{fulltruncatedchargehamiltonianeq}\end{aligned}\ ] ] where @xmath110 is the microwave mode - qubit coupling @xcite , and where @xmath111 is the single cooper pair charging energy for the total capacitance @xmath112 and @xmath113 is the effective flux - dependent josephson energy , with @xmath114 the josephson energy of a single jj . note that we have also made the canonical replacement @xmath115 , @xmath116 . we now incorporate the mechanical degree of freedom , supposing that a segment of the squid loop directly opposite the centre conductor forms a mechanically compliant , doubly - clamped beam . for a small in plane displacement with centre - of - mass amplitude @xmath117 , the position dependent mutual capacitance is @xmath118 , where @xmath119 is the static equilibrium gap between the centre conductor and opposite facing squid loop segment and @xmath120 is a geometrical factor of order one accounting for the non - uniform beam displacement . substituting this @xmath121 into the hamiltonian ( [ fulltruncatedchargehamiltonianeq ] ) and assuming that @xmath122 , we see that to leading order the coupling to the mechanical resonator arises through a modulation of the @xmath62 induced polarization @xmath123 , as well as through a modulation of the microwave mode - qubit coupling @xmath124 . express the applied voltage bias as @xmath125 , where @xmath126 is a constant dc voltage and @xmath127 is slowly varying as compared with the microwave mode frequency @xmath63 . we obtain for the mechanical resonator contribution to the hamiltonian : @xmath128 where @xmath129 with @xmath130 the zero - point uncertainty and we assume that it is the frequency @xmath131 fundamental flexural mode of the mechanical resonator that predominantly couples to the qubit with strength @xmath132 hamiltonian ( [ mechanicalhamiltoniancapeq ] ) neglects direct microwave - mechanical oscillator interaction terms , assumed to have a small effect on the coupled dynamics since the two modes are significantly off - resonance : @xmath133 . we also assume that @xmath134 , necessary in order to achieve strong mechanical oscillator - qubit coupling , and that the mechanical oscillator driving voltage @xmath135 . later , we shall see that the typical large mechanical quality factors ensure that such ac driving voltages are adequate for exciting the mechanical oscillator to sufficiently large amplitude as required by the qubit control protocol @xcite . the control protocol also requires operating at the charge degeneracy point @xmath136 , where the cpb qubit coherence time is a maximum @xcite . thus , the full microwave - qubit - mechanical oscillator hamiltonian becomes @xmath137 where we have changed the pauli matrix basis : @xmath138 , @xmath139 , and @xmath140 is the qubit transition frequency . from ( [ concisechargehamiltonianeq ] ) , we see that the mechanical oscillator couples to the qubit and microwave oscillator through two interaction terms of the form @xmath141 and @xmath142 , respectively . given the wide separation of timescales , @xmath143 , an adiabatic approximation can be performed to show that the @xmath141 interaction term gives rise to an approximate , dispersive interaction @xmath144 between the mechanical oscillator and qubit @xcite . thus , a dominant effect of the qubit on the mechanical oscillator is to shift its frequency up or down depending on the state of the qubit . by preparing the qubit in a superposition state , the mechanical oscillator is in turn driven into a superposition state via the dispersive interaction , as analyzed in reference @xcite . provided the frequency difference @xmath145 is much larger than the coupling @xmath124 , the microwave - qubit interaction @xmath146 can similarly be replaced by an approximate dispersive interaction @xmath147 @xcite . supposing that the qubit remains in its ground state , the microwave mode then exerts a ponderomotive - type force on the mechanical oscillator described by the interaction @xmath148 . thus , passive cooling or alternatively amplification of the mechanical motion can in principle be implemented with appropriate red or blue detunings of the microwave drive frequency with respect to the microwave mode frequency @xmath63 @xcite . we discuss these possibilities in section [ comparison ] . in this section , we obtain the truncated flux qubit hamiltonian . while the analysis is well known and involves basic schrdinger wave mechanics , it does serve a purpose in bringing to the fore the differences in the flux and charge qubit parameter regimes . readers not interested in the details of the derivation may skip directly to the resulting qubit hamiltonian ( [ concisefluxhamiltonianeq ] ) . from equation ( [ hamiltonianindeq ] ) , the microwave mode effective mass is @xmath149 and the microwave mode fundamental frequency is @xmath150 . in terms of the raising and lowering operators , the hamiltonian ( [ hamiltonianindeq ] ) is then @xmath151\right)^2\right]\cr & & + \hbar\omega_w a^+_w a_w -\frac{m_m}{l } \sqrt{2 \hbar\omega_w \frac{(\phi_0/2\pi)^2}{l_w l } } \gamma_- ( a_w^+ + a_w ) . \label{hamiltonianind2eq}\end{aligned}\ ] ] examining the interaction hamiltonian , we see that the coupling constant in units of @xmath93 depends essentially on the ratio of the cpw single flux quantum addition energy to the cpw mode single photon energy . the eigenstates and eigenvalues of the squid part of the hamiltonian ( [ hamiltonianind2eq ] ) can be obtained , e.g. , by working in the schrdinger wavefunction form . in the representation @xmath152 and @xmath153 ( where we have dropped the ` @xmath154 ' subscript on @xmath155 for notational convenience ) , the squid schrdinger equation becomes @xmath156\right)^2 -\cos\gamma\right]\psi(\gamma)\cr & & = e\psi(\gamma ) , \label{schreq}\end{aligned}\ ] ] where @xmath157 is the cooper pair charging energy of the jj capacitor , @xmath158 is the josephson energy , and we have suppressed for the present the slowly time - varying contribution @xmath159 to the current bias @xmath160 . provided @xmath161 and @xmath162 , @xmath163 , then the potential energy function admits a symmetric double well centered at @xmath164 . figure [ doublewell ] shows an example plot of the potential @xmath165 in region of its minimum , indicating a double - well . also shown is the resulting asymmetric double well when the external flux @xmath31 is detuned slightly from the above symmetric well condition value . in order that the lowest energy eigenvalues lie ` within ' the double well and hence be strongly anharmonic , we require that the harmonic ground state energy @xmath166 neglecting the cosine term in the potential @xmath165 be smaller than the central barrier maximum ( @xmath167 for the degenerate , symmetric well see figure [ doublewell ] ) . thus , we require that @xmath168 , or @xmath169 . however , @xmath170 can not be too small , for otherwise the energy spacing between the ground and first excited states will correspondingly be too small , making the resulting qubit susceptible to damping and decoherence by thermal and other low energy noise sources . we have seen that , in contrast to the charge qubit ( see section 2.1 ) , the flux qubit requires a nonnegligible @xmath50 , i.e , the flux qubit requires a much larger squid loop self - inductance @xmath24 than the cpb . adequate effective self - inductances can be achieved by using small loop , multiple josephson junction squid devices @xcite , where the additional jj s compensate for the small loop area , hence avoiding potential increases in flux noise that come with using larger loop areas . however , in the present work , we consider only a double jj , dc squid for simplicity ; the relevant multiple jj squid devices do not involve any conceptually new features and can be analyzed along similar lines . it is quite informative to solve the schrdinger equation ( [ schreq ] ) approximately in terms of the two harmonic ground basis functions for the quadratic well expansions about the two minima . we have : @xmath171 ( \gamma-\gamma_i)^2 , \label{harmonicpoteq}\ ] ] where the double well minima , @xmath172 , are solutions to @xmath173\right ) + \sin\gamma_i=0 . \label{wellmininimaeq}\ ] ] the normalized ground eigenfunctions of the harmonic approximation hamiltonians with @xmath165 replaced by @xmath174 are @xmath175^{1/8}\cr & & \times\exp\left[-\frac{1}{2}\sqrt{\frac{e_j}{e_{c_j}}\left[\beta_l^{-1}+\cos(\gamma_i)/2\right]}(\gamma-\gamma_i)^2\right ] . \label{harmonicfunctioneq}\end{aligned}\ ] ] the lowest two eigenfunctions of the double well hamiltonian @xmath176 are then approximately expressed as linear combinations of @xmath177 : @xmath178 , where the constants @xmath179 , @xmath180 are solutions to @xmath181 \left [ \begin{array}{c } a\\ b \end{array}\right]= \left [ \begin{array}{c } 0\\ 0 \end{array } \right ] \label{evaluefunctioneq}\ ] ] and the solutions @xmath182 give the approximate energy eigenvalues of @xmath176 . note that the two basis functions are not orthogonal ; the solutions are a reasonable approximation provided @xmath183 . let us now consider a concrete example . taking @xmath184 ph and @xmath185 , we have @xmath186 and @xmath187 mev . taking @xmath188 ff , we have @xmath189 mev , so that @xmath190 . considering a symmetric well with @xmath162 , we find that @xmath191 , and the approximate energy eigenvalues are : @xmath192 mev and @xmath193 mev with difference @xmath194 ( and @xmath195 ) . the approximate normalized eigenstates are @xmath196 , i.e. , ( anti)symmetric combinations of the basis states , with the symmetric combination associated with the lower , ground eigenvalue @xmath197 . for this example , we have @xmath198 . on the other hand , for a slight external flux bias detuning @xmath199 , making the double well potential asymmetrical , we have @xmath200 mev and @xmath201 mev with difference @xmath202 , and associated eigenstates @xmath203 and @xmath204 . thus , the lowest two eigenstates correspond to the phase being approximately localized to one or the other of the two wells . proceeding formally with the truncation of the hamiltonian ( [ hamiltonianind2eq ] ) to the subspace spanned by exact eigenstates @xmath205 and incorporating the mechanical degree of freedom through a position - dependent mutual inductance @xmath206 ( with @xmath207 a geometrical factor of order one ) , we have : @xmath208\right ) \cr & & + \eta_l \frac{x_{zp}}{d}\frac{m_m}{l}\hbar g_l ( a_w+a_w^+)(a_m+a_m^+ ) \gamma\cr & & + \eta_l \frac{x_{zp}}{d}\frac{m_m}{l}\frac{\phi_0}{2\pi } i_{ac}(t ) ( a_m+a_m^+ ) \left(\gamma + \frac{\pi}{\phi_0 } [ \phi_{\mathrm{ext}}-2m_m i_{dc}]\right ) , \label{fulltruncatedfluxhamiltonianeq}\end{aligned}\ ] ] where the microwave mode - qubit coupling is @xmath209 and the truncated phase operator @xmath210 is @xmath211 with @xmath212 , etc . note that @xmath213 and the truncated matrix elements @xmath214 are @xmath31 and @xmath215-dependent . for the example parameters considered above with symmetric double well potential [ @xmath162 ] , we have : @xmath217 , and @xmath218 , so that the @xmath219 operator does not arise in the couplings . on the other hand , for the above asymmetric well example [ @xmath199 ] , we have @xmath221 , and @xmath222 , so that the @xmath219 operator term is now present with much larger coupling as compared with the @xmath223 operator term . we assume that the flux qubit s coherence time is optimized for the symmetric well case , i.e. , the bias conditions are : @xmath162 . the hamiltonian ( [ fulltruncatedfluxhamiltonianeq ] ) then simplifies : @xmath224 where @xmath225 , @xmath226 , and the qubit - mechanical oscillator coupling is @xmath227 in the hamiltonian simplification ( [ concisefluxhamiltonianeq ] ) , we neglect direct microwave oscillator - mechanical oscillator interaction terms and also assume small mechanical oscillator drive currents @xmath228 . note that the microwave oscillator - flux qubit - mechanical oscillator hamiltonian ( [ concisefluxhamiltonianeq ] ) is formally identical to the microwave oscillator - charge qubit - mechanical oscillator hamiltonian ( [ concisechargehamiltonianeq ] ) , the differences occuring only in their respective coupling strengths . this close identification will enable a straightforward comparison of the relative merits of the two hamiltonians in section [ comparison ] . we now include model environments for the cpw - qubit - mechanical resonator system . for the cpw , we suppose that the dominant direct dissipation mechanism is due to coupling to the microwave pump - probe line used to drive and read out the state of the cpw . assuming weak mechanical oscillator - bath and cpw mode - pump / probe line couplings , with the baths modeled as a continuous spectrum of free harmonic oscillators , we obtain the following langevin equations for the mechanical and cpw mode operators @xmath229 and @xmath230 , respectively : @xmath231-\gamma_m a_m -i\sqrt{2\gamma_m } e^{i\phi_b } a_b^{\mathrm{in } } \label{ameq}\ ] ] and @xmath232-\gamma_w a_w -i\sqrt{2\gamma_w } e^{i\phi_p } a_p^{\mathrm{in } } , \label{aweq}\ ] ] where @xmath233 is the charge qubit hamiltonian ( [ concisechargehamiltonianeq ] ) or flux qubit hamiltonian ( [ concisefluxhamiltonianeq ] ) . the @xmath234 s are the various mode amplitude damping rates and the ` in ' and ` out ' bath operators are related as follows @xcite : @xmath235 this identity is used to determine the measured quantities expectation values involving @xmath236 in terms of the solutions to the mode operators @xmath237 and the prescribed initial @xmath238 s . the qubits themselves also directly couple to environments other than the damped microwave and mechanical oscillators . for completeness , langevin equations describing the dissipative qubit dynamics are given in [ qubitlangevin ] , although they are not used in the present paper . in the previous section , we derived hamiltonians describing a charge qubit ( [ concisechargehamiltonianeq ] ) , or alternatively flux qubit ( [ concisefluxhamiltonianeq ] ) , coupled to a microwave oscillator and mechanical oscillator . for convenience , we reproduce the two hamiltonian expressions here : @xmath239 and @xmath240 an important question concerns the relative merits of these two implementations . in this section , we address this by comparing realisable microwave - charge qubit and microwave - flux qubit coupling strengths @xmath124 ( [ gccoupling ] ) and @xmath241 ( [ glcoupling ] ) , qubit - mechanical oscillator coupling strengths @xmath242 ( [ cpboscillatorcoupling ] ) and @xmath243 ( [ fluxoscillatorcoupling ] ) , and realisable mechanical displacement amplitudes due to achievable drive strengths @xmath127 and @xmath159 . refering to the cpw circuit in figure [ schemefig ] , we consider a geometry modeled on the circuit qed device of reference @xcite dictated by the requirement that the cpw impedance @xmath244 matches as closely as possible the typical @xmath245 impedances of coaxial feedlines . furthermore , the resonant frequences of the first ( voltage node at midpoint ) and second ( voltage antinode at midpoint ) modes are required to be in the @xmath246 ghz range in order be in the quantum regime for the cpw at low dilution fridge temperatures , as well as to match the level separations of the flux or charge qubit . we consider the centre conductor of the cpw to be uniformly @xmath247 m wide and separated from the lateral ground planes by a @xmath248 m gap . the doubly - clamped , suspended beam segment of the dc squid directly opposite the centre conductor is assumed to be @xmath249 nm wide , with a gap of @xmath250 nm between the two . all the described elements of the cpw and squid are @xmath249 nm thick . the substrate is assumed to be silicon , with an etched out trench beneath the mechanical resonator that extends up to the adjacent centre conductor edge . the mutual capacitance per unit length between the centre conductor and lateral ground planes , @xmath13 , is obtained as follows . apply two linearly independent static voltage configurations to the centre and lateral planes and for each configuration evaluate ( numerically ) the total stored electric field energy per unit length : @xmath251 . next , invert to obtain the capacitance per unit length matrix @xmath252 . finally , obtain the mutual capacitance per unit length in terms of the capacitance matrix using @xmath253 on the other hand , to obtain the self inductance per unit length of the centre inductor , @xmath12 , apply some non - zero steady state current @xmath254 to the centre conductor , with zero applied current through the lateral conductors and then evaluate ( numerically ) the total stored magnetic field energy per unit length @xmath255 . the self inductance is then simply @xmath256 . carrying out these procedures , we obtain @xmath257 and @xmath258 . these values give an impedance @xmath259 and @xmath260 for the phase velocity . therefore , the first mode frequency is @xmath261 and second mode frequency is twice this : @xmath262 . thus , the cpw should be about @xmath263 long in order to have the first two microwave modes in the desired frequency range . in order to evaluate the mutual capacitance @xmath6 and mutual inductance @xmath5 between the centre conductor and suspended squid beam , we assume for simplicity an infinitely long beam and obtain : @xmath264 and @xmath265 . for the microwave mode - qubit coupling strengths , we also require the squid qubit jj capacitance values ( for the antinode case ) and self inductance values ( for the node case ) . as example values , we refer to the cpb qubits of the yale group @xcite and the flux qubits of the delft group @xcite . we shall consider only order of magnitude estimates . beginning with the inductively coupled flux qubit , we first note that the requirement @xmath266 gives @xmath267 . thus , in order that the squid inductance @xmath24 and hence loop area not have to be too large , we require jj s with large critical current @xmath26 values . the delft group jj s have @xmath268a , giving @xmath269 . considering a cpw length of @xmath270 gives a fundamental ( node ) mode frequency @xmath271 , comparable to the rabi oscillation frequency at the symmetry point of the flux qubit in reference @xcite . a few micron long mechanical resonator gives @xmath272 , so that we have for the inductance ratio @xmath273 . for the capacitive coupling in the case of the cpb qubit , a cpw length of @xmath274 gives a second ( antinode ) mode frequency @xmath275 , comparable to the single electron charging energy ( in units @xmath276 ) of the original cpb qubit in reference @xcite . therefore , @xmath277 and a few micron long mechanical resonator gives a coupling ratio @xmath278 . with all the capacitance and inductance values in hand , we are now ready to compare the coupling strengths @xmath124 and @xmath241 . we have : @xmath279 and @xmath280 where note that the cpw is twice as long ( @xmath281 ) as in the voltage node case ( @xmath1 ) so that the 2nd mode and 1st mode frequencies are the same , allowing a direct comparison . interestingly , although the inductance ratio @xmath282 is about an order of magnitude smaller than the capacitance ratio @xmath283 , the flux quantum addition energy term is about an order of magnitude larger than the charging energy term , bringing the two coupling strength terms ( [ indcoupstrengtheq ] ) and ( [ capcoupstrengtheq ] ) into line . the reason for the difference in the flux quantum addition energy and charging energy magnitudes can be seen more clearly by taking the ratio of the square root terms : @xmath284 note that this is essentially just the ratio of the universal quantum of resistance ( i.e. , von klitzing constant @xmath285 ) to the cpw impedance and so with typical microwave device impedances in the several tens of ohms together with the @xmath286 factor in the denominator , we expect this ratio to be typically much larger than one . thus , with regards to the microwave mode - qubit coupling strengths , there is no advantage to be gained by implementing one qubit scheme over the other . comparing the qubit - mechanical oscillator coupling strengths ( [ cpboscillatorcoupling ] ) and ( [ fluxoscillatorcoupling ] ) , we have @xmath287 where we have assumed geometrical factors @xmath288 . thus , applying a cpw current bias @xmath289 in the flux qubit device gives the same order mechanical coupling strength as applying a cpw voltage bias @xmath290 in the charge qubit device . for a 5 micron long mechanical resonator with fundamental flexural frequency @xmath291 , zeropoint displacement @xmath292 , and mechanical beam - centre conductor gap @xmath293 , we have for the absolute , charge qubit - mechanical coupling strength @xmath294 therefore , a voltage bias @xmath290 provides sufficiently strong charge qubit - mechanical oscillator coupling strength @xmath295 for the scheme of reference @xcite . the @xmath127 and @xmath159 terms in hamiltonians ( [ concisechargehamiltonianeq ] ) and ( [ concisefluxhamiltonianeq ] ) are used to drive the mechanical oscillator into a non - zero displacement amplitude state , as required by the scheme of the companion paper @xcite to rapidly generate a mechanical superposition state . working with hamiltonian ( [ concisechargehamiltonianeq ] ) and assuming a sinusoidal voltage drive that is resonant with the mechanical frequency , the steady state mechanical amplitude @xmath296 in units of the zeropoint displacement is @xmath297 where @xmath298 is the ac voltage drive amplitude and @xmath23 is the mechanical quality factor . for the same mechanical oscillator parameters as above and quality factor @xmath299 and dc voltage bias @xmath300 , we have @xmath301 thus , according to the mechanical superposition generation scheme @xcite , ac voltage drive amplitudes of no more than a few microvolts are sufficient to obtain the required displacement amplitudes . such ac voltage amplitudes do not significantly perturb the dc bias conditions on the cpb , since they are considerably smaller than the voltage bias difference @xmath302 separating neighbouring cpb charge degeneracy points . alternatively , working with the hamiltonian ( [ concisefluxhamiltonianeq ] ) and assuming a sinusoidal current drive that is resonant with the mechanical frequency , the steady state mechanical amplitude @xmath296 in units of the zeropoint displacement is @xmath303 where @xmath304 is the ac current drive amplitude . using the same mechanical resonator parameters as above and a dc current bias @xmath305 ( for which the flux qubit - mechanical oscillator coupling strength @xmath243 coincides with charge qubit - oscillator strength @xmath242 see above ) , we have @xmath306 thus , ac current drive amplitudes of no more than a few tens of microamps are sufficient to obtain the required displacement amplitudes for the mechanical superposition generation scheme @xcite . such ac current amplitudes do not significantly perturb the dc bias conditions on the flux qubit , since they are considerably smaller than the current bias difference @xmath307 separating neighbouring symmetric double well configurations . the @xmath308 interaction terms in hamiltonians ( [ concisechargehamiltonianeq ] ) and ( [ concisefluxhamiltonianeq ] ) can in principle be employed to passively cool the mechanical oscillator ( i.e. , flexural mode ) by driving the microwave mode off - resonance on the red - detuned side @xcite . provided @xmath309 , the described interaction term can be approximately replaced with the dispersive interaction : @xmath310 if the qubit is prepared and subsequently remains in its ground state during the driving interval , then we obtain the familiar ponderomotive interaction between the microwave and mechanical oscillator modes . for the above considered parameter values , we have for the dispersive coupling strength in either qubit case : @xmath311 interestingly , this upper bound on the achievable coupling strength is still orders of magnitude smaller than what is possible in a related scheme where the squid with mechanically compliant loop segment is instead imbedded within the planar waveguide , interrupting the centre stripline conductor @xcite . applying the theoretical analysis of reference @xcite assuming a dispersive coupling strength @xmath312 , microwave mode of @xmath313 and quality factor @xmath314 , and mechanical flexural mode of @xmath315 ( thus we are in the good cavity limit : @xmath316 @xcite ) and quality factor @xmath299 , we find that pump drive powers of magnitude @xmath317 are required in order to cool the flexural mode down to a few energy quanta starting from a temperature of , say , @xmath318 . such drive power strengths are unrealistic given the need to substantially attenuate the thermal johnson noise in the original ghz drive signal entering the rf line at room temperature @xcite . thus , the dispersive coupling in the presently considered scheme is not strong enough in order to achieve significant passive cooling . nevertheless , as emphasized in the companion paper @xcite , there is no upper thermal threshold to verifying mechanical superposition states , although for the considered device parameters and temperatures higher than a few tens of mk , we expect it to become difficult to tease out the recoherences in an echo experiment that are the signature of mechanical superposition states . as long as the mechanical resonator temperature can be maintained to within a few tens of mk ( i.e. , dilution fridge temperatures ) during the drive and subsequent qubit control stages , there should be no problem observing mechanical superposition states ; cooling is not essential for the scheme . we have described an implementation of the nanomechanical quantum superposition generation and verification scheme presented in the companion reference @xcite . the implementation is based on the circuit qed set - up with the incorporation of a mechanical degree of freedom formed out of a suspended squid loop segment located opposite the centre conductor of a coplanar waveguide . two qubit realizations were investigated , namely the capacitively coupled charge qubit with voltage - biased qubit - mechanical resonator coupling , and the inductively coupled flux qubit with current - biased qubit - mechanical resonator coupling . both qubit realizations were found to have comparable and feasible resonator coupling strengths that are adequate for the quantum superposition generation scheme @xcite . ultimately , less predictable qubit properties such as their coherence times will decide which implementation is the more promising of the two . nevertheless , the present considerations suggest that both implementations are equally worth pursuing in experiment . we especially thank a. rimberg and also e. buks , j. hertzberg , and k. schwab for very helpful discussions . mpb was partly supported by the national science foundation ( nsf ) under nirt grant no . cms-040431 , the us - israel binational science foundation ( bsf ) and by the foundational questions institute ( fqxi ) . ada was supported by the epsrc under grant ep / e03442x/1 . the most general solution to the wave equation ( [ waveeq ] ) for @xmath0 that satisfies the boundary conditions @xmath319 takes the following fourier integral form for @xmath8 : @xmath320\cr & & + \int dkb(k ) \sin(kvt ) \left[\sin(kz+\theta(k))-\sin(kz+kl-\theta(k))\right ] , \label{phisolneq}\end{aligned}\ ] ] with @xmath321 , @xmath9 , for a voltage antinode ( node ) solution . substituting eq . ( [ phisolneq ] ) evaluated at @xmath7 into eqs . ( [ voltsum2eq ] ) , ( [ gammaplus2eq ] ) , ( [ gammaminus2eq ] ) , and ( [ voltdiff2eq ] ) can serve as a starting point for analyzing the classical dynamics of the coupled cpw - squid system . because of the nonlinear force " terms in the squid part of the equations of motion ( [ gammaplus2eq ] ) and ( [ gammaminus2eq ] ) for @xmath322 , the coupled dynamics is expected to be quite nontrivial and rich in general . on the other hand , if the force terms had been linear , then a simpler normal mode analysis could be used with only two distinct @xmath323 values in ( [ phisolneq ] ) necessary for constructing the normal mode solutions . furthermore , provided the two normal mode @xmath323 values differ by much less than @xmath324 [ so that the dispersive differences in the @xmath325-dependent terms of ( [ phisolneq ] ) can be neglected ] , then the wave equation describing the cpw can be approximately replaced by a much simpler , single mode harmonic oscillator equation with linear coupling to the linearized squid . in other words , the cpw - linearized squid system can be modelled simply as two coupled oscillators . nevertheless , under conditions of weak coupling between the cpw and squid , i.e. , @xmath326 and @xmath327 , it may still be possible to approximately replace the cpw with a single microwave frequency harmonic oscillator bilinearly coupled to the nonlinear squid . we shall assume this is the case and will now proceed in this appendix to derive the approximate single harmonic oscillator - squid coupled equations of motion for the voltage antinode , capacitively coupled case . we stress , however , that there still remains the issue of establishing the full domain of accuracy of the solutions to these approximate equations by comparing with the classical solutions assuming the full multimode fourier integral form ( [ phisolneq ] ) . restricting to single mode forms of equation ( [ phisolneq ] ) , we have @xmath328\cos(kvt ) , \label{phifieldeq}\ ] ] where we have fixed the phase in equation ( [ phisolneq ] ) to be @xmath329 . defining @xmath330 , the phase amplitude at the squid location @xmath7 , we have @xmath331 and @xmath332 substituting equations ( [ phieq ] ) and ( [ phipeq ] ) into equation ( [ voltsum2eq ] ) and linearized equation ( [ gammaplus2eq ] ) and expressing the latter two using more concise notation , we have @xmath333 and @xmath334 where @xmath335 , @xmath336 , @xmath337 and @xmath338 . from the form of eqs . ( [ gammaconciseeq ] ) and ( [ phiconciseeq ] ) , we make a guess that the coupled oscillator equations of motion that yield approximately the same solutions take the form : @xmath339 however , because of the @xmath340 term in eq . ( [ phiconciseeq ] ) , it will only be possible to match the solutions to those of the coupled oscillator equations with to - be - determined constant coefficients @xmath341 , @xmath342 , @xmath97 and @xmath119 , provided that the two normal mode frequencies are sufficiently close to each other so that dispersive differences can be neglected . in order to determine the oscillator equation coefficients in terms of @xmath343 , @xmath344 and @xmath345 , we match the normal mode frequencies of eqs . ( [ gammaconciseeq ] ) and ( [ phiconciseeq ] ) to those of eq . ( [ canonicalosceq ] ) . solving first for the normal frequencies of eq . ( [ canonicalosceq ] ) to second order in @xmath345 , we obtain @xmath346 and @xmath347 the secular equation following from eqs . ( [ gammaconciseeq ] ) and ( [ phiconciseeq ] ) is @xmath348+\frac{\omega^4}{2}\epsilon^2=0 , \label{seculareq}\ ] ] where we have used @xmath349 . solving eq . ( [ seculareq ] ) for the smallest mode @xmath350 to second order in @xmath345 , we obtain : @xmath351 and @xmath352 comparing eqs . ( [ omegaplus2eq ] ) and ( [ omegaminus2eq ] ) with eqs . ( [ omegapluseq ] ) and ( [ omegaminuseq ] ) , we can clearly see that the cpw - linearized squid mode frequency @xmath353 has a different form than the @xmath353 arising from the coupled oscillator equations ( [ canonicalosceq ] ) . this is because the boundary conditions on the cpw result in mode solutions with nonlinear dispersion , as signified by the presence of trigonometric terms . however , if we assume that the modes @xmath353 and @xmath354 are sufficiently close to each other , then we can expand the cotangent term in the mode frequency difference . ( [ omegaplus2eq ] ) then becomes approximately @xmath355 validity of the second order in @xmath345 expansion together with this expansion in the frequency difference demands the following condition : @xmath356 the cpw - linearized squid mode frequencies now take the same form as those for the coupled harmonic oscillators . matching to determine the oscillator equation coefficients , we have : @xmath357 , @xmath358 , @xmath359 , and @xmath360 . substituting these values into eq . ( [ canonicalosceq ] ) and restoring the nonlinear @xmath48 force term , we finally obtain the desired coupled single oscillator - squid equations of motion ( [ gammaplus3eq ] ) and ( [ phiosceq ] ) . from eq . ( [ validitycondeq ] ) , we assume these equations are a good approximation to the full equations provided the difference in the characteristic frequencies for the squid and cpw dynamics , @xmath361 , satisfies @xmath362 the derivation of the single oscillator - squid equations for the voltage node , inductively coupled case run along similar lines , yielding equations ( [ gammaminus3eq ] ) and ( [ phiosc2eq ] ) . we assume these equations are a good approximation to the full equations provided the difference in the characteristic frequencies for the squid and cpw dynamics , @xmath361 , satisfies @xmath363 in this appendix , we derive the qubit langevin equations . this operator formulation provides a complementary approach for addressing the dissipative qubit dynamics to that of the more commonly employed fokker - planck , master equation approach which solves for the time - dependence of the qubit density matrix @xcite . the langevin equation formulation of the qubit dynamics naturally belongs to the ` in - out ' quantum optics approach to quantum measurement and control @xcite and can be straightforwardly applied to analyzing circuit qed set - ups @xcite . the qubit baths are modeled as a dense spectrum of oscillator modes and a rotating wave approximation is made for the assumed weak qubit - bath couplings : @xmath364 \cr & & + \int d\omega [ k_2^*(\omega ) a_2^+(\omega ) \sigma_z + k_2(\omega ) \sigma_z a_2(\omega ) ] , \label{qbbatheq}\end{aligned}\ ] ] where @xmath365 and @xmath366 are the qubit ( spin ) raising and lowering operators ( @xmath367 ) and the @xmath368 , @xmath369 are independent oscillator bath modes , the first coupling @xmath370 induces qubit decay through spin flips and where the second coupling @xmath371 causes pure dephasing . the equations for the bath modes are @xmath372 for the qubit , it suffices to consider the equations for @xmath219 and @xmath365 : @xmath373 - 2i\int d\omega ( k_1 \sigma^+ a_1-k_1^ * a_1^+ \sigma^- ) \label{sigmazeq}\ ] ] and @xmath374-i\int d\omega k_1^ * a_1^+ \sigma_z + 2i\int d\omega ( k_2^ * a_2^+ \sigma^+ + k_2\sigma^+ a_2 ) , \label{sigmapeq}\ ] ] where @xmath233 is the charge qubit hamiltonian ( [ concisechargehamiltonianeq ] ) or flux qubit hamiltonian ( [ concisefluxhamiltonianeq ] ) . integrating the bath equations of motion ( [ batheqs ] ) , we have @xmath375 substituting these solutions into equations ( [ sigmazeq ] ) and ( [ sigmapeq ] ) for @xmath219 and @xmath365 , we have @xmath376\cr & & -2\int_{t_0}^t dt ' \left[\sigma^+(t)\sigma^-(t ' ) \int d\omega |k_1|^2 e^{-i\omega ( t - t')}\right.\cr & & \left.+\sigma^+(t')\sigma^-(t ) \int d\omega |k_1|^2 e^{+i\omega ( t - t')}\right]\cr & & -2i \sigma^+ ( t)\int d\omega k_1 e^{-i\omega ( t - t_0 ) } a_1(\omega , t_0)\cr & & + 2i \int d\omega k_1^ * e^{+i\omega ( t - t_0 ) } a^+_1(\omega , t_0 ) \sigma^- ( t ) \label{zeffectivebatheq}\end{aligned}\ ] ] and @xmath377+\int_{t_0}^t dt ' \sigma^+(t ' ) \sigma_z ( t ) \int d\omega |k_1|^2 e^{i\omega ( t - t')}\cr & & -i\int d\omega k_1^ * e^{i\omega(t - t_0 ) } a_1^+ ( \omega , t_0)\sigma_z ( t)\cr & & -2\int_{t_0}^t dt ' \left[\sigma_z(t')\sigma^+ ( t)\int d\omega |k_2|^2 e^{i\omega ( t - t')}\right.\cr & & \left.-\sigma^+(t)\sigma_z ( t ' ) \int d\omega |k_2|^2 e^{-i\omega ( t - t')}\right]\cr & & + 2i\int d\omega k_2^ * e^{i\omega ( t - t_0 ) } a_2^+ ( \omega , t_0)\sigma^+ ( t)\cr & & + 2i\sigma^+ ( t)\int d\omega k_2 e^{-i\omega ( t - t_0 ) } a_2 ( \omega , t_0 ) . \label{peffectivebatheq}\end{aligned}\ ] ] we now make the so - called first markov approximation @xcite , neglecting the frequency dependences of the bath couplings @xmath378 to obtain @xmath373-\gamma_1 ( \sigma_z + 1 ) -2i\sqrt{\gamma_1 } ( e^{i\phi_1 } \sigma^+ a_1^{\mathrm{in}}-e^{-i\phi_1}a_1^{\mathrm{in}+ } \sigma^- ) \label{finalsigmazeq}\ ] ] and @xmath377-(\gamma_1/2+\gamma_{\varphi})\sigma^+\cr & & -i\sqrt{\gamma_1 } e^{-i\phi_1 } a_1^{\mathrm{in}+}\sigma_z + i\sqrt{2\gamma_{\varphi } } ( e^{-i\varphi } a_2^{\mathrm{in}+}\sigma^++e^{i\varphi}\sigma^+ a_2^{\mathrm{in } } ) , \label{finalsigmapeq}\end{aligned}\ ] ] where we have reparametrized @xmath379 in terms of the decay rate @xmath380 , phase @xmath3 and @xmath381 in terms of the pure dephasing rate @xmath382 and phase @xmath383 . we have also used the identities @xmath384 and @xmath385 . the ` in ' bath operators are defined as @xmath386 . 99 armour a d and blencowe m p 2008 _ new . j. phys . _ * xx * xxx wallraff a , schuster d i , blais a , frunzio l , huang r - s , majer j , kumar s , girvin s m , and schoelkopf r j 2004 _ nature _ * 431 * 162 blais a , huang r - s , wallraff a , girvin s m and schoelkopf r j 2004 _ phys . rev . _ a * 69 * 062320 bouchiat v , vion d , joyez p , esteve d and devoret m h 1998 _ phys . _ * t76 * 165 nakamura y , pashkin yu a and tsai j s 1999 _ nature _ * 398 * 786 chiorescu i , bertet p , semba k , nakamura y , harmans c j p m and mooij j e 2004 _ nature _ * 431 * 159 buks e and blencowe m p 2006 _ phys . _ b * 74 * 174504 zhou x x and mizel a 2006 _ phys . lett . _ * 97 * 267201 xue f , wang w d , sun c p , okamoto h , yamaguchi h and semba k 2007 _ new j. phys . _ * 9 * 35 blencowe m p and buks e 2007 _ phys . b _ * 76 * 014511 buks e , zaitsev s , segev e , abdo b and blencowe m p 2007 _ phys . e. _ * 76 * 026217 regal c a , teufel j d and lehnert k w 2008 _ preprint _ arxiv:0801.1827 arbel - segev e , abdo b , shtempluck o and buks e 2006 _ ieee trans . . supercond _ * 16 * 1943 schuster d i 2007 _ phd thesis _ section 6.1 devoret m h , wallraff a and martinis j m 2004 _ preprint _ cond - mat/0411174 wendin g and shumeiko v s 2006 _ handbook of theoretical and computational nanotechnology _ ed m rieth and w. schommers ( california : american scientific ) vol 3 ( _ preprint _ cond - mat/0508729 ) johansson g , tornberg l , shumeiko v s and wendin g 2006 _ j. phys . condens . matter _ * 18 * s901 geller m r , pritchett e j , sornborger a t and wilhelm f k 2006 ( _ preprint _ quant - ph/0603224 ) makhlin y , schn g and shnirman a 2001 _ rev . mod . phys . _ * 73 * 357 ithier g , collin e , joyez p , meeson p j , vion d , esteve d , chiarello f , shnirman a , makhlin y , schriefl j and schn g 2005 _ phys . _ b * 72 * 134519 frunzio l , wallraff a , schuster d , majer j and schoelkopf r 2005 _ ieee trans . appl . supercond _ * 15 * 860 marquardt f , chen j p , clerk aa and girvin s m 2007 _ phys . lett . _ * 99 * 093902 gardiner c w and collett m j 1985 _ phys . rev . _ a * 31 * 3761

we present here the results of applying a new chemo - evolutionary stellar population model developed by ourselves in a previous paper ( vazdekis _ et al . _ 1996 ) to new high quality observational data of the nuclear regions of two representative elliptical galaxies and the bulge of the sombrero galaxy . here we fit in detail absorption lines and 6 optical and near - infrared colors following two approaches : fitting a single - age single - metallicity model and fitting our full chemical evolutionary model . we find that all of the iron lines are weaker than the best fitting models predict , indicating that the iron - abundance is anomalous and deficient . we also find that the ca index at is much lower than predicted by the models . we can obtain good fits for all the other lines and observed colors with models of old and metal - rich stellar populations , and can show that the observed radial gradients are due to metallicity decreasing outward . we find that good fits are obtained both with fully evolutionary models and with single - age single - metallicity models . this is due to the fact that in the evolutionary model more than 80% of stars form within 1.5 gyr after the formation of the galaxies . the fact that slightly better fits are obtained with evolutionary models indicates these galaxies contain a small spread in metallicity .

astro-ph9701036

the study of the stellar populations and the distribution of metallicities plays an important role in our understanding of the star formation history of the galaxies . their stellar populations are expected to be more complex than those of , for example , globular clusters , which are thought to be composed of a single stellar population . in fact , burstein _ et al . _ ( 1984 ) found differences when comparing colors and spectroscopic features of globular clusters with galaxies . key parameters in the interpretation of the observed colors and the line - strengths are the metallicity and the age . the problem is that even in the simplest unresolved stellar systems their effects are very difficult to separate using only colors ( oconnell 1986 , renzini 1986 , buzzoni _ et al . _ 1992 ) . using colors together with absorption lines more accurate conclusions can be drawn . although every absorption line strength is dependent on different kinds of stars , in principle it should be possible to determine average metallicities or ages by carefully selecting features which more sensitive to the metallicity and others which are more sensitive to the age ( e.g. worthey _ et al . however the abundances of some elements may well evolve differently from those of others ( e.g. @xmath3-enhancement ) , and the conversion of ages and metallicities through models to observed colors and indices may be not unique due to problems in e.g. stellar evolution theory . finally , the large velocity broadening in giant elliptical galaxies implies that only the strongest lines can be used to obtain physical information from their spectra . in the process of understanding the stellar population of early - type galaxies we first developed a new spectrophotometric model , which can be used to interpret observed colors and absorption lines of galaxies ( vazdekis _ et al . _ 1996 , hereafter paper i ) . the model is based on the latest improvements in stellar evolution theory and on the most recent stellar libraries . instead of studying a large sample of galaxies using a few lines indices , as has been done before ( e.g. worthey _ et al . _ 1992 , gonzalez 1993 ) , we preferred to obtain high quality observations of three representative early - type galaxies ( two giant ellipticals and the bulge of the sombrero galaxy ) , but in many colors and absorption lines , and to make very detailed fits to each index , to understand better global ages and metallicities , and also to follow the abundances of individual elements . such analysis now is possible , since we could calibrate our observations using the large sample of stars from the extended lick - system ( worthey _ et al . _ ( 1994 ) hereafter wfgb ) . in this paper we have applied our spectrophotometric population synthesis model following both the single - age single - metallicity and the chemical evolutionary approaches . we address here the problem of whether the conclusions we obtain depend on the stellar population synthesis method we use . in the end we find that the use of many indices does yield interesting information , and we show that we can learn more than by using only a few indices , as has been done in the past . at the same time we study the stellar population gradients in the three galaxies . this paper is organized as follows : in section 2 we explain our observations and the method we use to derive the line - strengths . in section 3 we fit our population synthesis model and discuss the results obtained by fitting the data . finally in section 4 we present our conclusions . long - slit spectra of three well know early - type galaxies were obtained with the isis spectrograph on the 4.2 m wht in march , 1995 , at the observatorio del roque de los muchachos , la palma . the spectra were taken using both arms of the instrument , with a large format tek windowed to @xmath4 pixels , each with a size of @xmath5 ccd chip . in the _ blue _ we used a grating of @xmath6 giving a sampling of @xmath7 while in the _ red _ we used a @xmath8 grating giving a sampling of @xmath9 . our spectra were taken in the range @xmath10 . this configuration allowed us to cover almost the whole set of the absorption features contained in wfgb in addition to some uv features as defined in pickles ( 1985 ) . in fact , in the blue arm we covered the range 3700 to 4500 @xmath11 , while in the red arm we covered the range 4800 to 6300 @xmath12 . we could not observe the ca4455 , fe4531 and fe4668 features because they fall in the crossover region of the dichroic . the measured resolution was @xmath13 in the blue and @xmath14 in the red spectra . we also lost the tio@xmath15 since this index falls at the limit of the range covered . a set of stars of the sample of wfgb was also observed to calibrate our line - strength measurements . we positioned the slit on the major axis for ngc 4472 , at @xmath16 for ngc 3379 ( the major axis is at @xmath17 , see peletier _ 1990a ) and on the minor axis for the sombrero galaxy . the exposure times were @xmath18 for both frames . all data reduction was done with the iraf software package . the first step was subtracting the _ bias _ value calculated from the unilluminated portion of each frame . after this we flat - fielded using tungsten lamp exposures . next , the spectra were wavelength calibrated using cuar+cune calibration lamp exposures . the obtained pixel scale was @xmath19 @xmath20 for the blue and @xmath21 @xmath22 for the red spectra . the following step was the _ sky subtraction _ , for which the outer parts of each galaxy frame ( chosen at approximately @xmath23 from the center of the galaxy ) were averaged to produce a mean sky spectrum , which then was subtracted from each frame . the last step is the elimination of pixels affected by _ cosmic rays _ in each frame . we need a well - defined way to assign values to the strengths of features at different radii in the galaxy . for this purpose we used the expanded lick - system ( wfgb ) . here some indices ( the atomic features ) are defined as equivalent widths and some ( the molecular bands ) as ratios of line - depth to continuum in magnitudes and we have maintained these definitions . to measure these line - strength indices along the major axis we had to de - redshift each feature and the continuum on each side of it using the recession velocity corresponding to each spectrum . to calculate the rotation curve the spectra at each radius were cross - correlated with the observed spectrum of a stellar velocity standard star which looks most like the galaxy ( with spectral type k iii , see section 2.3.1 ) . this method is described in bottema ( 1988 ) , following the paper of tonry & davis ( 1979 ) . then we calculated the indices by co - adding a sufficient number of spectra in the spatial direction , so that a satisfactory continuum level was reached across the whole wavelength range . more details about the indices can be found in wfgb and burstein _ et al . _ ( 1984 ) . l|cc + index&vel . disp.&conv . to lick + uv cn & 1.002&- + ca h+k & 1.016&- + fei+cn & 1.030&- + cn1 & 1.036&-0.028 + cn2 & 1.046&-0.041 + ca4227 & 1.334&-0.35 + g - band & 1.032 & 0.11 + fe4383 & 1.074&-2.95 + h@xmath24&0.988&0.0 + fe5015 & 1.132&0.37 + mg@xmath25 & 1.016&-0.058 + mg@xmath15 & 1.006&-0.053 + mg@xmath26 & 1.106&0.16 + fe5270 & 1.134&0.22 + fe5335 & 1.253&0.10 + fe5406 & 1.234&-0.05 + fe5709 & 1.129&0.0 + fe5782 & 1.242&0.0 + nad & 1.068&0.29 + tio@xmath25 & 1.067&0.006 + l|cccc + index & poisson&rot . disp.&conv . lick + uv cn & 0.87 & 0.35 & 0.03 & - + ca h+k & 0.44 & 0.06 & 0.02 & - + fei+cn & 0.45 & 0.15 & 0.05 & - + cn1 & 0.005&0.001 & 0.001&0.017 + cn2 & 0.005&0.001 & 0.001&0.017 + ca4227 & 0.11 & 0.06 & 0.04 & 0.23 + g - band & 0.15 & 0.09 & 0.03 & 0.30 + fe4383 & 0.25 & 0.09 & 0.0 & 0.46 + h@xmath24&0.17 & 0.06 & 0.0 & 0.16 + fe5015 & 0.36 & 0.30 & 0.02 & 0.33 + mg@xmath25 & 0.003&0.002 & 0.0 & 0.005 + mg@xmath15 & 0.002&0.003 & 0.001&0.006 + mg@xmath26 & 0.17 & 0.09 & 0.06 & 0.17 + fe5270 & 0.17 & 0.08 & 0.02 & 0.20 + fe5335 & 0.18 & 0.08 & 0.02 & 0.21 + fe5406 & 0.15 & 0.03 & 0.01 & 0.14 + fe5709 & 0.09 & 0.08 & 0.01 & 0.13 + fe5782 & 0.09 & 0.15 & 0.0 & 0.15 + nad & 0.11 & 0.22 & 0.05 & 0.21 + tio@xmath25 & 0.003&0.001 & 0.001&0.005 + -3 cm l|cccc + index & a & @xmath27(a ) & b & @xmath27(b ) + uv cn & 33.6322 & 0.2918 & -3.6227 & 0.3837 + ca h+k & 25.4420 & 0.3357 & -2.9515 & 0.4413 + fei+cn & 13.1787 & 0.2556 & -4.6013 & 0.3229 + cn1 & 0.1302 & 0.0023 & -0.0639 & 0.0029 + cn2 & 0.1474 & 0.0032 & -0.0733 & 0.0040 + ca4227 & 0.7547 & 0.0411 & -0.0238 & 0.0519 + g - band & 5.3519 & 0.0775 & -0.3650 & 0.1019 + fe4383 & 5.5943 & 0.2005 & -1.2002 & 0.2532 + h@xmath28 & 1.5859 & 0.0597 & -0.3210 & 0.0746 + fe5015 & 5.4273 & 0.1132 & -1.0814 & 0.1454 + mg@xmath25 & 0.1765 & 0.0025 & -0.0455 & 0.0032 + mg@xmath15 & 0.3419 & 0.0038 & -0.0791 & 0.0047 + mg@xmath26 & 5.0203 & 0.0822 & -1.0044 & 0.1029 + fe5270 & 3.2075 & 0.0583 & -0.5363 & 0.0729 + fe5335 & 2.6646 & 0.0394 & -0.4522 & 0.0493 + fe5406 & 1.7023 & 0.0308 & -0.3325 & 0.0385 + fe5709 & 0.7610 & 0.0277 & -0.1625 & 0.0346 + fe5782 & 0.7529 & 0.0364 & -0.2218 & 0.0455 + nad & 5.3155 & 0.0434 & -1.3884 & 0.0542 + tio@xmath25 & 0.0477 & 0.0012 & -0.0096 & 0.0015 + + index & a & @xmath27(a ) & b & @xmath27(b ) + uv cn & 31.1736 & 0.4250 & -3.3511 & 0.4730 + ca h+k & 24.9050 & 0.2165 & -1.2801 & 0.2484 + fei+cn & 13.3326 & 0.1865 & -3.3352 & 0.2076 + cn1 & 0.1332 & 0.0020 & -0.0436 & 0.0022 + cn2 & 0.1589 & 0.0025 & -0.0480 & 0.0028 + ca4227 & 1.1783 & 0.0530 & -0.1077 & 0.0590 + g - band & 5.5165 & 0.0811 & -0.0531 & 0.0903 + fe4383 & 6.1080 & 0.1604 & -1.0448 & 0.1839 + h@xmath28 & 1.6280 & 0.0441 & -0.2115 & 0.0491 + fe5015 & 5.4281 & 0.1392 & -0.6750 & 0.1550 + mg@xmath25 & 0.1749 & 0.0018 & -0.0229 & 0.0020 + mg@xmath15 & 0.3445 & 0.0022 & -0.0454 & 0.0025 + mg@xmath26 & 4.8062 & 0.0524 & -0.3174 & 0.0583 + fe5270 & 3.2897 & 0.0595 & -0.4047 & 0.0662 + fe5335 & 2.4941 & 0.0664 & -0.1417 & 0.0761 + fe5406 & 1.6290 & 0.0462 & -0.2629 & 0.0514 + fe5709 & 0.7112 & 0.0346 & -0.0079 & 0.0385 + fe5782 & 0.7444 & 0.0267 & -0.1918 & 0.0298 + nad & 5.9980 & 0.0540 & -1.1499 & 0.0601 + tio@xmath25 & 0.0560 & 0.0011 & -0.0109 & 0.0012 + + index & a & @xmath27(a ) & b & @xmath27(b ) + uv cn & 33.4517 & 0.4922 & -6.8335 & 0.6023 + ca h+k & 24.8937 & 0.2858 & -2.0476 & 0.3734 + fei+cn & 13.9587 & 0.2868 & -6.3573 & 0.3510 + cn1 & 0.1414 & 0.0040 & -0.0901 & 0.0049 + cn2 & 0.1566 & 0.0040 & -0.1015 & 0.0049 + ca4227 & 0.7825 & 0.0477 & -0.1832 & 0.0584 + g - band & 4.9575 & 0.0521 & -0.3513 & 0.0680 + fe4383 & 6.5563 & 0.1437 & -1.3583 & 0.1831 + h@xmath28 & 1.0519 & 0.0456 & 0.2782 & 0.0545 + fe5015 & 4.8142 & 0.1603 & -0.7205 & 0.1916 + mg@xmath25 & 0.1847 & 0.0030 & -0.0577 & 0.0036 + mg@xmath15 & 0.3466 & 0.0040 & -0.0903 & 0.0048 + mg@xmath26 & 5.1513 & 0.0762 & -1.3318 & 0.0910 + fe5270 & 3.2815 & 0.0520 & -0.6962 & 0.0621 + fe5335 & 2.7584 & 0.0490 & -0.6646 & 0.0586 + fe5406 & 1.6711 & 0.0284 & -0.4386 & 0.0351 + fe5709 & 0.7455 & 0.0329 & -0.0401 & 0.0394 + fe5782 & 0.7162 & 0.0195 & -0.3548 & 0.0233 + nad & 5.7204 & 0.0832 & -2.0836 & 0.0994 + tio@xmath25 & 0.0568 & 0.0016 & -0.0168 & 0.0019 + since we want to compare our data with the spectrophotometric model we developed in paper i , which is based on the lick - system , we need to transform our results to that system . among the problems encountered in achieving this are the fact that the instrument we used ( isis ) has a different spectral response , the fact that wfgb did not flux - calibrate their stars and the higher resolution of our data ( @xmath29 in the blue spectra and @xmath30 in the red spectra ) compared to theirs ( @xmath31 ) . while the effect of a different instrumental response is almost negligible in the narrow indices it is important in the broader ones such as mg@xmath15 since both the index itself and the two pseudocontinua cover a wide range in wavelength . on the other hand the effect of having a higher resolution is important only in the narrower indices , which are affected in the same way as by the velocity dispersion broadening . therefore , to transform our indices to the expanded lick - system we first pre - broadened both spectra , the galaxy and the reference stars , so that to match the resolution of the lick - system . next we compared the thus obtained line - strength measurements of our stars with those given in wfgb to find an empirical average correction constant for each feature ( see table 1 ) . in particular , the stars used for this conversion were hr 3461 ( k0 iii ) , hr 4521 ( k3 iii ) and hr 4932 ( g8 iii ) ( see for details wfgb ) . we attribute the large correction factor found for the fe4383 feature to the fact that its right - hand pseudocontinuum falls at the edge of our observed spectra , entering the cross - over wavelength region of the dichroic used in the observations . the large correction constant of ca4227 is mainly due to the fact that this line is weak and the number of stars used for the conversion is small . as expected , we see that the most important corrections constants are those obtained for the molecular features . to correct the measured indices for instrumental resolution and velocity dispersion of the galaxy we used the spectra of a few k giants of the sample of wfgb to calibrate their effects on the galaxy indices in the same way as has been done before by e.g. davies _ et al . _ an auto - correlation of the central spectrum of each galaxy gave its corresponding broadening , which includes both the instrumental and the real velocity dispersion . after this we convolved the pre - broadened stellar spectrum with a gaussian of width @xmath32 calculated to match that observed in the galaxy . the velocity dispersion of this gaussian was ( in pixels ) : @xmath33 where @xmath34 is the observed velocity dispersion of the galaxy , @xmath35 is the measured instrumental profile ( obtained by the cross - correlation of the star with itself ) and @xmath36 is the conversion factor in @xmath37 ( see section 2.2 ) . for each index , _ i _ , an empirical correction factor , @xmath38 , defined as @xmath39 , was determined for all these stars . we performed this step by calculating all the indices in ew and never in @xmath40 . the resulting @xmath38 was found by taking the mean of the measured correction factors ( see table 1 ) . from a quick look at these factors we see that , in general , the higher the resolution and weaker the feature the higher the correction . for example , the weakest line in the blue spectra , the ca4227 , shows the highest correction ( see also fig . 13 ) , while in the red spectra this is the case for some of the iron features . finally , the calculated velocity dispersions amounted to @xmath41 for ngc 3379 , @xmath42 for ngc 4472 and @xmath43 for the sombrero . these values are not very different from those obtained by davies _ et al . _ ( 1983 ) who obtained @xmath44 for ngc 3379 and @xmath45 for ngc 4472 if we take into account that our error is about @xmath46 . for the sombrero galaxy kormendy ( 1988 ) obtained @xmath41 . we attribute the differences to the moderate seeing we had during the observations . the line - strength measurements for the three galaxies are shown in figs . 1 to 6 . in the three galaxies we see appreciable gradients for most of the indices . however we detect only weak gradients for h@xmath28 , the g - band and the weakest lines : ca4227 , fe5709 and fe5782 . the low slopes in h@xmath28 were found useful for constraining the galaxy formation scenarios by fisher _ ( 1995 ) . in the next sections we will concentrate on the study on the nuclear regions of these galaxies as well as on their gradients . to study the inner regions we have selected values for the indices corresponding to 5 from the center , and to study radial gradients we also selected values at 15 . we did not go further inwards because our seeing was poor ( @xmath03 ) , to avoid possible nuclear emission lines ( e.g. goudfrooij & emsellem 1996 , boroson & thompson 1991 ) , and because not much surface photometry is available in the inner regions . in fig . 7 we give a comparison with the data from the current literature for ngc 4472 . our mg@xmath15 index is slightly lower than the values of the other authors but we are in a better agreement with the data of saglia _ et al . _ for h@xmath28 the agreement is generally good while our values fall in the middle range of other observations . for the sake of brevity , we have not shown other comparisons but , for example , for ngc 3379 our mg@xmath15 and h@xmath28 indices obtained are in a very good agreement with davies _ et al . _ ( 1993 ) , while their is higher than ours . the main sources of errors are : the poisson noise , the error in the adopted zero - point of the calculated rotation curve ( used to de - redshift the spectra ) , the chosen velocity dispersion correction factors and the transformation to the extended lick - system . a quantitative estimate of the photon counting statistics has been carried out following the error analysis of spectroscopic features by rich ( 1988 ) . the error in equivalent width is @xmath47^{1/2}\ ] ] where @xmath48 is the dispersion , @xmath49 is the error in fixing the continuum , @xmath50 is the total number of counts in the line bandpass ( including negative counts ) , and @xmath51 is the mean continuum at the feature , defined as the value of the continuum point interpolated between the two continuum bands , at either side of the feature . the @xmath52 translates to a magnitude error @xmath53 : @xmath54 ^{1/2}\ ] ] the uncertainty caused by the adopted zero - point of the rotation curve was estimated by calculating the rotation curves separately for each star and then comparing the differences in the obtained line - strengths . the errors due to the correction for velocity dispersion and the conversion to the lick - system were estimated in the same way , by looking at the dispersion in the line - strengths obtained from different stars . typical errors of the various types are tabulated in table 2 . here one can notice that in general the conversion to the expanded lick - system introduces the most important uncertainty . this error is mainly due to the low resolution of the lick - data , and therefore we are assuming their uncertainties . in figs . 1 to 6 and in fig . 7 this error is given separately from the others in the corners of the plots . in this section we apply the spectrophotometric population synthesis model developed in paper i , which was especially designed to study early - type galaxies . briefly , the model makes predictions for the optical and ir colors and 25 absorption line indices . it is based on the new theoretical isochrones of bertelli _ et al . _ ( 1994 ) ( calculated with solar abundance ratios ) , but converted to the observational plane by using empirical calibrations of individual stars ( for details see paper i ) . to calculate line - strengths it uses the latest stellar spectral libraries ( mainly wfgb ) . the model calculates the properties of a stellar system , starting from a primordial gas cloud and calculating the chemical evolution in a way which broadly follows the mathematical formalism of arimoto & yoshii ( 1986 ) . the model can also be used to obtain the integrated colors of single - age single - metallicity stellar population ( ssp ) . in this work we will use both schemes . first we will apply a ssp model and later the full chemical evolutionary model . we took the colors of ngc 3379 and ngc 4472 from peletier _ et al . _ ( 1990a ) and peletier _ et al . _ ( 1990b ) while those for ngc 4594 are taken from he s & peletier ( 1993 ) . as was explained in section 2.3.3 , we took the colors and line - strengths of two different regions of the galaxies : at 5 and 15 from the center ( see table 4 ) . all these observables were calculated by linear interpolation between the neighbouring points and averaging between the two sides , except for the sombrero galaxy at 15 where we only took into account the observables of the dust free side . for ngc 3379 we took into account the fact that our spectra were taken with the slit positioned at @xmath55 . for example , a major axis position of 5 corresponds to a distance of 4.8 from the center along the slit . c|cc + + arcsec&@xmath56&fraction r@xmath57 + 5.0&0.29&0.13 + 15.0&0.86&0.40 + + 5.0&0.32&0.04 + 15.0&0.95&0.13 + + 5.0&0.35&0.08 + 15.0&1.05&0.24 + using the v - k - mg@xmath15 and b - v - mg@xmath15 diagrams we showed in paper i that to fit this set of galaxies solar metallicities or larger values are required . in fig . 8 we plot a number of color - color diagrams . in fig . 9 we have selected two key colors , @xmath58 and @xmath59 , and plotted a representative feature of each element versus these colors . finally , in fig . 10 we have plotted various index - index diagrams and since the number of features is large we selected as references for these plots three of the most commonly used indices in the literature : h@xmath24 , mg@xmath15 and . to obtain these figures we used the two forms of the imf defined in paper i : the _ unimodal imf _ with a power law of slope @xmath60 as a free parameter ( where 1.35 corresponds to the salpeter value ) , and a _ bimodal imf _ , which is equal to the unimodal imf above @xmath61 , but reduces the influence of stars with masses below @xmath61 . the two imf s used here have a lower mass - cutoff of 0.1 m@xmath62 and an upper mass - cutoff of 72 m@xmath62 . looking at the respective diagrams in these figures we see that a difference between models using the two imf s is seen only in the redder spectral indices such as tio@xmath25 or the redder colors such as @xmath59 . of course , the unimodal imf gives us higher line - strengths since the number of low - mass stars is higher and therefore the relative number of _ diluting _ blue stars is lower . looking at all these figures , but excluding those diagrams that contain iron features , we infer that to fit this set of galaxies we need either solar metallicity and very high ages ( say 15 gyr or even more ) or metallicities that are higher than solar and lower ages ( around say 8 gyr ) but never very low ages . we also see that the colors seem to be best fitted using metallicities lower than those used for line - strengths as one can see when comparing the best fits of the color - color diagrams with those of the index - index diagrams . this shows how important is the combination of colors and indices in this kind of studies ( see also sections 3.1.4 and 3.1.6 ) . from the index vs. h@xmath24 diagram one can conclude that the observed gradients inside each galaxy must be attributed to the metallicity rather than age variations . from the index - index diagrams we see that while most of the observed indices can be easlily fitted when plotted versus mg@xmath15 or h@xmath24 , this is not the case when plotted versus the index . in the following we will discuss various aspects of the fits . in fig . 9 we see that all the iron lines from the model are stronger than those observed , and in fig . 10 we see that the models are always able to fit the different index - index plots , except those against . this suggest that iron is anomalous and deficient compared to the other indices and colors . this shows that solar ratios are not adequate to fit this kind of galaxies , confirming the results of peletier ( 1989 ) , worthey _ et al . _ ( 1992 ) , gonzalez ( 1993 ) and davies _ et al . _ these results do not depend on whether we are working with a unique ssp or with a mixture of ssp s of different ages and/or metallicities ( as in section 3.2 ) as can be easily inferred from the different fe vs. mg@xmath15 plots . to try to explain these results we applied some simple models based on the hypothesis of @xmath3-enhancement . in reality , this phenomenon in principle affects all the parameters of stellar evolution and therefore a new set of isochrones ought to be obtained calculating observables for integrated stellar populations . as a first approximation we follow the conclusions of weiss _ et al . _ ( 1995 ) that for the calculation of @xmath3-enhanced indices one can use the isochrones calculated for solar abundance ratios , keeping the global metallicity constant but changing the ratios . in fig . 11 we find better solutions for the mg and fe indices , but worse for na ( at least with the adopted ratios ) . for an imf slope of 1.35 it does not make any difference whether a unimodal or bimodal imf is chosen . from the fe vs. mg@xmath15 plots we see that to fit the data we need [ mg / fe ] in the range 0.3 to 0.7 , in agreement with the results obtained by weiss _ this number does not seem to vary inside each galaxy . to conclude , we find that the metallicity ( in terms of z ) determined from the fe lines is different from that determined from the mg lines . we left the discussion of the ca4227 line for a separate section , since this is the worst fitting line . in fig . 12 we plot different representative features and the color @xmath59 versus this feature . apart from our models we also plot those of worthey ( 1994 ) . we see here that the discrepancy between predictions and observations is more dramatic than was found for the iron lines . independently of the imf slope , age or metallicity , the ca4227 line in the three galaxies is much lower than predicted by the models . it has been suggested before by oconnell ( 1976 ) and worthey ( 1994 ) that calcium tracks iron . in fact in this figure we see that the fits obtained when plotting this line vs. are not as bad as when plotting it vs. the other features . even so this line can not be fitted in an acceptable way . neither is the result due to a deficiency in the supporting library of stars , since for all the other indices ( except those dominated by fe ) reasonable fits can be achieved . it looks as if ca is depleted in this type of galaxies . if however we look at other ca lines we can not easily confirm this . we could not observe the ca4455 feature ( a blend containing an important contribution of ca@xmath1 ) as a result of the wavelength position of the dichroic used for the observations , and the ca ii near - ir triplet seems to give only a small depletion . ( 1990 ) for ngc 4472 found a total ew of @xmath63 for the two strongest lines , corrected for velocity dispersion . if we model this feature , using the stellar library of daz _ et al . _ ( 1989 ) and preserving their definitions ( for details see paper i ) , we see that our best fitting models predict an ew of around @xmath64 ( see tables 7 and 8) . this means that we find some depletion , but much less than for the ca4227 feature . as can be seen in the ca4227 vs. mg@xmath15 diagram of fig . 11 it is easy to understand that an @xmath3-enhanced scenario does not fit the observed ca4227 since this feature is almost all due to calcium , another @xmath3-element . the same plot also shows us that , even considering this scenario , the small contribution to this feature by fe@xmath1 ( around @xmath65 by inspection in the arcturus atlas ) can not explain the observed data . since the @xmath66 isotope is much less abundant than the @xmath67 we also can not attribute this effect to the fact that @xmath66 isotope is not produced in quantities appropriate to its solar abundance and because it is made in type ia supernovae ( those that ignite a carbon deflagration very near the chandrasekhar mass ) as recently suggested by woosley & weaver ( 1995 ) . rose ( 1984 ) measured ca@xmath68 h and k indices in stars , finding that hyades and pleiades dwarfs present strong ca@xmath68-emision compared with field dwarfs . he suggested that many of the principal absorption features in the blue spectral region are affected by stellar activity . however , in another paper ( rose 1994 ) , he himself measured the ca4227 line in 47 tuc and m32 , and reached the opposite conclusion to that which we find here . this suggests that this effect , if it is real , will be present only in bright early - type galaxies in the same way as mg tends to be enhanced with respect to fe in this kind of galaxies . if this result is general it will have important consequences for our knowledge of element - synthesis in type ii supernovae . however we can not strongly affirm the anomalous behavior of ca for two principal reasons . firstly , ca4227 is a weak line and secondly because , as one can see from table 1 , its conversion to the extended lick - system entails one of the highest corrections ( -0.35 ) . the error in this correction is quite large , as a result of which this depletion formally only is a @xmath69 result . to see whether the depletion is really present , we show the spectrum of ngc 4472 together with the star hr 3461 , for which wfgb measure an ew for the ca4227 of @xmath70 and we measure @xmath71 , while for the galaxy we find a value @xmath72 ( the last two ew converted to the lick - system ) . for the present we can say only that this will need more study and that observations of the ca@xmath68 triplet feature in the near - ir may give additional help . since the number of observables is large , we have implemented a _ merit - function _ , _ m _ , defined as @xmath73 where @xmath74 , @xmath75 are the synthetic and observed variable @xmath76 respectively and @xmath77 is the corresponding observational error . all these quantities must be expressed in magnitudes to perform the calculation of the _ m_. finally , @xmath78 is the assigned weight . in practice we determine a separate merit figure for the set of colors @xmath79 and for the set of line - strengths @xmath80 . the merit function determines the goodness of a fit for all observables taken together . the problem of optimizing a fit is now reduced to finding a minimum for this merit function . since the observational errors are non - zero the code calculates the maximum acceptable value for the merit , _ m@xmath81 _ using the same equation , but using the observational error instead of @xmath82 . any solution that gives a value of merit smaller than _ m@xmath81 _ is in principle acceptable . in this paper we have assigned the same total weight to the colors and to the lines . therefore the final global merit is calculated as follows @xmath83 to find the best solutions we first have to give weights to each color and index . as has been shown before it is clear that whatever the metallicities , ages or imf slopes are we can not fit the iron lines , possibly because of fe deficiency . for that reason in this paper we give a weight of 0 to indices composed mainly by iron . we also assign a weight of 0 to the ca4227 aborption line due to its apparently strong _ depletion_. to the other observables we give a weight of 1 except for the cn1 and cn2 , because they are the same index ( they only differ in the definition of the blue pseudocontinuum bandpass ) and therefore we assing a weight of 0.7 to each instead of 1 , so that their global weight is higher than 1 but lower than 2 . the given weights are summarized in table 5 ( case a ) . on the basis of the observational errors given in table 2 ( excluding the error from the conversion to the expanded lick - system ) and on the assigned weights , only merits smaller than 10.0 can be considered as acceptable fits . however , as one can see later , the obtained merits are often as high as 20 . these numbers are probably still acceptable , since we did not include the systematic uncertainties in converting our indices to the expanded lick system and systematic errors in the theoretical models . using this merit function we scanned the 3-parameter space ( z , age,@xmath60 ) for each galaxy . since our ssp models make use of the isochrones with a large step in metallicities ( see paper i ) we made a new grid of synthetic values by interpolating linearly between the output obtained for z=0.008 , z=0.02 and z=0.05 to obtain the synthetic observables corresponding to z=0.012 , z=0.016 , z=0.03 and z=0.04 . 14 shows the merit values obtained for ngc 3379 at the two selected positions 5 and 15 . to obtain this figure we used a unimodal imf with slopes varying from 0 to 2.3 , while the age was varied from 1 to 17 gyr . we plot the contours and the grey - levels for merits which are lower than @xmath84 . we choose this maximum acceptable limit to be safe and to avoid excluding even moderately possible solutions . in fig . 15 we scanned the same parameter space to fit this galaxy but using a bimodal imf . comparing the two figures it is clear that the unimodal imf gives worse fits and for that reason we will work throughout this paper with a bimodal imf . the same results were found in paper i using our full chemo - evolutionary population synthesis model . the main difference that we found looking at the fits obtained with the two imf s , is the fact that the best fits obtained using a unimodal imf require lower slopes than those obtained with the bimodal imf . we expect this since the unimodal imf yields a higher number of low - mass stars ( see paper i ) . as a general conclusion we can say that these galaxies can not be fitted assuming low values for the age . the inner regions of the three galaxies must be metal - rich since merit figures lower than 25.0 are only found for metallicities higher than solar . in particular ngc 4472 is even more metal - rich than the other two galaxies . looking at figs . 14 and 15 we see that to fit the two selected positions of ngc 3379 we must keep the age almost constant with values around 13 gyr , while the metallicity varies from values higher than solar ( z@xmath00.03 ) at 5 to lower than solar ( z@xmath00.016 ) at 15 . we also notice that it is not possible to obtain acceptable fits if we maintain the metallicity constant but change the age and/or the imf slope . however this is not the case for ngc 4472 ( see fig . 16 ) for which we find acceptable solutions if we decrease the age from 10 gyr to 8 gyr going outward and maintain the metallicity constant at @xmath0z=0.04 . of course we also find a fit if we keep the age constant around 10 gyr but decrease the metallicity from z=0.04 to z=0.03 going outward . for the bulge of the sombrero galaxy ( see fig . 17 ) a decrease in the metallicity is required when going outward and no age - variable metallicity - constant solution will do . we see that this galaxy seems to behave in the same way as ngc 3379 but with ages slightly lower . we can conclude that the observed radial index gradients are attributable to metallicity gradients rather than gradients in age . the fact that this conclusion can not be completely verified in ngc 4472 may be explained if we take into account that this galaxy is much larger than the other two , and thus the measured positions represent relatively small fractions of its effective radius ( see table 4 ) . [ [ testing - for - the - presence - of - dust . ] ] testing for the presence of dust . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to test for the presence of dust in these galaxies we have used the simple reddening law of rieke & lebofsky ( 1985 ) and applied it to the synthetic colors using steps in @xmath85 of 0.1 @xmath86 . we scanned again the whole parameter space ( z,@xmath60,age ) and examined the fits using the merit function . we found that the inclusion of dust does not improve the fits , except for the sombrero galaxy at 15 where we obtained a better fit including @xmath87 . from the previous color - index and index - index diagrams we have inferred that the iron lines are not well fitted by the models . however to be more complete one might think to include them in the fitting procedure and neglect the magnesium features or take other weighting distributions to check the validity of our selection and for a better interpretation of our fits . for that purpose , together with the choice of weights used throughout this work ( case a ) in which we have not given any weight to the fe and ca lines , we defined four alternative merit functions . in case b we have taken into account all the iron features and assigned each a weight of 0.43 , so that their global weight is 3 , the same as the sum of the magnesium features . in case c instead of neglecting the iron lines as in case a , we gave a weight of 0 to the three magnesium lines . in case d we only took into account the iron lines and h@xmath24 and have neglected all the other lines . finally , in case e we neglected both the iron and magnesium as well as the ca4227 line and gave a weight of 1 to each of the other features . then , as an illustrative example , we used our ssp models with a bimodal imf to fit ngc 4472 at 5 . in table 6 we summarize the best fits obtained with these merit functions , showing us that the required metallicities , ages and imf slopes vary little amongst the cases except for case d where e.g. the metallicity obtained is much lower than for the others . the fact that the merit worsens considerably when combining fe with other indices , and the fact that in case d , where we excluded all the indices except the iron lines , a fit is obtained which is not better than in case a ( where we kept most of the indices ) encourages us to use case a throughout this paper . from this excercise we can conclude that the iron lines yield worse fits and that these indices should not be used in combined fits with either colors or other indices , indicating that in this set of early - type galaxies the iron is in fact _ deficient_. another conclusion is that the global metallicity inferred must depend on whether we use magnesium or iron lines as the prime indicators . this result also shows how important it is to use colors and line - strengths together , since in this case the number of constraints is much higher than if we use colors alone . l|ccccc + & + color / index & a&b&c&d&e + u - v & 1&1&1&1&1 + b - v & 1&1&1&1&1 + v - r & 1&1&1&1&1 + v - i & 1&1&1&1&1 + v - j & 1&1&1&1&1 + v - k & 1&1&1&1&1 + cn1 & 0.70&0.70&0.70&0 & 0.70 + cn2 & 0.70&0.70&0.70&0 & 0.70 + ca4227 & 0 & 0 & 0 & 0 & 0 + g - band & 1 & 1 & 1 & 0 & 1 + fe4383 & 0 & 0.43&0.43&0.43&0 + h@xmath24&1 & 1 & 1 & 1 & 1 + fe5015 & 0 & 0.43&0.43&0.43&0 + mg@xmath25 & 1 & 1 & 0 & 0 & 0 + mg@xmath15 & 1 & 1 & 0 & 0 & 0 + mg@xmath26 & 1 & 1 & 0 & 0 & 0 + fe5270 & 0 & 0.43&0.43&0.43&0 + fe5335 & 0 & 0.43&0.43&0.43&0 + fe5406 & 0 & 0.43&0.43&0.43&0 + fe5709 & 0 & 0.43&0.43&0.43&0 + fe5782 & 0 & 0.43&0.43&0.43&0 + na d & 1 & 1 & 1 & 0 & 1 + tio@xmath1 & 1 & 1 & 1 & 0 & 1 + c|ccc|c + case & z & age(@xmath88 ) & @xmath60 & merit + a&0.04 & 10 & 2.3&16.4 + b&0.03 & 14 & 2.3&30.9 + c&0.04 & 8 & 2.3&37.1 + d&0.016&12&2.3&17.8 + e&0.05 & 8 & 2.3&17.8 + using v - k - mg@xmath15 and u - v - h@xmath24 diagrams we found in paper i that for a _ closed - box _ approximation these metal rich galaxies can not be fitted with a single imf which is constant in time using the full chemical evolutionary model . the reason for this is the impossibility of producing a dominant old but metal - rich population in our observed galaxies . the same result was found in casuso _ ( 1996 ) on the basis of the mg@xmath15 index . to solve the problem , we proposed a scenario invoking an imf skewed towards high - mass stars during a short initial period @xmath89 ( smaller than 1 gyr ) , followed by preferential low - mass star formation during the remaining time . in that paper the favored level of metallicity reached after this short period was around @xmath90 for fitting the inner regions of the three galaxies . we will use here only the bimodal imf , since we showed in section 3.1.4 that this works slightly better than the unimodal imf for a unique ssp model . this was also shown in paper i for the evolutionary model for a limited set of indices . once again we work with our merit function to account for the whole set of colors and line - strengths contained in the observed data . the parameter space to be scanned is therefore bounded by the values proposed in table 8 of paper i for the pairs ( @xmath91,@xmath92 ) which drive the gas to higher than solar metallicity at @xmath89 . however since in the present paper we not only want to fit the three galaxies at 5 , but also at 15 , it is also required to use the pairs of parameters which drive the metallicity to solar after the initial period , as in fact was shown in section 3.1.6 using the ssp models . we scanned this initial period of time with values of 0.2 , 0.5 and 1 gyr . finally we choose to scan the remaining imf slope @xmath60 between 1.3 and 2.3 , while the age of the galaxy is varied between 1 to 17 gyr . to illustrate our best fits , in figs . 18 to 20 , we have presented those merit figures using the weights of case a ( see section 3.1.6 ) which are lower than @xmath93 ( 25.0 ) for @xmath89 equal to @xmath94 . for the sake of brevity we did not plot here the cases for @xmath89 equal to @xmath95 and @xmath96 because , even if they show better fits than those in figs . 14 to 17 , they are in general slightly worse than those obtained with @xmath94 . for this reason we concentrate on this latter value for the initial period . we point out that both the final metallicity as well as the average metallicity are now non free input parameters as in ssp models , because they depend on the selected @xmath89 , @xmath91 , @xmath92 , @xmath60 and the age of the galaxy which determine the followed chemical evolution . we also point out that in these plots we selected the pairs ( @xmath91,@xmath92 ) to drive the metallicity to @xmath97 at @xmath89 except for the last two rows of merit diagrams of each figure in which the metallicity raises only to values around solar at @xmath89 . however , different chemical evolutions could be achieved by varying @xmath60 and the age of the galaxy , driving the metallicity to values different from those obtained at @xmath89 . for example , if we make the imf slope low and the age high the metallicity continues raising ( see for details paper i ) . a general trend which can be found looking at figs . 18 to 20 is that acceptable fits for the three galaxies are in fact possible . also one can notice that the grey levels are now darker than in figs . 14 to 17 , indicating lower merits and hence better fits than the ones obtained with ssp models . we see that the prediction of high ages for this set of galaxies is now robust . in fact , ages above @xmath98 are preferred to match the data . we also see that to fit the three galaxies we require values of @xmath60 in the second phase always greater than @xmath99 , indicating that after the first rapid formation the formed stars were primarily of lower mass than in the solar neighborhood ( @xmath100 ) . solutions involving @xmath101 for the second period are not acceptable , because then we can not stabilize the chemical evolution which continues raising the metallicity to higher values ( even higher than z=0.1 ) . all this shows that low - mass stars are very important in this kind of systems as was pointed out by e.g. faber & french ( 1980 ) . another general conclusion is that we require the sfr coefficient ( @xmath91 ) to be in the range @xmath102 ( with low @xmath89 ) meaning that our values are high compared to the solar neighborhood , estimated to be @xmath103 ( arimoto & yoshii 1986 ) , but much lower than the values proposed by these authors in their second paper ( 1987 ) to fit the ellipticals ( @xmath104 ) . the reason for this difference is that they stop the star formation after the occurrence of the galactic winds i.e. after 1 gyr of their model . in our study such high values do not emerge because we do not stop the star formation ( unless all the gas has been consumed , precisely when using such high values of @xmath91 with normal or even higher imf slope ) and therefore unacceptable values for the metallicity for the remaining gas are obtained . this happens especially when working at low imf slopes . this difference in input physics has important consequences to the fits obtained , since models which stop the star formation after a short initial period of time only allow the formation of stars with a mixture of metallicities but never with a mixture of ages . we also differ from arimoto & yoshii ( 1987 ) in the prediction of the slope of the imf where they assume to be @xmath105 ( slightly lower than salpeter ) . to compare our results with theirs we have to look at the results obtained with the unimodal imf . the prediction obtained by ourselves in paper i ( as can be deduced from fig . 20 of that paper ) is around @xmath106 , therefore higher than salpeter and than their estimate . we attribute this difference to the fact that they need this combination of lower than salpeter imf and very high star formation rates to yield the required high metallicities in their period previous to the occurrence of the galactic wind ( 1 gyr ) , while we obtain this in a shorter time by means of lower imf slopes than they require . bressan et al . ( 1994 ) also follow the galactic - wind scheme , without changing the salpeter imf , but requiring a shorter period of time before the appearence of the galactic winds . these short periods of initial different behaviour are in better agreement with our best fits . now turning our attention to the differences within galaxies and between them , we see that ngc 3379 and the bulge of sombrero galaxy appear to lie in the same regions of ( @xmath60,age ) parameter space for a given ( @xmath91,@xmath92 ) . we also see that , for the same set of parameters @xmath91 and @xmath92 , ngc 4472 at 5 gives better fits for ages which are greater by @xmath107 than the ones obtained for the other two galaxies at the same distances from their centers . however , by looking at the results obtained with the ssp models and the metallicities obtained in the best fits ( given in table 7 ) we are biased to think that after the initial period , @xmath89 , ngc 3379 and ngc 4594 reached lower metallicities than ngc 4472 , as also was shown in paper i. this also holds when varying the value of the imf slope for the remaining time . the full chemo - evolutionary models also produce the same conclusion as the ssp models when fitting the observed gradients , i.e. the metallicities of the outer regions are lower for ngc 3379 and ngc 4594 but not noticeably lower for ngc 4472 , probably because the latter is too large and therefore the radial separation of the observed points does not represent an important fraction of its effective radius . in fact in figs . 18 and 20 we see that the outer regions of ngc 3379 and ngc 4594 are only fitted in an acceptable way for the last two rows of merit figures , where the metallicity does not raise above solar at @xmath89 . for more details about the final and average metallicities see table 8 , where we summarize the most representative fits . we also point out that in this table comparing the b - v and v - k colors of ngc 3379 and sombrero galaxy we are inclined to think that the latter could be affected by a small amount of dust , since its colors are too red while the lines seem to be well fitted . we think it is likely that a reddening correction of e(b - v)@xmath00.05 should be applied . however , the improvement in our fit will not be very large , given the size of the error - bars . finally , in figs . 21 and 22 we show the chemical and fractional gas mass evolution as well as the distribution of the _ live _ stars as a funcion of metallicity , age and temperature for the two selected positions of ngc 3379 . we see that at 5 the stars mainly have metallicities larger than solar ( weighting in the v band ) but with approximately @xmath108 of the light coming from stars with metallicity of solar or lower . however at 15 we see that most of the stars ( around @xmath109 ) have solar metallicity . from the diagrams giving the predicted distributions of stars as functions of time one can also notice that the bulk of the stars were formed at an early - stage of the evolution of the galaxy , at ages lower than some @xmath110 . in general , for the three galaxies we obtain that the contribution to the light in the u band is around @xmath111 for giants and @xmath112 for dwarfs , in the v band it is @xmath113 for giants and @xmath113 for dwarfs , while in the k band the proportions are @xmath114 for giants and @xmath115 for dwarfs respectively . we find that using the evolutionary model most of the stars were formed in the early stages ( @xmath116 ) of the galactic evolution , as seen in figs . 21 and 22 . therefore , it is not surprising that qualitatively the fits from the evolutionary model are not very different from those of the single - age , single - metallicity models . somewhat better fits are obtained however , indicating that these galaxies probably have stars with a mixture of metallicities , as has been found also in our galactic bulge ( rich 1988 ) . it is also easy to understand that the evolutionary model gives somewhat larger ages , since the stars of lower - metallicities have to be compensated by giving the whole population a greater age . we have obtained high quality observations of almost the whole set of line - indices of the extended lick - system for three representative early - type galaxies , and have applied a new spectrophotometric chemo - evolutionary population synthesis model developed by ourselves in a previous paper ( paper i ) . we can make models which give good fits in all the colors and many of the most important line - indices . these fits however can not synthesize quantitatively a number of lines primarily from fe and ca . we find that 6 independent fe lines are too weak compared to lines of all other elements indicating that the iron - abundance is anomalous and deficient in the radial range of the galaxies that we studied . this implies that the global metallicity inferred must depend on whether we use magnesium or iron lines as the prime indicators . invoking @xmath3-enhancement one can obtain better fits for the iron lines , but other features such as nad then become worse if we follow the abundance ratios given in weiss _ et al _ ( 1995 ) . finally we find that the ca4227 is much fainter than predicted by the models . in general we find that the three galaxies require metallicities higher than solar for the inner regions while the ages are older than 10 gyr , and the observed radial gradients are due to metallicity decreasing outward . we also find that ngc 4472 is more metal - rich , than the other two galaxies . to fit this set of galaxies with the full chemical evolutionary population synthesis model we used the variable imf scenario ( defined in paper i ) which invokes an imf skewed towards high - mass stars in the begining , during a short period of time ( smaller than 1 gyr . ) , and towards low - mass stars later for the remaining time . the best fits indicate that dwarfs contribute @xmath117 to the u band , @xmath118 in v band and @xmath119 in k band . we find that slightly better fits are being obtained with the chemical evolutionary model than with the single - age , single - metallicity model , justifying the extra complications . however , since the predicted spread in metallicities is not very large , and since the bulk of the stars were formed at the very early stages of the galactic evolution ( at age lower than @xmath120 ) we conclude that the single - age single - metallicity stellar population models offer reasonable first order fits to this kind of stellar systems , especially if one wishes to avoid comprobational complexity . this study shows that it would be useful to extend the present analysis to include other features at shorter wavelengths in the uv region such as the indices of rose ( 1994 ) , and to the near - ir with indices such as na@xmath1 at @xmath121 , the ca@xmath68 triplet and the co or h@xmath15o features . to understand the stellar populations of the early - type galaxies and to e.g. disentangle age and metallicity ( jones & worthey 1995 ) , bressan _ et al . _ 1995 ) it will be important to introduce as many constraints as possible , by observing the galaxies in many calibrated absorption lines . we are indebted to the referee , j. gonzalez , who have done a very thorough work on our first version as a result of which the present paper have been greatly improved . we are grateful to j. gorgas for providing us with a set of observational results before publication . we also thanks a. weiss for his comments . this work was partially supported by grants pb91 - 0510 and pb94 - 1107 of the spanish dgicyt . arimoto , n. , & yoshii , y. , 1986 , , 164 , 260 arimoto , n. , & yoshii , y. , 1987 , , 173 , 23 bertelli , g. , bressan , a. , chiosi , c. , fagotto , f. & nasi , e. , 1994 , , 106 , 275 boroson , t. & thompson , i. , 1991 , , 101,111 bressan , a. , chiosi , c. , fagotto , f. , 1994 , , 94 , 63 bressan , a. , chiosi , c. & tantalo , r. , 1996 , , 311 , 425 bottema , r. , 1988 , , 197 , 105 burstein , d. , faber , s. m. , gaskell , c. m. & krumm , n. , 1984 , , 287 , 586 burstein , d. , davies , r.l . , dressler , a. , faber , s.m . , stone , r.p.s . lynden - bell , d. , terlevich , r. & wegner , g. , 1987 , 1987 , , 64 , 601 buzzoni , a. , gariboldi , g. & mantegazza l. , 1992 , , 103 , 1814 casuso , e. , vazdekis , a. , peletier , r. & beckman , j.e . , 1996 , , 458 , 533 davies , r.l . , efstathiou , g. , fall , s.m . , illingworth , g. , schechter , p.l . , 1983 , , 266 , 41 davies , r.l . , sadler , e.m . & peletier , r.f . , 1993 , , 262 , 650 daz , a. , terlevich , e. & terlevich , r. , 1989 , , 239 , 325 faber , s.m . & french , h.b . , 1980 , , 235 , 405 fisher , d. , franx , m. & illingworth , g. , 1995 , , 448 , 119 gonzalez , j.j , 1993 , phd . thesis , university of lick , santa cruz , california gonzalez , j.j . , gorgas , j. , 1996 , in _ fresh views on elliptical galaxies _ , eds . buzzoni , a. , renzini , a. , a. serrano , asp conf . ser . , 86 , 225 . goudfrooij , p. & emsellem , e. , 1996 , a&a , 306 , l45 he s , r. & peletier , r. , , 268 , 539 jones , l. & worthey , g. , 1995 , , 446 , 31 kormendy , j. , 1988 , , 335 , 40 oconnell , r. , 1976 , , 206 , 370 oconnell , r. , 1986 , in stellar populations , ed . c. norman , a. renzini and m. tosi , cambridge university press , p. 167 peletier , r.f . , 1989 , d. thesis , univ . of groningen peletier , r.f . , davies , r.l . , illingworth , g.d . , davis , l.e . & cawson , m. , 1990a , , 100 , 1091 peletier , r.f . , valentijn , e.a . & jameson , r.f . , 1990b , , 233 , 62 pickles , a.j . , 1985 , , 296 , 340 rich , r.m . , 1988 , , 95 , 828 renzini , a. , 1986 , in stellar populations , ed . c. norman , a. renzini and m. tosi , cambridge university press , p. 213 rieke , g.h . & lebofsky , m.j . , 1985 , , 288 , 618 rose , j.a . , 1984 , , 89,1238 rose , j.a . , 1994 , , 107,206 saglia , r.p . , bertin , g. , bertola , f. , danziger , j. , dejonghe , h. , sadler , e.m . , stiavelli , m. , de zeeuw , p.t . , zeilinger , w.w . , 1993 , , 403 , 567 terlevich , e. , daz , a. & terlevich , r. , 1990 , 242 , 271 tonry , j. & davis , m. 1979 , , 84 , 1511 vazdekis , a. , casuso , e. , peletier , r. f. & beckman , j. e. , 1996 , , 106 , 307 . * paper i*. weiss , a. , peletier , r.f . & matteucci , f. , 1995 , , 296 , 73 woosley , s.e . & weaver , t. , 1995 , , 101 , 181 worthey , g. , faber , s. & gonzlez , j. , 1992 , , 398 , 69 worthey , g. , faber , s. , gonzlez , j. & burstein , d. , 1994 , , 94 , 687 . * wfgb*. worthey , g. , 1994 , , 95 , 107

for any $ ] , the series converges almost everywhere on $ ] by a result of hardy - littlewood concerning the growth of the sums , but not everywhere . however , there does not yet exist an intrinsic description of the set of convergence for . in this paper , we define in terms of even continued fractions a subset of points of $ ] of full measure where the series converges . as an intermediate step , we prove that , for , the sequence of functions converges when to a function continuous on\setminus\{0\}$ ] with ( at most ) a singularity at of type or a logarithmic singularity ( ) . we provide an explicit expression for and the error term . finally , we study thoroughly the convergence properties of certain series defined in term of the convergents of the even continued fraction of an irrational number .

1211.5426

the famous lacunary fourier series @xmath13 was proposed by riemann in the 1850 s as an example of continuous but nowhere differentiable function . since then , this series has drawn much attention from many mathematicians ( amongst them , hardy and littlewood ) , and its complete local study was finally achieved by gerver in @xcite and jaffard in @xcite . in particular , its local regularity at a point @xmath14 depends on the diophantine type of @xmath14 , and it is differentiable only at rationals @xmath15 where @xmath16 and @xmath17 are both odd . in this article , we study the series defined for @xmath18 and @xmath19 by @xmath20 and we denote by @xmath21 its @xmath22-th partial sum . both are periodic functions of period @xmath23 in @xmath14 and @xmath24 in @xmath25 . for @xmath26 and @xmath27 the imaginary part of is . for any fixed @xmath25 , if @xmath28 , @xmath4 is in @xmath29 and it converges almost everywhere by carleson s theorem . it is not everywhere convergent however . one of the aim of this paper is to understand better the convergence of @xmath30 especially when @xmath27 . we set @xmath31 , @xmath32 , resp . @xmath33 if @xmath34 , resp . @xmath35 , and @xmath36 . we define @xmath37 with @xmath38 . we denote by @xmath39 and @xmath40 the integer part and fractional part respectively of a real number @xmath14 . for @xmath34 , @xmath41 and @xmath42 , we set @xmath43 this function is well - defined and if @xmath44 , the second integral is equal to @xmath45 ( because the series vanishes ) . we then define a function @xmath46 as follows : @xmath47 for simplicity , given a function @xmath48 , we will write @xmath49 for @xmath50 . the function @xmath46 will be particularly important in this paper . our first result is a consequence of the celebrated `` approximate functional equation for the theta series '' of hardy and littlewood ( proposition [ prop : h - l ] in section [ sec : hl ] ) , which corresponds exactly to the case @xmath44 in our theorem [ theo:1 ] below ; see @xcite where many references and historical notes are given . [ theo:4 ] let @xmath14 be an irrational number in @xmath51 whose ( regular ) continued fraction is denoted by @xmath52 , and let @xmath41 . @xmath53 if @xmath54 and @xmath55 then @xmath30 is absolutely convergent . @xmath56 if @xmath12 and @xmath57 then @xmath58 is absolutely convergent . conditions and hold for lebesgue - almost all @xmath14 . it is possible to prove a quantitative version of therorem [ theo:4 ] . we need to introduce more notations . we introduce the two transformations @xmath59-\frac tx \right\}.\ ] ] then , for all @xmath14 satisfying at least the conditions or , it can be proved ( and this would add nearly ten more pages to the paper . ] ) that @xmath60 where @xmath61 if @xmath62 is even , @xmath63 if @xmath62 is odd , and @xmath64 eq . holds very generally , the right - hand side converges quickly and the appearence of gauss transform @xmath65 is a nice feature . but this is at the cost of the simultaneous appearance of the operator @xmath66 and this makes looks very complicated , even when @xmath27 because @xmath67 in general . however , the underlying modular nature of @xmath30 implies that the transformation of @xmath2\setminus\{0\}$ ] given by @xmath68 is more natural than gauss in this specific study , and in particular it leads to another expression ( i.e , below ) for @xmath30 which is formally similar to but simpler . the comparison of both approaches is one of our motivations . our next theorem below explains what we mean by `` the modular nature of @xmath30 '' and the subsequent theorems are devoted to convergence conditions of @xmath30 ( mainly when @xmath27 ) in terms of series defined by the operator @xmath69 , as well as their relations with theorem [ theo:4 ] . [ theo:1 ] @xmath53 for any @xmath70\setminus \{0\}$ ] , @xmath41 , @xmath42 , we have the estimate @xmath71 when @xmath22 tends to infinity . the implicit constant depends on @xmath72 and @xmath25 , but not on @xmath14 . @xmath56 when @xmath73 , the function @xmath74 is continuous on @xmath75 , differentiable at any rational number @xmath15 with @xmath76 both odd , and @xmath77 are bounded on @xmath78 . @xmath79 when @xmath80 , the function @xmath74 is differentiable on @xmath81 and continuous at @xmath45 . the function @xmath82 is the same as the one used by cellarosi @xcite and fedotov - klopp @xcite . when @xmath80 , the proof of item @xmath56 yields only that @xmath8 is bounded around @xmath45 . see figures [ fig3 ] and [ fig4 ] for an illustration of theorem [ theo:1 ] . as @xmath83 , the left hand side of tends to @xmath46 when @xmath34 . the resulting `` modular '' equation @xmath84 holds a priori at least _ almost everywhere _ for @xmath85 for any fixed @xmath86 and @xmath87 , and theorem [ theo:1 ] shows in which sense we can say it holds _ everywhere_. for other examples of this phenomenon , see @xcite for instance . of course , if @xmath80 , holds for all @xmath14 . in fact , we will obtain a more precise estimate for the error term in , uniform in @xmath88\setminus\{0\}$ ] , @xmath89 and @xmath41 : @xmath90 where the constants in the @xmath91 on the right - hand side depend now at most on @xmath72 and are effective . we need this refinement to prove theorem [ theo:2 ] below . theorem [ theo:1 ] will be used to get informations of the convergence of @xmath92 in terms of the diophantine properties of @xmath14 . in the sequel @xmath93 denotes the @xmath94-th iterate of @xmath14 by @xmath69 . by @xmath23-periodicity of @xmath69 , eq . can be rewritten as follows ( when @xmath27 ) : for any @xmath70\setminus \{0\}$ ] , @xmath95 and integer @xmath89 , @xmath96 ( when @xmath27 , the error term @xmath97 in is absorbed by the error term in , for some constant that depends only on @xmath72 ; this is enough for the application we have in mind . ) the second sum @xmath98 in involves less terms than the first one for any irrational number in @xmath99 , because @xmath100 . hence for any fixed @xmath22 and @xmath14 , we can iterate because @xmath101 . after a finite number of steps ( say @xmath102 , which depends on @xmath14 ) , we get an empty sum @xmath103 on the right hand side together with a finite sum defined in terms of iterates of @xmath74 and a quantity we expect to be an error term ( i.e. , that tends to @xmath45 as @xmath22 tends to infinity under suitable condition on @xmath14 ) . we prove the following result . [ theo:2 ] let @xmath104 be an irrational number . + @xmath53 if @xmath105 and if @xmath106 then @xmath92 is also convergent and the following identity holds : @xmath107 @xmath56 if @xmath108 then @xmath109 is also convergent and the following identity holds : @xmath110 the series in , and do not converge everywhere . it was an open question whether such series converge lebesgue - almost everywhere . note that the results in the cited papers of cellarosi @xcite , kraaikamp - lopes @xcite , schweiger @xcite , and sinai @xcite give estimates for the average behavior of @xmath111 when @xmath62 tends to infinity , but these estimates are not sharp enough ( is ergodic with respect to a measure @xmath112 supported on @xmath2 $ ] but , in contrast with the ergodic theory of gauss transformation @xmath65 , the measure @xmath112 is infinite . as a consequence , the analogue of birkoff s ergodic theorem is not known and one must content with `` convergence in probability '' results ( see @xcite ) , which are not well suited to our study . ] ) to guarantee the almost everywhere convergence . it would be interesting to know if @xmath109 converges under the assumption of convergence of only . we now precise the diophantine content of theorem [ theo:2 ] . [ theo:3 ] let @xmath113 , @xmath114 , and set @xmath115 . @xmath53 if @xmath116 , then the series @xmath117 converges if @xmath118 @xmath56 if @xmath119 , then the series converges if @xmath120 @xmath79 if @xmath121 , then and impose the convergence of . @xmath122 if @xmath123 and @xmath124 then the series @xmath125 converges . the condition can be simplified according to the values of @xmath126 and @xmath127 , in terms of the irrationality exponent @xmath128 of an irrational @xmath129 , defined as @xmath130 it is well - known that @xmath131 for almost every real numbers @xmath14 . when @xmath12 , choosing @xmath132 , @xmath133 , one sees that implies . when @xmath134 , putting @xmath135 and @xmath136 , a simple computation shows that @xmath137 in theorem [ theo:3 ] . [ coro:4 ] @xmath53 if @xmath138 and @xmath139 is convergent , then the identity holds true . the series converges for every @xmath14 such that @xmath140 . @xmath56 if @xmath141 converges , then the equality holds true . the series converges for every @xmath14 such that @xmath142 . corollary [ coro:4 ] is not entirely satisfying , since the series @xmath92 converges ( even absolutely ) when a weaker condition on the standard convergents of @xmath14 holds ( theorem [ theo:4 ] , conditions and ) . the main reason for this discrepency is the factor @xmath143 : it is present in but not in , respectively in but not in - . our next result shows that the role of this factor is very important , even though its modulus is @xmath24 . we explain after the theorem why we are not able to keep track of it in our proof of theorem [ theo:2 ] . [ theo:5 ] @xmath53 let @xmath144 be a bounded function , differentiable at @xmath145 and @xmath146 ( in particular , if @xmath147 ) . then for any @xmath113 and any irrational number @xmath85 , the series @xmath148 converges . @xmath56 for any @xmath149 , any @xmath150 and any irrational number @xmath85 , the series @xmath151 converges if @xmath152 @xmath79 for any @xmath149 and any irrational number @xmath85 , the series @xmath153 converges if @xmath154 these convergence properties are essentially optimal . they are very different from the absolute convergence properties ( which are also essentially optimal ) , as stated in theorem [ theo:2 ] . in fact , in our proof of theorem [ theo:2 ] , only one technical detail impedes us to prove that formulas and hold true when conditions and are satisfied ( and not only when the more constraining conditions in @xmath53 and @xmath56 of corollary [ coro:4 ] hold ) . indeed , we show that the convergence of the series and is equivalent to the convergence of three auxiliary simpler series ( see equations and ) . for two of these series , their convergence follows from theorem [ theo:5 ] , and the convergence conditions are optimal ( i.e. , when conditions and are satisfied ) . for the third series , which contains heuristically a sort of `` error '' term , we do not have an estimate precise enough to apply theorem [ theo:5 ] , and we can only use theorem [ theo:3 ] and the conditions ensuring absolute convergence of the series and , which are stronger . nevertheless , we do believe that conditions and imply the identities and respectively . for @xmath25 not necessarily an integer , the iteration of leads to an identity , for which we have to introduce the following sequences of operators : @xmath155 then , for any fixed @xmath156 $ ] and @xmath95 , the following identity ( but , again , we skip the details to shorten the length of the paper . ] ) holds for almost every @xmath14 : @xmath157 this new representation of @xmath30 is similar to identity displayed after theorem [ theo:4 ] . for @xmath14 unspecified , eq . is simpler than when @xmath27 , because @xmath158 for all @xmath62 while neither @xmath159 nor @xmath160 necessarily vanish finally , it is easy to see that when @xmath14 has _ even _ partial quotients and @xmath27 , then the summands of and are equal . the simplicity of ( relatively to ) when @xmath27 was our motivation to make the detailed study of various series defined in term of the operator @xmath69 , for which apparently nothing was done in the direction of our results . finally , when @xmath80 , the series @xmath30 obviously converges absolutely for any real numbers @xmath14 and @xmath25 . it turns out that identities and hold for any @xmath25 and any irrational number @xmath14 , and with minor modification for any rational number @xmath14 as well . on the one hand , this is not difficult to prove for , whose right hand side converges very quickly . on the other hand , the convergence of the right - hand side of for all irrational @xmath14 is a consequence of theorem [ theo:5]@xmath53 applied with @xmath135 because for @xmath80 , @xmath74 is bounded on @xmath2 $ ] and differentiable at @xmath161 by theorem [ theo:1]@xmath79 . we note that @xmath69 is closely related to the theta group , a subgroup of @xmath162 of the matrices @xmath163 with @xmath164 , @xmath165 ; see @xcite . this relation has been used in the papers @xcite to study the ( non)-derivability of riemann series @xmath166 , culminating with jaffard s determination of its spectrum of singularities @xcite . it would be very interesting to know if jaffard s results can be recovered by a direct study of in the case @xmath26 and @xmath27 , which reads @xmath167 where @xmath168 is differentiable on @xmath2\setminus\{0\}$ ] , and continuous at 0 . the paper is organized as follows . we start by recalling some facts on regular and even continued fractions in section [ sec:1 ] . then , in section [ sec : hl ] , we prove theorem [ theo:4 ] . section [ sec : newthm ] , which is rather long , contains the proof of the approximate functional equation for @xmath169 ( part @xmath53 of theorem [ theo:1 ] ) . the second part @xmath56 of theorem [ theo:1 ] , i.e. the regularity properties of the functions @xmath8 , is dealt with in section [ sec:2 ] . theorem [ theo:2 ] and the diophantine identities and are proven in section [ sec:4 ] . finally , the standard and absolute convergence properties of the series and ( theorems [ theo:5 ] and [ theo:3 ] ) are studied in sections [ sec:10_2 ] and [ sec:10_1 ] . the regular theory of continued fractions is well - known , and is related to gauss dynamical system @xmath170\mapsto 1/x \mod 1 $ ] . in the following , we use capital letters when we refer to the regular convergent @xmath171 of an irrational number @xmath129 . we set @xmath172 where @xmath173 is the @xmath22-th iterate of @xmath65 . we write the rcf ( regular continued fraction ) of @xmath14 as @xmath174_r.\ ] ] recall that one has the recurrence relations @xmath175 it is also classical that in this case , for every irrational @xmath176 , @xmath177 this guarantees the convergence for every @xmath113 and @xmath129 of the series @xmath178 in theorems [ theo:2 ] and [ theo:3 ] , the natural underlying dynamical system is the one generated by the map @xmath179\setminus \{0\ } \mapsto [ -1,1 ] $ ] defined by @xmath180 . as is classical with the gauss map @xmath65 , using this transformation @xmath69 one can associate with each irrational real number @xmath70\setminus \{0\ } $ ] a kind of continued fraction : for every @xmath181 , denote by @xmath182 the unique even number such that @xmath183 , and set @xmath184 . then one has the unique decomposition called the even continued fraction ( ecf ) expansion ( see @xcite for instance ) @xmath185 the difference with the rcf expansion is twofold : first , only even integers @xmath186 are allowed in the decomposition , and second the integers @xmath187 , etc , may take the values @xmath24 and @xmath33 ( not only @xmath24 ) . for compactness , we write for @xmath188 @xmath189_e.\ ] ] now , given the ecf , we define the @xmath22-th convergent and the @xmath22-th remainder respectively as @xmath190 we use small letters for ecf , and capital letters for rcf . ecf expansions are obtained from the rcf expansions via the following iterative method . observe that for any positive integers @xmath191 and any positive real number @xmath192 , one has @xmath193 where the term @xmath194 appears exactly @xmath195 times . the procedure is then as follows : write the rcf expansion of a real number @xmath196_r$ ] as @xmath197_e$ ] , which looks like an ecf expansion , except that all @xmath198 are 1 and some integers @xmath199 may be odd . if all @xmath199 are even , then this expansion is indeed the ecf of @xmath14 . otherwise , consider the smallest index @xmath22 such that @xmath199 is odd , and apply to transform @xmath200_e\ ] ] into @xmath201_e.\ ] ] the odd number @xmath199 has been removed , and one iterates the procedure with this new ecf - like expansion , whose coefficients before @xmath202 are even . by uniqueness of the ecf expansion for irrational numbers , the expansion obtained as a limit is indeed the ecf expansion of @xmath14 . from the above construction , one also derives some useful properties between the even and the regular convergents . next proposition is contained in @xcite . [ prop : ecf2 ] let @xmath14 be an irrational number in @xmath51 . @xmath53 for any @xmath203 , if the regular convergent @xmath171 is not an even convergent @xmath204 , then necessarily @xmath205 is an even convergent . @xmath56 if the regular convergent @xmath171 is equal to the even convergent @xmath204 for some @xmath181 , and if @xmath205 is also an even convergent , then @xmath206 . @xmath79 if the regular convergent @xmath171 is equal to the even convergent @xmath204 for some @xmath181 , and if @xmath205 is not an even convergent , then for every @xmath207 , @xmath208 in particular , @xmath209 . hence , amongst two consecutive regular convergents , there is at least one even convergent , and an even convergent @xmath204 is either a principal or a median convergent of the regular continued fraction . this will be useful to prove theorem [ theo:3 ] . the next proposition gathers some useful informations about even continued fraction ( references include @xcite , ( * ? ? ? * , eq . ( 1.13 ) ) , and @xcite ) . [ prop : ecf1 ] for every irrational @xmath176 and every @xmath181 , we have @xmath210 and @xmath211 ( we will freely use these properties without necessarily quoting proposition [ prop : ecf1 ] . ) unfortunately there is no uniform convergence rate for this sequence that guarantees the convergence of series of the form for every @xmath14 . this is in sharp contrast with the classical continued fractions and equation . nevertheless we have found some optimal condition to guarantee the convergence of the sum @xmath212 , see theorem [ theo:3 ] . in this section , we obtain sufficient conditions of convergence of @xmath30 expressed in term of the usual regular continued fraction of @xmath14 . these conditions are simple consequences of the following proposition , due to hardy and littlewood @xcite . at the end of this section , we present an identity which , in principle , would be a qualitative version of theorem [ theo:4 ] . [ prop : h - l ] for any irrational number @xmath14 in @xmath51 with regular continued fraction @xmath52 and any @xmath41 , we have @xmath213 for any integers @xmath214 , where the implicit constant is absolute . eq . is a corollary of the approximate functional equation of hardy - littlewood for the theta series , which is exactly the case @xmath44 of our theorem [ theo:1 ] . in this case , the error term reduces to @xmath215 where the constant is absolute . the most precise version of the functional equation is given in @xcite . we do not reproduce the proof of : it is obtained by iteration of , see @xcite . we now prove theorem [ theo:4 ] . fix an integer @xmath216 . by abel summation , @xmath217 we set @xmath218 and let @xmath219 be the unique integer such that @xmath220 . we denote by @xmath221 the sum of the modulus of the summands of @xmath222 . then , for some constant @xmath223 independent of @xmath224 , @xmath225 for any @xmath226 , @xmath227 but the behavior of @xmath228 depends on whether @xmath12 or @xmath229 . if @xmath12 , then @xmath230 so that @xmath231 and the condition @xmath232 ensures the absolute convergence of @xmath233 . if @xmath234 , then @xmath235 hence , @xmath236 and the condition @xmath237 ensures the absolute convergence of @xmath238 . our choice @xmath218 is not arbitrary . indeed , it is such that @xmath239 both converge / diverge simultaneously when @xmath138 . its importance is lesser when @xmath12 where we could simply take @xmath240 . in the introduction , we presented identity obtained by iteration of with the operator @xmath69 . it is also possible to iterate with gauss operator @xmath65 . let us assume that @xmath241 , @xmath156 $ ] , @xmath95 are such that @xmath30 is convergent . then , letting @xmath83 in , we obtain @xmath242 with @xmath243-\frac tx\right\}.\ ] ] skipping all the details , it can be proved that the iteration of ( the finite version of ) yields the identity stated in the introduction , which holds for all @xmath14 satisfying at least the conditions - . the proof is rather long and intricate . we define the following functions , which are building blocks of the function @xmath46 : @xmath244 @xmath245 @xmath246 and @xmath247 due to the identity @xmath248 ( @xmath34 ) , it is easy to see that @xmath249 in this section , we present all the details of the proof of the theorem except the proofs of four lemmas , which are postponed to section [ ssec : lemmas ] . throughout , we assume that @xmath34 and explain in the end how to get the case @xmath35 . the method is borrowed to mordell @xcite in the version presented in @xcite . we introduce the parameters @xmath250 and @xmath251 . note that @xmath252 and @xmath253 . let us define the function @xmath254 which is holomorphic as a function of @xmath255 in @xmath256 $ ] . [ lem:1 ] for all @xmath257 , @xmath42 , @xmath34 and @xmath41 , we have @xmath258 we focus on the first integral in lemma [ lem:1 ] , i.e. @xmath259 we have @xmath260 where the assigned value of @xmath261 is in fact irrelevant . if @xmath262 , the sum is empty , equal to @xmath45 and is a tautology ; to avoid talking about empty sums , we assume from now on that @xmath263 but the results also hold when @xmath262 . now @xmath264 we set @xmath265 . we have that @xmath266 for each integer @xmath267 , so that @xmath268 where the second equality holds by cauchy theorem because @xmath269 is never in the closed horizontal strip defined by the lines @xmath270 and @xmath271 . ( indeed , we have simultaneously @xmath272 and @xmath273 . ) we now rewrite as @xmath274 where @xmath275 is the integral on the second line . we observe here that @xmath276 where the implicit constant is absolute . integrating , we thus obtain @xmath277 we will treat the three expressions in separately . the third term is simple because our choice of @xmath278 ensures that @xmath279 is independent of @xmath22 and is equal to the function @xmath280 defined in . to study the first term in , we change the summation index @xmath281 to @xmath282 : @xmath283 the sum is a partial sum of the series @xmath284 defined in . if @xmath44 , then the summand is equal to @xmath45 . if @xmath5 , we need the following lemma . [ lem : riv1 ] for all @xmath285 , @xmath5 , @xmath34 and @xmath41 , we have @xmath286 where the implicit constant is absolute . therefore , @xmath287 where the implicit constant depends on @xmath72 . to study the second term in , we do similar formal manipulations and obtain the representation @xmath288 where the sum in the integrand is a partial sum of @xmath289 . if @xmath44 , then the summand is equal to @xmath45 . if @xmath5 , we need the following lemma . [ lem : riv2 ] for all @xmath285 , @xmath5 , @xmath34 and @xmath41 , we have @xmath290 where the implicit constant is effective and depends at most on @xmath72 . it follows that @xmath291 because @xmath292 we now use the estimates , , in together with lemma [ lem:3 ] . this gives us @xmath293 since @xmath294 , the first error term absorbs the second one . we also want to replace @xmath295 by @xmath296 . this can be done at the cost of an error @xmath297 because @xmath298 . hence @xmath299 it remains to deal with the second integral in lemma [ lem:1 ] . for this , we have to distinguish between the case @xmath12 and the case @xmath300 . [ lem:3 ] if @xmath301 , for any @xmath302 , @xmath34 , @xmath303 , we have @xmath304 where the implicit constant depends on @xmath72 . if @xmath305 for any @xmath257 , @xmath34 , @xmath303 , we have @xmath306 where the implicit constant is absolute . in the case @xmath300 , the threshold @xmath307 determine which bound is the best . we now replace @xmath22 by @xmath308 and set @xmath309 . we get @xmath310 for any @xmath34 , any @xmath89 and any @xmath311 , @xmath312 and @xmath313 are both @xmath314 so that we can simplify the error term in to just @xmath315 to deal with the case @xmath35 , we simply take the complex conjugate of both sides of and change @xmath14 to @xmath316 and @xmath25 to @xmath317 . we finally deduce from all this discussion that _ for any @xmath104 , @xmath318 , any @xmath41 , any @xmath42 and any @xmath89 _ , we have @xmath319 where @xmath320 is @xmath34 and @xmath321 if @xmath35 . the implicit constants are effective and depend at most on @xmath72 . we obtain the expression for @xmath322 given in the introduction by means of the expression for @xmath323 . we follow mordell s method as it is presented in the book @xcite . we define @xmath324 and we integrate it over the parallelogram @xmath325 ( positively oriented ) defined by @xmath326 , @xmath327 , @xmath328 , @xmath329 and @xmath330 is a parameter . clearly , @xmath331 if @xmath255 is on @xmath332 or @xmath333 , we have @xmath334 where the implicit constant does not depend on @xmath335 . moreover , @xmath336 with @xmath337 on @xmath338 and @xmath339 on @xmath340 , so that @xmath341 it follows from and that @xmath342 thus @xmath343 where @xmath344 is for @xmath345 . we observe that @xmath346 , so that @xmath347 we now let @xmath345 to get the lemma . by definition of @xmath284 , the point is to estimate the series @xmath348 the modulus of which is obviously bounded by @xmath349 we observe that , since @xmath350 and @xmath285 , @xmath351 the lemma follows . we want to estimate the modulus of @xmath352 which is bounded above by @xmath353 this problem is similar to the one we dealt with in the proof of lemma [ lem : riv1 ] but this is a bit more difficult here because @xmath354 is not real . we set @xmath355 . we observe that @xmath356 where the implicit constant depends on @xmath72 only . we denote by @xmath358 and @xmath359 these two series and we write @xmath360 for some @xmath361 . ( note that @xmath362 . ) if @xmath363 , then @xmath364 for some implicit constant that depends only on @xmath72 . moreover , @xmath365 for some implicit constant that depends only on @xmath72 . adding the two upper bounds for @xmath358 and @xmath359 , we get @xmath366 which proves the lemma in this case . let us now consider the case @xmath367 . if @xmath368 , then @xmath369 so that @xmath370 for some implicit constants that depend only on @xmath72 . if @xmath371 , then @xmath372 so that again @xmath373 for some implicit constants that depend on @xmath72 . we conclude exactly as above . again , we follow mordell s method in the book @xcite . for any @xmath42 and @xmath374 , we have @xmath375 with @xmath376 . by definition of @xmath377 , we have @xmath378 and it follows that @xmath379 for any @xmath380 . moreover , for any @xmath381 and any @xmath257 , we have @xmath382 for some effective constant that depends only on @xmath72 . hence , for any @xmath42 , @xmath383 where the implicit constant depends on @xmath72 . since @xmath384 , we readily deduce that @xmath385 for some effective constant @xmath386 . to get the second upper bound , we need to distinguish the case @xmath12 and the case @xmath300 . * case @xmath301 . * we assume for the moment that @xmath387 and explain below how to remove this assumption . we set @xmath388 ; with @xmath389 , we get @xmath390 first , @xmath391 for some effective constant @xmath392 because @xmath393 . second , @xmath394 for some effective constant @xmath395 because @xmath393 . hence , @xmath396 with @xmath397 . in summary , we have obtained so far : @xmath398 with @xmath399 and since @xmath400 when @xmath387 , we deduce that @xmath401 for all @xmath402 . this completes the proof in this case . * case @xmath12 . * we set @xmath403 for simplicity . we then have ( with @xmath389 ) @xmath404 first , @xmath405 second , @xmath406 collecting both estimates , we obtain @xmath407 for some absolute constant . if @xmath408 , then @xmath409 and if @xmath410 , then @xmath411 . consequently , there exists an absolute constant @xmath223 such that @xmath412 for any @xmath257 and any @xmath34 . in this section , we prove that * the function @xmath74 is continuous on @xmath413 for any @xmath42 . * the function @xmath74 is differentiable at any rational number @xmath15 with @xmath76 both odd , for any @xmath42 . * the function @xmath414 is bounded on @xmath415 when @xmath416 . * the function @xmath417 is bounded on @xmath415 . * the function @xmath74 is differentiable on @xmath413 and continuous at @xmath9 for any @xmath80 . given the definition of @xmath74 by means of the function @xmath418 ( see ) , it is enough to prove these facts for @xmath419 . note that @xmath421 and thus @xmath422 where @xmath423 and @xmath424 it is clear that @xmath425 defines a differentiable function on @xmath420 . the series @xmath426 defines a continuous function of @xmath427 on @xmath428 . moreover , by lemma [ lem : riv2 ] applied with @xmath429 , we have @xmath430 this guarantees the continuity @xmath431 on @xmath432 , and that @xmath433 on @xmath432 . hence , @xmath74 is continuous on @xmath420 . to prove that @xmath74 is differentiable at any rational number @xmath434 with @xmath76 both odd , it remains to prove that this is the case of @xmath435 . integrating by parts , we get @xmath436 where differentiation under the sum is allowed by uniform convergence of @xmath437 . we observe that @xmath438 for some constant @xmath439 independent of @xmath281 and @xmath14 . therefore , @xmath440 on the one hand , the series @xmath441 can be termwise differentiated with respect to @xmath14 and it follows easily that the first term on the right hand side of is differentiable for any @xmath34 . on the other hand , the second term on the right hand side of is @xmath442 a result of luther @xcite ensures that the series @xmath443 is differentiable at any rational number @xmath434 , @xmath76 odd , for any fixed @xmath444 ( and this is no longer true when @xmath445 ) . since @xmath42 , this result is more than needed to complete the proof that @xmath435 is differentiable at any rational number @xmath434 , @xmath76 odd . moreover , when @xmath80 , the series is everywhere differentiable ( except at 0 ) by uniform convergence , hence @xmath74 is differentiable on @xmath420 . it was proved above that @xmath433 on @xmath432 and @xmath425 is continuous on @xmath420 . we will now establish that * @xmath425 is continuous at @xmath9 for any @xmath80 . * @xmath446 on @xmath420 for any @xmath416 . * @xmath447 on @xmath420 . the case @xmath80 is easy because @xmath448 is finite , by lebesgue dominated convergence theorem . this implies that @xmath74 is continuous at @xmath9 when @xmath80 . we assume from now on that @xmath449 $ ] . the change of variable @xmath450 gives @xmath451 if @xmath452 , then since @xmath453 we have @xmath454 and thus @xmath455 since the integral @xmath456 is convergent , by lebesgue dominated convergence theorem , we get that @xmath457 if @xmath458 , @xmath459 when @xmath460 and we can not invoke lebesgue s theorem because @xmath461 is not integrable over @xmath432 . @xmath462 so that @xmath463 on @xmath432 , and lebesgue s theorem entails that @xmath464 it thus remains to study the simpler integral @xmath465 because @xmath466 as @xmath467 . setting @xmath468 , we obtain @xmath469 we see here that as @xmath467 , the integral is `` close '' to the integral @xmath470 . the latter is convergent if @xmath471 but divergent if @xmath12 and we now have to distinguish between both possibilities . * case @xmath471*. observe that @xmath472 for @xmath473 and any @xmath419 . ( this is true when @xmath474 by using the taylor polynomial , and it goes to zero when @xmath474 tends to @xmath475 . ) moreover , @xmath476 . hence , @xmath477 therefore , @xmath478 we now prove that @xmath479 indeed , with @xmath480 , we have @xmath481 where we have used that @xmath482 using in gives @xmath483 and finally @xmath484 as @xmath467 . * case @xmath12*. we start in a similar way : @xmath485 therefore @xmath486 integrating by parts , we get @xmath487_0^{+\infty}+2\pi \int\limits_0^{+\infty } v \log\big(v+\rhob\frac{\sqrt{x}}{2}\big)e^{-\pi v^2}\dd v \notag \\ & = \log(1/\sqrt{x})+ \mathcal{o}(1 ) .\end{aligned}\ ] ] thus , @xmath488 when @xmath467 and the same estimate holds for @xmath489 as claimed in item 2 . for simplicity , we set @xmath490 . for any integer @xmath491 , we define an integer @xmath492 as follows : @xmath493 and @xmath494 for any integer @xmath495 . for instance , @xmath496 and @xmath497 . we also set @xmath498 . this integer is well defined . indeed , by proposition [ prop : ecf2 ] , it is obvious that @xmath499 and the right hand side tends to @xmath45 as @xmath500 . by definition , we have @xmath501 . it is clear that @xmath502 because otherwise if @xmath503 were bounded , we would have @xmath504 which is false . we write @xmath505 where @xmath506 and the function @xmath507 is bounded on @xmath2 $ ] by theorem [ theo:1]@xmath56 . let us define @xmath508 as the error term in , which is such that @xmath509 for any irrational number @xmath104 , eq . reads @xmath510 since @xmath511 is also an irrational number in @xmath99 , we can use with @xmath14 and @xmath22 replaced by @xmath511 and @xmath512 respectively , and iterate again the result . formally , we find that for any integer @xmath513 , @xmath514 with @xmath515 , becomes @xmath516 because @xmath517 , being an empty sum . under hypothesis of theorem [ theo:2]@xmath53 , the series @xmath518 converge absolutely ( recall that @xmath519 ) and this also forces the absolute convergence of the series @xmath520 because @xmath521 and @xmath522 is bounded on @xmath2 $ ] . hence , since @xmath523 with @xmath22 and by , the convergence of @xmath524 will follow from the proof that @xmath525 tends to @xmath45 as @xmath83 . for this , we observe that ( by theorem [ theo:1 ] ) @xmath526 @xmath527 and @xmath528 by hypothesis . hence , tannery s theorem can be applied and it yields that @xmath529 as expected . we now let @xmath530 in to get the identity @xmath531 the proof is more complicated than when @xmath229 . we can write a bound similar to but the corresponding right hand side depends on @xmath22 and we can not invoke tannery s theorem . instead we use an ad hoc inelegant method . for this , we first need more informations on @xmath532 . by multiple applications of the trivial inequality @xmath533 , we get @xmath534 we define @xmath535 as the maximal integer such that @xmath536 ; it is clear that @xmath537 with @xmath22 . we claim that @xmath538 for all @xmath539 . indeed , the sequence @xmath540 is non - increasing ( because @xmath541 ) and thus @xmath542 ( note that the sequence @xmath543 is increasing . ) therefore , yields that @xmath544 for any @xmath545 . moreover , for those @xmath62 , we also have @xmath546 . we write @xmath547 where the function @xmath548 is bounded on @xmath2 $ ] by theorem [ theo:1]@xmath56 . let us define @xmath549 as the error term in for @xmath12 , i.e. @xmath550 for any irrational number @xmath104 , eq . reads @xmath551 since @xmath511 is also an irrational number in @xmath99 , we can use with @xmath14 and @xmath22 replaced by @xmath511 and @xmath512 respectively , and iterate again the result . formally , we find that for any integer @xmath513 , @xmath552 from now on , the letter @xmath553 will stand exclusively for @xmath503 ( defined above ) , in which case becomes @xmath554 because @xmath555 , being an empty sum . under hypothesis of theorem [ theo:2]@xmath56 , the series @xmath556 converges absolutely and hypothesis implies the absolute convergence @xmath557 which in turn forces the absolute convergence of @xmath558 by boundedness of @xmath559 on @xmath2 $ ] . hence , since @xmath523 with @xmath22 and by , the convergence of @xmath560 will follow from the proof that @xmath561 tends to @xmath45 as @xmath83 . with @xmath562 , we split @xmath563 into three parts : @xmath564 the sum from @xmath565 to @xmath566 might be empty if @xmath567 , which would simplify the proof below . we assume that it is not empty , i.e that @xmath568 . for the two sums from @xmath569 to @xmath570 , we use the fact that ( for any @xmath89 , any @xmath34 ) @xmath571 first @xmath572 by definition of @xmath573 . second , for the sum from @xmath574 to @xmath570 , we observe that @xmath575 because @xmath576 for such @xmath62 by definition of @xmath553 . hence , @xmath577 the series with term @xmath578 is convergent ( by hypothesis ) , thus the expression in tends to @xmath45 as @xmath22 tends to infinity because @xmath565 tends to @xmath475 with @xmath22 . it remains to consider the case for @xmath579 . here , we use the fact that ( for any @xmath89 , any @xmath34 ) @xmath580 ( in this case , @xmath22 and @xmath14 are replaced by @xmath581 and @xmath582 respectively . ) hence , @xmath583 the properties @xmath584 and @xmath585 together imply that @xmath586 , so that @xmath587 therefore , becomes @xmath588 hypothesis and in theorem [ theo:2]@xmath56 now ensure that the right hand side of tends to @xmath45 when @xmath523 . this concludes the proof that @xmath563 tends to @xmath45 when @xmath83 . we now let @xmath530 in to get the identity @xmath589 we introduce a second dynamical system more adapted to our study ( see the work of schweiger @xcite ) . let us partition the interval @xmath590 $ ] into the double - indexed sequence of intervals @xmath591 \ \mbox { and } \ b(-1,k)= \left(\frac{1}{2k+1},\frac{1}{2k } \right ] .\ ] ] consider the map @xmath592 by @xmath593 and if @xmath318 @xmath594 it is trivial to check that for every irrational @xmath176 , @xmath595 . so we will prove theorem [ theo:5 ] using this transformation @xmath596 instead of @xmath69 . a key property is that @xmath596 is an ergodic transformation with a @xmath597-finite invariant measure @xmath598 with infinite mass ( due to the parabolic point @xmath24 for the mapping @xmath596 ) , whose density relatively to the lebesgue measure is given by @xmath599 further we need to study in details the orbit of a typical point @xmath14 under the action of the dynamical system @xmath600 . for this , we introduce the points @xmath601 , for every integer @xmath602 . observe that for every @xmath603 , @xmath604 [ lem:33 ] let @xmath603 be an integer , and let @xmath605 be such that @xmath606 then for every @xmath607 , one has @xmath608 equation follows from and the monotonicity of @xmath596 on the interval @xmath609 . further , @xmath610 the same holds for the lower bound . recall that @xmath611 stands for the sign of @xmath129 . [ lem:22 ] for every irrational @xmath176 , there are two possibilities : either there exists an integer @xmath612 such that for every @xmath613 , @xmath614 , or there exist two sequences of integers @xmath615 and @xmath616 satisfying the following : for every @xmath617 , @xmath618 , @xmath619 , and if we set @xmath620 , then : * for every @xmath621 , @xmath622 , * for every @xmath623 , @xmath624 , and the @xmath625 are all equal ( i.e. the @xmath626 have all the same sign ) . * @xmath627 is the unique integer greater than 1 such that @xmath628 assume that we are not in the first case . hence @xmath629 for infinitely many integers @xmath62 . consider the first integer @xmath62 such that @xmath629 , and call this integer @xmath630 . observe that it is possible that @xmath631 , if @xmath14 itself is greater than @xmath632 . on the interval @xmath633 $ ] , as already stated in lemma [ lem:33 ] , the map @xmath596 is strictly decreasing and concave , and has a derivative strictly less than one when @xmath634 . consequently , as long as @xmath635 stays in the interval @xmath636 , the sequence @xmath635 is strictly decreasing . moreover , using , there is a first integer @xmath637 such that @xmath638 and @xmath639 . iterating this scheme allows to find the sequences @xmath640 and @xmath641 . the fact that the @xmath626 have all the same sign follows from the fact that @xmath642 ) \subset \pm [ 1/2,1)$ ] . the third item follows from the definition of @xmath627 and lemma [ lem:33 ] . the rest of this section is devoted to the proof of part @xmath53 of theorem [ theo:5 ] . we denote @xmath643 a positive constant such that @xmath644 for every @xmath14 . we fix an irrational number @xmath85 , and for convenience we will denote @xmath645 if there exists @xmath612 such that @xmath646 for every @xmath613 , then the series converges . thus , we assume that @xmath647 for infinitely many integers @xmath62 . adapting lemma [ lem:22 ] , we immediately get : [ lem:22bis ] for every irrational @xmath176 , there are two possibilities : either there exists an integer @xmath612 such that for every @xmath613 , @xmath648 , or there exist two sequences of integers @xmath615 and @xmath616 satisfying the following : for every @xmath617 , @xmath618 , @xmath649 , and if we set @xmath620 , then : * for every @xmath621 , @xmath650 , * for every @xmath651 , @xmath652 . * @xmath627 is the unique integer greater than 1 such that @xmath653 the difference with lemma [ lem:22 ] is that we impose @xmath654 for every @xmath281 . this follows from the fact that the sequence @xmath655 is slowly decreasing when @xmath656 $ ] . it is clear that the convergence of the sequence does not depend on the first terms . hence , without loss of generality , we assume that @xmath657 . we now bound by above the partial sums @xmath658 * step 1 : * we separate the cases where @xmath659 and @xmath660 . @xmath661 if @xmath662 , then we have @xmath663 hence @xmath664 @xmath661 we now consider @xmath665 . we observe that @xmath666 by the second item of lemma [ lem:22 ] , all the @xmath667 are equal . we assume , without loss of generality , that they are equal to 1 . hence , we need to take care of the sum @xmath668 now , we use that the @xmath669 are all close to 1 . since @xmath144 is differentiable at 1 , we have for every @xmath670 @xmath671 using this equation , we split the sum into two parts : @xmath53 the sum @xmath672 can be rewritten ( since @xmath673 ) @xmath674 @xmath675 the second sum contains at most 7 terms all bounded in absolute value by 1 , hence it is less than 7@xmath676 . we now consider the sum between parenthesis above . fix @xmath677 , and let us write for every @xmath678 @xmath679 obviously from their definition , the @xmath680 are small positive quantities , and they are increasing with @xmath94 . we thus have @xmath681 in absolute value , this sum is less than @xmath682 ( since the @xmath680 are small and increasing ) . in addition , since the terms @xmath683 belong to the interval @xmath684 , one also has @xmath685 . finally , we get @xmath686 using lemma [ lem:33 ] ( in particular that @xmath687 ) , we can bound by above this sum by @xmath688 for some constant @xmath689 independent of the problem . @xmath56 the second sum @xmath690 is bounded in absolute value , by using again lemma [ lem:33 ] to find explicit upper bounds for the terms @xmath691 and @xmath692 . we find , for some constant @xmath689 depending on @xmath144 only , @xmath693 hence , going back to , we obtain @xmath694 for some constant depending on @xmath144 . the same holds if the @xmath626 are all close to -1 and not to 1 . * step 2 : * we separate the cases where @xmath695 and @xmath696 . @xmath661 if @xmath697 : all the terms @xmath698 satisfy @xmath699 . we deduce that @xmath700 @xmath661 we now consider @xmath701 . the same procedure as above ( in step 1 ) yields @xmath702 * step @xmath703 : * by an immediate recurrence , one obtains that for every @xmath704 , we have the following properties : @xmath705 we can now conclude regarding the convergence of the series . obviously by construction @xmath706 , and the first series @xmath707 converges , since the ratio between two consecutive terms @xmath708 is less than @xmath709 . for the second series @xmath710 , the same argument applies ( the ratio between two consecutive terms @xmath711 is less than @xmath712 ) . we consider the two sums and . using theorem [ theo:5]@xmath53 , the only problem may come from the terms with @xmath713 , since one can write @xmath714 , where @xmath144 is bounded and differentiable at @xmath24 and @xmath33 . the same holds for @xmath715 . thus we will focus on the convergence of the series @xmath716 and @xmath717 we will prove the absolute convergence because the complex terms do not play any role any more in this convergence . the next lemma is key in the proof . [ lemmefinal ] if @xmath718 , then there exists an integer @xmath22 such that @xmath719 moreover , with two different @xmath62 s such that @xmath713 correspond two different @xmath22 s . this follows from a fine analysis of the even and regular continued fractions . if @xmath14 is an irrational number whose rcf expansion @xmath720_r$ ] contains only even numbers , then one knows that the ecf expansion of @xmath14 is @xmath721_e$ ] , so @xmath722 for every @xmath181 , and the lemma follows from and the definition of @xmath65 . we assume that this is not the case . let us start with two observations on the shape of the mapping @xmath69 : ( r1 ) @xmath723 . when @xmath724 , then @xmath725 is given by @xmath726 . similarly , when @xmath727 , @xmath728 . subsequently , if @xmath729 for some @xmath94 , then , recalling the algorithm producing the ecf from the rcf , the @xmath94-term in the ecf expansion of @xmath730_e$ ] will be @xmath731 . ( r2 ) @xmath732 . if @xmath733 , then @xmath734 is necessarily even . in this case , the @xmath94-th term in the expansion of @xmath14 is @xmath735 for some integer @xmath22 . if @xmath733 , then @xmath734 is odd and @xmath736 is necessarily equal to @xmath737 . this simply follows from the form of the mapping @xmath69 . again using the algorithm , this amounts to change the rcf terms @xmath738 into @xmath739 in order to get the ecf expansion of @xmath14 , whose @xmath94-th term is necessarily @xmath740 for some integer @xmath22 . we treat the case where @xmath741 , the other case is symmetric . there are two possibilities : @xmath661 if @xmath742 : necessarily , we are at a step @xmath62 where , in the ecf expansion @xmath743_e$ ] of @xmath14 , there was a sequence of @xmath744 for some time before the index @xmath62 . then , this sequence of @xmath744 stops at the @xmath745-th iterate , since for @xmath746 , @xmath747 is not defined by @xmath726 any more . by our remark ( r1 ) above , this can be translated to the ecf and rcf expansions as follows : there is an integer @xmath22 such that @xmath748 where @xmath749 . in particular , this implies that @xmath750 . in this case , we know by propositions [ prop : ecf2 ] and [ prop : ecf1 ] ( as in section [ sec:10_1 ] ) that @xmath751 . this yields @xmath752 moreover , we have @xmath753 , with @xmath754 . we deduce that @xmath755 @xmath661 if @xmath756 : then one can apply our remark ( r2 ) above : for some integer @xmath22 , one has @xmath757 with either @xmath758 or @xmath759 for some integer @xmath22 . in both case we have @xmath760 for some integer @xmath22 , and the same estimates as above hold true . it is obvious from the construction that the integers @xmath22 corresponding to different integers @xmath62 are pairwise distinct . theorem [ theo:5]@xmath56-@xmath79 follows immediately . indeed , from lemma [ lemmefinal ] , we have @xmath761 and @xmath762 where we have used that @xmath763 . @xmath661 if @xmath768 for some integer @xmath22 and @xmath769 , then @xmath770 is either equal to @xmath771 or to @xmath772 . one also has @xmath773 . since in this configuration , @xmath199 is necessarily even , we deduce that @xmath774 . hence , @xmath775 @xmath661 if @xmath778 for some integers @xmath203 and @xmath779 , then necessarily @xmath780 , with equality when @xmath781 ( i.e. when @xmath782 ) . moreover , @xmath770 is equal to @xmath783 . hence , @xmath784 we deduce from this analysis that for some constant @xmath689 depending on @xmath785 and @xmath127 only , @xmath786 the first sum is clearly bounded by the second sum . for the third one , one sees that @xmath787 where we used that @xmath788 . finally , @xmath789 @xmath661 @xmath791 . we rewrite the general term of @xmath359 as @xmath792 then , observe that @xmath793 where @xmath794 because of our choice of @xmath127 , @xmath795 . hence , the convergence of the series @xmath358 implies the convergence of @xmath359 ( because the series @xmath796 converges as soon as @xmath795 for all real numbers @xmath14 ) . in terms of diophantine properties , @xmath359 converges for all real numbers whose irrationality exponent @xmath128 is smaller than @xmath801 . indeed , for such an @xmath14 , one has @xmath802 . writing @xmath803 we deduce that the series @xmath359 converges , since our choice for @xmath128 implies @xmath804 . @xmath661 @xmath805 . the convergence of the two series are close to be equivalent , but they are not , depending on the values of @xmath126 ( and @xmath806 ) . it is simpler to indicate that for all real numbers @xmath14 with irrationality exponent less than @xmath807 ( which coincides with @xmath808 ) , the two series @xmath358 and @xmath359 converge . the analysis is very similar to the one we performed in the last section , so we mention the main steps of the proof . we write for an irrational @xmath85 @xmath809 the first term in the right hand - side in dominated in absolute value by the second term , so we focus on the convergence of this term . equation yields @xmath810 the same distinction in three cases as in the previous section yields that @xmath811 the second term dominates the first one . then we use that @xmath812 to prove that @xmath813 using that @xmath788 , we finally get .

the sharp change in slope of the ultrahigh energy cosmic ray ( uhecr ) spectrum around ( the ankle ) , combined with evidence of a light but extragalactic component near and below the ankle and intermediate composition above , has proved exceedingly challenging to understand theoretically , without fine - tuning . we propose a mechanism whereby photo - disintegration of ultrahigh energy nuclei in the region surrounding a uhecr accelerator accounts for the observed spectrum and inferred composition at earth . for suitable source conditions , the model reproduces the spectrum and the composition over the entire extragalactic cosmic ray energy range , i.e. above . predictions for the spectrum and flavors of neutrinos resulting from this process are also presented .

1505.02153

the cosmic ray spectrum spans roughly eleven decades of energy , @xmath2 and has three major features : the steepening of the spectrum dubbed the `` knee '' at @xmath3@xmath4 @xcite , a pronounced hardening of the spectrum at @xmath5 , the so - called `` ankle '' feature @xcite , and finally a cutoff around @xmath6 @xcite . three additional more subtle features have been reported between the knee and the ankle : a hardening of the spectrum at around @xmath7 @xcite followed by two softenings at @xmath8 @xcite and and @xmath1 @xcite . the latter is traditionally referred to as the `` second knee '' . the variations of the spectral index reflect various aspects of cosmic ray production , source distribution and propagation . the first and second knee have straightforward explanations , as reflecting the maximum energy of galactic magnetic confinement or acceleration capability of the sources , both of which grow linearly in the charge @xmath9 of the nucleus ; the first knee being where protons drop out and the second knee where the highest-@xmath9 galactic cosmic rays drop out . as the energy increases above the second knee to the ankle , the composition evolves from heavy to light @xcite while the cosmic ray arrival directions are isotropic to high accuracy throughout the range @xcite . finally , as the energy increases above the ankle , not only does the spectrum harden significantly , but the composition gradually becomes heavier ( interpreting the data using conventional extrapolations of accelerator - constrained particle physics models ) @xcite . this observed evolution in the extragalactic cosmic ray composition and spectral index presents a major conundrum . a pure proton composition might be compatible with the observed spectrum of extragalactic cosmic rays @xcite when allowance is made for experimental uncertainties in the energy scale and the fact that the real local source distribution is not homogeneous and continuous @xcite ( although the sharpness of the ankle is difficult to accommodate ) , but a pure proton composition is incompatible with the depth - of - shower - maximum ( @xmath10 ) distributions observed by auger @xcite unless current extrapolations of particle physics are incorrect . moreover , a fit of the spectrum with a pure proton composition seems to require a very strong source evolution @xcite which leads to a predicted neutrino flux in excess of experimental limits @xcite . on the other hand , models which fit the spectrum and composition at highest energies , predict a deep gap between the end of the galactic cosmic rays and the onset of the extragalactic cosmic rays @xcite . models can be devised to fill this gap , but fine - tuning is required to position this new population so as to just fit and fill the gap @xcite . here we offer a resolution to this conundrum , by showing that `` post - processing '' of uhecrs via photo - disintegration in the environment surrounding the source , can naturally explain the entire spectrum and composition . in our model , extragalactic cosmic rays below the ankle are predominantly protons from nucleons knocked off higher energy nuclei in the region surrounding the accelerator , and the spectrum and composition above the ankle are predominantly dictated by the accelerator and propagation to earth . the model makes distinctive predictions about the spectrum and flavor ratios of neutrinos , which should enable it to be tested . if the ankle and the protons below it arise on account of our mechanism , we obtain a new constraint on uhecr sources beyond the hillas criterion and total - energy - injection requirements , namely that the environment around the source has the conditions giving rise to the required amount of photo - disintegration . up until now , photo - disintegration ( pd ) has been mainly considered as a danger inside the accelerator , as it would cut off the cosmic ray spectrum at energies such that the pd interaction length and the acceleration length are comparable . since the acceleration length increases with energy , whereas the pd interaction length generally decreases with energy , photo - dissociation acts as a _ low - pass filter_. the insight underlying the mechanism we propose , is that if the primary locus of pd is _ outside _ the accelerator , pd generally acts as a _ high - pass filter _ , permitting the highest energy cosmic rays to escape unscathed while the lower energy ones are disintegrated inside the source region , generating nucleons with energy @xmath11 of the original nucleus of mass @xmath12 . as we shall see , these spallated nucleons naturally produce the ankle feature , explain why extragalactic cosmic rays below the ankle are protonic , and account for the spectral index below the ankle . examples of systems in which the accelerator is embedded in a photon field and the cosmic rays are trapped by magnetic fields in that environment could be the dusty torus surrounding an active galactic nucleus or the interstellar medium of the star - forming region surrounding most young pulsars ; see also @xcite . the basic setup of our phenomenological model is illustrated in fig.[fig : illustration ] . the layout of the paper is as follows . in sec.[sec : ankle ] we introduce our model and in sec.[sec : datacomparison ] we compare its predictions with experimental data . details about particle propagation and the calculation of multi - messenger signatures are given in the appendices . section [ sec : conclusions ] contains our conclusions . to illustrate the mechanism we have identified to create the ankle and generate protons below , consider a system in which the accelerator ( also referred to as the source ) is embedded in an environment in which the cosmic rays are confined for some time by magnetic fields while interacting with the ambient radiation field . our essential simplifications are : _ ( i ) _ a fast acceleration mechanism and/or a low photon density inside the accelerator , _ no energy is lost except through an interaction , and whenever a nucleus interacts it loses one or more nucleons by photo - disintegration or photo - pion production ( in this case the nucleus loses a fraction of its energy corresponding to the reduction in its nuclear mass ) ; _ ( iii ) _ a cosmic ray either escapes without changing energy , with a rate @xmath13 , or the cosmic ray interacts one or more times before escaping ; _ ( iv ) _ @xmath13 and @xmath14 are independent of position in the source environment and depend only on @xmath15 of the nucleus . in this approximation the number of nuclei in a given energy range and with a specified @xmath16 decreases exponentially with time , with @xmath17 a fraction @xmath18 of the particles escape without interaction and the rest interact before escaping , so @xmath19 . note that @xmath20 and @xmath21 depend only on the ratio of the escape and interaction times , but not on the absolute value of either of them . a simple analytic treatment is instructive . to illustrate the low / high - pass filter mechanism , consider the case that the escape and interaction times are both power laws in energy , @xmath22 then @xmath23 where @xmath24 is the ratio of the escape and interaction time at reference energy @xmath25 . when @xmath26 , the source environment acts as a _ low - pass filter _ on the particles injected from the accelerator , leading to a cutoff in the escaping spectrum at high energies . this situation is typical of leaky box models of diffuse acceleration at time - independent shocks @xcite where @xmath27 because the higher the energy of the particle , the longer it needs to stay in the accelerator to reach its energy . by contrast , if the escape time decreases with energy , as in the case of diffusion in turbulent magnetic fields outside the accelerator , then it is possible to have @xmath28 leading to a _ high - pass filter _ on the energy spectrum of injected nuclei : the lower the energy , the more time the nuclei have to interact before escaping , leading to a hardening of the spectrum and lightening of the composition of nuclei escaping the region surrounding the source . the spallated nucleons have energies of @xmath29 ; these nucleons are most abundant at low energies and have a steeper spectrum @xmath30 . thus the high - pass scenario leads naturally to an ankle - like feature separating the nucleonic fragments from the remaining nuclei . the normalization and slope of the spectrum of spallated nucleons relative to that of the primary nuclei is determined by how thoroughly the primary nuclei are disintegrated , which is governed by the ratio of escape and interaction lengths of the most abundant primaries . to obtain a more realistic treatment of the interaction time , we must specify the shape of the spectrum of the target photons . in our work to date we have considered : _ ( i ) _ a broken power - law ( bpl ) , characterized by its peak energy @xmath31 and lower and upper spectral indices @xmath32 ( this is a simplified representative of non - thermal emission that allows for analytic calculation as discussed below and in appendix [ app : appendixb ] ) ; _ ( ii ) _ a black - body spectrum ; _ ( iii ) _ two types of modified black - body spectrum , which result from a reprocessed black - body in a dusty environment @xcite . details are given in appendix [ app : appendixa ] . for such peaky photon spectra the interaction time does not have the simple representation of ( [ taus ] ) but it does have a rather universal structure . in our actual calculations we adopt a numerical integration of talys @xcite and sophia @xcite cross sections using @xcite , but the analytic expression for @xmath33 derived in appendix [ app : appendixb ] for the bpl in the narrow - width approximation for the interaction cross sections , is qualitatively similar and useful for understanding . as can be seen in fig.[fig : vwl ] the folding of a single resonance with a broken power - law spectrum leads to a `` v '' shape curve for @xmath33 in a log - log plot for both photo - disintegration ( top panel ) and photopion production ( middle panel ) . combining both processes in narrow - resonance approximation yields an interaction time with a `` w '' shape , while numerical integration including the plateau for multi - pion production softens the `` w '' to what we shall refer to as an `` l '' shape for brevity , a shown in the bottom panel of fig.[fig : vwl ] . as evident from fig.[fig : vwl ] , below the inflection point for photodisintegration @xmath34 , the narrow - resonance approximation is good , while from the full numerical integration in the high - energy region @xmath33 is roughly constant , so using the bpl spectrum , we have the approximate representation : @xmath35 where formulae for @xmath36 and @xmath37 are given in appendix [ app : appendixb ] , and the parameter values for photodisintegration are to be used . returning to the discussion of @xmath38 in ( [ taus ] ) with ( [ tautau ] ) yields the fraction of nuclei which escape without interaction in a peaky photon spectrum . it is straightforward to see that if @xmath39 and the interaction time is described by an l - shaped curve , then @xmath40 has the properties of a high - pass filter . these conclusions do not depend on the exact shape of the photon spectrum . as can be seen in fig.[app : photon - spectra ] of appendix [ app : appendixa ] , the interaction times flatten to an l - curve as well if the photon density is assumed to follow a ( modified ) black body spectrum . as our fiducial example , we adopt a broken power - law photon spectrum as a simplified representative of non - thermal emission given by @xmath41 where @xmath42 is the photon energy , the maximum photon number density is at an energy of @xmath43 and following @xcite we take the slope parameters @xmath44 and @xmath45 . as we shall see later , any peaky spectrum gives similar results , with the position of the peak , @xmath43 , being the most important parameter besides the peak photon density . inspired by the energy dependence of the diffusion coefficient for propagation in a turbulent magnetic field , we model @xmath46 as a power law in rigidity @xmath47 , @xmath48 since only the ratio of escape and interaction times matters , and the @xmath15 dependence of this ratio is entirely determined once the spectral index of the escape time @xmath49 is specified , the remaining freedom in characterizing the source environment can be encoded by specifying the ratio of escape to interaction time for a particular choice of @xmath15 , which we take at @xmath50 for iron nuclei , denoted @xmath51 . in application to a particular source candidate , @xmath51 depends on the density of photons and the properties of the turbulent magnetic field that delays the escape of the uhecrs from the environment of their source . figure [ fig : tau&escspec ] , upper panel , shows the escape and interaction times in the fiducial source environment , as a function of the cosmic ray energy , for proton , he , n , si and fe ; the interaction times are calculated including both photo - disintegration and photo - pion production . the gross features of the energy dependence of the interaction times can be understood in the approximation of resonant interactions in the nucleus rest frame @xmath52 . at low cosmic - ray energies , reaching @xmath52 requires high photon energy ( @xmath53 ) , so that the interaction time decreases with increasing cosmic - ray energy as @xmath54 . however for high enough cosmic ray energy , the resonance can be reached in collisions with photons of @xmath55 . from here , as the cosmic ray energy increases , the photon density at the resonant energy decreases as @xmath56 , and correspondingly the interaction times increase . the laboratory energy of the inflection point of the interaction times for a cosmic ray nucleus of mass @xmath57 is at @xmath58 . the inflection point of the photo - dissociation times can be seen as a dip in the plot in the upper panel of fig.[fig : tau&escspec ] , e.g. , at around @xmath59 for iron nuclei . at slightly higher energy , photo - pion production becomes important , with the result that the energy dependence of the interaction time is roughly speaking an l - shaped curve in a log - log presentation . using these energy - dependent interaction and escape times , we propagate nuclei through the source environment with the procedure described in appendix [ app : appendixc ] . cosmic rays of some given composition are injected from the accelerator into the source environment with a power law spectrum and an exponential cutoff at some maximum rigidity . to keep the complexity of the fiducial model to a minimum , we inject only a single nuclear species and fix the injection spectral index @xmath60 , as expected for acceleration in young neutron stars @xcite . the particles escaping the source environment are then propagated through the intergalactic medium using the procedure explained in appendix [ app : appendixd ] . in total the fiducial model has 14 parameters , with 8 parameters allowed to float freely in the fit , as indicated in table [ table : fiducial ] . the spectral index and normalization of the galactic spectrum are free `` nuisance '' parameters with the best fit giving a spectral index of @xmath61 . this should be understood as an effective spectral index describing the cutoff of the galactic cosmic ray population , and hence can not be directly compared with the parameter reported by the kascade - grande collaboration @xcite , because their single - power law fit is driven by the `` low - energy '' data . the fraction of galactic cosmic rays at @xmath1 is 55% . the best description of the data is obtained with @xmath62si of maximum energy @xmath63 ; the impact of allowing other parameters to vary is discussed in following sections . normalizing this model to the observed flux at earth , we infer a comoving volumetric energy injection rate in crs at @xmath64 , above @xmath1 , of @xmath65 . the unmodified injection spectrum and the spectrum of escaping nuclei for this fiducial model are shown in the lower panel of fig.[fig : tau&escspec ] . at low energies , the nuclei are depleted relative to the injected flux because @xmath66 , but the escaping nuclei follow the original spectral index because in this example the interaction and escape times are parallel , as to be expected for @xmath67 . once the corner of the l - shape is reached , the fraction of escaping nuclei grows , leading to an apparent hardening of the spectral index . even for the simple case in which a single nuclear species is injected into the source environment , we obtain a complex evolution of the mass composition with energy . at low energies the composition is dominated by knock - off nucleons whereas at high energies the composition becomes heavier as the ratio of escape to interaction time drops and more heavy nuclei can escape before interacting . this fiducial model of interactions in the source environment is a very simple one , yet even so it offers a remarkably good accounting for the flux and composition at earth as determined by the pierre auger observatory . ( data from the telescope array ( ta ) are consistent with the auger results within systematic and statistical uncertainties @xcite and also can be well - fit ; we come to ta separately below . ) in fig.[fig : fit ] we compare the fiducial model prediction to the auger measured flux , from @xmath1 to above @xmath68 @xcite and to the mean and variance of the distribution of the logarithm of mass on top of the atmosphere , @xmath69 and @xmath70 @xcite . there is a good overall agreement between the model and the data . the shape of the spectrum is described well , including the ankle and the flux suppression . the model also qualitatively reproduces the increase of the average logarithmic mass with energy and the decrease of its variance . the neutrino signals of the fiducial model are shown in fig.[fig : neutrinos ] ; details of the calculation are given in appendix [ app : appendixe ] . an exciting aspect of our model for the ankle is the presence of a detectable anti - electron - neutrino flux from neutron @xmath71-decay , with a rate consistent with the nave estimate of @xcite . the right panel of fig.[fig : neutrinos ] shows the number of events as a function of energy predicted in ten years of ic86 , using the icecube acceptance for different neutrino flavors given in @xcite . in total , the fiducial model predicts 3.5 events in the range @xmath72 after 10 years of operation of icecube ( corresponding to about one year of operation for an upgraded icecube - gen2 detector @xcite ) . we emphasize the distinctive @xmath73 enrichment due to beta decay of spallated neutrons . the associated photon flux from nuclear de - excitation in our model is well below the fermi - lat data ( see appendix [ app : appendixe ] for more detail ) . photo - pion interactions at the source and during propagation produce an additional flux of photons via @xmath74-decay ; this is consistent with fermi - lat data , as follows : if the origin of the photons measured by fermi - lat is exclusively from these interactions , then from @xcite the associated diffuse neutrino flux saturates the icecube upper limit @xcite . since the neutrino flux in the fiducial model is below the icecube limit , it follows that also the associated photon flux is consistent with fermi - lat data . a more sophisticated realization of our mechanism than in the fiducial model must also respect the icecube limits , and therefore the fermi - lat data as well . [ cols="<,^,<,<",options="header " , ] in this section we discuss the impact of theoretical and experimental uncertainties on our model , as well as different choices for the fiducial parameters . to study the influence of the experimental systematic uncertainties on our fit , we have repeated the fit for all combinations of altering the measurements by @xmath75 , @xmath76 and @xmath77 of the quoted uncertainties on the energy and composition scale . we find that the best fit is obtained within the experimental systematics when shifting the energy scale up by @xmath78 and by shifting @xmath69 and @xmath79 corresponding to a shift of the shower maximum by @xmath80 . the best - fit values after the application of these shifts are shown in brackets in table [ table : fiducial ] . most notably , the peak energy of the photon spectrum decreases from 110 to 70 and the best - fit value of the spectral index of the escape time decreases from @xmath81 to almost @xmath82 . the neutrino flux at earth obtained for this fit is about 30% smaller than in case of the fiducial model . this is mainly due to the difference in the best - fit peak energy of the photon field in the source environment . the sensitivity of the neutrino flux to @xmath43 will be further discussed in sec.[sec : modelvarphotonspec ] . the overall description of the spectrum and composition is considerably improved , as can be seen in fig.[fig : fitsys ] . the model variations discussed below will therefore be performed based on shifted data . the interpretation of experimental air shower data in terms of mass composition relies on the validity of extrapolations of the properties of hadronic interactions to ultrahigh energies . using alternative models for this interpretation ( @xcite or @xcite instead of @xcite ) , decreases the value of the @xmath83 data points by about @xmath84 and leads to a worse fit of the data . if this difference between models gives a fair estimate of the uncertainties of the mass determination in both directions , @xmath85 , then a hadronic interaction model that leads to a heavier interpretation of auger data than would make the fit with the fiducial model even better , similar to the systematic shift in the composition scale discussed in the previous section . + it is remarkable that a good description of both the spectrum and mass composition at earth is possible by assuming only a single injected species at the source as assumed for simplicity in the fiducial model . however , depending on the astrophysical scenario , this might be an unrealistic assumption . in fig.[fig : massscan ] we explore the capability of our model to incorporate additional flux components of mass @xmath86 below and above the mass @xmath87 that gives the best fit for the fiducial single - mass model . as can be seen , our calculation allows for an additional proton or helium component as large as 80% and up to 70% for nitrogen . for an additional flux component with a heavy mass , the model is more restrictive as illustrated in the lower left panel of fig.[fig : massscan ] using @xmath88 . in this case , the description of the data considerably deteriorates for fractions above 10% . the reason for this behavior is twofold . firstly , the injection of too much iron at the source leads to a too heavy composition at earth as compared to the estimates from the pierre auger observatory . secondly , if the end of the cosmic ray spectrum is to be described by the maximum rigidity of iron nuclei , then the energy of secondary nucleons needed to populate the flux at and below the ankle is too small to describe the data ( the maximum energy of secondary nucleons is @xmath11 of the maximum energy of nuclei ) . if the cut - off of the flux is at higher energies , as suggested by the measurement of ta @xcite , then a larger fraction of iron primaries at the source can be incorporated , as shown in the lower right panel of fig.[fig : massscan ] . when using the ta data in the fit , as shown in fig.[fig : fitta ] , the spectrum can be described reasonably well even for an injected flux consisting of 100% iron nuclei . but in this scenario the composition at earth at ultrahigh energies is heavier than suggested by the interpretation of the @xmath89 data of auger . as an illustration of a more complex composition model , we use the abundances of galactic nuclei at a nucleus energy of 1 , which we read from fig.28.1 in @xcite . the flux fractions are 0.365 , 0.309 , 0.044 , 0.077 , 0.019 , 0.039 , 0.039 , 0.0096 , 0.014 , 0.084 for h , he , c , o , ne , mg , si , s , ar+ca , fe , respectively . the resulting fit is shown in fig.[fig : fitgalactic ] ( @xmath90 and @xmath91 ) . this example demonstrates that our mechanism for producing the ankle is working even when considering a complicated mix of primaries . + to have a concrete fiducial model , we needed to specify how the production of uhecrs varied over cosmological time scales . this is known as the source evolution , which we took to be in direct proportion to the star - formation - rate as would be expected in a source scenario such as young magnetars . in this section , we consider alternative evolutions of the source luminosity density described by the simple one - parameter functional form @xmath92 with @xmath93 and @xmath94 ranging from @xmath95 to @xmath96 . @xmath97 yields a uniform source luminosity distribution , @xmath98 corresponds to a strong evolution similar to the one of active galactic nuclei , and negative values result in sources that are most abundant or most luminous within the low - redshift universe as suggested in @xcite . the resulting fit parameters are displayed in fig.[fig : gammafits ] for three choices of the spectral index @xmath99 of the injected flux : @xmath82 , as in the fiducial model , @xmath100 as expected for stochastic shock acceleration and for letting @xmath99 float freely in the fit . as can be seen in fig.[fig : gamma0 ] , @xmath101 gives a poor description of the data for @xmath102 , but is a viable choice for closeby sources , in accordance to the findings of @xcite . for positive values of @xmath94 , a fixed value of @xmath103 gives a similar fit quality as the freely floating @xmath99 , but the latter converges to values larger than @xmath82 for source evolutions with @xmath104 ( cf . fig.[fig : gamma2 ] ) . for the `` traditional '' source evolutions with @xmath105 and the fit with @xmath103 we find that most of the parameters exhibit only a minor variation with @xmath94 , with the exception of the power - law index of the escape time @xmath49 ( fig.[fig : gamma4 ] ) and the power density @xmath106 ( fig.[fig : gamma4 ] ) . we conclude that our model for the ankle does not critically depend on the choice of the source evolution , but that for a given choice of @xmath94 we can constrain the allowed values of @xmath99 , @xmath49 and @xmath106 . + we repeated the model fits using alternative energy distributions of the photon density instead of the broken power law used in the fiducial model : a black body spectrum and two modified black body spectra . all four spectra are normalized to the same integral photon density and depend only on one parameter , the peak energy @xmath43 ( see appendix [ app : appendixa ] ) . the resulting fit results are shown in fig.[fig : photonfits ] for a freely floating spectral index @xmath99 and for source evolutions with @xmath107 . as can be seen , all four photon spectra describe the data equally well ( fig.[fig : photon0 ] ) . the best - fit values of the free model parameters are very similar and in particular the obtained peak values are within @xmath108 . we conclude that as long as the photon spectrum is `` peaky '' , the particular details of its shape do not influence the parameters of our model . the sensitivity of the fit to the peak energy is shown in the left panel of fig.[fig : chi2andnu ] . as can be seen , the @xmath109 deteriorates very quickly at low values of @xmath43 , but it is almost flat above the minimum . this feature can be easily understood recalling @xmath43 in the `` l - curve '' approximation introduced in sec.[sec : ankle ] : the smaller @xmath43 , the larger is the energy of inflection of the interaction length , @xmath37 . for too - small values of @xmath43 , the interaction and escape times are parallel over the full energy range and thus no high - pass filter is created . on the other hand , once @xmath37 is small enough , a further decrease changes only the flux at low energy , where the escaping spectrum is dominated by low - mass nuclei from spallation ( see e.g. fig.[fig : tau&escspec ] ) which can be compensated by adjusting other parameters such as @xmath51 . + to first order , our model can therefore only give a lower limit on the peak energy of the photon flux in the source environment . however , future limits or observations of neutrinos in the 10 - 100 range will help to constrain this important source property , because the number of predicted neutrinos strongly depends on @xmath43 , as shown in the left panel of fig.[fig : chi2andnu ] by the superimposed open symbols . a larger peak energy of the ambient photon environment increases neutrino production at the source in two ways . firstly , shifting @xmath37 to lower energies ( and compensating as necessary by adjustment of @xmath110 ) moves the interaction times of protons closer to the escape time and correspondingly additional neutrinos are produced via photo - pion production of protons ( compare e.g. the red curves at around @xmath111 in the upper panel of fig.[fig : tau&escspec ] ( @xmath112 ) to the ones in fig.[fig : tausys ] ( @xmath113 ) ) . secondly , increasing @xmath43 moves the minimum of the interaction time for photo - pion production of nuclei to lower energies . since the neutrinos from photo - pion production carry a larger fraction of the nucleon energy than the neutrinos from neutron decay after photo - dissociation , this increases the neutrino flux as well . it is tempting to give a quantitative interpretation of the @xmath114-curve of fig.[fig : chi2andnu ] in terms of a lower limit on @xmath43 and the number of neutrinos . however , the minimum of @xmath114 is far away from @xmath115 which assuming this model is correct is indicative of experimental systematics or an under - estimation of the experimental uncertainties or of deficiencies in the modeling of hadronic interactions in the atmosphere needed to interpret the data in terms of mass composition ( see above ) . in the absence of a concrete explanation we follow the pdg @xcite and rescale the uncertainties by a common factor @xmath116 to bring the rescaled @xmath117 to 1 . this rescales the @xmath114 value of any given model so that the number of standard - deviations it is from the minimum is given by @xmath118 . this yields an approximate lower limit on @xmath43 at @xmath119 of @xmath120 and @xmath121 assuming the validity of the fixed fiducial parameters given in table [ table : fiducial ] . the corresponding lower temperature limits are 180 k , 125 k and 100 k for the black body spectra with @xmath122 , 1 and 2 respectively . the lower limit on the neutrino spectrum is shown in the right panel of fig.[fig : chi2andnu ] . in addition to interactions with the background photon field , nucleons and nuclei can also scatter off hadrons in the source environment . in this paper we assume that the density of hadronic matter in the source environment is low enough that such hadronic interactions can be neglected . for any concrete astrophysical realization of our scenario , one must check and if necessary include hadronic interactions in the source environment . production of @xmath123 s and @xmath74 s in hadronic collisions could significantly increase the fluxes of neutrinos and photons emitted in the eev energy range . fast - spinning newborn neutron stars provide a particular example @xcite . precise estimates of the impact of hadronic collisions on the predictions of our model will be presented in a separate publication . the results presented here are valid for all astrophysical systems in which the interactions are dominated by photo - nuclear processes . in this paper we have proposed a new explanation for the ankle in the cosmic ray spectrum , and for the evolution with energy of the composition of extragalactic cosmic rays : from light below the ankle to increasingly heavy above . when nuclei are trapped in the turbulent magnetic field of the source environment , their escape time can decrease faster with increasing energy than does their interaction time . under these conditions , only the highest energy particles can escape the source environment unscathed , and the source environment acts as a high - pass filter on uhecrs . nuclei below the crossover energy such that @xmath124 interact with photons in the environment around the source , with ejection of nucleons or alpha particles and consequent production of a steep spectrum of secondary nucleons . the superposition of this steeply falling nucleon spectrum with the harder spectrum of the surviving nuclear fragments creates an ankle - like feature in the total source emission spectrum . above the ankle , the spectrum emerging from the source environment exhibits a progressive transition to heavier nuclei , as the escape of non - interacting nuclei becomes efficient . abundant production of @xmath125 s is a signature of this mechanism . we illustrated the high quality of the fit which can be obtained to the auger data , with a fiducial model in which nuclei are accelerated up to a maximum rigidity found to be @xmath126 v , with spectrum @xmath127 , and are then subject to photo - disintegration in the vicinity of the accelerator before escaping for their journey to earth . we showed that the details of the photon spectrum around the accelerator are unimportant , except for its peak energy . the other important characteristic of the environment is the photon density relative to the magnetic diffusivity , which we characterized in a very simplistic way ( through a single parameter ) in this initial study . we studied the sensitivity of the mechanism to the energy - scale uncertainty and hadronic - interaction - modeling uncertainty , which affects the composition inferred from the atmospheric shower observations , and also used the ta spectrum instead of the auger spectrum . the conclusion of these studies is that a good quality fit can be obtained in most cases , but details of the fit parameters such as the composition and maximum energy characterizing the accelerator change . a corollary is that until these systematic uncertainties in the observations and their interpretation are reduced , such details of the accelerator can not be reliably inferred from the data . the fiducial model parameters needed in the fits are such that the scenario can be reasonably achieved in at least one type of proposed astrophysical source , as will be discussed in a future publication . our mechanism has two predictions beyond fitting the shape of the spectrum and composition evolution , which are independent of many environmental variables and can be used to test the validity of this scenario for production of the ankle . i ) the spectral cutoff of spallated nucleons emerging from the source environment is @xmath128 , where @xmath129 is the rigidity cutoff of the accelerator , because @xmath130 while @xmath131 , and finally @xmath132 , largely independent of composition . this relation holds prior to the extragalactic propagation from the source , thus giving complementary information on the accelerator to that obtained from the spectrum and composition above the ankle alone . ii ) there is a one - to - one relation between the spectrum of spallated nucleons and the anti - electron - neutrinos produced by beta decay of neutrons , unless the spallated nucleons lose energy by interacting with hadronic material in the source environment . independent of other properties of the environment or the source evolution , @xmath125 s will have an identical spectral shape , shifted down by a factor @xmath133 from the kinematics of @xmath134 and reduced by a factor-2 in normalization because only half the nucleons are neutrons . this follows because propagation energy losses are small for nucleons of such low energy , and redshift impacts both nucleons and neutrinos identically . thus , detailed comparison of the @xmath125 and spallated nucleon spectra will reveal if hadronic interactions in the source environment are important , which would imply a correlated production of photo - pion produced neutrinos . after this work was presented at the icecube particle astrophysics symposium a paper appeared on the arxiv exploring another mechanism for producing the ankle , arising in the context of gamma - ray bursts @xcite . we would like to acknowledge many useful discussions with our colleagues of the pierre auger collaboration . furthermore we thank david walz for his support regarding questions about crpropaand benoit marchand for his help running the calculations on the butinah high performance computing cluster of nyu abu dhabi . mu acknowledges the financial support from the eu - funded marie curie outgoing fellowship , grant piof - ga-2013 - 624803 . the research of grf is supported in part by the u.s . national science foundation ( nsf ) , grant phy-1212538 and the james simons foundation ; she thanks kipac / slac for their hospitality . the research of laa is supported by nsf ( grant career phy-1053663 ) and nasa ( grant nnx13ah52 g ) ; he thanks the center for cosmology and particle physics at new york university for its hospitality . in this paper we explore the propagation effects of the four types of photon spectra shown in fig.[app : photon - spectra ] . the first consists of broken power - law ( see e.g. @xcite ) as a simplified representative of non - thermal emission given by @xmath135 where @xmath42 is the photon energy and the maximum of the number density is at an energy of @xmath43 . we also consider modified black - body spectra using the functional form @xmath136 where @xmath137 denotes the temperature in the case of pure black - body , and the absorption factor is given by @xmath138 ( see e.g. @xcite ) . @xmath139 , @xmath140 and @xmath141 are the planck constant , boltzmann constant and speed of light respectively . for @xmath122 and @xmath142 the unmodified black - body spectrum is obtained . the relation between the peak energy and temperature parameter is given by a modified wien s displacement law , @xmath143 k \ , t,\ ] ] where @xmath144 is the lambert function ( see e.g. @xcite ) and @xmath145 . + for the study of the effect of using different functional forms of photon spectra in our model ( cf . sec.[sec : modelvarphotonspec ] ) , it is useful to use a common normalization for all spectra . the integral photon density of eq.([app : eq : photonfield ] ) is @xmath146 and for eq.([app : eq : blackbody ] ) it is @xmath147 where @xmath148 denotes the riemann zeta function and @xmath149 is the gamma function . choosing the photon density of the unmodified black body spectrum as reference we use the following normalization constants , @xmath150 and @xmath151 an example of the four photon spectra after normalization and for the same peak energy of @xmath152 is shown in the left panel of fig.[app : photon - spectra ] . the corresponding interaction time for the sum of photo - dissociation and photo - pion production is shown in the right panel . the interaction between photons and high energy nuclei has been extensively discussed in the literature @xcite . in this appendix , we describe how we implement the photon - nucleus collisions in our analysis . the interaction time for a highly relativistic nucleus with energy @xmath153 ( where @xmath99 is the lorentz factor ) propagating through an isotropic photon background with energy @xmath42 and spectrum @xmath154 , normalized so that the total number density of photons is @xmath155 , is given by @xcite @xmath156 where @xmath157 is the photo - nuclear interaction cross section of a nucleus of mass @xmath158 by a photon of energy @xmath159 in the rest frame of the nucleus . detailed tables of @xmath157 are available in crpropa @xcite . we use the numerical tools provided at @xcite to calculate the interaction times for the photon field given by eqs . ( [ app : eq : photonfield ] ) and ( [ app : eq : blackbody ] ) . for illustrative purposes , the cross section can be approximated by a single pole in the narrow - width approximation , @xmath160 where @xmath161 is the resonance peak , @xmath162 its width , and @xmath163 the pole in the rest frame of the nucleus . the factor of @xmath164 is introduced to match the integral ( i.e. total cross section ) of the breit - wigner and the delta function @xcite . the mean interaction time is obtained substituting eq.([sigma ] ) into eq.([app : eq : interaction ] ) , @xmath165 substituting ( [ app : eq : photonfield ] ) into ( [ a1 ] ) yields : @xmath166 + \left(e_b / e\right)^2 & ~ e > e_b \end{array } \right . \ , , \ ] ] where @xmath167 the parameters characterizing the photo - disintegration cross section are : @xmath168 , @xmath169 , and @xmath170 for @xmath171 ( @xmath172 ) @xcite . the parameters for the photo - pion production cross section are : @xmath173 , @xmath174 , and @xmath175 @xcite . in our simple model we consider interactions and escape of particles treating the source environment as a leaky box . if at a given time , @xmath176 , @xmath177 particles are injected at random into the source environment , then the number of particles @xmath178 remaining in the source at any later time @xmath179 changes as @xmath180 where @xmath46 and @xmath181 are the escape and interaction times respectively . integration yields the time evolution of @xmath178 as @xmath182 where @xmath183 the total number of escaping particles is given by @xmath184 and likewise the number of particles suffering interactions is @xmath185 with @xmath186 . as can be seen , @xmath187 and @xmath188 depend only on the ratio of the escape and interaction times , but not on the absolute value of either of them . in the following we consider sources at steady state , i.e. sources which are active long enough to justify integrating to infinity in eq.([app : eq : nesc ] ) and for which the injected flux equals the escaping flux . interacting particles constitute the source for secondary particles of lower mass number . since the particle trajectory in the source is treated as a random walk starting from a random position , the escape time of a secondary does not depend on the time it was produced . therefore we can apply eqs . ( [ app : eq : sourceprop1 ] ) to ( [ app : eq : nint ] ) also to the secondary particle production , which greatly simplifies the equations with respect to previous analytic approaches that had been developed for the extra - galactic propagation of cosmic - ray nuclei @xcite . the basic principle of the analytic calculation can be best illustrated by firstly describing the case where interactions with the photon field lead to the knock - out of a single nucleon , @xmath189 and the nucleon carries away a fraction of @xmath11 of the initial energy of the nucleus . this approach has been successfully applied to the photo - disintegration ( pd ) during the extra - galactic propagation of nuclei ( see e.g. @xcite ) . it can also serve as a good approximation for the losses due to photo - pion production ( pp ) if nuclei are treated as the superposition of @xmath12 individual nucleons ( see e.g. @xcite ) . the interaction time is therefore the combination of the two processes , i.e. @xmath190 in this simplified propagation scheme , secondaries with mass @xmath12 and energy @xmath191 originate from nuclei with energy @xmath192 and mass @xmath193 . they are produced at a rate @xmath194 where the factor @xmath195 is the jacobian determinant needed to transform the differential injection rate from the primary to secondary energy . in analogy to eq.([app : eq : nesc ] ) , a fraction of the secondaries escapes the source environment , @xmath196 and the remaining particles interact again at a rate of @xmath197 this assumes that the escape probability of a secondary is independent of the time or position it got produced in the source environment . this calculation can be iterated to obtain the escape rate of any remnant with mass @xmath198 produced during the propagation of a nucleus of mass @xmath199 : @xmath200 the rate of nucleons being knocked out of nuclei during propagation via either of the considered processes @xmath201 is @xmath202 with elasticities of the knock out nucleon given by @xmath203 and @xmath204 and the fractional contribution from pd and pp given by @xmath205 the total escape rate of particles of mass @xmath198 from injected nuclei of mass @xmath199 is @xmath206 \ , , \label{app : eq : esctot}\ ] ] where @xmath207 is the kronecker delta . the propagation scheme described in the last section can be easily extended to take into account the emission of several nucleons or light nuclei in photo - nuclear reactions . we use the total interaction time and branching ratios for photo - dissociation from talys @xcite as available in crpropaand neglect multi - nucleon emission for photo - pion production since it can be safely neglected at the energies relevant here . instead of the closed formulae derived above for the single - nucleon case , we now have the following recursive relation for the rate of produced remnant nuclei of mass @xmath198 . @xmath208 with @xmath209 , @xmath210 and @xmath211 . @xmath212 is the branching ratio that gives the probability that a nucleus of mass @xmath213 with energy @xmath214 will have a remnant mass of @xmath198 after the interaction . the @xmath215 knocked out nucleons and nuclei are calculated the same way but replacing the branching fraction by @xmath216 , i.e. the probability to produce @xmath215 fragments of mass @xmath198 , and summing over @xmath215 . the rest of the calculation proceeds as in the case of single - nucleon emission . once nucleons are generated in photo - disintegration they are assumed to either escape immediately in the case of neutrons , or to interact further via photo - pion production . the average elasticity of this process is @xmath204 and corresponds to a shift in energy of @xmath217 . since we perform the calculation in logarithmic bins of this width , proton interactions can be treated similarly to photo - disintegration as a `` trickle - down '' of particles fluxes subsequently shifted by one energy bin . for the reaction @xmath218 , the neutron escapes and the interaction chain is finished . in case of @xmath219 , the secondary proton has a reduced energy ; it may also interact again . the neutron , proton and positive pion fluxes in an energy bin @xmath140 are calculated from the recursive relations @xmath220 @xmath221 and @xmath222 where @xmath223 is the branching fraction of the process @xmath224 and the un - primed fluxes are the sum of the knocked - out nucleons from eq.([app : eq : multinuc ] ) and primary protons . the offset of 7 in the equation for the pion flux is due to the energy shift of the pions , @xmath225 . to compare the spectra obtained in the last section , the particles need to be propagated to earth . the number of cosmic rays per unit volume and energy in the present universe is equal to the number of particles accumulated during the entire history of the universe and is comprised of both primary particles emitted by the sources and secondaries produced in the photo - disintegration process . herein , the variable @xmath179 characterizes a particular age of the universe and @xmath226 indicates its present age . we adopt the usual concordance cosmology of a flat universe dominated by a cosmological constant , with @xmath227 and a cold dark matter plus baryon component @xmath228 @xcite . the hubble parameter as a function of redshift @xmath229 is given by @xmath230 $ ] , normalized to its value today , @xmath231 , with @xmath232 @xcite . the dependence of the cosmological time with redshift can be expressed via @xmath233 . the co - moving space density of cosmic rays @xmath234 of mass @xmath12 from a population of uniformly distributed sources with ( possibly age - dependent ) emission rate per volume @xmath235 is given by @xmath236 where @xmath237 is the expectation value for the number of nuclei of mass @xmath12 in the energy interval ( @xmath238 which derive from a parent of mass @xmath199 and energy @xmath239 emitted at time @xmath179 , and @xmath240 is the ratio of the product of co - moving source density and @xmath235 , relative to the value of that product today . note that @xmath237 includes propagation effects both at the source environment and _ en route _ to earth . we assume that the emission rate of cosmic rays is the same for all sources and the spectrum and composition is independent of the age of the universe , so that evolution of the volumetric emission rate with cosmological time can be described by an overall source evolution factor , @xmath240 discussed below . ( it need not be specified whether this is due to an evolution of the number of sources or their intrinsic power . ) we further assume , as per usual practice , that emission rate is fairly well described by a power - law spectrum . under these general assumptions the source emission rate per volume takes the form @xmath241 where @xmath242 is the maximal energy of emitted protons , i.e. , maximum rigidity of the accelerator , @xmath243 is the nucleus atomic number , @xmath25 is some reference energy , and @xmath244 here , @xmath245 is the number of bursts per unit volume per unit time and @xmath246 is the spectrum of particles produced by each burst , or for a steady source @xmath247 is the number density of sources at @xmath64 , and @xmath248 is the uhecr production rate per unit energy per source . the cosmic ray power density above a certain energy @xmath249 is given by @xmath250 where @xmath251 denotes the upper incomplete gamma function . the cosmological evolution of the source density per co - moving volume is parametrized as @xmath252 with @xmath253 . we adopt for the fiducial model that the evolution of sources follows the star formation rate with @xmath254^{c } } \label{eq : evolution}\ ] ] where @xmath255 , @xmath256 and @xmath257 @xcite . additionally we consider the family of evolution models parameterized as @xmath258 . to propagate the particles escaping the source environment to earth we use the crpropaframework @xcite . for this purpose , we generate a library of propagated nuclei with @xmath259 injected uniformly in light - travel distance . the latter corresponds to a non - evolving source distribution in comoving distance after accounting for the cosmological time dilation . we simulated particles up to @xmath260 . given this library of simulated particles , we can construct the propagation matrix @xmath261 for arbitrary source evolutions for each nuclear mass @xmath262 escaping the source and secondary mass @xmath263 at earth . the elements of the propagation matrix give the expected number of secondaries in an energy interval @xmath264 $ ] at earth originating from nuclei at the source at an energy @xmath265 $ ] for a given source evolution @xmath266 and a uniform logarithmic spacing in energy with @xmath267 . numerically , the elements are constructed via discretization of eq.([app : ncr ] ) @xmath268 where @xmath269 and @xmath270 for a non - evolving injection rate per unit volume , the number of generated events per bin is constant , @xmath271 . then , for any source evolution @xmath272 $ ] , ( [ app : auno ] ) can be rewritten as @xmath273 \ \ \delta t_a \nonumber \\ & = & \sum_{a^*_\mu = a_\nu}^{a ' } \sum_{i = j}^{n_i } { \cal q}_{\rm esc}^{\rm tot } ( e^*_i , a^*_\mu , a_\nu ) \frac{\delta e^*_i}{\delta e_j } \ t_h \ \frac { \sum_{a=0}^{n_a } n_{i j \mu \nu a}^{\rm earth } \ \xi[z(t_a)]}{n_a \ k^{\rm gen}_{i \mu } } \nonumber \\ & = & \sum_{a^*_\mu = a_\nu}^{a ' } \sum_{i = j}^{n_i } { \cal q}_{\rm esc}^{\rm tot } ( e^*_i , a^*_\mu , a ' ) \ \frac{\delta e^*_i}{\delta e_j } \ t_h \ \frac { \sum_{a=0}^{n_a } n_{i j \mu \nu a}^{\rm earth } \ \xi [ z(t_a)]}{\sum_{a=0}^{n_a } n^{\rm gen}_{i \mu a } } \nonumber \\ & = & \sum_{a^*_\mu = a_\nu}^{a ' } \sum_{i = j}^{n_i } { \cal q}_{\rm esc}^{\rm tot } ( e^*_i , a^*_\mu , a ' ) \ \frac{\delta e^*_i}{\delta e_j } \ t_h \ \frac{\sum_{p=0}^{n^{\rm earth}_{i j \mu \nu } } \xi[z(t_p ) ] } { n^{\rm gen}_{i \mu } } \nonumber \\ & = & \sum_{a^*_\mu = a_\nu}^{a ' } \sum_{i = j}^{n_i } \frac{\delta e^*_i}{\delta e_j } \ t_h \ { \cal m}_{i j \mu \nu } \ , { \cal q}_{i \mu } \ , , \label{app : atres}\end{aligned}\ ] ] where @xmath274 $ ] denotes the @xmath275-weighted sum over all events generated with @xmath276 arriving at earth with @xmath277 and @xmath278 is the total number of generated events with @xmath279 . note that if binned in @xmath280 , then @xmath281 , and hence ( [ app : atres ] ) can be rewritten as @xmath282 where @xmath283 and @xmath284 . for a given spectrum of injected nuclei of mass @xmath199 we obtain the space density of cosmic rays at earth with energy @xmath285 and mass @xmath12 , @xmath286 for an isotropic arrival direction distribution ( which is an excellent approximation based on current observations ) the relation between the spectrum and the cosmic ray density is @xmath287 the total flux at earth of particles of mass @xmath263 is @xmath288 where @xmath289 denotes the fraction of particles of mass @xmath290 injected at the source . the results of the last two sections can be readily applied to obtain the flux of neutrinos at earth from the decay of neutrons and charged pions . we approximate the emission rate of pions from photo - pion production by using @xmath291 in eq.([app : qpesc ] ) . the energies of neutrinos escaping from the source are given by the kinematics of the two - body decay of pions and the subsequent muon decay which we treat approximately by assigning a third of the muon energy to each of the decay products . in this way we construct a propagation matrix for pions , @xmath261 with @xmath292 and @xmath293 . similarly , a propagation matrix for neutrons is obtained with @xmath294 and @xmath295 . neutrino oscillation over astronomical distances modifies the initial flavor distribution of fluxes , @xmath296 , in calculable ways . the relevant parameters for such a calculation are the three euler rotations ( @xmath297 , @xmath298 , @xmath299 ) and the @xmath300-violating dirac phase @xmath49 . the current best - fit values as well as the allowed ranges of the mixing parameters at the @xmath301 level are : @xmath302 , , @xmath303 , @xmath304 @xcite . the mixing probabilities are given by @xmath305 @xmath306 @xmath307 and the unitarity relations @xmath308 where @xmath309 , @xmath310 , and @xmath311 @xcite . the measurable neutrino flux at earth is given by @xmath312 in addition to neutrinos , photons are produced from @xmath74 production and decay @xcite , and by photo - disintegration of high - energy nuclei followed by immediate photo - emission from the excited daughter nuclei @xcite . the @xmath99-rays , electrons , and positrons produced in the decay of @xmath74 and @xmath123 trigger an electromagnetic ( em ) cascade on the cosmic microwave background , which develops via repeated @xmath313 pair production and inverse compton scattering . other contributions to the cascade are provided by bethe - heitler production of @xmath313 pairs and @xmath99-rays emitted during the photo - disintegration process , after the photo - dissociated nuclear fragments de - excite . the net result is a pile up of @xmath99-rays at @xmath314 , just below the threshold for further pair production on the diffuse optical backgrounds . the em energy then gets recycled into the so - called fermi - lat region , which is bounded by observation @xcite to not exceed @xmath315 @xcite . the latest fermi - lat limts @xcite do not significantly influence the determination of the @xmath316 upper bound of @xcite , because that bound is not very sensitive to the high energy bins added by @xcite . the photons coming from photo - pion production in the source environment were shown to be below the fermi - lat bound in section [ sec : fidmodel ] . to place a bound on the contribution of photons from nuclear de - excitation to the fermi - lat diffuse gammas , without performing an explicit calculation , we can turn to the estimate of @xcite which found @xmath317 assuming the high - energy icecube spectrum to be entirely due to neutron beta - decay . since in our model the neutrino flux from neutron decay is significantly below the icecube spectrum , the corresponding de - excitation photon contribution to the fermi - lat data must be far below the limit . 94ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1016/j.astropartphys.2005.04.001 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.71.3401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.101101 [ * * , ( ) ] , link:\doibase 10.1016/j.physletb.2010.02.013 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.101.061101 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2012.05.023 [ * * , ( ) ] link:\doibase 10.1103/physrevd.88.042004 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1016/j.nima.2013.09.018 [ * * , ( ) ] link:\doibase 10.1086/322240 [ * * , ( ) ] , link:\doibase 10.1088/0954 - 3899/34/10/r01 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2012.02.004 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2010.12.007 [ * * , ( ) ] , link:\doibase 10.1088/0067 - 0049/203/2/34 [ * * , ( ) ] , link:\doibase 10.1088/0004 - 637x/802/2/111 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.90.122005 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.90.122006 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.74.043005 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.87.023004 [ * * , ( ) ] , link:\doibase 10.1051/epjconf/20159904004 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1016/j.astropartphys.2006.09.006 [ * * , ( ) ] , \doibase 10.1051/0004 - 6361:20077478 [ * * , ( ) ] , link:\doibase 10.1088/1475 - 7516/2008/10/033 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2009.09.004 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2013.11.006 [ * * , ( ) ] , link:\doibase 10.1016/j.crhy.2014.02.009 [ * * , ( ) ] , link:\doibase 10.1007/s11467 - 013 - 0319 - 7 [ * * , ( ) ] , link:\doibase 10.1088/1475 - 7516/2014/10/020 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.91.083009 [ * * , ( ) ] , link:\doibase 10.1051/0004 - 6361/200911839 [ * * , ( ) ] , link:\doibase 10.1088/0004 - 637x/707/1/370 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.80.123018 [ * * , ( ) ] , link:\doibase 10.1088/0004 - 637x/750/2/118 [ * * , ( ) ] , link:\doibase 10.1088/1475 - 7516/2013/03/010 [ * * , ( ) ] , link:\doibase 10.1093/mnras / stv893 [ * * , ( ) ] , link:\doibase 10.1016/j.nuclphysbps.2014.10.023 [ * * , ( ) ] , link:\doibase 10.1088/1475 - 7516/2015/08/026 [ * * , ( ) ] , link:\doibase 10.1016/0927 - 6505(94)90027 - 2 [ * * , ( ) ] , link:\doibase 10.1016/s0927 - 6505(98)00055 - 3 [ * * , ( ) ] , @noop ( ) , , @noop _ _ ( , ) @noop @noop @noop ( ) , link:\doibase 10.1016/s0010 - 4655(99)00446 - 4 [ * * , ( ) ] , @noop link:\doibase 10.1086/312626 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.87.081101 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1051/epjconf/20135301005 [ * * , ( ) ] , in @noop _ _ ( ) link:\doibase 10.1088/1475 - 7516/2013/02/026 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1103/physrevd.91.022001 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.88.112008 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.91.027301 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1016/j.astropartphys.2010.06.003 [ * * , ( ) ] , link:\doibase 10.1088/2041 - 8205/802/2/l19 [ * * , ( ) ] , link:\doibase 10.1111/j.1365 - 2966.2012.20841.x [ * * , ( ) ] , link:\doibase 10.1103/physrevd.80.094003 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.83.014018 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1088/2041 - 8205/768/1/l1 [ * * , ( ) ] , link:\doibase 10.1088/1674 - 1137/38/9/090001 [ * * , ( ) ] @noop ( ) , link:\doibase 10.1146/annurev.ns.25.120175.003011 [ * * , ( ) ] link:\doibase 10.1103/physrevd.90.103005 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.92.021302 [ * * , ( ) ] , link:\doibase 10.1016/j.cpc.2012.07.008 [ * * , ( ) ] , link:\doibase 10.1103/physrev.180.1264 [ * * , ( ) ] link:\doibase 10.1086/154321 [ * * , ( ) ] link:\doibase 10.1016/0927 - 6505(93)90023 - 7 [ * * , ( ) ] link:\doibase 10.1103/physrevd.57.7103 [ * * , ( ) ] , link:\doibase 10.1088/1126 - 6708/1998/10/009 [ * * , ( ) ] , link:\doibase 10.1086/306816 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2004.12.007 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.77.103007 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2012.07.010 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2012.06.003 [ * * , ( ) ] , link:\doibase 10.1103/physrevd.82.123005 [ * * , ( ) ] , link:\doibase 10.1016/j.astropartphys.2012.12.001 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1103/physrevd.75.063001 [ * * , ( ) ] , @noop ( ) , link:\doibase 10.1016/j.dark.2014.04.002 [ * * , ( ) ] link:\doibase 10.1088/1126 - 6708/2008/02/005 [ * * , ( ) ] , @noop _ _ ( , ) link:\doibase 10.1103/physrevlett.98.121101 [ * * , ( ) ] , link:\doibase 10.1103/physrevlett.104.101101 [ * * , ( ) ] , link:\doibase 10.1088/0004 - 637x/799/1/86 [ * * , ( ) ] , link:\doibase 10.1016/j.physletb.2010.11.019 [ * * , ( ) ] ,

time - resolved electronic spectra can be obtained as the fourier transform of a special type of time correlation function known as fidelity amplitude , which , in turn , can be evaluated approximately and efficiently with the dephasing representation . here we improve both the accuracy of this approximation with an amplitude correction derived from the phase - space propagator and its efficiency with an improved cellular scheme employing inverse weierstrass transform and optimal scaling of the cell size . we demonstrate the advantages of the new methodology by computing dispersed time - resolved stimulated emission spectra in the harmonic potential , pyrazine , and the nco molecule . in contrast , we show that in strongly chaotic systems such as the quartic oscillator the original dephasing representation is more appropriate than either the cellular or prefactor - corrected methods .

1306.0721

ultrafast spectroscopy with a time resolution as high as @xmath0s is essential for understanding many quantum dynamical processes in chemical physics.@xcite although short time scales should simplify theoretical studies by requiring shorter simulations , solving the time - dependent schrdinger equation ( tdse ) is challenging even for short times due to the exponential scaling with the number of degrees of freedom . an attractive approach offering a compromise between accuracy and computational efficiency is provided by the semiclassical initial value representation methods , miller:1970,heller:1981,herman_kluk:1984,miller:2001,herman:1994,*thoss_wang:2004,*kay:2005,*ceotto_zhuang:2013 which benefit from the ultrafast character of the dynamics not only because of lower computational cost , but also because their accuracy deteriorates with increasing time . the so - called dephasing representation@xcite ( dr ) , is an efficient initial - value - type semiclassical approximation particularly fitted for calculations of time - resolved electronic spectra.@xcite the dr improves on a previous method@xcite inspired by the semiclassical perturbation theory of miller and coworkers.@xcite in electronic spectroscopy , the dr and closely related approximations are known as mukamel s phase averaging method@xcite or wigner - averaged classical limit , and were used by various authors.shemetulskis_loring:1992,rost:1995,wang_sun:1998,li_fang:1996,*egorov_rabani:1998,*egorov_rabani:1999,shi_geva:2005 shi and geva@xcite derived this approximation without invoking the semiclassical propagator by linearizingpoulsen_nyman:2003,bonella_coker:2005 the path integral quantum propagator . although the original formulation of the dr pertains to a single pair of potential energy surfaces , the generalization to multiple surfaces , and hence to nonadiabatic dynamics , exists.zimmermann_vanicek:2012,*zimmermann_vanicek:2012a the dr has many other applications ; the method successfully described , e.g. , the local density of states and the transition from the fermi - golden - rule to the lyapunov regime of fidelity decay.wang_casati:2005,*ares_wisniacki:2009,*wisniacki_ares:2010,*garcia-mata_wisniacki:2011 yet the most attractive feature of the dr is its efficiency : motivated by numerical comparisons with other semiclassical methods , wehrle_sulc:2011 it has been recently proved analyticallymollica_vanicek:2011 that the number of trajectories required for convergence of the dr is independent of the system s dimensionality , hamiltonian , or total evolution time . the efficiency was further increased in the cellular version of the dr,@xcite which was inspired by heller s cellular dynamics@xcite and which can significantly reduce the required number of trajectories . the original implementation of the _ cellular dr _ ( cdr ) , however , does not converge to the dr in the limit of infinite number of trajectories . unlike its efficiency , the accuracy of the dr is not always sufficient . the dr is exact in displaced harmonic oscillatorsmukamel:1982,*book_mukamel and often accurate in chaotic systems , vanicek:2004,*vanicek:2006 but it breaks down in as simple systems as harmonic oscillators with different force constants . this breakdown can be partially remedied by augmenting the dr with a prefactor , zambrano_almeida:2011 which , however , leads to a much higher computational cost per trajectory and also typically requires more trajectories to achieve convergence . the first goal of the present paper is to describe a general numerical implementation of the prefactor correction and apply it to the calculation of time - resolved electronic spectra . as the numerical evaluation of the cdr requires , incidentally , the same ingredients as the prefactor correction , the second goal is to combine the advantages of the cellular approach and prefactor correction into a single formula , and show that the resulting method , _ cellular dr with prefactor _ ( cdrp ) , is able to increase both the efficiency and accuracy of the dr . our third goal is presenting a major improvement of the cellularization process by employing the inverse weierstrass transform of the initial state as the optimal sampling weight instead of the widely used wigner or husimi functions , and by correlating the size of the cells with their number and the number of degrees of freedom , which guarantees the convergence of the cdr to the original dr in the limit of infinite number of trajectories . the remainder of the paper is organized as follows : the correlation function approach and the dr approximation for evaluating time - resolved stimulated emission spectra is reviewed in section [ sec : theory ] ; in particular , the dr , its prefactor correction , and its cellular version are deduced . after explaining how the new cellular approach provides optimal choices of the sampling weight and width of gaussian cells , we derive the cdrp , i.e. , a method combining the prefactor correction and cellularization into a single framework . section [ sec : examples ] contains several analytical and numerical results testing the theory developed in section [ sec : theory ] , while section [ sec : final_remarks ] provides conclusions . to be specific , we will present the methodology for time - resolved stimulated emission ( trse ) ; modification to other ultrafast processes is straightforward . within the electric dipole approximation , time - dependent perturbation theory , and ultrashort pulse approximation , the dispersedpollard_lee:1990,stock_domcke:1992 trse spectrum can be computed as a fourier transform of the following correlation function : pollard_lee:1990,stock_domcke:1992,wehrle_sulc:2011,sulc_vanicek:2012 @xmath1 . \end{split } \label{eq : c trse}\ ] ] here @xmath2 and @xmath3 denote the amplitudes of the pump and probe laser pulses , @xmath4 represents the nuclear density operator in the electronic ground state at temperature @xmath5 , @xmath6 is the transition dipole moment operator coupling electronic states @xmath7 and @xmath8 , @xmath9 stands for the time delay between the pump and probe pulses , and @xmath10 is the time elapsed after the probe pulse . finally , @xmath11 denotes the nuclear quantum evolution operator @xmath12with hamiltonian @xmath13 where @xmath14 is the nuclear kinetic energy and @xmath15 is the @xmath16th potential energy surface ( pes ) . in all expressions , the hat denotes operators in the hilbert space of nuclei . within the franck - condon approximation and zero - temperature limit , correlation function ( [ eq : c trse ] ) reduces to @xmath17where @xmath18is a specific time correlation function and the initial state @xmath19 is typically the vibrational ground state of the ground pes . the trse spectrum , given by@xcite @xmath20is proportional to the so - called wave packet spectrum @xmath21 obtainedbook_tannor as @xmath22correlation function ( [ eq : f(t , tau ) ] ) for the stimulated emission is a particular example of a more general concept of _ fidelity amplitude_,@xcite defined as @xmath23where @xmath24 , @xmath25 , is the time evolution operator for a time - dependent hamiltonian @xmath26 : @xmath27 \quad \quad ( j=1,2 ) , \label{eq : u_j}\]]where @xmath28 denotes the time - ordering operator . correlation function ( [ eq : f(t , tau ) ] ) for trse is obtained from the general fidelity amplitude ( [ eq : fid_amplitude ] ) by substituting the following time - dependent hamiltonians @xmath26 into eq . ( [ eq : u_j]):@xmath29 note that @xmath30 if @xmath31 . besides electronic spectroscopy applications , li_fang:1996,*egorov_rabani:1998,*egorov_rabani:1999,shi_geva:2005,rost:1995,shemetulskis_loring:1992 correlation function ( [ eq : fid_amplitude ] ) proved useful , e.g. , in nmr spin echo experiments@xcite and in the theories of quantum computation , decoherence , cucchietti_dalvit:2003,*gorin_prosen:2004 and inelastic neutron scattering.@xcite fidelity amplitude was also used as a measure of the dynamical importance of diabatic , nonadiabatic , or spin - orbit couplings , zimmermann_vanicek:2010,zimmermann_vanicek:2012,*zimmermann_vanicek:2012a and of the accuracy of quantum molecular dynamics on an approximate pes.li_mollica:2009,zimmermann_ruppen:2010 in practical calculations , correlation function ( [ eq : fid_amplitude ] ) must usually be approximated , and dr provides an efficient semiclassical approximation.vanicek:2004,*vanicek:2006,li_fang:1996,*egorov_rabani:1998,*egorov_rabani:1999,shi_geva:2005,rost:1995,shemetulskis_loring:1992 if we denote by @xmath32 the phase - space coordinates at time @xmath33 of a point along a classical trajectory of the _ _ average__wehrle_sulc:2011,mukamel:1982,*book_mukamel , zambrano_almeida:2011 hamiltonian @xmath34 , the dr of fidelity amplitude ( eq : fid_amplitude ) can be written as @xmath35with @xmath36here @xmath37 is the number of degrees of freedom , @xmath38 denotes the wigner transform of the initial density operator @xmath39 , and @xmath40 is the action due to the difference @xmath41 along trajectory @xmath42 : @xmath43for trse , @xmath44 is given by @xmath45denoting the phase - space average of a quantity @xmath46 with respect to a weight function @xmath47 by @xmath48time correlation function ( [ eq : dr formula integral ] ) can be written in a compact way as @xmath49formula ( [ eq : dr formula ] ) can be evaluated efficiently by monte carlo integration . indeed , because the convergence of the dr is independent of dimensionality , the dr is in many - dimensional systems much more efficient than other quantum or classical algorithms for computing the fidelity amplitude.@xcite the accuracy of the dr typically improves with decreasing @xmath44 and increasing complexity of hamiltonians @xmath50 and @xmath51 . while the dr is exact in displaced harmonic oscillators with arbitrary displacement , this perturbative approximation breaks down in some singular cases , such as when hamiltonians @xmath52 and @xmath51 represent harmonic oscillators with significantly different force constants.@xcite the above - mentioned breakdown of the dr can be partially corrected by including a prefactor in the dr formula ( [ eq : dr formula]).zambrano_almeida:2011 we now briefly derive this improved version of the dr . fidelity amplitude ( [ eq : fid_amplitude ] ) can be expressed as the expectation value of the _ _ echo operator__gorin_prosen:2006,*jacquod_petitjean:2009 @xmath53 : @xmath54 = \left\langle \mathcal{e}_{w}(x^{0},t)\right\rangle _ { \rho_{w}(x^{0 } ) } , \label{eq : f trace}\]]where @xmath55 is the wigner transform of the echo operator . note that @xmath56 itself can be interpreted as a single forward - backward evolution operator describing propagation driven by @xmath51 for time @xmath10 followed by a propagation driven by @xmath57 from time @xmath10 to @xmath58 . the path labeled by @xmath59 in fig . [ fig : drp figure ] is a classical analog of such a forward - backward propagation . , the path @xmath60 is determined by two requirements : ( _ i _ ) for @xmath61 it is driven by @xmath51 ( dashed path ) , while for @xmath62 it is driven by @xmath57 ( continuous path ) ; and ( _ ii _ ) @xmath63 . geometrical part of the phase @xmath64 is the shadowed area and the dotted line is the chord between @xmath65 and @xmath66 . ] a semiclassical approximation to the wigner transform @xmath67 consists in replacing it by a single phase - space semiclassical propagator,@xcite @xmath68with the constraint @xmath69/2 $ ] . here @xmath70 is the identity matrix in @xmath71 dimensions and @xmath72 is the standard symplectic matrix in @xmath71 dimensions , @xmath73where the subscripts specify the dimensionality of each square block . more details about this semiclassical phase - space propagator are presented in appendix [ ap : ps propagator ] . in eq . ( [ eq : echo_symbol ] ) , phase @xmath74 is the so - called _ center - action _ of the path @xmath75 at time @xmath10 ; explicitly , this function is defined as @xmath76where the closed integral is evaluated along the path consisting of @xmath75 and of the straight line connecting @xmath77 and @xmath78 , as shown in fig . [ fig : drp figure ] , and @xmath79center - action ( [ eq : s_fb ] ) appears naturally in the weyl representation of quantum mechanics.@xcite as mentioned in appendix [ ap : ps propagator ] , the center - action is a function of the center @xmath80 and , in general , is multivalued : a given center @xmath80 may be the midpoint between the initial and final points for two or more paths ( see , e.g. , fig . fig : le - areas u(x ) in appendix [ ap : ps propagator ] ) . nevertheless , as shown in appendix [ ap : ps propagator ] , for our purposes , we can assume that @xmath81 has only a single branch . approximating the center - action in the semiclassical echo operator ( eq : echo_symbol ) by the dr action , @xmath82 , which is valid up to the first order in perturbation theory,@xcite yields an improved approximation for fidelity amplitude given by @xmath83 , where @xmath84 with @xmath85 we will refer to expression ( [ eq : drp formula ] ) as the _ dr with prefactor _ or drp : it corresponds to including a prefactor to the contribution of each trajectory in the dr formula ( [ eq : dr formula ] ) . the drp is free of caustics because the prefactor ( [ eq : a_drp ] ) can not diverge . however , the prefactor is the most expensive part of the drp evaluation because it depends on the hessian of the dr phase @xmath40 with respect to the initial conditions ; in appendix ap : derivatives we show how to compute this hessian from the derivatives of the stability matrix of the classical trajectory . finally note that switching the pess in the definition ( [ eq : fid_amplitude ] ) of fidelity amplitude is equivalent to taking the complex conjugate of this equation . drp preserves this property because of the identity @xmath86 , proven in appendix ap : proof . the _ cellular dephasing representation _ ( cdr ) was developed in ref . in order to further accelerate the convergence of the dr in the spirit of heller s _ cellular dynamics_.@xcite the main idea of the cdr consists in decomposing the wigner transform of the initial state into phase - space cells and evaluating the contribution of an entire cell of nearby trajectories approximately , using the dynamical information collected along a single , central trajectory . here we describe a simpler and more rigorous cellularization process than that used in the original cdr ( ref . ) and other cellularizationheller:1991,walton_manolopoulos:1996 or filinov filteringfilinov:1986,makri_miller:1987,wang:2001 schemes . in particular , the new methodology provides both a natural criterion for cell size [ see eq . ( eq : lambda_scaling ) ] and a natural sampling weight for the cell centers [ given by inverse weierstrass transform ( [ eq : weierstrass trans ] ) ] . most importantly , unlike the previous approaches , in the limit of infinite number of trajectories , the new methodology converges to the original , noncellular method ( in our case , the dr ) . in standard cellularization or filinov filtering procedures , filinov:1986,makri_miller:1987,heller:1991,walton_manolopoulos:1996,wang:2001,sulc_vanicek:2012 the initial state is covered with phase - space gaussians as in fig . fig : cellularization(a ) , the centers of these gaussians being sampled from a given distribution ( denoted with a black circle ) , typically a wigner or husimi transform of the initial state , which is independent of the size and number of cells . then one decreases the cell size ( measured by parameter @xmath87 , defined so that each cell has phase - space volume @xmath88 ) until the approximate treatment of contributions of neighboring trajectories ( typically involving quadratic expansion of the action ) becomes sufficiently accurate . independently , the number of cells @xmath89 is increased until convergence . there are several problems with this standard approach : first , decreasing the size of the cell to zero ( @xmath90 ) for a fixed number of cells @xmath89 eventually results in the initial state not being fully covered [ see the middle row of fig . [ fig : cellularization](a ) ] . second , in case that the quadratic expansion of the action is accurate , taking the limit @xmath91 for a fixed nonzero width @xmath92 is wasteful since many cells are overlapping [ see the middle column of fig . fig : cellularization(a ) ] . third , if the quadratic expansion is inaccurate , taking the limit @xmath93 for a fixed width @xmath92 converges to a result different from the original noncellular method . fourth , for gaussian initial states and @xmath94 , the optimal choice of a single cell is clearly the initial state , but in the standard approach the width and position of the cell are uncorrelated with the number of cells [ see the top row of fig . [ fig : cellularization](a ) ] . the solution of the first three problems is simple and provided by scaling the size of the cell with the number of cells and dimensions according to@xmath95guaranteeing that the phase - space volume of the initial state is equal to the total volume of all cells [ fig . [ fig : cellularization](b ) ] . this avoids an _ ad hoc _ choice of the width of the cell , replacing two limiting processes @xmath96 and @xmath93 with a single process @xmath93 , and pictorially corresponds to going along the diagonal from the top left to the bottom right corner of fig . [ fig : cellularization](a ) . in the derivation presented below it is shown that the fourth problem is solved by sampling the centers of the cells from the inverse weierstrass instead of the wigner transform of the initial state . as we shall see , this inverse weierstrass transform , represented by red circles in fig . [ fig : cellularization](b ) is a natural sampling weight , which is correlated to the size of the cell . if the initial state is a gaussian , for @xmath94 , the single cell has uniquely defined size and position , equal to the size and position of the initial state . in the limit of infinitely many very small cells , their centers are sampled from the wigner transform . all together , @xmath89 determines both the size of each cell and the sampling weight for their centers . to put the above ideas into a precise mathematical form , consider a phase - space gaussian function centered at the origin , @xmath97where @xmath98 is a @xmath99 real , symmetric , positive definite matrix , whose determinant is inversely proportional to the square of the phase - space volume occupied by @xmath100 , while the prefactor in eq . ( [ eq : weyl gaussian ] ) ensures normalization of @xmath100 : @xmath101 . in particular , if @xmath102 and @xmath103 ( for @xmath104 and @xmath105 ) , then @xmath106 coincides with the wigner transform of a @xmath37-dimensional gaussian wave packet with the same width @xmath21 in all @xmath37 coordinate directions . however , @xmath106 of eq . ( [ eq : weyl gaussian ] ) is , in general , not required to be a wigner transform of any physical quantum state . most importantly , @xmath106 can be arbitrarily narrow both in position and momentum , and hence does not have to satisfy the heisenberg uncertainty principle.@xcite employing sufficiently narrow gaussian functions ( [ eq : weyl gaussian ] ) with fixed @xmath98 as our cells , the wigner transform of a general state can be expanded as @xmath107where the asterisk denotes the convolution of @xmath100 with @xmath108 . function @xmath109 , playing a role of continuous expansion coefficient , is known as the _ inverse weierstrass transformation _ of @xmath110.widder:1954 thanks to normalization of @xmath110 and @xmath100 , integrating eq . ( [ eq : gaussian diag expansion ] ) over @xmath111 implies that @xmath112 is also normalized : @xmath113 . equation ( [ eq : gaussian diag expansion ] ) can be inverted via the convolution theorem to obtain @xmath114/\mathcal{f% } [ { g}_{\sigma } ] ] \notag \\ & \equiv h^{-d}\int \!d\eta \,e^{\eta ^{\mathsf{t}}\cdot \sigma ^{-1}\cdot { % \eta } /2\hbar ^{2}}e^{iz^{\mathsf{t}}\cdot { \eta } /\hbar } \mathcal{f}[\rho _ { w}]({\eta } ) , \label{eq : weierstrass trans}\end{aligned}\]]where @xmath115 $ ] denotes the phase - space fourier transform,@xmath116({\eta } ) : = h^{-d}\int \!dx\,\rho_{w}(x)\,e^{-ix^{% \mathsf{t}}\cdot \eta /\hbar } , \]]while @xmath117 $ ] stands for its inverse . the fourier transform of @xmath100 can be evaluated analytically as @xmath118(\eta ) = e^{-\eta ^{\mathsf{t}}\cdot \sigma ^{-1}\cdot \eta /2\hbar ^{2}}.\]]from eq . ( [ eq : weierstrass trans ] ) we see that @xmath119 is well - defined only if @xmath120 $ ] decays sufficiently faster than @xmath121 $ ] . in other words , the gaussian cells must be sufficiently narrow in order that the integral ( eq : weierstrass trans ) over @xmath122 converges . if the initial state is a gaussian , i.e. , @xmath123 , the cell functions @xmath100 in eq . ( [ eq : gaussian diag expansion ] ) can be conveniently chosen as scaled versions of @xmath124 with widths in all coordinate and momentum directions multiplied by a factor @xmath92 , where @xmath125 , which is equivalent to setting @xmath126 . the width of cell @xmath100 may vary from zero ( a delta function ) for @xmath127 to the width of the initial state @xmath128 for @xmath129 . the inverse weierstrass transform ( [ eq : weierstrass trans ] ) can be evaluated analytically for all admissible @xmath92 ( i.e. , @xmath130 ) as @xmath131where @xmath132note that for @xmath133 the inverse weierstrass transform ( eq : weierstrass trans ) diverges . the limiting cases of the sampling weight ( [ eq : weier_scaled ] ) are @xmath134 g_{\sigma ^{0}}(z - z^{0 } ) , & \lambda \rightarrow 0^{+},% \end{array}% \right . \label{eq : weierstrass for gaussian}\]]and are represented , respectively , by the red dot at the top and red circle at the bottom of fig . [ fig : cellularization](b ) . indeed , for @xmath129 , there is no freedom in the choice of the center of the single cell , whereas in the limit @xmath90 , the sampling weight converges to @xmath38 . inserting the cellular expansion ( [ eq : gaussian diag expansion ] ) into the dr formula ( [ eq : dr formula ] ) yields @xmath135 in order to carry out the integration over @xmath80 analytically , one expands the dr phase about point @xmath136 as @xmath137 , where the cdr action is @xmath138 in the last equation , @xmath139 , @xmath140 is the gradient of @xmath141 at @xmath136 , and @xmath142 , already defined in eq . ( [ eq : ds_expansion_b ] ) , is , up to a factor @xmath143 , the hessian of @xmath144 at @xmath136 . using the quadratic expansion ( [ eq : s_cdr ] ) , the integral over @xmath80 in the double integral representation ( [ eq : dr with weiss ] ) of the dr is performed analytically to yield the final result cdr : @xmath145with @xmath146 straightforward numerical implementation evaluates @xmath147 in eq . ( [ eq : cdr formula ] ) by monte carlo importance sampling . this means arithmetically averaging the estimator @xmath148 over the set of @xmath89 initial conditions sampled from the weight @xmath149 using the box - muller algorithm for gaussian initial states or metropolis algorithm for general states . [ the positivity of @xmath150 is for gaussian initial states guaranteed by eq . ( [ eq : weier_scaled ] ) . ] equivalently , one can think of this procedure as expanding the wigner transform @xmath38 of the initial state into a finite set of gaussians , i.e. , @xmath151where @xmath152 and centers @xmath153 are sampled from @xmath154 . this expansion is then combined with the quadratic expansion ( [ eq : s_cdr ] ) of @xmath141 and substituted into the dr formula ( [ eq : dr formula integral ] ) . as mentioned above , a natural value of the scaling parameter is @xmath155 for which the @xmath89 cells @xmath156 cover essentially the same phase - space volume as the initial state @xmath157 . moreover , for @xmath94 , eq . ( eq : lambda_scaling ) gives @xmath129 . from eq . ( [ eq : weierstrass for gaussian ] ) we see that @xmath158 degenerates to a delta function and the single cell is identical to @xmath110 . on the other hand , @xmath93 implies @xmath159 and eq . ( eq : sigma_k ) yields @xmath160 and @xmath161 since for @xmath159 , @xmath162 and @xmath163 , comparison of eqs . ( [ eq : cdr formula ] ) and ( [ eq : dr formula ] ) confirms that the cdr reduces in the limit @xmath91 to the original dr , as promised . note that this desirable property was satisfied neither by the original cdr nor by standard cellularization or filinov filtering procedures for the van vleck or herman - kluk propagators . several further improvements are possible : first , a significant boost in computational efficiency could be gained with ideas implemented in the _ generalized _ filinov filtering@xcite or stationary phase monte carlo method.@xcite motivated by the generalized filinov method , for instance , one would add a complex linear term to the exponent of the gaussian cell to ensure that the overall phase of the integrand of the @xmath80 integral in eq . ( [ eq : dr with weiss ] ) were approximately stationary , making the original integral more amenable to monte carlo integration . this is in contrast to the original filinov approach,@xcite which does not employ an additional phase . another improvement relies on sobol sampling , press_numericalrecipes:2010 which actively seeks different initial conditions while preserving the normal distribution , and was used , e.g. , by walton and manolopoulos.@xcite finally , it is advantageous to allow the expansion coefficients @xmath164 in eq . ( eq : rho_w = sum ) to differ from @xmath165 . specifically , one finds the optimal coefficients @xmath164 for given , already sampled , gaussian centers @xmath153 by minimizing the residual @xmath166 error of the expansion ( [ eq : rho_w = sum ] ) under the constraints [ eq : def_cn ] @xmath167which guarantee that @xmath168 and @xmath169 . from numerical point of view , this amounts to solving a convex quadratic program.@xcite as demonstrated in sec . [ sec : examples ] , this procedure further enhances efficiency , nevertheless the acceleration due to the cellularization procedure itself is dominant . in practice , one should always use all five tricks , i.e. , sampling ( [ eq : cdr formula ] ) from the inverse weierstrass transform , scaling ( [ eq : lambda_scaling ] ) of the cells with @xmath89 , generalized filinov filtering , makri_miller:1987,wang:2001 sobol sampling , press_numericalrecipes:2010 and optimal coefficients ( [ eq : def_cn_i])-([eq : def_cn_ii ] ) . although clearly beneficial , generalized filinov filtering and sobol sampling were not employed here , in order to clearly separate the effect of the three new ideas presented : sampling ( [ eq : cdr formula ] ) from the inverse weierstrass transform , scaling ( eq : lambda_scaling ) of the cells with @xmath89 , and optimal coefficients ( eq : def_cn_i)-([eq : def_cn_ii ] ) . the numerical prerequisites of the cdr ( subsec . [ subsec : cdr ] ) and drp ( subsec . [ subsec : drp ] ) are the same the cost per trajectory is determined by evaluating the hessian of @xmath141 with respect to initial conditions . this allows for a straightforward combination of the methods , without increasing the cost per trajectory , by multiplying the contribution ( [ eq : cdr formula ] ) of each trajectory with the prefactor ( [ eq : a_drp ] ) and thus obtaining the _ cellular dephasing representation with prefactor _ ( cdrp ) : @xmath170 where @xmath171 in principle , the cdrp should benefit both from the enhanced efficiency of the cdr and improved accuracy of the drp , as depicted in fig . [ fig : cdrp diagram ] . ) . typically , the accuracy increases along the horizontal arrows , corresponding to adding the prefactor ( [ eq : drp formula ] ) , while the efficiency improves in the downward direction , corresponding to the cellularization procedure ( [ eq : cdr formula ] ) . ] as for the asymptotic computational complexity of eq . ( [ eq : cdrp formula ] ) per trajectory , a straightforward implementation scales with system s dimensionality @xmath37 and total propagation time @xmath10 as @xmath172 . linear scaling with time is easily verified by direct inspection of eq . ( eq : cdrp formula ) , while the cubic dependence on @xmath37 is due to the necessity to propagate the stability matrix and due to the matrix operations implicit in eqs . ( [ eq : cdr formula ] ) and ( [ eq : a_cdr ] ) . the cdrp is thus cheaper than , e.g. , the popular _ _ forward backward initial value representation__@xcite which would scale as @xmath173 . in this section we will show how the cdrp approximation improves the accuracy of the time correlation function ( [ eq : fid_amplitude ] ) and stimulated emission spectrum ( [ eq : sigma_wp ] ) for several well - known systems . as the first example we consider two quadratic hamiltonians in @xmath37 dimensions : [ eq : h harmonic ] @xmath174where @xmath175 is the gap between the two potential wells , @xmath176 is the @xmath99 hessian matrix of @xmath177 , @xmath178 being the force - constant matrix , @xmath179 is the @xmath180 matrix of masses , and @xmath181 is the phase - space displacement of the two hamiltonians : e.g. , @xmath182 is the coordinate distance between the two potential minima . the hessian of the average hamiltonian is given by the ( invertible ) @xmath99 matrix @xmath183where @xmath184 . the path driven by the average hamiltonian is @xmath185where @xmath186 is the stability matrix for @xmath187 and @xmath188 . since the hamiltonians ( [ eq : h harmonic ] ) are quadratic , it is possible to evaluate the dr phase analytically for an arbitrary initial condition @xmath189 as @xmath190where @xmath191with @xmath192 and @xmath193 . note that in the harmonic systems , the cellular schemes are exactly equal to their noncellular analogs , e.g.,@xmath194[however , if a discrete gaussian expansion ( [ eq : rho_w = sum ] ) is used , the accuracy of the results will be limited by the error inherent in eq . ( eq : rho_w = sum ) . ] since @xmath195 and hence @xmath196 are in this case independent of @xmath80 , the drp and cdrp can be calculated for an arbitrary initial state as @xmath197 explicit formulas for one degree of freedom are @xmath198 t.\end{aligned}\ ] ] here , @xmath199 , @xmath200 , and @xmath201 , i.e. , @xmath202 has only position components . additionally , the determinant prefactor is given by @xmath203 figure [ fig : sho ] shows the fully converged time correlation functions for zero time delay in one - dimensional harmonic oscillator ( [ eq : h harmonic ] ) using a gaussian initial state . we observe the effect of the prefactor ( [ eq : a_drp ] ) : it enhances the accuracy compared with the dr , so that the approximate time correlation function does not decay with increasing time . note that the fourier transforms of time correlations shown in fig . [ fig : sho ] can be interpreted both as trse spectra with zero time delay and as _ continuous - wave absorption spectra_. now we consider a two - dimensional harmonic system ( [ eq : h harmonic ] ) with @xmath204 , @xmath205 , @xmath206which is a prototype of the breakdown of the dr in simple molecular systems . while the dr describes exactly the behavior of the excited mode corresponding to displaced simple harmonic oscillators,@xcite this agreement is lost due to the decay of the dr in the silent mode , corresponding to harmonic oscillators with different force constants [ as in fig . [ fig : sho](a ) ] , in which the dr breaks down . in other words , the breakdown of the dr for the uninteresting mode covers up the accurate information about the interesting mode . figure [ fig : sho2d ] shows the time correlation function for time delay @xmath207the drp can in this system almost completely remove the error introduced by the dr . time correlation function for time - resolved stimulated emission spectrum in a two - dimensional harmonic oscillator model . displacements are @xmath208 with @xmath209 and @xmath210 , @xmath211 , and @xmath212 , and force constants [ according to eq . ( [ eq : sho2d ] ) ] are @xmath213 and @xmath214 . the initial state is the ground state of the ground pes . time delay @xmath215 . ] the next system is based on the four - dimensional vibronic coupling model taking into account normal modes @xmath216 , @xmath217 , @xmath218 , and @xmath219 of pyrazine.@xcite we employ the @xmath220 and @xmath221 surfaces from ref . , but disregard the nonadiabatic coupling between states @xmath221 and @xmath222 since for the @xmath223 excitation this coupling is much less important than for the @xmath224 excitation and since nonadiabatic dynamics is not our primary focus . however , even this simplified model requires a nontrivial duschinsky rotation@xcite connecting normal modes of the @xmath220 and @xmath221 states . . initial state is the ground state of the @xmath225 surface , the delay time @xmath226 . ( a ) time correlation function [ damped by @xmath227 of eq . ( [ eq : damping ] ) , shown as a dash - dotted line ] . ( b ) corresponding spectrum . ( c ) convergence error @xmath228 [ defined in eq . ( [ eq : err_eta ] ) ] of the damped correlation function as a function of the number of trajectories @xmath89 . the points labeled by `` opt . '' were computed with optimized expansion coefficients @xmath229 of eq . ( [ eq : def_cn ] ) ( see subsec . [ subsec : cdr ] ) . [ fig : pyrazine ] ] since the pyrazine model is globally quadratic , the action expansion in eq . ( [ eq : s_cdr ] ) is exact ( as discussed in subsec . [ subsec : lho ] ) and thus the fully converged dr and drp correlation functions can be obtained by the cellular variants cdr@xmath230 and cdrp@xmath230 of these methods obtained with a single trajectory . figure [ fig : pyrazine](a ) shows pyrazine trse correlation function @xmath231 , calculated for a particular delay time @xmath232 fs and multiplied by a phenomenological damping function@xcite @xmath233\,\theta ( t - t ) , \label{eq : damping}\]]where @xmath5 denotes the total propagation time . parameters of the calculation are summarized in the caption of fig . [ fig : pyrazine ] . the drp is shown in fig . [ fig : pyrazine](a ) to yield an excellent agreement with the quantum calculation . this is also confirmed in the corresponding spectrum [ fig . [ fig : pyrazine](b ) ] , computed as the fourier transform ( eq : sigma_wp ) of the damped correlation function . finally , fig . [ fig : pyrazine](c ) compares the convergence behavior of individual methods . the convergence is quantified by the relative @xmath166 error achieved for @xmath234 trajectories : @xmath235where @xmath236 . the subscript @xmath89 of @xmath237 in eq . ( eq : err_eta ) emphasizes that the quantity @xmath237 was computed with @xmath89 trajectories , while the fully converged results are denoted by @xmath238 . time integrals appearing implicitly in eq . ( [ eq : err_eta ] ) are evaluated with simpson s method . the cellularization accelerates convergence by lowering the number of trajectories required to achieve the same statistical error by about two orders of magnitude [ fig . [ fig : pyrazine](c ) ] . additional minor improvement is achieved by optimizing the expansion coefficients in eq . ( eq : rho_w = sum ) using constraints ( [ eq : def_cn ] ) . after discussing harmonic systems , which are rather simple even in high dimensions , let us turn to the opposite limit of chaotic dynamics , which can present difficulties even in few dimensions . in particular , we consider a two - dimensional chaotic quartic oscillator.meyer:1986,*waterland_yuan:1988,*eckhardt_hose:1989,*bohigas_tomsovic:1993,*revuelta_vergini:2012 the two potential energy surfaces , @xmath239differ only in the parameter @xmath240 . chaotic behavior is due to the coupling term @xmath241 since in the limit @xmath242 , the hamiltonian @xmath243 becomes separable and hence integrable . due to the chaotic character of this system , one expects that the central ingredient of the cellularization , i.e. , the quadratic expansion of the action difference in eq . ( [ eq : s_cdr ] ) will be poor and hinder convergence . this is indeed confirmed in fig . [ fig : ds](a ) , showing the difference of the dr action ( [ eq : dr phase ] ) for two neighboring trajectories specified by initial conditions @xmath136 and @xmath244 , i.e. , @xmath245this quantity is then compared with predictions based on the quadratic expansion ( [ eq : s_cdr ] ) and its linear part . the expansion order denoted linear + @xmath143 is a widely used approximation@xcite to the quadratic expansion ( [ eq : s_cdr ] ) within which one neglects the third derivatives of the potential ( see appendix [ ap : derivatives ] ) . figure [ fig : ds](a ) shows clearly that in the quartic oscillator the quadratic expansion ( eq : s_cdr ) is reliable only for short times and that the linear expansion is superior to the presumably more accurate linear + @xmath143 approach . time dependence of the dr action difference @xmath246 of eq . ( [ eq : delta_ds ] ) calculated for two neighboring trajectories ( initial conditions @xmath247 and @xmath248 ) . ( a ) quartic oscillator ( fig . [ fig : quartic ] ) . ( b ) collinear nco molecule ( fig . [ fig : nco ] ) . delay times are as in figs . [ fig : quartic ] and [ fig : nco ] , @xmath247 is the phase - space center of the gaussian initial state ( of width @xmath249 ) and @xmath250 . the order of the expansion ( [ eq : s_cdr ] ) is distinguished by line type : `` linear '' ( dashed ) , `` quadratic '' ( dotted ) , and `` linear + @xmath143 '' ( dash - dotted ) . the symbol @xmath143 signifies that the derivatives of the stability matrix in eq . ( [ eq : ds_hessian ] ) are neglected . solid line shows numerically exact @xmath246 . ] as a consequence , fig . [ fig : quartic ] , comparing the trse correlation functions , shows that the method of choice for the quartic oscillator is the bare dr [ fig . [ fig : quartic](a ) ] , since the cdr [ fig . [ fig : quartic](c ) ] converges more slowly , while the drp and cdrp are reliable only for short times since the prefactor ( eq : a_drp ) ( understood as a function of time for fixed initial conditions ) grows quickly and oscillates widely at later times . typical chemical systems are neither globally harmonic as our pyrazine - based model , nor fortunately as strongly chaotic as the quartic oscillator . in our last example we therefore consider a realistic , anharmonic system , in order to see how the cdr , drp , and cdrp might perform in typical situations . for this purpose , we chose a two - dimensional model of the collinear nco molecule based on the @xmath251 ( ground ) and @xmath252 ( excited ) pess.@xcite the pess are given in ref . in a form of a polynomial fitted to _ ab initio _ calculations on the domain @xmath253\,$]a.u . and @xmath254 $ ] , specified in @xmath255 ( n c ) , @xmath256 ( c o ) bond - length coordinates and the bond angle @xmath257 . we set @xmath258 ( equilibrium value ) and describe the reduced two - dimensional surfaces in the @xmath255 and @xmath256 coordinates by a simplified two - term form @xmath259\right\ } ^{2}\ ! , \label{eq : nco_pot}\]]where the equilibrium bond lengths @xmath260 are the same as in ref . , while the parameters @xmath175 , @xmath261 , and @xmath262 were obtained by fitting potential ( [ eq : nco_pot ] ) to the functional form of ref . on the domain @xmath263 $ ] with @xmath264a.u . resulting values are summarized in tab . tab : nco_pot_params . these parameters differ from the values employed in our earlier work@xcite and should better reflect the dynamics of this system . frequency - mass - scaled normal mode coordinates of the @xmath251 pes were used so that the vibrational ground state is in the harmonic approximation described by a gaussian with unit widths centered at the origin . . ld4.3d1.3d1.3d1.3d1.3d1.3d1.3 & & & & & & & + @xmath265&-167.653&0.150&1.698&2.302&0.333&1.160&2.246 + @xmath266&-167.549&0.144&1.984&2.234&0.398&1.140&2.232 + the initial state for the trse calculation was prepared by the following procedure.@xcite first , we computed the @xmath251 ground vibrational state by imaginary - time propagation . this state was then pumped to the @xmath267 pes , propagated there for a net time of @xmath268 , dumped to @xmath251 , and propagated for additional @xmath269 . in order to facilitate computation of @xmath270 in eq . ( eq : weierstrass trans ) , we approximated the resulting state by a single gaussian . an independent quantum calculation confirmed that this does not impact the spectrum significantly . the trse correlation function for a delay time of @xmath271 fs is displayed in fig . [ fig : nco](a ) , confirming that the prefactor correction extends the agreement of the dr with the quantum correlation function to longer times . as a consequence , the prefactor correction yields sharper peaks in the corresponding spectrum , shown in fig . [ fig : nco](b ) . finally , fig . fig : nco(c ) , comparing the statistical convergence of the dr , cdr , drp , and cdrp , confirms that in nco the cellularization increases numerical efficiency , although the effect is as expected smaller than in the harmonic pyrazine model [ fig . [ fig : pyrazine](c ) ] . . initial state is a non - stationary state prepared by a pump - dump procedure@xcite discussed in the text , the delay time @xmath272a.u . @xmath273 ( a ) time correlation function [ damped by @xmath227 of eq . ( [ eq : damping ] ) , shown as a dash - dotted line ] . ( b ) corresponding spectrum . ( c ) convergence error @xmath228 [ defined in eq . ( [ eq : err_eta ] ) ] of the damped correlation function as a function of the number of trajectories @xmath89 . the points labeled by `` opt . '' were computed with optimized expansion coefficients @xmath229 of eq . ( [ eq : def_cn ] ) ( see subsec . [ subsec : cdr ] ) . [ fig : nco ] ] classical trajectories needed in the dr , cdr , drp , and cdrp were calculated with a fourth - order symplectic integrator , while quantum calculations employed the corresponding fourth - order split - operator method.wehrle_sulc:2011 time steps used for the pyrazine , quartic oscillator , and collinear nco models were @xmath275 , @xmath276 , and @xmath277 , respectively . also note that the branch of the square root in the prefactor in eq . ( [ eq : a_cdr ] ) was gradually adjusted in the course of the propagation in order to ensure that the phase of the prefactor be continuous in time . we have introduced the cdrp , a rather accurate and efficient semiclassical method for computing ultrafast time - resolved electronic spectra . the cdrp is a two - stage refinement of the dr of fidelity amplitude : a prefactor correction , which typically increases accuracy , is followed by a cellularization procedure increasing efficiency ( see fig . [ fig : cdrp diagram ] ) . the new method has the same computational cost per trajectory as the two intermediate refinements , cdr and drp ; this cost is determined by propagating the stability matrix and its derivatives . while the cost per trajectory is significantly higher than the cost of each dr trajectory , the reduction in the required number of trajectories can in many situations result in higher efficiency compared with the dr . the new methodology has been tested on several systems . in harmonic potentials ( figs . [ fig : sho ] and [ fig : sho2d ] ) , pyrazine - based model ( fig . [ fig : pyrazine ] ) , and collinear nco molecule ( fig . [ fig : nco ] ) , the trse correlation functions and spectra computed with the cdrp were more accurate and required fewer trajectories than the corresponding quantities computed with the original dr . for harmonic potentials , analytical formulas have been derived ; particularly , we have shown that cellularized calculations using a single trajectory are identical to the fully converged noncellular methods since the second - order expansion of the dr phase is exact . moreover , in harmonic potentials the prefactor is the same for all trajectories . in contrast , in systems with highly nonlinear or chaotic dynamics , such as the quartic oscillator , the second - order approximation to the semiclassical action @xmath141 breaks down and its use can decrease both the accuracy and efficiency . interestingly , in such systems the bare dr can perform rather well [ see fig . [ fig : quartic](a ) ] , in agreement with previously published results.@xcite an important result in its own right is the new simple , yet rigorous cellularization scheme for the dr , in which the size and the sampling weight of the gaussian cells changes with their number . a similar cellularization scheme using the inverse weierstrass transform should be useful also for more general quantum dynamics using semiclassical initial value representations such as the heller - herman - kluk - kay propagator . this research was supported by the swiss nsf with grant no . 200021_124936/1 and nccr must ( molecular ultrafast science & technology ) , and by the epfl . we would like to thank m. wehrle and t. zimmermann for discussions . semiclassical propagator in position representation , known as the _ van vleck propagator _ , is given by the expression@xmath278where the summation is performed over all trajectories @xmath16 of the classical hamiltonian @xmath279 starting from @xmath280 and arriving at @xmath281 after time @xmath10 , @xmath282 is the classical action along the @xmath16th path , and @xmath283 is its morse index . the _ phase - space propagator _ is the wigner transform of the evolution operator , @xmath284the integrand in the last equation includes the position propagator between @xmath285 to @xmath286 . using the van vleck propagator , we can obtain the semiclassical expression for eq . ( [ eq : wigner transform of u]):berry:1989,almeida:1998 @xmath287 , \label{eq : u semi}\]]where the sum runs over all paths @xmath16 centered at @xmath288 , i.e. , paths for which @xmath289 [ see fig . [ fig : le - areas u(x ) ] ] , @xmath290 is the _ stability matrix _ of the flow @xmath291 , and the function @xmath292 , called _ center - action _ , is defined as @xmath293where the first term is the symplectic area enclosed by a closed path consisting of a trajectory centered at @xmath288 and the chord connecting this trajectory s final and initial points . and @xmath294 ) contributing to @xmath295 are shown ; @xmath296 is the midpoint of both . geometrical parts of the center - actions @xmath297 and @xmath298 are displayed as filled and hatched areas , respectively . ] fig : le - areas u(x ) in general , the center - action is multivalued and each of its branches is associated with a classical trajectory centered at @xmath288 , as shown in fig . [ fig : le - areas u(x ) ] . the stability matrix , defined as @xmath299 , defines the local linearization of the classical path in the tangent phase - space and the phase - space propagator has caustics whenever @xmath290 has an eigenvalue @xmath300.@xcite moreover , @xmath290 is equal to the cayley transform of @xmath301 , @xmath302and the determinant in phase - space propagator ( [ eq : u semi ] ) can be written in terms of the center action as@xcite @xmath303 ^{-1}=\det \left ( i+\frac{j}{2}\cdot \frac{\partial ^{2}s_{\text{c}}}{\partial \bar{x}^{2}}\right ) .\]]the last relation follows from the fact that both @xmath290 and the hessian of @xmath304 define the same local linearization of the classical equations of motion in a neighborhood of a classical trajectory @xmath305 . this linearization is described by the mapping : @xmath306 for short times , the wigner transformation @xmath307 of the echo operator can be approximated by a propagator ( [ eq : u semi ] ) with a single classical trajectory,@xcite as in eq . ( [ eq : echo_symbol ] ) . one of the main numerical prerequisites of both the drp and cdr is the second order expansion of the dr phase , @xmath308 , as indicated in eq . ( [ eq : s_cdr ] ) . here we describe a symplectic numerical procedure for obtaining the time derivatives of the phase - space derivatives @xmath309 for @xmath310 ( multi - index notation was used ) . as in other semiclassical methods , the knowledge of the hessian of the potential is required for propagating the stability matrix @xmath290 . below we show that in order to obtain the hessian of @xmath308 with respect to @xmath80 , third derivatives of the potential , @xmath311 , are also needed . although the third derivative is in principle required also in cellular dynamics@xcite and cellularized frozen gaussian approximation,@xcite the associated computational cost has led the authors of these methods to neglect the contribution of terms depending on @xmath311 . however , as demonstrated in fig . fig : ds(a ) , this contribution can be essential even in simple realistic models such as the collinear nco molecule . first , consider components of the gradient of @xmath141 , @xmath312where @xmath313 , @xmath314 is the force difference vector , and repeated indexes imply summation . similarly , the components of the hessian of @xmath141 are @xmath315where @xmath316 denotes the hessian of @xmath44 at time @xmath317 and @xmath318while the time integrals in eqs . ( [ eq : ds_gradient ] ) and ( eq : ds_hessian ) are evaluated using composite newton - cotes formulas , the integrands can be propagated symplectically . the algorithm for @xmath319 propagation , e.g. , is obtained by applying the chain rule to the preceding equation , @xmath320whereas the symplectic propagation scheme for the stability matrix was described previously:@xcite @xmath321derivatives of phase - space coordinates in eqs . ( [ eq : n ] ) and ( [ eq : m ] ) are obtained from symplectic integrators for @xmath322 and @xmath323 , which are for standard hamiltonians of the form @xmath324 based on a lie - trotter - type@xcite decomposition of a short - time propagator into elementary steps within which the system is propagated under the influence of either the kinetic or the potential term only . action of the kinetic term @xmath325 for time @xmath326 results in a phase - space shear preserving the momentum , @xmath327whereas the action of the potential term @xmath328 changes momentum and preserves position : @xmath329since the only nonlinear dependence of @xmath330 on @xmath331 stems from the presence of the potential gradient in eq . ( [ eq : p_prop ] ) , the second derivative terms in eq . ( [ eq : n ] ) are nonzero only during the p - propagation ( [ eq : p_prop ] ) and explicitly involve derivatives of the hessian : @xmath332as already mentioned , these third derivatives of the potential , which should appear in other semiclassical propagation schemesheller:1991,walton_manolopoulos:1996 as well , are usually neglected in order to reduce computational cost . yet , in section [ subsec : nco ] we have shown that they can play an essential role even in rather simple systems such as the nco . as discussed in subsec . [ subsec : drp ] , switching the roles of the pess in eq . ( [ eq : f(t , tau ) ] ) for the correlation function corresponds ( for @xmath333 ) to taking the complex conjugate of this equation . likewise , when one interchanges the pess , the dr phase and hence the matrix @xmath334 change the sign . however , since the prefactor @xmath335 in eq . ( eq : drp formula ) is real , it might seem that the drp is incompatible with this operation . here we demonstrate that this is not the case by proving that @xmath336 . to this end , consider a general , symmetric , @xmath99 matrix @xmath337 and let @xmath338 denote any of the eigenvalues of @xmath339 . then @xmath340where we have used the properties @xmath341 , @xmath342 , and that taking the transpose of an arbitrary square matrix does not affect its determinant . equation ( [ eq : app_b_proof ] ) shows that the eigenvalues of @xmath339 come in pairs @xmath343 . this directly implies that @xmath344 . [ this also follows from setting @xmath345 in eq . ( eq : app_b_proof ) and using the fact that the matrix @xmath339 is of even order . ] 75ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1126/science.1169183 [ * * , ( ) ] link:\doibase 10.1126/science.1165733 [ * * , ( ) ] link:\doibase 10.1126/science.1175005 [ * * , ( ) ] link:\doibase 10.1063/1.1674535 [ * * , ( ) ] link:\doibase 10.1063/1.442382 [ * * , ( ) ] link:\doibase 10.1016/0301 - 0104(84)80039 - 7 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1146/annurev.pc.45.100194.000503 [ * * , ( ) ] link:\doibase 10.1146/annurev.physchem.55.091602.094429 [ * * , ( ) ] link:\doibase 10.1146/annurev.physchem.56.092503.141257 [ * * , ( ) ] link:\doibase 10.1063/1.4789759 [ * * , ( ) ] link:\doibase 10.1103/physreve.70.055201 [ * * , ( ) ] link:\doibase 10.1103/physreve.73.046204 [ * * , ( ) ] link:\doibase 10.2533/chimia.2011.334 [ * * , ( ) ] link:\doibase 10.1080/00268976.2012.668971 [ * * , ( ) ] link:\doibase 10.1103/physreve.68.056208 [ * * , ( ) ] link:\doibase 10.1103/physreva.17.939 [ * * , ( ) ] link:\doibase 10.1063/1.444976 [ * * , ( ) ] link:\doibase 10.1063/1.443638 [ * * , ( ) ] @noop _ _ , ed . ( , , ) link:\doibase 10.1063/1.463248 [ * * , ( ) ] http://stacks.iop.org/0953-4075/28/i=19/a=002 [ * * , ( ) ] link:\doibase 10.1063/1.476447 [ * * , ( ) ] link:\doibase 10.1063/1.471407 [ * * , ( ) ] link:\doibase 10.1063/1.475512 [ * * , ( ) ] link:\doibase 10.1063/1.478420 [ * * , ( ) ] link:\doibase 10.1063/1.1843813 [ * * , ( ) ] link:\doibase 10.1063/1.1626631 [ * * , ( ) ] link:\doibase 10.1063/1.1896948 [ * * , ( ) ] link:\doibase 10.1063/1.3690458 [ * * , ( ) ] link:\doibase 10.1063/1.4738878 [ * * , ( ) ] link:\doibase 10.1103/physreve.71.037202 [ * * , ( ) ] link:\doibase 10.1103/physreve.80.046216 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.254101 [ * * , ( ) ] http://stacks.iop.org/1751-8121/44/i=31/a=315101 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.214101 [ * * , ( ) ] link:\doibase 10.1063/1.459848 [ * * , ( ) ] link:\doibase 10.1103/physreve.84.045201 [ * * , ( ) ] link:\doibase 10.1063/1.457815 [ * * , ( ) ] link:\doibase 10.1103/physreva.45.3032 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1016/j.physrep.2006.09.003 [ * * , ( ) ] link:\doibase 10.1080/00018730902831009 [ * * , ( ) ] link:\doibase 10.1016/s0378 - 4371(00)00146 - 1 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.91.210403 [ * * , ( ) ] http://stacks.iop.org/1367-2630/6/i=1/a=020 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.164101 [ * * , ( ) ] link:\doibase 10.1063/1.3451266 [ * * , ( ) ] link:\doibase 10.1063/1.3187240 [ * * , ( ) ] link:\doibase 10.1002/qua.22730 [ * * , ( ) ] link:\doibase 10.1098/rspa.1989.0052 [ * * , ( ) ] link:\doibase 10.1016/s0370 - 1573(97)00070 - 7 [ * * , ( ) ] link:\doibase 10.1088/0951 - 7715/8/2/005 [ * * , ( ) ] link:\doibase 10.1080/00268979600100651 [ * * , ( ) ] link:\doibase 10.1016/s0550 - 3213(86)80034 - 7 [ * * , ( ) ] link:\doibase 10.1016/0009 - 2614(87)80142 - 2 [ * * , ( ) ] link:\doibase 10.1063/1.1402992 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1016/0009 - 2614(88)87380 - 9 [ * * , ( ) ] http://www.cambridge.org/gb/knowledge/isbn/item1174754/?site_locale=en_gb[__ ] , ed . ( , ) @noop _ _ , ed . ( , ) link:\doibase 10.1063/1.478571 [ * * , ( ) ] link:\doibase 10.1063/1.470689 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1016/0022 - 2852(90)90247-n [ * * , ( ) ] , , and , eds . , @noop _ _ , ed . ( , , ) link:\doibase 10.1063/1.450296 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.61.2733 [ * * , ( ) ] link:\doibase 10.1103/physreva.39.3776 [ * * , ( ) ] link:\doibase 10.1016/0370 - 1573(93)90109-q [ * * , ( ) ] link:\doibase 10.1103/physreve.85.026214 [ * * , ( ) ] link:\doibase 10.1080/00268979300102131 [ * * , ( ) ] link:\doibase 10.1021/j100009a011 [ * * , ( ) ] link:\doibase 10.1063/1.473532 [ * * , ( ) ] link:\doibase 10.1090/s0002 - 9939 - 1959 - 0108732 - 6 [ * * , ( ) ]

accurately calibrated effective field theories are used to compute atomic parity non - conserving ( apnc ) observables . although accurately calibrated , these effective field theories predict a large spread in the neutron skin of heavy nuclei . while the neutron skin is strongly correlated to a large number of physical observables , in this contribution we focus on its impact on _ new physics _ through apnc observables . the addition of an isoscalar - isovector coupling constant to the effective lagrangian generates a wide range of values for the neutron skin of heavy nuclei without compromising the success of the model in reproducing well constrained nuclear observables . earlier studies have suggested that the use of isotopic ratios of apnc observables may eliminate their sensitivity to atomic structure . this leaves nuclear structure uncertainties as the main impediment for identifying physics beyond the standard model . we establish that uncertainties in the neutron skin of heavy nuclei are at present too large to measure isotopic ratios to better than the 0.1% accuracy required to test the standard model . however , we argue that such uncertainties will be significantly reduced by the upcoming measurement of the neutron radius inpb at the jefferson laboratory .

nucl-th0501014

the quest for discovering fundamental characteristics of atomic nuclei , such as their sizes and shapes , is as old as nuclear physics itself . although for the most part this challenge has been met with enormous success , precise knowledge of certain fundamental nuclear - structure properties among them the spatial distribution of neutrons is still lacking . while the original motivation for measuring the neutron density of atomic nuclei was deeply rooted in nuclear structure , the quest has been recently revived in response to the widespread impact that such a measurement will have over seemingly unrelated areas of physics . even though the present study concentrates exclusively on atomic parity non - conservation ( apnc ) @xcite , a brief review of areas that could benefit from an accurate measurement of the neutron density is also presented . unquestionably , the accurate determination of the neutron density and in particular its root - mean - square ( rms ) radius remains a top priority in nuclear - structure physics . this inadequate state of affairs is in stark contrast to our present knowledge of proton densities . indeed , the proton distribution in nuclei has already been mapped with exquisite precision , often much better than 1% @xcite , due to the availability of state - of - the - art electron - scattering facilities all over the world . a particularly illustrative example is @xmath0pb , a nucleus whose charge radius is presently known with an extremely high accuracy ( @xmath1 fm @xcite ) that exceeds at least by two orders of magnitude that of its neutron radius . clearly , an accurate determination of neutron radii is a pressing issue . further , without such knowledge for stable nuclei , the prospects of real progress in the future domain of nuclear structure , namely , _ exotic nuclei _ , may be seriously compromised . precision studies of the nucleon - nucleon ( nn ) force is an area that has been directly affected by our poor knowledge of neutron radii . at the most fundamental level , quantum chromodynamics predicts a small flavor violation in the strong force due to the different charges and masses of the up and down quarks . in turn , this flavor violation induces a small ( of the order of 1% ) breaking in the isospin symmetry of the nn force . however , to accurately quantify these `` novel '' isospin violations , it is necessary to have all `` conventional '' effects under control . a good example of this interplay between novel and conventional is provided by the nolen - schiffer ( or coulomb energy ) anomaly . the anomaly consists in the residual discrepancy observed in the ground - state energies of mirror nuclei with a single nucleon added ( or removed ) from a closed shell ( e.g. , @xmath2sc@xmath2ca ) . however , some of the conventional effects depend sensitively on our knowledge of neutron radii . indeed , it has been argued that by using neutron skins defined as the difference between the neutron and proton rms radii significantly smaller than those predicted by theory ( and apparently justified by experiment ) the anomaly disappears @xcite . knowledge of neutron radii could also remarkably improve our understanding of the equation of state of neutron - rich matter . although existing ground - state observables appear to constrain the symmetry energy of nuclear matter at a neutron density around @xmath3 @xmath4 ( as a reference value typical in nuclei ) @xcite , the density dependence of the symmetry energy is unknown . recently , however , it has been found that the neutron skin of a heavy nucleus , like @xmath5pb , calculated with different non - relativistic and relativistic mean field parametrizations displays a tight linear relationship with the slope of the equation of state of neutron matter ( or , analogously , with the derivative of the symmetry energy ) , evaluated at @xmath6 @xmath4 @xcite . in particular , models with a `` stiffer '' symmetry energy predict larger neutron skins . for instance , the neutron skin in @xmath5pb computed with relativistic models is larger by typically 0.10.2 fm than the value obtained with skyrme forces . in turn , the neutron skin of @xmath5pb is linearly correlated with the neutron skin of other relatively heavy nuclei such as @xmath7sn or @xmath8ba . thus , a measurement of the neutron radius of @xmath0pb , or indeed of any other heavy nucleus , could fix a fundamental property of the equation of state . fixing the slope of the symmetry energy at @xmath9 will also impact favorably on astrophysical observables , particularly on the structure and dynamics of neutron stars . indeed , strong correlations between the neutron skin of @xmath5pb and various neutron - star properties have been established @xcite . among these , a model - independent relation between the neutron skin of @xmath5pb and the transition density from uniform neutron - rich matter ( in the mantle ) to a non - uniform phase ( in the crust ) was observed . other neutron - star properties sensitive to the neutron skin of @xmath5pb include the radius of the star , its composition , and its cooling mechanism @xcite . ultimately , these correlations emerge as a result of the similar composition of the neutron skin of a heavy nucleus and the crust of a neutron star , namely , neutron - rich matter at similar densities . while the above arguments are already compelling enough to justify the commission of new experiments to measure the neutron radius of @xmath5pb with unprecedented accuracy @xcite , this manuscript will focus on the crucial role that such a measurement could have on precision studies of the standard model , vis - - vis , atomic parity non - conservation . the road to new physics beyond the standard model in low - energy tests in atoms , which stems from violations of fundamental symmetries that occur in the weak interaction , passes through the observation of deviations between the values measured in the laboratory and the predictions of the standard model . the effects are inherently small and , traditionally , the analysis of apnc experiments has been hindered by uncertainties in both atomic and nuclear - structure theory . however , a fruitful strategy has been devised to remove the sensitivity to atomic theory . this strategy consists in measuring _ ratios _ of parity - violating observables along an isotopic chain @xcite . in this manner , uncertainties in the atomic theory factor out from the ratios . based on their long chains of several naturally occurring isotopes , cesium , barium , dysprosium , and ytterbium appear as ideal candidates @xcite . as a result , nuclear - structure uncertainties primarily in the form of neutron radii remain as the limiting factor in the search for physics beyond the standard model @xcite . in an earlier study , a strong correlation between the neutron radius of @xmath5pb and the neutron radii of @xmath8ba , @xmath10dy , and @xmath11yb was established @xcite . in that study the density dependence of the symmetry energy and consequently the neutron radius of @xmath5pb was modified through the addition of a new coupling constant ( @xmath12 ) to the underlying lagrangian @xcite . this new coupling allows one to modify the density dependence of the symmetry energy without changing the saturation properties of symmetric nuclear matter . in a heavy nucleus @xmath12 modifies the rms radius of the neutron density while leaving the rms radius of the proton density and the total binding energy practically unchanged @xcite . however , the nuclear - structure model used lacked both deformation and pairing effects that are important for the nuclei considered in apnc experiments . thus , in ref . @xcite the calculations were limited to the study of a single member of each isotopic chain ( the one with a closed neutron shell ) . this shortcoming is resolved in the present contribution . as a result , one is now able to map the neutron radii of these long isotopic chains as a function of both the neutron number @xmath13 and @xmath12 . there are previous studies of the isotopic dependence of the neutron radius in the relativistic mean field ( rmf ) approximation @xcite and of its dependence on @xmath14 @xcite . yet the merit of the present work is that uncertainties in the neutron radius are studied for the first time as a function of both @xmath13 and @xmath12 within a unified `` best - fit '' model . in addition , it has been recently pointed out that the accuracy in the measurements of apnc effects can be improved by using high-@xmath15 atoms , such as francium , where the parity violating effects are expected to be an order of magnitude larger than in cesium @xcite . thus we also include , in anticipation of future experiments , the analysis of the nuclear structure corrections to the weak charge in francium isotopes . the manuscript has been organized as follows . in sec . [ theory ] a short description of the nuclear model is given , with special emphasis paid to the treatment of deformation and pairing correlations as well as to the predictions of the model applied to cesium , barium , dysprosium , and ytterbium . in sec . [ results ] a brief review of atomic parity violation is given , with the main goal of addressing uncertainties in apnc observables emerging from our poor knowledge of neutron radii ; a detailed analysis of such uncertainties within our model is also presented in the same section . finally the conclusions are reserved for sec . [ conclusions ] . we may summarize the two main conclusions of our work as follows . first , it appears that even when using the wealth of existing ground - state observables to constrain our nuclear - structure models , the uncertainties in the neutron radius of heavy nuclei and consequently on apnc observables may have been underestimated . second , we have found , as others have before us , a tight correlation between the neutron radius of @xmath5pb and the neutron radii of barium , dysprosium , and ytterbium , which is shown here to hold also in deformed systems and in the presence of pairing . this correlation suggests that the nuclear structure observables relevant for apnc will profit positively from the upcoming high precision measurement of the neutron radius of @xmath5pb at the jefferson laboratory @xcite . the mean field treatment of effective field theories of hadrons , generally known as quantum hadrodynamics ( qhd ) , is well established as a successful approach for describing diverse bulk and single - particle properties of finite nuclei and uniform nuclear matter @xcite . these effective field theories are based on a lagrangian density that contains the nucleon ( as an elementary dirac particle ) together with an isoscalar - scalar ( @xmath16 ) meson , an isoscalar - vector ( @xmath17 ) meson , an isovector - vector ( @xmath18 ) meson , and the photon as the relevant degrees of freedom for describing the nuclear many - body problem . in the mean field approximation the meson fields are replaced by their ground - state expectation values , thereby becoming classical fields . the quantum structure of the theory is carried by the nucleon field . for systems with time reversal symmetry , only the time - like component of the vector fields contribute . at the mean field ( `` tadpole '' ) level , charge conservation implies that only the third component ( in isospin space ) of the isovector @xmath18-meson field does not vanish . as a final product , one obtains the following hamiltonian density : @xmath19 \varphi_\alpha \nonumber \\[3 mm ] & & + \frac{1}{2g_{\rm s}^2 } \left [ \left ( \mbox{\boldmath$\nabla$}\phi\right ) ^2 + m_{\rm s}^2\phi^2\right ] + \frac{\kappa}{3!}{\phi^3}+\frac{\lambda}{4!}{\phi^4 } -\frac{1}{2g_{\rm v}^2 } \left [ \left ( \mbox{\boldmath$\nabla$ } w\right ) ^2 + m_{\rm v}^2 w^2\right ] \nonumber \\[3 mm ] & & -\frac{1}{2g_\rho^2 } \left [ \left ( \mbox{\boldmath$\nabla$ } b\right ) ^2 + m_\rho^2 b ^2\right ] -\lambda_{\rm v } w^2b^2 -\frac{1}{2 } \left ( \mbox{\boldmath$\nabla$ } a\right ) ^2,\end{aligned}\ ] ] where the summation runs over the occupied nucleon states @xmath20 of positive energy and @xmath21 stands for the third component of the isospin operator . the scaled meson fields associated with the @xmath16 , @xmath17 , and @xmath18 mesons are , respectively , @xmath22 , @xmath23 , and @xmath24 . finally , @xmath25 represents the time - like component of the photon field . the variation of eq . ( [ pv1 ] ) with respect to the dirac spinors yields the dirac equation satisfied by the nucleons , while the variations with respect to the meson fields lead to the klein - gordon equations for the corresponding mesons @xcite . the functional in eq . ( [ pv1 ] ) contains cubic and quartic scalar meson self - interactions ( couplings @xmath26 and @xmath27 , respectively ) that are tuned to bring the value of the compression modulus of symmetric nuclear matter down to the empirical value ( @xmath28300 mev ) @xcite . further , through the inclusion of these terms one can accurately reproduce the systematics of finite nuclei throughout the periodic table as , e.g. , in the case of the celebrated relativistic mean field ( rmf ) parameter set nl3 @xcite . the additional isoscalar - isovector cross coupling @xmath12 enables one to modify the density dependence of the symmetry energy without compromising the success of the model . insofar as we are interested in the properties of some relatively heavy nuclei that have been identified as possible candidates for apnc experiments , we extend the model of ref . @xcite to include nuclear deformation and pairing correlations in the present rmf study . only quadrupole deformation ( @xmath29 ) is included though , as it is known that hexadecapole deformation is generally small for the nuclei under consideration @xcite . a bcs framework is employed to deal with the pairing correlations . note that because the nuclei to be studied lie relatively close to the @xmath30-stability line , the use of the bcs approximation is justified as the fermi level is sufficiently bound and the mixing with continuum states essential for drip line nuclei is not relevant here . as in most bcs calculations , the strength of the neutron - neutron and proton - proton pairing interactions is determined from the neutron and proton pairing gaps ( respectively , @xmath31 and @xmath32 ) evaluated from the experimental odd - even mass differences , using a five - point formula @xcite . the dirac equation and the klein - gordon equations for the meson fields are treated using an axially deformed basis expansion method as described in detail in refs . the calculations presented here have been carried out by expanding the fermionic wave functions and the bosonic fields in 12 and 20 oscillator shells , respectively . the solution of the equations in an axially deformed basis for the odd - even and odd - odd nuclei is more complicated . indeed , in these cases time reversal symmetry is broken and the odd particle induces polarization currents and time - odd components in the mean fields . however , the impact of these effects on deformation and binding energies is small @xcite and will be neglected henceforth . in the pairing calculation of the odd nucleon we use the blocking approximation @xcite , thereby restoring time reversal invariance in the intrinsic frame . for the isotopic chains under consideration for apnc experiments , namely , cs , ba , dy , and yb , the nl3 parameter set yields ground - state properties , such as binding energies and deformations , in good agreement with the available experimental data @xcite . for our calculations we adopt the nl3 parameter set and suitably modify it with the addition of the isoscalar - isovector coupling @xmath14 , prompted by the fact that the slope of the nuclear symmetry energy at ( or near ) the saturation density is at present unknown . thus , we utilize the nonlinear isoscalar - isovector term @xmath14 to change the density dependence of the symmetry energy , which in turn modifies the neutron radius of neutron - rich nuclei . this can be done without a significant change in those ground - state properties of finite nuclei that are well constrained experimentally @xcite . in practice , for a given @xmath12 value we readjust the nucleon - rho coupling constant @xmath33 so that the symmetry energy of nuclear matter at a density of @xmath34 be 25.68 mev @xcite . the choice is motivated by the fact that the symmetry energy of uniform nuclear matter at saturation density is not well constrained by the known properties of finite nuclei . rather , it is some average between the bulk symmetry energy at saturation and the surface symmetry energy which is constrained by the binding energy of finite nuclei @xcite . for example , typical effective nuclear forces fitted to the empirical energies of nuclei predict a liquid drop model symmetry energy coefficient , which contains a surface correction , of 2224 mev for @xmath5pb @xcite , far smaller than the value of the bulk symmetry energy coefficient for that interaction . although it has been established that a variety of ground - state properties of spherical nuclei are insensitive to the choice of @xmath12 @xcite , this is not the case for the neutron skin . the neutron skin @xmath35 , where @xmath36 and @xmath37 are the neutron and proton rms radii ( @xmath38 ) , is strongly sensitive to the density dependence of the symmetry energy and hence , to the isoscalar - isovector coupling @xmath12 . this is because the slope ( i.e. , pressure ) of the symmetry energy controls how far is the neutron skin pushed out relative to the symmetric core . in order to show that this scenario holds true even when the spherical symmetry is broken and pairing correlations are important , we display in fig . [ fig1 ] the charge radius , binding energy , quadrupole deformation , and the neutron skin as a function of @xmath12 for the representative heavy nucleus @xmath39yb . note that all observables in fig . [ fig1 ] have been normalized to the corresponding nl3 values ( @xmath40 ) . it is seen that the charge radius , binding energy and quadrupole deformation are insensitive to the value of @xmath12 in the 00.025 range ; the quadrupole deformation displays the largest change , but this amounts to only 2% . in contrast , the neutron skin decreases rapidly with increasing @xmath12 and it varies by about 25% when @xmath12 is changed from 0 to 0.025 . this reflects the softening of the symmetry energy as @xmath12 increases . therefore , we conclude that neither pairing correlations nor deformation alter our earlier conclusions about the role of @xmath12 in the systematics of finite nuclei . the neutron skin of a nucleus is an important component in the study of atomic parity non - conservation . so is its charge radius . while neutron radii of heavy nuclei ( @xmath36 ) are poorly known , high precision data for the charge radii are available . to test the reliability of our model , we now compare in fig . [ fig2 ] calculated charge radii with experimental data for some of the relevant isotopes used in apnc experiments ; these include cs , ba , dy and yb . the experimental values are taken from the recent compilation of ref . @xcite , which combines measured data from electron scattering , from muonic x - rays , and from @xmath41 and optical isotope shifts for the purpose of providing a unified set of nuclear charge radii . theoretical calculations are presented with the original nl3 parametrization ( @xmath40 ) and with one for which the density dependence of the symmetry energy has been softened by fixing the value of the isoscalar - isovector coupling to @xmath42 . it can be seen that both sets of calculated charge radii are in excellent agreement with the experimental values for _ all _ of the considered nuclei . at a more quantitative level , we find that the relative @xmath43 value between the theoretical and experimental charge radii for the isotopic chains of cs , ba , dy , and yb is always smaller than 0.25% for both ( @xmath40 and @xmath44 ) parameter sets . this result validates the use of both parameter sets in computing key nuclear observables of relevance to the apnc program . note that the binding energies and deformations for all the nuclei considered here are also in good agreement with the experimental data ( not shown ) . moreover , and as alluded earlier , these observables are highly insensitive to the choice of @xmath12 . a more daring challenge for a theoretical model than the charge radii themselves is the reproduction of the observed charge isotope shifts . one famous case is the pronounced kink exhibited by the experimental charge radii of the pb nuclei at the magic neutron number @xmath45 . the kink is not reproduced by conventional non - relativistic theories ( either skyrme or gogny forces ) . in contrast , the rmf calculations show a remarkable agreement with experiment @xcite . the origin for the deviations between the two models has been traced back , largely , to the isospin dependence of the spin - orbit term . we present in fig . [ fig2new ] results for the isotope shifts in two of our flagship chains for the subsequent apnc analysis : the cesium and francium alkali metals . the shifts are referred to @xmath46cs and @xmath47fr , respectively , as are the data of ref . the experimental method of preference for the measurement of nuclear charge isotope shifts is high - resolution laser spectroscopy , which exploits the isotopic effects on the hyperfine structure of atomic optical transitions . in addition to the rmf calculations with @xmath48 and @xmath49 we have drawn in fig . [ fig2new ] the skm * and siii results of the non - relativistic hartree - fock calculations of ref . @xcite for cs . one immediately recognizes the ability of the rmf model in reproducing the trend of the changes of the experimental data when neutrons are added ; conspicuously , the change of slope at the @xmath50 shell closure for cs and at the @xmath45 shell closure for fr . in spite of this , the theoretical predictions may be off the experiment by a factor of @xmath51 in some specific isotopes . one should keep in mind that this happens in cases where the isotope shift is a small quantity and that it emerges from the cancellation of two much larger numbers ( the square charge radii of two isotopes ) . similarly to the observables discussed above , the calculated isotope shift is a quantity that remains unaffected by the @xmath12 coupling . as has been mentioned in earlier publications @xcite , typical relativistic parametrizations predict a neutron skin in @xmath5pb of about @xmath52 fm . this should be contrasted with the case of non - relativistic forces of the skyrme or gogny type that yield values that are significantly smaller ( between 0.1 and 0.2 fm ) . as the proton rms radius in @xmath5pb is reproduced by both relativistic and non - relativistic models , the spread in the neutron rms radius in @xmath5pb owing to the different density dependence of the symmetry energy in different theoretical models is about 0.10.2 fm . within the relativistic mean - field models a spread ( i.e. , a theoretical uncertainty ) of the neutron skin for a specific nucleus can be simulated in a controlled manner through the addition of the isoscalar - isovector coupling @xmath12 @xcite . starting from the original nl3 parameter set and changing the isoscalar - isovector coupling constant from @xmath40 ( pure nl3 interaction ) to @xmath44 , the neutron skin in @xmath5pb varies from 0.280 fm to 0.209 fm @xcite . hence , in this formalism the induced theoretical uncertainty of the neutron radius of @xmath5pb between these two extreme values of @xmath14 is approximately @xmath53 fm . this spread for @xmath5pb is only slightly larger than the expected experimental error in the measurement of the neutron radius in @xmath5pb via the parity violating electron - scattering experiment at the jefferson laboratory ( known as the prex experiment ) @xcite . thus we take the theoretical spread in the neutron radius of @xmath5pb as a baseline to estimate a realistic uncertainty in those nuclei involved in apnc experiments . the neutron skins calculated with both of the relativistic parametrizations ( namely , @xmath48 and @xmath49 ) for stable and long - lived isotopes of cs , ba , dy , and yb are displayed in fig . [ fig3 ] . for a specific isotope , the shift between the two curves displayed in the figure gives the model spread ( due to @xmath12 ) of the neutron skin of the given nucleus . for all the considered isotopic chains the neutron skins calculated with @xmath40 ( the stiffest symmetry energy ) and @xmath44 ( the softest symmetry energy ) lie roughly along two parallel lines . in general , the effect of @xmath12 on the neutron skin is more prominent for the more asymmetric and heavier nuclei . this can be observed for the relatively long chain of cs isotopes in fig . [ fig3 ] , where the spread of the neutron skin varies approximately from 0.03 fm to 0.06 fm in passing from @xmath54cs to @xmath55cs . to the extent that in the rmf model used in this work the proton rms radius of a given nucleus is almost independent of @xmath12 , the theoretical spread of its neutron skin is therefore approximately equal to the one for its neutron rms radius . at present , the situation regarding our ignorance of the neutron skin of heavy nuclei is unlikely to improve because the existing experimental data can not reduce the spread in the neutron radii of cs , ba , dy , or yb any further . thus , we now examine whether the prex experiment on the neutron radius of @xmath5pb may place important constraints on the neutron skin of apnc nuclei , including those deformed isotopes with an odd number of nucleons . figure [ fig4 ] evidences the strong linear correlation @xcite between the neutron skin of @xmath46cs , @xmath8ba , @xmath56dy , and @xmath57yb and that of @xmath5pb . the calculations include values for the isoscalar - isovector coupling constant @xmath12 ranging from 0 to 0.025 . note that the correlation holds even for the case of deformed nuclei having an odd number of nucleons . with the culmination of the prex experiment , the present theoretical spread in the neutron radius of @xmath5pb of @xmath58 fm ( relativistic and non - relativistic models altogether ) @xcite will be replaced by a genuine experimental error five times smaller ; that is , @xmath59 fm @xcite . this experimental error in the neutron radius of @xmath5pb will also be the experimental error in the neutron skin , as the proton radius is known with a much higher accuracy . if the expected experimental accuracy is attained , then the spread in the neutron skin of the several apnc isotopes will also be ( indirectly ) reduced . in turn , this result will impact favorably on apnc observables by appreciably reducing the nuclear structure uncertainty . parity violation in atomic systems arises from the interference between the parity conserving electromagnetic interaction and the parity violating weak interaction . although most of the binding energy of the atomic electrons comes from their attractive coulomb interaction with the @xmath15 protons in the nucleus , the weak charge of the nucleus @xmath60 induces a small correction in the binding energy and parity of the electronic wave functions . these small corrections can be measured experimentally . the experimentally measured quantity is the electric dipole amplitude between two electronic states that in the absence of parity violating effects would have the same parity , thereby making the amplitude vanish . the observable related to the dipole amplitude can be parametrized as follows @xcite : @xmath61\;,\ ] ] where @xmath62 embodies the atomic structure contribution and @xmath63 represents the weak charge of a nucleus of neutron number @xmath13 and electric charge @xmath15 . the experimentally measured weak charge of the nucleus ( without radiative corrections @xcite ) @xmath63 differs from the standard model prediction , @xmath64 by a nuclear correction factor that arises from the difference between proton and neutron one - body densities . that is , @xmath65 with @xmath66 note that in eq . ( [ pv4 ] ) @xmath67 is the weinberg ( or weak mixing ) angle , while @xmath68 in eq . ( [ pv6 ] ) is the neutron ( proton ) density normalized to one , and @xmath69 is an electronic form factor that describes the spatial variation of the electronic axial - vector matrix element over the size of the nucleus . the atomic parity formalism `` starts '' by assuming that neutrons and protons have the same spatial distribution and then adds a nuclear - structure correction factor to quantify the differences between the actual neutron and proton densities . this is motivated by the fact that the proton density , which is needed to determine various parity conserving observables , is very accurately known . during the course of this presentation and as has been done elsewhere @xcite an additional quantity that parametrizes the nuclear structure corrections will be extensively used . this quantity is defined as follows : @xmath70 note that , in terms of this quantity , the weak charge of the nucleus may be written as @xmath71 \nonumber \\ & = & \big[-nq_n+zq_{p}(1 - 4\sin^{2}\theta_{w})\big ] \frac{1}{q_{p } } \nonumber \\ & = & \big[q_w^{\rm sm}(n , z)+q_w^{\rm nucl}(n , z)\big ] \frac{1}{q_{p } } \;.\end{aligned}\ ] ] to compute the axial form factor @xmath69 one expands the electronic dirac wave functions in a power series about the origin @xcite ( the coulomb potential is assumed to be that of a uniform nuclear charge distribution ) . using this prescription yields the electronic form factor in closed form : @xmath72,\ ] ] where @xmath73 is the cutoff radius of a sharp nuclear charge density . effects from the finite nuclear size are computed from the ground - state expectation value of the time - like component of the weak vector current by considering the neutron and proton as point - like particles . again , we assume sharp proton and neutron densities with rms radii equal to those predicted by self - consistent rmf models . note that the relation between the sharp cutoff radii @xmath74 and the corresponding rms radii @xmath75 is given by @xmath76 estimates for the weak matrix elements @xmath77 and @xmath78 are then given by the following expressions : @xmath79 \nonumber\\ & = & 1-(z\alpha)^2\left[\frac{817}{3150}+\frac{232}{525 } \frac{t}{r_p}+{\cal o}\big((t / r_p)^{2}\big)\right].\end{aligned}\ ] ] although in the present approximation @xmath77 does not depend on nuclear structure effects , the sensitivity of @xmath78 to nuclear structure uncertainties appears in the form of the rms proton radius which is well known and the poorly known neutron skin @xmath80 . the measurement of the weak charge in apnc experiments is plagued by theoretical uncertainties in both atomic and nuclear structure . fortunately , uncertainties in atomic structure may be eliminated , or at least considerably reduced , by studying parity violation along a chain of isotopes and taking ratios of apnc measurements @xcite . according to eq . ( [ pv2 ] ) , the dependence of the parity violating amplitude @xmath81 on the atomic theory contribution @xmath62 will cancel out in the ratio of two measurements performed in two different isotopes of the same element , provided @xmath62 does not change appreciably along that isotopic chain . thus , we now address the role of the remaining nuclear structure uncertainty in apnc observables along the isotopic chains of cs , ba , dy , and yb . the observables of interest are expressed in the form of the following two ratios : @xmath82 and @xmath83 the approximate sign ( @xmath84 ) in the above two equations follows from the assumption that the overlap @xmath85 appearing in eq . ( [ pv7 ] ) remains constant along the whole isotopic chain . as can be realized from tables [ table2 ] and [ table3 ] ( see below ) , this is an excellent approximation . it has been argued in ref . @xcite that corrections to standard model predictions induced by new physics or uncertainties in nuclear structure are essentially the same whether one uses @xmath86 and @xmath87 . thus , we focus here exclusively on @xmath86 . using eqs . ( [ pv4 ] ) and ( [ pv5 ] ) we can write the ratio @xmath86 approximately as @xmath88 \right],\ ] ] where @xmath89 ( @xmath13 ) is the largest ( smallest ) neutron number and @xmath90 represents the difference in neutron number between the two extremes of the isotopic chain . further , ref . @xcite also established that a significant test of the standard model on the basis of isotope ratios requires a measurement of @xmath86 with an accuracy better than @xmath91 . otherwise , a less precise determination of isotope ratios would not be worthwhile to compete with the sensitivity to new physics of the presently known data for the cesium isotope @xmath46cs . the measurement of the electron - nucleon parity violating effect in @xmath46cs to 0.35% accuracy @xcite remains to date as the paradigm of apnc experimental precision . because atomic uncertainties have been eliminated from the ratio ( [ pv8 ] ) , the remaining uncertainties in @xmath86 reflect the known accuracy in the neutron and proton rms radii . while proton densities have been determined from electron - scattering experiments with remarkable accuracy , knowledge of the neutron densities to a comparable level of accuracy is lacking . thus , the main nuclear structure uncertainty in the isotopic ratio @xmath86 comes from our poor knowledge of the neutron radii ( or neutron skin ) of heavy nuclei . indeed , by inserting eq . ( [ pv11 ] ) into eq . ( [ pv8 ] ) , the relative uncertainty in @xmath86 may be approximated by @xmath92\;.\ ] ] in sec . [ theory ] we showed that a spread in the neutron radius of @xmath5pb is obtained by using parametrizations with isoscalar - isovector couplings @xmath40 and @xmath44 , which is representative of the experimental error ( @xmath93 ) expected from the jefferson laboratory experiment . note that the quoted error in the neutron radius of @xmath5pb is also the error in its neutron skin as the proton radius is known with a much higher accuracy . inasmuch as a linear correlation between the neutron skin of @xmath5pb and that of the apnc nuclei has been established ( see fig . [ fig4 ] ) , an estimate of the theoretical uncertainty in the weak charge can be obtained from the spread in the neutron skin of the apnc nuclei using the @xmath40 and the @xmath44 parameter sets . let us outline the procedure we shall follow to determine the relative error in @xmath86 ( namely , @xmath94 ) . first , we calculate the neutron skin @xmath95 in a certain isotopic chain using the @xmath40 parametrization . since @xmath37 is roughly similar for all the isotopes of the chain , the fractional neutron skin variation along the chain is estimated as @xmath96 where @xmath97 refers to the heaviest member and @xmath98 to the lightest member of the isotopic chain under consideration , and @xmath99 is the average proton rms radius of the nuclei in the chain . next , we compute the corresponding fractional neutron skin variation @xmath100 for the softer @xmath44 parameter set . finally , the model spread @xcite is obtained as the difference between the fractional neutron skin variations calculated in the two parameter sets . that is , @xmath101 in table [ table1 ] we display the relative uncertainty @xmath102 in the ratio @xmath86 for each one of the isotope chains ( cs , ba , dy , and yb ) together with the different components required to compute it . these are the largest neutron number of the isotopic chain @xmath103 , the largest difference in the neutron number of the isotopic chain @xmath104 , the average proton rms radius @xmath99 , and the model spread @xmath105 . the latter turns out to be a small quantity , not only in the present relativistic calculations but also in calculations with non - relativistic nuclear energy functionals ( see ref . @xcite and references quoted therein ) . the reason resides in the approximately linear relationship of the neutron skin along a chain of isotopes with the neutron number . while the model dependence in the value of the intercept is obvious ( see fig . [ fig3 ] ) , the slope is only mildly model dependent . and it is the difference in slope that the model spread @xmath105 is sensitive to . it is seen that the estimated relative uncertainty of the apnc ratios @xmath86 in the four isotopic chains considered is around 0.25%0.35% , with a slight increase with increasing atomic number . the quoted uncertainty appears somewhat larger than the @xmath91 value that would be desirable to extract new physics limits beyond the present reach of the @xmath46cs experiment . note that the neutron skin along a given isotopic chain varies ( almost ) linearly with neutron number ( see fig . [ fig3 ] ) . this makes the model spread @xmath105 roughly proportional to the neutron difference @xmath104 . as the combination @xmath106 is therefore nearly independent of @xmath104 @xcite , it is unlikely that the precision in @xmath86 can be improved by enlarging the range of the neutron difference @xmath104 between the two extremes of the isotopic chain [ see eq . ( [ pv13 ] ) ] . table [ table1 ] also displays the model spread @xmath107_{0.1\%}$ ] that would be required to reach the sought - after relative uncertainty of 0.1% in the isotopic ratio @xmath86 . in all cases this number is smaller than the value provided by the rmf model . recall that we have tuned the effective interaction by means of @xmath12 to mimic the purported 1% accuracy of the prex experiment . hence , it appears that under the present and future situation , nuclear structure uncertainties affecting the variation of the neutron distribution along an isotopic chain are too large to make @xmath86 a useful probe for physics beyond the standard model . however , note that the differences are not dramatic : a factor 2.5 for cs and ba , 2.75 for dy , and 3 for yb . thus , second generation experiments and/or novel facilities may significantly aid in this quest . for most of the apnc nuclei particularly cs and fr , with one and no stable isotopes , respectively one can not envisage obtaining precision data on several members of the isotopic chain from experiments at the present time . for these cases , and to provide useful hints for future experiments , accurate estimates of the uncertainty in the neutron skin are essential . thus , we now compute the weak charge including nuclear and one - loop electroweak radiative corrections for some cs isotopes using both relativistic ( @xmath40 and @xmath44 ) parametrizations . the relevant formulas to be employed are given by eqs . ( [ pv4])([pv5 ] ) suitably modified by the inclusion of radiative corrections derived by marciano and rosner @xcite ; that is , @xmath108 \ ; , \label{qwsmrad } \\ & & \hspace{-0.4 in } q_w^{\rm nucl}(n , z)=(0.9857 \pm 0.0004 ) \left[-n(q_{n}-1)+z\big(1-(4.012\pm 0.010)\bar{x}\big ) ( q_{p}-1)\right ] , \label{qwnucrad}\end{aligned}\ ] ] where we have adopted the shorthand notation @xmath109 notice that corrections due to new physics have not been included in the above expressions ; these may be found in refs . @xcite . to be able to compare more precisely our results with those in the available literature , the weak matrix elements @xmath77 and @xmath78 appearing in eq . ( [ qwnucrad ] ) have been slightly modified from those in eqs . ( [ pv10 ] ) and ( [ pv11 ] ) . we now take into account the corrections due to the finite size of the proton and neutron charge distributions , which originate additive correction terms @xmath110 and @xmath111 to eqs . ( [ pv10 ] ) and ( [ pv11 ] ) , respectively . these terms are given by @xcite @xmath112 where @xmath113 are the intrinsic nucleon weak rms radii , @xmath114 are the density distributions of nucleon centers , and @xmath115 are the nucleon weak charges . assuming uniform nucleon distributions , we can cast the corrective terms @xmath116 due to the finite size of the nucleon as @xcite @xmath117 where @xmath118 denotes the sharp cutoff charge radius ( which now includes the finite size of the proton ) and a contribution of 0.14 @xmath119 from the strangeness radius of the nucleon is included . the analysis of the data of a clean measurement by wood _ et al . _ @xcite of the amplitude of the parity non - conserving transition between the 6@xmath120 and 7@xmath120 states of @xmath46cs , the only naturally occurring cesium isotope , established a value of @xmath121 for the weak charge of this element @xcite . this experimentally extracted result differed by 2.5@xmath122 from the standard model prediction with radiative corrections of @xmath123 [ ref . @xcite and eq . ( [ qwsmrad ] ) ] . let us mention that @xmath16 is evaluated by adding in quadrature the experimental and theoretical errors quoted for @xmath124 . the excitement over this apparent disagreement has faded as a consequence of recent reports that demonstrate that the deviation between the experimental results and the standard model prediction can be removed by taking into account self - energy and vertex qed radiative corrections @xcite ( see also ref . @xcite for an up - to - date review of the status of measurements and calculations of parity violation in atoms ) . the newly reported values of the weak charge for @xmath125cs which reconcile the experiment with the standard model prediction are @xmath126 @xcite and @xmath127 @xcite . it should be noted that in the extraction of the experimental values of the weak charge @xmath124 , one assumes equal proton and neutron densities ( normalized to unity ) , so that the overall nuclear correction ( namely @xmath85 ) factors out . an efficient way to account for the difference between neutron and proton densities and thus between @xmath78 and @xmath77has been indicated in eq . ( [ pv4b ] ) . including radiative corrections , this nuclear structure correction factor becomes @xmath128 and the total weak charge of the nucleus is computed as @xmath129 . results for the weak charge together with other relevant quantities , such as @xmath130 [ including the intrinsic structure corrections ( [ qpmod ] ) and ( [ qnmod ] ) ] , the proton rms radius @xmath37 , and the ratio @xmath131 of neutron to proton rms radii are displayed in table [ table2 ] for cs isotopes . as expected , we find that @xmath77 is constant , model independent , and within 0.04% of the sharp - radius value expressed in eq . ( [ pv10 ] ) . however , both the changes and the model dependence of the neutron radii along the isotopic chain are noteworthy . clearly , the ratio @xmath131 increases with increasing neutron number @xmath13 . further , for a given neutron number @xmath13 , @xmath131 decreases with increasing @xmath12 , owing to the softening of the symmetry energy . hence , @xmath78 decreases with increasing neutron number and with a decreasing @xmath12 . this ultimately leads to an increase in the nuclear part of the weak charge @xmath132 [ see eq . ( [ qwnucrad ] ) ] . in summary , while @xmath77 is largely independent of @xmath12 , @xmath78 shows a moderate increase with @xmath12 . thus , through variations in @xmath12 one has a margin to adjust the nuclear contribution to the weak charge . in particular , with the proposed model dependence a change of @xmath133 is induced in @xmath134 for the @xmath125cs atom . the values of the nuclear structure correction factor @xmath135 for the studied cs isotopes , calculated from ( [ pv15 ] ) , have been tabulated in table [ table2 ] . for the specific case of @xmath125cs , we find a nuclear structure correction factor of @xmath136 ( 0.278 ) for the @xmath40 ( @xmath44 ) parameter set . addition of this nuclear correction term to the standard model prediction with radiative corrections ( @xmath137 , third column of table [ table2 ] ) ends in a total weak charge @xmath138 which is in satisfactory agreement with the measured values according to the recent revisions @xcite . it has been discussed in the literature @xcite that @xmath135 calculated with rmf parametrizations is almost twice as large as the corresponding values predicted by skyrme interactions @xcite . this is a reflection of the stiffer symmetry energy of the relativistic models . for example , by using the sharp - cutoff - radius approximation of eqs . ( [ pv10 ] ) and ( [ pv11 ] ) one can write @xmath139 because the above nuclear correction factor is directly proportional to the neutron skin @xmath80 , rmf parametrizations with stiffer symmetry energies and consequently larger neutron skins yield larger values for @xmath135 relative to skyrme interactions . equation ( [ pv16 ] ) may also be used to estimate the model spread in @xmath140 for @xmath125cs . this is given by @xmath141 where the model spread for @xmath142 has been derived from the @xmath131 values calculated with @xmath40 and @xmath143 for @xmath125cs ( listed in table [ table2 ] ) . a straightforward estimate of the spread computed as the difference of the values of @xmath135 between the @xmath40 and the @xmath44 parameter sets yields @xmath144 which compares well with our previous value . in ref . @xcite it has been argued that for the purpose of significantly reducing the nuclear structure uncertainties , relative to those arising from radiative corrections , the neutron radius of @xmath125cs must be known within 2% . the promise of a 1% measurement of the neutron radius of @xmath5pb at the jefferson laboratory @xcite , together with theoretical correlations of the type displayed in fig . [ fig4 ] , will be of considerable help in reaching the desired goal . another strategy proposed as a means for attempting to improve the accuracy in apnc experiments is the utilization of high-@xmath15 atoms that enhance parity violating effects owing to both the increase in the number of nucleons and the increased electronic density in the neighborhood the nucleus . further , alkali atoms have the simplest atomic structure , thereby minimizing theoretical uncertainties . this makes francium , the heaviest `` simple '' atom , an attractive candidate for apnc experiments . indeed , the increased number of nucleons and the increased electronic density in the region of the nucleus makes the apnc effect in fr some 15 to 20 times larger than in cs @xcite . unfortunately , francium has no stable naturally occurring isotopes . although remarkable experimental progress has already been achieved in the field @xcite , the production of intense beams of radioactive fr isotopes represents an exciting challenge for the future of the rare - isotope facilities all over the world . we present estimates for apnc observables in francium along an isotopic chain that ranges from @xmath145fr to @xmath146fr and that includes the two isotopes with the longest half - life ( of about 20 min ) ; these are @xmath47fr and @xmath147fr . in fig . [ fig5 ] we display the neutron skin , charge radii , binding energy per nucleon , and deformation for six isotopes in the chain computed with the two extremes of the values of the isoscalar - isovector coupling constant used in this work ( @xmath40 and @xmath44 ) . as established earlier for @xmath39yb in fig . [ fig1 ] , the calculated values for the charge radii , binding energies , and deformation show little model dependence and reproduce the existing experimental data ( where available ) @xcite rather well . as expected , it is only the ( experimentally unknown ) neutron skins that are sensitive to the value of @xmath12 . note that with the exception of the semi - magic nucleus @xmath148fr , the ground state of all other isotopes is found to be deformed . as was done for the cs isotopes in table [ table2 ] , we display in table [ table3 ] the calculated weak charge for the six considered fr isotopes , alongside other relevant observables . according to eq . ( [ qwsmrad ] ) , the standard model predictions for the weak charges of @xmath47fr and @xmath147fr result in @xmath149 and @xmath150 , respectively , where the uncertainty in @xmath151 has been estimated in a similar way to that of @xmath46cs in ref . the nuclear corrections to the weak charge of @xmath47fr ( @xmath147fr ) calculated from @xmath78 and @xmath77 through eq . ( [ pv15 ] ) are given as follows : @xmath152 these results suggest that nuclear - structure uncertainties in our model , arising from differences in the shape of the neutron density relative to that of the proton , are 0.266 and 0.347 for the francium isotopes @xmath47fr and @xmath147fr , respectively . the numbers are similar to the uncertainties 0.264 and 0.345 that one would obtain with the sharp - radius approximation of eqs . ( [ pv16 ] ) and ( [ pv17 ] ) . thus , with the expected precision to be attained by the prex experiment , which we mimic here through a change in @xmath12 , our predictions for the weak charge @xmath153 of the @xmath47fr and @xmath147fr isotopes vary between @xmath154 ( @xmath48 ) and @xmath155 ( @xmath156 ) for @xmath47fr , and between @xmath157 and @xmath158 for @xmath147fr . note that the uncertainty due to @xmath12 should be augmented by a @xmath159 error from @xmath160 . motivated by the prospects of a high - precision ( 1% ) measurement of the neutron radius of @xmath5pb at the thomas jefferson laboratory , we have examined the impact of such experiment on atomic parity non - conserving observables . while such a measurement will have far reaching consequences in fields as diverse as nuclear structure , flavor violations in the strong interactions , and nuclear astrophysics , we have focused in this contribution exclusively on apnc observables . our theoretical framework is based on the highly successful relativistic nl3 parameter set , suitably modified by the addition of an isoscalar - isovector term ( @xmath12 ) . the virtue of such a term is that it enables one to change the neutron radius of heavy nuclei which at present is poorly known without compromising the success of the model in reproducing a variety of ground state properties of nuclei throughout the periodic table . relative to an earlier study based closely on this formalism , the present study improves on it in one essential aspect : the inclusion of deformation and pairing correlations . without adding these effects , the predictions for the structure of most nuclei along the isotopic chains of relevance to the apnc program would be unreliable . as a result , in the present work we have mapped the neutron radii of the relevant isotopic chains as a function of both @xmath13 and @xmath12 . as established by others before us , we have found a strong correlation between the neutron radius of @xmath0pb and the neutron radius of apnc nuclei even for the case of odd - even and odd - odd nuclei . employing this correlation , we have used a range of values for @xmath12 that closely matches the 1% ( or @xmath161 fm ) uncertainty expected to be achieved by the jefferson laboratory experiment . with this information at hand , we proceeded to study the impact of such a 1% measurement on two different combinations of apnc observables . the first set of observables ( @xmath86 and @xmath162 ) are formed from _ ratios _ of weak nuclear charges along isotopic chains . the merit of such observables is that the ratios are largely insensitive to uncertainties in atomic structure , leaving nuclear uncertainties in the form of the neutron skin as the main source of theoretical error . the second set of observables involves a direct determination of the weak charge of various alkali metals , such as cesium and francium . unfortunately , these elements have very few stable or long - lived isotopes , so in their case the accurate determination of ratios of weak nuclear charges is more difficult . in the case of the isotopic ratio @xmath86 , it has been claimed that a significant test of the standard model requires a determination of @xmath86 to better than 0.1% . this precision would be required for @xmath86 to supersede the @xmath125cs experiment as the most stringent test for new physics within the apnc program . unfortunately , our results indicate that the projected 1% accuracy in the measurement of the neutron radius of @xmath0pb at the jefferson laboratory appears unlikely to translate into the required 0.1% ( or lower ) uncertainty in @xmath86 . instead , we have established an uncertainty in @xmath86 that is two to three times larger ( of the order of 0.250.35% ) . although the jefferson laboratory experiment is unlikely to achieve the desired accuracy , it is plausible that second - generation experiments may reach this goal . in the case of the weak nuclear charge of the alkali metals , where few stable or long - lived isotopes exist , one must consider theoretical uncertainties arising from both atomic and nuclear structure . although it is no longer possible to fully eliminate uncertainties in atomic structure ( as was done for @xmath86 ) alkali metals at least enjoy the simplest atomic structure . as far as the nuclear structure is concerned , the uncertainty is fully subsumed into a single factor : the neutron skin of the nucleus of interest . it has been argued that to reduce significantly the nuclear - structure uncertainties relative to radiative corrections , the neutron radius of @xmath125cs must be known with an accuracy of 2% or better . similar accuracy should be expected for the case of other alkali metals . indeed , in anticipation of the commissioning of rare - isotope accelerators all over the world and the ongoing advances in the field of production and magneto - optical trapping of radioactive atoms , we have computed the weak nuclear charge of the two longest lived francium isotopes : @xmath47fr ( 20 min ) and @xmath147fr ( 21.8 min ) . our assumed uncertainty in the neutron radii of @xmath47fr and @xmath147fr translated into a theoretical uncertainty in the value of their weak nuclear charges of 0.2% and 0.3% , respectively . in summary , we have studied the impact of a high - precision measurement of the neutron radius of @xmath0pb on atomic parity non - conservation experiments . however , the relevance of such a measurement on a plethora of other research areas such as nuclear structure , flavor violations in qcd , and neutron - star structure has been strongly emphasized here and elsewhere . thus , the neutron radius of @xmath0pb stands as one of the most fundamental nuclear - physics quantities yet to be accurately determined . this unfortunate situation should be promptly corrected . 99 m. a. bouchiat and c. bouchiat , rep . * 60 * , 1351 ( 1997 ) . j. s. m. ginges and v. v. flambaum , phys . rep . * 397 * , 63 ( 2004 ) . i. angeli , at . data nucl . data tables * 87 * , 185 ( 2004 ) . s. shlomo and e. friedman , phys . rev . lett . * 39 * , 1180 ( 1977 ) . b. a. brown , phys . lett . * 85 * , 5296 ( 2000 ) . r. j. furnstahl , nucl . phys . * a706 * , 85 ( 2002 ) . s. typel and b. a. brown , phys . rev . * c64 * , 027302 ( 2001 ) . m. centelles , m. del estal , x. vias , and s. k. patra , in _ the nuclear many - body problem _ 2001 _ _ , vol . 53 of nato advanced studies institute series b : physics , edited by w. nazarewicz and d. vretenar ( kluwer , dordrecht , 2002 ) , p. 97 . m. baldo , c. maieron , p. schuck and x. vias , nucl . a736 * , 241 ( 2004 ) . c. j. horowitz and j. piekarewicz , phys . lett . * 86 * , 5647 ( 2001 ) . c. j. horowitz and j. piekarewicz , phys * c64 * , 062802(r ) ( 2001 ) . c. j. horowitz and j. piekarewicz , phys * c66 * , 055803 ( 2002 ) . r. michaels , p. a. souder , and g. m. urciuoli , spokespersons , jefferson laboratory experiment e-00 - 003 . v. a. dzuba , v. v. flambaum , and i. b. khriplovich , z. phys . d * 1 * , 243 ( 1986 ) . s. c. bennett and c. e. wieman , phys . lett . * 82 * , 2484 ( 1999 ) ; phys . lett . * 82 * , 4153 ( 1999 ) ; phys . rev . lett . * 83 * , 889 ( 1999 ) . x. xiong , m. he , y. zhao , and z. zhang , j. phys . * b23 * , 4239 ( 1990 ) . d. demille , phys . 74 * , 4165 ( 1995 ) . d. budker , in _ proceedings of wein _ 98 _ _ , edited by p. herczeg , c. m. hoffman , and h. v. klapdor - kleingrothaus ( world scientific , singapore , 1999 ) . e. n. fortson , y. pang , and l. wilets , phys . lett . * 65 * , 2857 ( 1990 ) ; phys . * 66 * , 677(e ) ( 1991 ) . s. j. pollock , e. n. fortson , and l. wilets , phys . rev . * c46 * , 2587 ( 1992 ) . b. q. chen and p. vogel , phys . rev . * c48 * , 1392 ( 1993 ) . m. j. ramsey - musolf , phys . rev . * c60 * , 015501 ( 1999 ) . b. g. todd and j. piekarewicz , phys . rev . * c67 * , 044317 ( 2003 ) . d. vretenar , g.a . lalazissis , and p. ring , phys . rev . * c62 * , 045502 ( 2000 ) . p. k. panda and b. p. das , phys * c62 * , 065501 ( 2000 ) . g. d. sprouse and l. a. orozco , annu . nucl . part . sci . * 47 * , 429 ( 1997 ) . b. d. serot and j. d. walecka , adv . * 16 * , 1 ( 1986 ) . p. ring , prog . . phys . * 37 * , 193 ( 1996 ) . b. d. serot and j. d. walecka , int e6 * , 515 ( 1997 ) . j. boguta and a. r. bodmer , nucl . * a292 * , 413 ( 1977 ) . lalazissis , j. knig and p. ring , phys . rev . * c55 * , 540 ( 1997 ) . m. bender , k. rutz , p .- reinhard and j. a. maruhn , eur . j. * a8 * , 59 ( 2000 ) . p. ring , y. k. gambhir and g.a . lalazissis , comput . . commun . * 105 * , 77 ( 1997 ) . t. sil , b. k. agrawal , j. n. de and s. k. samaddar , phys . rev . * c63 * , 064302 ( 2001 ) , and references therein . lalazissis , d. vretenar and p. ring , nucl . a650 * , 133 ( 1999 ) . p. ring and p. schuck , _ the nuclear many - body problem _ , ( springer - verlag , berlin , 1980 ) m. centelles , m. del estal , and x. vias , nucl . a635 * , 193 ( 1998 ) . s. j. pollock and m. c. welliver , phys . lett . * b464 * , 177 ( 1999 ) . c. s. wood , s. c. bennett , d. cho , b. p. masterson , j. l. roberts , c. e. tanner , and c. e. wieman , science * 275 * , 1759 ( 1997 ) . w. j. marciano and j. l. rosner , phys . * 65 * , 2963 ( 1990 ) ; phys . * 68 * , 898(e ) ( 1992 ) . m. yu . kuchiev and v. v. flambaum , phys . * 89 * , 283002 ( 2002 ) . a. i. milstein , o. p. sushkov , and i. s. terekhov , phys . . lett . * 89 * , 283003 ( 2002 ) . c. j. horowitz , s. j. pollock , p. a. souder , and r. michaels , phys . rev . * c63 * , 025501 ( 2001 ) . v. a. dzuba , v. v. flambaum , and o. p. sushkov , phys . rev . * a51 * , 3454 ( 1995 ) . s. sanguinetti , j. guna , m. lintz , ph . jacquier , a. wasan , and m. a. bouchiat , eur . j. d * 25 * , 3 ( 2003 ) . g. d. sprouse , s. aubin , e. gomez , j. s. grossman , l. a. orozco , m. r. pearson , and m. true , eur . j. a * 13 * , 239 ( 2002 ) ; s. aubin , e. gomez , l. a. orozco , and g. d. sprouse , phys . a * 70 * , 042504 ( 2004 ) . g. audi and a. h. wapstra , nucl . phys a * 595 * , 409 ( 1995 ) . .the relative uncertainty @xmath94 in the apnc isotope ratio @xmath86 and the various components needed to evaluate it , as described in the text . the last row denotes the model spread @xmath163 , eq.([modelspread ] ) , that would be required to achieve @xmath164 . [ cols="^,^,^,^,^",options="header " , ] the dependence of the ground - state charge radius , binding energy , quadrupole deformation , and neutron skin on the isoscalar - isovector coupling @xmath12 is exemplified for the @xmath57yb nucleus . the observables have been normalized to the values of the nl3 interaction ( @xmath40 ) . ] charge radii calculated with the models considered in the text are compared with the available experimental data @xcite in isotope chains of interest for apnc studies . the actual experimental error bar for @xmath165dy and @xmath10dy has been reduced by a factor of 10 for display . ] variation of the neutron skin in four isotope chains of possible relevance to apnc isotope ratios measurements , for the two effective field theory models considered in this work . note that the change in the neutron skin of @xmath5pb predicted by these two models , otherwise accurately calibrated for binding energies and charge radii , mocks up the uncertainty that will be left after the projected precision measurement of the neutron radius of @xmath5pb at the jefferson laboratory @xcite . ] model correlation of the neutron skin of @xmath46cs , @xmath8ba , @xmath56dy , and @xmath57yb to the neutron skin of @xmath5pb . the isoscalar - isovector coupling varies between @xmath48 ( right side of the figure ) and @xmath156 ( left side ) . ] calculated ground - state properties ( neutron skin , charge radius , binding energy per particle , and quadrupole deformation ) of some francium isotopes . experimental data are shown for comparison when available @xcite . ]

we discuss the effective interactions between two localized perturbations in one - dimensional ( 1d ) quantum liquids . for non - interacting fermions , the interactions exhibit friedel oscillations , giving rise to a rkky - type interaction familiar from impurity spins in metals . in the interacting case , at low energies , a luttinger liquid description applies . in the case of repulsive fermions , the friedel oscillations of the interacting system are replaced , at long distances , by a universal casimir - type interaction which depends only on the sound velocity and decays inversely with the separation . the casimir - type interaction between localized perturbations embedded in a fermionic environment gives rise to a long range coupling between quantum dots in ultracold fermi gases , opening a novel alternative to couple qubits with neutral atoms . we also briefly discuss the case of bosonic quantum liquids in which the interaction between weak impurities turns out to be short ranged , decaying exponentially on the scale of the healing length .

cond-mat0505420

interactions between localized defects which are mediated by the continuum they are embedded in , play an important role in many areas of physics . typical examples are the rkky interaction between spins in a fermi liquid or the interaction between vortices in superfluids . in the present work , we discuss interactions between impurities in 1d quantum liquids . this study is motivated by the recent realization of strongly interacting `` atomic quantum wires '' with ultracold gases of both bosonic @xcite and fermionic atoms @xcite and the proposal @xcite , that single atoms in optical traps which are embedded in a superfluid reservoir allow to realize an atomic analog of a quantum dot with a tunable coupling to the environment . such quantum dots may be used to store qubits , which , under certain conditions , can be completely decoupled from their environment . arrays of these dots thus appear as ideal candidates for quantum information processing . it is therefore of considerable interest to study the induced interactions of such dots , mediated by the environment they are embedded in . similar questions arise also for quantum dots in solid state realizations , e.g. in carbon nanotubes @xcite , where the interaction is mediated by electrons in the intervening wire . quite generally , for both bosons and fermions , the low energy properties of a gapless 1d quantum liquid are described by the so called luttinger liquid ( ll ) phenomenology @xcite : the effective theory is a hydrodynamic energy functional characterized by the velocity of sound @xmath0 and the luttinger interaction parameter @xmath1 . in particular , for fermions , @xmath2 corresponds to the non - interacting case , while @xmath3 for repulsion . for repulsive bosons , in turn , one has @xmath4 , with @xmath5 in the limit of weak interactions , where a gross - pitaevskii or bogoliubov approximation applies . as shown by kane and fisher @xcite ( see also @xcite for a recent discussion ) , the interaction of a single impurity with a ll depends crucially on the value of @xmath1 : for @xmath4 , the impurity is irrelevant for the low energy properties . a 1d bose liquid is therefore effectively superfluid , although there is no true condensate @xcite . for @xmath3 the impurity changes the ground state of the liquid in a non - perturbative way , effectively cutting it into two disconnected parts . in this case , we will see that the induced interaction between two impurities is essentially a casimir - like effect . indeed , at low energies , two impurities at distance @xmath6 define a box with reflecting boundary conditions for the phonon modes of the quantum liquid , which leads to an attractive casimir interaction energy proportional to @xmath7 with @xmath0 the sound velocity . the case @xmath2 is marginal and corresponds to a non - interacting fermi gas in 1d or - equivalently - , a system of hard core bosons , the tonks - girardeau gas @xcite . in the following we will study the interactions mediated by the 1d quantum liquid between two impurities for the various cases , including fermions with spin . we focus our analysis on the case of _ static _ impurities , while the situation of a slow time dependence , relevant for atomic quantum dots , where the interactions depend on the internal states is only discussed qualitatively at the end of the paper . before considering the generic situation of impurities embedded in a sea of interacting particles , we first address the marginal case @xmath2 of non - interacting fermions . for simplicity we start from a gas of @xmath8 non - interacting spinless fermions in the presence of two localized impurities separated by a distance @xmath6 . considering cold gases in atomic quantum wires , the solution of this problem is not just an academic exercise . indeed , since fermions in a single hyperfine state have no s - wave interactions due to the pauli principle , they realize an ideal fermi gas at sufficiently low temperatures . we assume that the particles are contained in a periodic box of length @xmath9 with average density @xmath10 . the ( grand canonical ) partition function of the liquid at a given temperature @xmath11 may be expressed in terms of a functional integral over the grassman fields @xmath12 representing the fermions : @xmath13 the corresponding action of an ideal gas is @xmath14 , \label{eq : sfl}\ ] ] where @xmath15 is the imaginary time running from @xmath16 to @xmath17 , and @xmath18 is the chemical potential ( we use units such that @xmath19 ) . the fields are anti - periodic in imaginary time . for short range interactions , appropriate for cold atoms , the interaction of the impurities with the liquid can be described by an additional contribution @xmath20 proportional to the local density at the impurity positions @xmath21 . here , the index @xmath22 labels the impurities , while the coupling constants @xmath23 describe the strength of collisions between the atoms in the liquid and the impurities . the expression for @xmath24 is based on assuming an effective pseudopotential for the interaction between the impurity and the quantum liquid . more precisely , the interaction should be replaced by a spatial integral of the detailed impurity potential with the microscopic density operator of the liquid . in the present section dealing with non - interacting fermions , there is no need of a high energy cutoff , as one can directly work with the well behaved microscopic theory . however , in order to discuss the low energy behavior and to make contact with the following sections dealing with interacting fermions using the luttinger liquid phenomenology , we introduce a high energy cutoff @xmath25 . its value can be estimated as @xmath26 , where @xmath0 is the characteristic velocity of excitations ( @xmath27 in an ideal fermi gas ) , @xmath28 is the impurity size and @xmath29 is the chemical potential ( at zero temperature ) with @xmath30 , the fermi momentum . as will become clear from our results below , the coupling constants @xmath23 are then - up to a factor @xmath31 - identical with the dimensionless backscattering amplitudes @xmath32 for fermions at the impurities . microscopically , they are thus determined by the solution of the single particle scattering problem off a single impurity . in practice , an appreciable value of the backscattering amplitude requires the impurity size to be smaller or of the order of the interparticle spacing , since otherwise the fourier component of the potential at @xmath33 is close to zero , and hence the dimensionless coupling constants @xmath34 vanish . therefore , in the following , we will take @xmath35 . the grassman fields @xmath12 are free everywhere apart from the points @xmath36 and hence can be easily integrated out by the following standard trick : first we formally introduce four @xmath37-functions into the integrand @xmath38\nonumber \\ & \times & \delta[\bar{\psi}(x_{\alpha},\tau)-\bar{\eta}_{\alpha}(\tau ) ] \,e^{-s_\mathrm{fl}-s_{i}},\end{aligned}\ ] ] where @xmath39 are the new grassman variables describing the fermions at the location of the individual impurities . then we introduce a set of auxiliary fields @xmath40 using the identity @xmath41 to rise the @xmath37-functions into the action . finally , we integrate out the fermionic fields @xmath12 , which appear only quadratically , by fourier transformation : @xmath42 where @xmath43 and @xmath44 where @xmath45 , and the summation is over the fermionic matsubara frequencies @xmath46 . the trivial prefactor @xmath47 arises from the integration over the fermionic fields in the absence of impurities , giving the grand partition function of the homogeneous liquid . the fields @xmath40 depend only on imaginary time @xmath15 , or frequency @xmath46 in the fourier representation and thus the integral over @xmath48 can be easily calculated . since for a sufficiently large separation @xmath49 , the interaction energy is small , the relevant frequencies are small compared to the fermi energy @xmath50 . an expansion to leading order in @xmath51 then gives @xmath52 where @xmath53 and @xmath54 . the characteristic frequency @xmath55 will play an important role in our subsequent discussions . physically it represents the inverse flight time for a characteristic excitation in the liquid between the locations of the two impurities , which naturally obeys the inequality @xmath56 provided the impurities are much further apart than the average distance between two fermions in the liquid . it is also the quantization energy between the two impurities . in order to obtain the impurity interaction directly from the partition function , we integrate out the auxiliary fields @xmath40 . this results in an action @xmath57 which only depends on the four time dependent grassman fields @xmath39 which describe the fermi field at the impurity positions . the coefficients @xmath58 and @xmath59 are defined by @xmath60 ) can now be written as @xmath61 where @xmath62 comes from the integration of the auxiliary fields : @xmath63 the total effective action @xmath64 is quadratic in the grassman fields @xmath39 and thus the integration can be done exactly to yield @xmath65 , with @xmath66\ .\ ] ] here , the @xmath67 are the dimensionless backscattering amplitudes , characterizing the interaction of the impurities and the liquid . we can now obtain the free energy of the 1d fermi gas in the presence of the impurities from @xmath68 as follows @xmath69\ } , \label{f12}\end{aligned}\ ] ] where @xmath70 is the free energy of the undisturbed , homogeneous liquid . the expression ( [ f12 ] ) is ill defined as it stands , since it contains both the energy of zero - point fluctuations in the gas , as well as the formally divergent self - energies of the separate impurities . the relevant interaction energy associated with a change of the separation of the two impurities is given by : @xmath71\nonumber \\ & = & f ( \gamma_{\alpha},r)- f ( \gamma_{\alpha},\infty ) . \label{renorm}\end{aligned}\ ] ] the renormalization thus requires subtracting first the free energy of the liquid without the impurities ( @xmath72 , vacuum energy ) and then the free energy of the system when the impurities are very far apart ( @xmath73 , self - energy of the impurities ) . while both the vacuum energy and the individual self - energies are infinite in the absence of a cutoff , the renormalized interaction ( [ renorm ] ) is finite and independent of the cutoff ( see also the discussion below in section [ classical ] ) . at low temperature @xmath74 we can switch from summation to integration according to @xmath75 , so that the effective interaction energy between the impurities can be expressed as : @xmath76 the integral can be performed analytically to yield our final result for the impurity interaction at @xmath77 : @xmath78 where @xmath79 is the di - logarithmic function @xcite and @xmath80 is the real part . obviously the interaction quite generally falls off very slowly like @xmath81 with an amplitude , which is a strictly periodic function . its period @xmath82 is equal to the average inter - particle distance . this is a characteristic property of degenerate fermions , essentially reflecting the well known friedel oscillations of the density ( see below ) . trivially , the interaction vanishes , if one of the scattering amplitudes @xmath83 is zero . a simple expression for the renormalized interaction energy @xmath84 is obtained in two limiting cases . first , if the interaction of the impurities with the liquid is weak @xmath85 , we can expand the di - logarithm in eq . ( [ eq : dilog ] ) to obtain : @xmath86 in the limit of strong impurities @xmath87 , we find in turn the result @xmath88 which is completely independent of the scattering amplitudes . in this case , the interaction energy @xmath84 can be represented as @xmath89 , where @xmath90 is a periodic function bounded as follows @xmath91 where : @xmath92 a simple way of understanding the slow @xmath81-decay and the oscillations with period @xmath93 may be obtained in the weak scattering limit eq . ( [ eq : v12idealweak ] ) . indeed , the density perturbation created by a single impurity of strength @xmath94 at position @xmath95 is asymptotically given by : @xmath96 this expression for the friedel oscillations in a spinless 1d non - interacting fermi gas is valid in the limit where @xmath97 and @xmath98 @xcite . since the impurities couple to the local density , the interaction energy ( excluding self - energies ) of the system of two weak impurities is simply obtained from @xmath99 where @xmath85 . when renormalized @xmath100 , this interaction energy coincides with eq . ( [ eq : v12idealweak ] ) . alternatively , the result may be derived by using the random - phase approximation ( rpa ) as shown in the appendix [ app : rpa ] . the analysis in this section is readily generalized to the case of a fermi gas with spin . in fact for non - magnetic impurities , such as considered in the current article , the two spin modes are decoupled and therefore the energy due to the presence of the two impurities is simply multiplied by a factor of two . it should be emphasized that the calculation above , can be immediately extended to the case of noninteracting fermions in two or three dimensions @xmath101 , giving rise to an interaction energy for weak coupling of the form @xmath102 where @xmath103 are the dimensionless backscattering amplitudes of the impurities . this result follows most simply by considering the density fluctuations @xmath104 induced by a single impurity at position @xmath105 . as discussed , e.g. , in ref . @xcite they exhibit friedel oscillations proportional to the dimensionless backscattering amplitude @xmath106 at the fermi energy . the resulting interaction energy is then simply given by @xmath107 . in fact , this is a special case of a general result @xcite , that the asymptotic interaction between two impurities is determined by the product of their backscattering amplitudes . in the presence of short range repulsive interactions between the fermions , we expect that the result ( [ 23d ] ) remains qualitatively correct in the case of two and three dimensions . this is based on the existence of a fermi liquid description in @xmath101 , which guarantees that the low energy properties are qualitatively unchanged from those of a fermi gas . for example , assuming that the static density response function at @xmath108 is given by the particle - hole bubble @xcite , the renormalization factor @xmath109 in the single - particle green function will give rise to a fermi liquid correction factor @xmath110 in @xmath84 . this argument , however , neglects possible vertex corrections in the density response which may lead to an enhancement rather than a suppression of the amplitude of the friedel oscillations . in fact this effectively happens in the one - dimensional case , where the vanishing @xmath111-factor gives rise to friedel oscillations , which decay more slowly than in the noninteracting case ( see below ) . while we are not aware of a quantitative calculation of the @xmath112-density response in fermi - liquids , it is very likely that they will give rise only to finite , multiplicative corrections to eq . ( [ 23d ] ) . as we will see below , however , the situation in one dimension is quite different from that in @xmath101 in the sense , that even qualitatively the asymptotic form of the interaction is _ not _ given by the friedel oscillation picture , even for very weak impurities . finally , we mention a recent work dealing with neutron matter . _ @xcite consider a neutron star crust , which is modeled as a degenerate non - interacting neutron gas ( i.e. an ideal 3d fermi gas ) containing various kinds of defects or inhomogeneities ( such as nuclei or bubbles ) immersed in it . these authors compute the interaction energy between two defects resulting from the quantum fluctuations of the fermi sea of neutrons . they obtain expressions similar to eq . ( [ 23d ] ) , which can be interpreted as rkky - like interactions between defects . in addition , they discuss the influence of the shape of the defects and consider situations with more than two defects . realistic fermi systems consist of interacting particles . in three and also in two dimensions , it is possible to describe even strong interactions by landau s fermi liquid theory . as is well known , however , this concept fails in one dimension . here we consider fermions with repulsive short - range interaction . at low - energy such a system exhibits a gapless excitation spectrum with a linear dispersion , characteristic for the universality class of luttinger liquids ( ll ) @xcite . for simplicity , we start by considering spinless fermions , for which the low - energy description is given by the following hydrodynamic action : @xmath113 . \label{eq : llaction}\ ] ] here @xmath0 is the sound velocity and @xmath1 the luttinger parameter . in a translationally invariant system , they obey the relation @xmath114 @xcite , with @xmath115 the fermi velocity of the associated non interacting spinless fermi gas . we consider repulsive interactions for which @xmath3 . since the luttinger liquid description only applies at low energies , the fields have to be cutoff at energy @xmath35 , where @xmath18 is the chemical potential . the associated cutoff length @xmath116 is of order @xmath117 . of course , for a quantitative calculation of the scale at which the low energy description applies , a microscopic model is needed , which allows to determine nonuniversal properties . for single impurities in luttinger liquids this problem has only recently been discussed , see @xcite . since we are concerned with the interaction at distances much longer that the average interparticle separation , only the low energy properties are relevant , which are well described by the hydrodynamic action ( [ eq : llaction ] ) . the corresponding field @xmath118 is related to the density of the liquid by @xmath119 , \label{density}\ ] ] where @xmath120 is the equilibrium density and only the first harmonics are taken into account @xcite . the interaction between the impurities and the luttinger liquid is taken to be of the form @xmath121 i.e. a coupling to the local density with phenomenological scattering amplitudes @xmath122 . inserting the expansion ( [ density ] ) into this interaction , gives rise to four different terms . the first term is just the constant hartree self - energy of the impurities , which - of course - does not contribute to the renormalized interaction energy @xmath84 . in addition , there are terms containing @xmath123 due to forward scattering . they describe quantum corrections to the self energies but again are irrelevant for the interaction @xmath84 between two widely separated impurities . the dominant term for this interaction is the contribution proportional to @xmath124 , which is due to backward scattering . in addition there are higher order terms like @xmath125 , however these are less relevant in the renormalization group ( rg ) sense @xcite . taking only the most relevant part of the interaction , the coupling between the impurities and the ll leads to the following nonlinear contribution to the action @xmath126\approx \int_0^{\beta } \mathrm{d}\tau \sum_{\alpha=1,2 } 2 \tilde{g}_{\alpha } \rho_0 \cos[2\sqrt{\pi k}\theta(x_\alpha)+ 2 p_{f}x_{\alpha}]\ , \label{eq : sitheta}\ ] ] where we introduced the renormalized field @xmath127 by @xmath128 . the complete statistical sum of the system can again be represented by a functional integral . to perform the integration over the field @xmath127 we use the same approach as for the ideal fermi gas : first we introduce the new variables @xmath129 and @xmath130 and then insert the two @xmath37-functions into the functional integral : @xmath131 e^{-s_\mathrm{ll}-s_{i}}.\end{aligned}\ ] ] we then transform the @xmath37 functions into the functional integrals over auxiliary fields and perform the gaussian integration @xmath132}\label{eq : smallkz}\ , \ ] ] where the effective action for the real fields @xmath133 is @xmath134 with the summation occurring over the bosonic matsubara frequencies @xmath46 , where @xmath135 and @xmath136 with @xmath137 . the factor @xmath138 comes from the integration over @xmath139 and is independent of the impurities , describing the homogeneous luttinger liquid . by contrast , the factor @xmath140 which comes from the integration over the auxiliary fields , describes the change in the phonon modes due to the constraint on the fermi fields at the positions of the two impurities . similar to the noninteracting situation , this contribution depends on the associated characteristic frequency @xmath55 and is crucial in obtaining a finite interaction energy @xmath84 which is independent of the cutoff . the remaining and now non - trivial functional integral over the time - dependent fields @xmath141 is of the same form as the one which appears in the context of quantum brownian motion in a periodic potential @xcite . indeed the effective action ( 29 ) basically describes two quantum particles subject to ohmic dissipation of dimensionless strength @xmath142 which move in a periodic potential generated by the backscattering amplitude . as has been shown in @xcite this problem leads to a localized ground state if @xmath143 with small fluctuations in the field @xmath127 . for a quantum liquid with sufficiently strong interactions between the fermions @xmath144 and strong impurities @xmath145 , the functional integral ( [ eq : smallkz ] ) over the time - dependent fields @xmath141 can thus be calculated using the stationary phase approximation ( spa ) : expanding the functions @xmath146 from eq . ( [ eq : sitheta ] ) to second order in the fields around one of its minimum , we approximate the interaction lagrangian in the form @xmath147 where @xmath148 . physically this means that the interaction between each of the impurities and the liquid is sufficiently strong to pin the local phase @xmath127 near the value minimizing the potential energy . the quantities @xmath149 play the role of effective frequencies for the evolution of the fields @xmath141 . the approximation of the original lagrangian ( [ eq : sitheta ] ) by the quadratic form ( [ eq : statphaseint ] ) is equivalent to an adiabatic approximation which describes physical processes occurring slower than a time scale given by @xmath150 . as the typical frequency of interest is @xmath151 , the stationary phase approximation is valid when @xmath152 . it is thus applicable in the case of strong impurities or - equivalently - long distances ( see below ) in a strongly repulsive liquid : @xmath153 . within the quadratic approximation ( [ eq : statphaseint ] ) , the full effective action in eq . ( [ eq : smallkz ] ) is quadratic in @xmath141 and hence can be evaluated exactly @xmath154 , where : @xmath155^{-1/2}\ .\ ] ] the associated free energy @xmath156 is given by @xmath157\}\ , \end{aligned}\ ] ] where @xmath158 again describes the undisturbed , homogeneous liquid . from the free energy we obtain the renormalized interaction energy between the two impurities @xmath84 in precisely the same manner as in eq . ( [ renorm ] ) : @xmath159 at sufficiently low temperatures @xmath74 , the summation may be replaced by an integral over the real frequency @xmath160 : @xmath161 in the limit of long distances ( or strong impurities ) , i.e. @xmath162 , the integral converges at @xmath163 and hence @xmath164 . for sufficiently large separations , we thus obtain the simple universal interaction @xmath165 which decays inversely with distance . the short distance regime , where @xmath166 , can , however , not be considered within the stationary phase approximation ( [ eq : statphaseint ] ) . indeed , the latter is only justified if the characteristic energy scale @xmath167 of excitations involved in the interaction does not exceed the effective frequencies , i.e. if @xmath168 . as a result , for intermediate and short distances , the interaction energy @xmath84 does not follow a simple @xmath81 -behavior . in particular , it is impossible to describe the limit @xmath169 without properly including a cutoff or working in a microscopic model from the beginning . it is only within such a more complete calculation , that the full energy @xmath170 of the two impurity problem approaches the self energy of the doubled single impurity case , as expected on physical grounds . for a single scalar field in 1d , as described by our hydrodynamic action ( 23 ) , this calculation has recently been done by jaffe @xcite . the validity of the quadratic expansion ( [ eq : statphaseint ] ) may be extended to the whole relevant range @xmath171 of the luttinger parameter with the help of the self - consistent harmonic approximation ( scha ) @xcite . in the context of friedel oscillations around a single impurity in a luttinger liquid , this has been used by egger and grabert @xcite . it is based on making a quadratic approximation ( [ eq : statphaseint ] ) for the backscattering term , however with frequencies @xmath149 which are determined from feynman s variational principle . using eq . ( [ eq : statphaseint ] ) as the trial action , one has to minimize the free energy @xmath172 where @xmath173 and @xmath174 . taking @xmath149 as variational parameters , we obtain @xmath175 where @xmath176 is the high - energy cutoff . following @xcite , we define the crossover scale @xmath177 by @xmath178 when both impurities have approximatively the same strength @xmath179 . the scha is a good approximation when @xmath3 and @xmath180 , i.e. at long distances @xmath181 . in the limit of strong impurities @xmath182 , the scha frequencies @xmath183 are the same as those obtained within the stationary phase approximation and the crossover scale is given by @xmath184 . in the opposite limit @xmath185 , they are given by @xmath186 and the crossover scale is @xmath187 where @xmath188 for repulsive fermions . note the singular behavior of the crossover scale @xmath189 in the limit @xmath190 of non - interacting fermions . this implies that the regime of validity of the scha is moved out to extremely long scales @xmath177 . quite generally , therefore , the long distance behavior @xmath191 of the interaction energy is always given by the casimir like expression ( [ eq : v12bosefin ] ) whenever @xmath3 . the scale , however , beyond which this simple result applies , strongly depends on the strength of the backscattering amplitude and the repulsive interaction . in order to study the limit of weak impurities and weak interactions ( @xmath185 and @xmath3 close to 1 ) in more detail , we use perturbation theory . at second order in @xmath192 , and when @xmath193 , we find @xmath194 where @xmath195 and @xmath196 is the gamma function . the function @xmath197 diverges as @xmath198 and approaches one at @xmath199 . the above equation was obtained under the assumption that @xmath200 . when @xmath2 , it reproduces exactly eq . ( [ eq : v12idealweak ] ) obtained for the non - interacting fermi gas , provided that we choose the following relation between the microscopic and phenomenological impurity strengths @xmath201 . perturbation theory breaks down when , in order of magnitude , the interaction energy ( [ eq : genfriedel ] ) reaches the strong impurity ( or long distances ) result ( [ eq : v12bosefin ] ) : @xmath202 . this occurs for @xmath203 , in agreement with eq . ( [ crossoverscale ] ) , because @xmath200 . therefore , the perturbative result ( [ eq : genfriedel ] ) is valid for intermediate distances @xmath204 . at long distances @xmath191 it is replaced by the casimir - type result ( [ eq : v12bosefin ] ) . finally , it is worth mentioning that the presence of two impurities implies the possibility of a tunneling resonance @xcite . the physics of such a resonance is , however , not captured by the scha . this issue will be discussed in more detail in the appendix [ app : coulomb ] . in this section we generalize the above analysis by including the spin degree of freedom , again using a luttinger liquid description . we consider @xmath8 fermions in an equal mixture of spin up and spin down components , i.e. , @xmath205 . here the ( non - interacting ) fermi velocity is @xmath206 with @xmath207 . taking spin into account the euclidean action of the luttinger liquid is generalized to @xmath208=\!\sum_{\mu=\rho,\sigma } \frac{1}{2\pi k_\mu}\int \!\!\mathrm{d } x\ , \mathrm{d}\tau\ , \left [ u_\mu ( \partial_{x}\theta_\mu)^{2}+\frac{1}{u_\mu}(\partial_{\tau}\theta_\mu)^{2 } \right]\ , \ ] ] where @xmath209 $ ] is an abbreviation for @xmath210 $ ] . this corresponds to the usual `` charge '' @xmath211 and `` spin '' @xmath212 fields that are linear combinations of the spin up and down fields : @xmath213 . the luttinger liquid parameters obey the relation @xmath214 and in addition @xmath215 where the first equation comes from considering repulsive interactions and the second from the @xmath216 symmetry of the model with spin - independent interactions @xcite . in addition @xmath217 when considering contact interactions between fermions , see e.g. @xcite . the action term that describes the interaction with the impurities is still given by eq . ( [ eq : sidens ] ) where the density is now simply the sum of the densities of the two spin modes that have the same form as eq . ( [ density ] ) . as stated in the previous section we only keep the most relevant term , which corresponds to backscattering by the impurity , and hence obtain : @xmath218 = \sum_{\alpha=1,2 } 2 \tilde{g}_{\alpha } \rho_0 \int_0^\beta \!\!\mathrm{d}\tau \ , & \cos [ \sqrt{2 } \theta_\rho ( x_{\alpha } ) + 2 p_f x_{\alpha } ] \nonumber \\ & \times \cos [ \sqrt{2 } \theta_\sigma ( x_{\alpha } ) ] \ .\end{aligned}\ ] ] now we follow the same procedure as in previous sections and obtain @xmath219 - s_i[\theta_{\mu \alpha } ] } \ , \ ] ] where we have rescaled the fields at the impurity positions by @xmath220 . the effective action @xmath221 is given by @xmath222 = \sum_{n } \sum_{\mu,\alpha,\delta } i^\mu_{\alpha \delta}\ \theta_{\mu \alpha ,- n } \ , \theta_{\mu \delta , n } \ , \ ] ] where the @xmath223 refers to the matsubara frequencies @xmath224 and @xmath225 are the elements of the matrix @xmath226 with @xmath227 , @xmath228 , and @xmath229 . in order to calculate the partition function , we again use the stationary phase approximation , which corresponds to expanding @xmath230 around its minima to second order in the fields , assuming that the impurities are strong , i.e. @xmath231 . the nonlinear action @xmath230 is thus replaced by a quadratic approximation @xmath232 . since the charge and spin fields are now completely decoupled , we have @xmath233 , with : @xmath234^{-1/2 } \ .\ ] ] the total free energy @xmath235 is then simply the sum of the charge and spin contribution . after renormalization the free energy is given by : @xmath236 \ .\ ] ] at low temperature @xmath237 we can again replace the sum over matsubara frequencies by an integral . in the limit of strong impurities @xmath238 we obtain @xmath239 this is the straightforward generalization for spin @xmath240 fermions of the result obtained in the previous section , see eq . ( [ eq : v12bosefin ] ) . as discussed there , the spa is valid only if @xmath238 . hence we are unable to calculate the interaction at shorter distances , where @xmath241 . as in the case of spinless fermions , we can go beyond the spa regime , using the scha . in particular we use @xmath173 and @xmath242 in eq . ( [ fvar ] ) ; and then we minimize @xmath243 with respect to @xmath244 . in the case of identical impurities , i.e. @xmath245 , the values of @xmath246 that minimize @xmath243 are such that @xmath247 . for large distances , i.e. , for @xmath248 , the scha is valid and @xmath249 is given by @xmath250 where @xmath25 is the high energy cutoff . as in the spinless case , we can define the crossover scale as @xmath251 , since @xmath3 . in the limit of very strong impurity backscattering @xmath252 , we recover the spa result , i.e. @xmath253 , and the crossover scale is @xmath254 . for intermediate impurity strength @xmath255 , we obtain @xmath256 and the crossover scale in this case is given by @xmath257 where @xmath258 is the short distance cutoff and @xmath259 . note that , as in the spinless case , the crossover scale diverges @xmath189 when @xmath190 . for weak interactions , therefore , the result ( [ eq : casspin ] ) is only valid at very large distances . the regime of weak impurity strength can be studied using perturbation theory . at second order in @xmath192 , we find @xmath260 where @xmath261 and @xmath262 is the hypergeometric function @xcite . the function @xmath197 is a smooth , monotonically decreasing function of @xmath1 which diverges like @xmath263 as @xmath264 and approaches one at @xmath199 . when @xmath2 , the preceding result reproduces exactly eq . ( [ eq : v12idealweak ] ) ( with an extra factor of two due to the spin degeneracy ) obtained for the non - interacting ( spinless ) fermi gas , provided that we choose the following relation between the microscopic and phenomenological impurity coupling constants @xmath201 . the perturbative result ( [ spinpert ] ) is valid for @xmath265 , with @xmath177 given in eq . ( [ spincross ] ) . at the crossover scale @xmath266 and for @xmath267 , @xmath268 as in eq . ( [ eq : casspin ] ) . in conclusion , therefore , the interaction between two impurities follows the behavior given by eq . ( [ spinpert ] ) only for intermediate distances and weak interactions . by contrast , for large distances , there is a crossover to the universal casimir - type interaction eq . ( [ eq : casspin ] ) which depends only on the velocities @xmath269 and @xmath270 . as our main result , we have shown that for separations much larger than the interparticle spacing , the interaction energy of two impurities in a luttinger liquid of repulsive fermions ( @xmath3 ) is a casimir - type interaction , given by a very simple universal relation , eq . ( [ eq : v12bosefin ] ) ( resp . ( [ eq : casspin ] ) ) for spinless ( resp . spin @xmath240 ) fermions . in contrast to the result eq . ( [ eq : v12limits ] ) obtained for strong impurities in a non - interacting fermi gas it does not contain friedel oscillations and is independent of both the impurity strengths and the interaction parameter @xmath1 . the physical origin of this long range force is thus quite different from the @xmath2 case . in the non - interacting gas , the long range force comes from the polarization of the ground state . in the strongly interacting case , the friedel oscillations of the ground state density around each of the independent impurity still exist @xcite but they are not relevant for the impurity interaction at long distances . instead the result eq . ( [ eq : v12bosefin ] ) is best understood as being the casimir interaction energy of two mirrors , i.e. impenetrable impurities , in a phononic bath @xcite . this interpretation is supported by the direct calculation of the interaction energy of two mirrors in the vacuum fluctuations of a 1d scalar field which represents the density modes of the intervening quantum liquid , see , e.g. , ref . in fact , a similar result has previously been found for the force between two infinite mass beads on a string @xcite . the friedel oscillations are relevant for the interaction between two impurities only in the non - interacting case or at intermediate distances in the interacting luttinger liquid ( @xmath3 ) , see eq . ( [ eq : genfriedel ] ) and ( [ spinpert ] ) . the resulting picture is consistent with the rg calculation of kane and fisher for a single impurity @xcite : when @xmath3 and @xmath271 , the backscattering amplitude is renormalized to strong coupling in the low energy ( or long distance @xmath191 ) limit . the liquid is thus effectively cut into pieces , with the impurities acting like perfect mirrors for the acoustic modes , resulting in a casimir force between them . the scale on which the impurities flow to strong coupling depends on : i ) the initial strength of the impurities @xmath272 and ii ) the flow velocity given by @xmath273 , see below . when the impurities are strong and the liquid is strongly interacting , the impurities flow quickly to strong coupling . the associated crossover scale @xmath177 is thus of the order of the interparticle distance . by contrast , when the impurities are weak and the liquid is almost non - interacting , it takes very long distances to reach the asymptotic regime . qualitatively , the crossover scale @xmath177 for two weak impurities can already be obtained from the scaling theory for a single impurity . indeed , the perturbative flow equation of kane and fisher @xcite gives as the running impurity strength : @xmath274 for spin @xmath240 fermions , the preceding equation holds provided that the exponent @xmath273 is replaced by @xmath275 . the crossover scale @xmath177 then corresponds to the distance at which the running impurity strength @xmath276 becomes of order one , i.e. @xmath277 , in agreement with eq . ( [ crossoverscale ] ) , because @xmath188 and @xmath200 for repulsive fermions with contact interactions . for longer distances , the impurity reaches strong coupling and cuts the liquid into disconnected pieces . in this section , we discuss the case of two identical impurities @xmath278 in a one - dimensional bose liquid . in particular , we consider 1d bosons with short range repulsive interactions @xmath279 . as was first shown by lieb and liniger @xcite , the dimensionless interaction parameter @xmath280 in this problem is inversely proportional to the density . the strong coupling , tonks - girardeau , limit @xmath281 is thus reached either for strong repulsion or at low densities . within a low - energy effective luttinger liquid description , the luttinger parameter @xmath1 for interacting bosons is larger than @xmath282 . it is related to the sound velocity @xmath0 by @xmath283 . in the weakly interacting , gross - pitaevskii limit @xmath284 , the luttinger parameter diverges like @xmath285 . for strong coupling @xmath286 in turn , one finds @xmath287 . the singular case of non - interacting bosons ( @xmath288 ) is discussed separately in appendix [ app : nibosons ] . before starting the explicit calculation of the renormalized impurity interaction energy @xmath289 in the case of bosons , we discuss a limitation of our quantum hydrodynamic approach when treating the bose case . we will shortly see , why this limitation was not discussed in the context of fermions . in quantum hydrodynamics , the ground state energy @xmath290 is obtained as the sum of two terms : the classical ground state energy @xmath291 and the quantum vacuum fluctuations @xmath292 , see e.g. ref . the classical ground state energy @xmath293 $ ] is a functional of the density profile @xmath294 . now , the presence of impurities or boundaries in a quantum liquid modifies the density profile @xmath294 over distances of the order of the healing length @xmath295 . this in turns modifies the classical ground state energy in addition to the change in the quantum vacuum fluctuations , which are responsible for the casimir - type interaction energy . in the ll approach , the classical ground state energy is usually neglected and only the fluctuations _ above _ the classical ground state are taken into account . in a homogeneous system , the classical ground state energy is just a constant in the hamiltonian and can therefore be safely ignored . provided that the healing length is much smaller than the system size , the effect of the boundaries disappears for bulk properties . in fermionic quantum liquids , the healing length is of the order of the interparticle distance @xmath117 . therefore , introducing two impurities with a separation @xmath296 leaves the classical ground state energy unchanged . however , in the case of bosons , the healing length is much larger than the interparticle spacing , diverging like @xmath297 for weak interactions @xmath298 . the condition @xmath299 for neglecting the contribution of the classical ground state energy in the calculation of the renormalized interaction energy between impurities @xmath84 thus becomes increasingly restrictive for small @xmath300 . as we will see , this leaves us only with an exponentially small interaction , quite in contrast to the case of fermions . in the case of a weakly interacting bose gas @xmath301 ( corresponding to _ high _ densities ) the bogoliubov approach is quantitatively applicable , as was shown long ago by lieb and liniger @xcite . the physical reason for that is , that the healing length @xmath302 in this limit is much larger than the average interparticle spacing . the situation is therefore essentially equivalent to the weak coupling limit of a 3d bose - einstein condensate at _ low _ densities , where interactions between two impurities have recently been discussed by klein and fleischhauer @xcite ( the interaction energy @xmath84 is called the conditional energy shift and denoted by @xmath303 in this work ) . the explicit calculation is based on the bogoliubov approach and assumes that the dimensionless impurity strength @xmath304 is much smaller than one . adapting the calculation of ref . @xcite to a one - dimensional gas , we obtain @xmath305 to lowest order in @xmath306 . the interaction energy between widely separated impurities in a weakly interacting bose gas thus vanishes exponentially on the scale set by the healing length @xmath307 . as discussed above , this calculation assumes that the impurities modify the density profile only locally , which is also the physical reason for the vanishing interaction at @xmath299 . the preceding result can again be qualitatively understood with the help of the scaling theory of kane and fisher @xcite for a single weak impurity in a ll . indeed , when @xmath4 , the effective coupling of a single impurity is renormalized to zero in the low - energy limit . the rg flow thus starts at high - energy @xmath308 ( corresponding to a short distance cutoff @xmath309 ) and ends at a much lower energy @xmath310 which is set by the separation @xmath6 of the two impurities . similar to the case of fermions , we define a crossover scale @xmath177 by the condition that the effective dimensionless impurity strength at this scale is of order one : @xmath311 since @xmath312 for weak impurities and @xmath313 due to the weak interaction condition , we find @xmath314 . this confirms that there is no interaction between two weak impurities embedded in a weakly interacting bose gas when they are further apart than a distance of the order of the healing length . more generally , the scaling theory indicates that there is no long range interaction between weak impurities even in a strongly interacting bose gas , in which @xmath315 . of course the limiting case @xmath2 of hard - core bosons is special as is the case @xmath316 of no interaction at all . the latter case is treated in appendix [ app : nibosons ] and shows , that there is no interaction between the impurities whatever the impurity strength . in strong contrast to that , the limit of a tonks - girardeau gas of hard - core bosons is equivalent to the case of non - interacting fermions for properties depending only on the modulus of the ground state wavefunction like the density distribution . on the basis of the calculations in section [ nonint ] , one thus expects a long - range interaction of the form ( [ eq : dilog ] ) between impurities in a tonks - girardeau gas which exhibits friedel - like oscillations . in view of the fact that the momentum distribution of hard - core bosons is quite different from that of a free fermi gas , showing no jump at @xmath317 , this is a quite remarkable result @xcite . another singular limit , where long - range interactions appear in a bose liquid is that of impenetrable impurities @xmath318 . for arbitrary values @xmath319 of the interactions , the interacting bose liquid is then cut in three disconnected pieces . the impurities thus act as perfect mirrors for the low energy phonon excitations of the intervening bose liquid , giving rise to a casimir interaction energy precisely as in eq . ( [ eq : v12bosefin ] ) for spinless fermions . it appears , however , that the limit of impenetrable impurities at arbitrary energies is nonphysical , imposing strict dirichlet boundary conditions on a scalar field @xcite . the casimir force is thus expected to be restricted to @xmath320 , while for any finite @xmath306 only short range interactions should survive . describing these crossovers in detail , clearly requires a quantitative theory of impurity interactions in 1d bose liquids at arbitrary values of @xmath1 and @xmath306 , an interesting problem for further study . the recent realization of one - dimensional ultracold fermi gases in a strong 2d optical lattice @xcite provides a novel opportunity to study luttinger liquid effects in a setup with cold gases , e.g. spin - charge separation @xcite . in order to study whether the casimir interactions discussed here might be observed in these systems , we consider an atomic gas of fermions in two hyperfine states . these two internal states play the role of ( iso)spin 1/2 states . in principle both the sign and the strength of the interaction can be controlled using scattering resonances , e.g. a confinement induced resonance as shown by olshanii @xcite . for simplicity , we assume that the two spin states are equally populated @xmath205 . we note that in order to have @xmath3 , one needs to consider a two - component atomic fermi gas . indeed , in a single component fermi gas , pauli s principle forbids @xmath321-wave collisions implying that @xmath2 for spinless fermions interacting via a contact potential . following several recent ideas @xcite which involve trapping single atoms in ultracold gases , we consider an atomic quantum dot ( aqd ) like configuration , which consists of single atoms confined in a tight trap created either magnetically or optically , e.g. by an additional optical lattice . we assume that the confining potential can be adjusted in such a way , that it does not affect the atoms of the bath . the impurity atom , which is trapped in a certain internal state @xmath322 interacts with the atoms of the bath through @xmath321-wave collisions . in the case where two such aqds are embedded in the bath and both impurity atoms are in state @xmath322 , the system precisely realizes the situation of two localized impurities interacting via a 1d quantum liquid . provided the liquid consists of spin-1/2 repulsive fermions , we expect that for distance @xmath6 much larger than the average interparticle spacing , the interaction is of the casimir form given in eq . ( [ eq : casspin ] ) . in principle , using a scattering resonance may allow to reach the strong impurity regime @xmath323 where the crossover scale @xmath177 is even smaller than the inter - particle distance @xmath117 . . here @xmath37 is the renormalized detuning and @xmath324 is the rabi frequency coming from a laser induced coupling , see section [ conclusion].,height=226 ] a possible way to detect the interaction energy @xmath289 , is to do spectroscopy of a single trapped atom as a function of the distance @xmath6 to a neighboring trapped atom . in addition to the mean field line shifts modifying the internal levels of the impurity atom , the casimir interaction produces a line shift depending on distance as @xmath81 . for a quantitative estimate of this effect , we compute the energy ( [ eq : casspin ] ) for the experimental situation realized in ref . there , about @xmath325 @xmath326k atoms ( per tube ) form an atomic wire of length @xmath327 @xmath18 m . the temperature can be as low as @xmath328 nk , which is about one tenth of the fermi temperature . the fermi velocity is of order @xmath329 m / s and we take @xmath330 . as the tube length is of the order of 10 @xmath18 m and the inter - particle distance @xmath331 @xmath18 m , we assume the inter - impurity distance to be @xmath332 @xmath18 m , which is larger than the crossover length for strong impurities . this gives a casimir - related line shift of the order of @xmath282 khz , which is in an experimentally accessible range . with the parameters given above , the characteristic frequency @xmath151 is of the order of @xmath333 , which is not much larger than @xmath11 as required for the validity of the zero temperature limit in which the casimir force is obtained . in conclusion , we studied the long - range interaction between two impurities mediated by the 1d quantum liquid in which they are embedded . we found that for repulsive fermions , the impurities interact via a rkky - like interaction at intermediate distances and via a casimir - like force at large distances . the crossover scale separating these two regimes depends on the strength of the impurities and on the interactions between fermions . we proposed an experimental realization of such a system with atomic quantum dots in an ultra - cold atomic gas and suggested a way to detect the casimir - type interaction by spectroscopy of a single atom in an aqd . an issue which is still open is to understand the interaction between impurities in a strongly interacting bose liquid . in particular , how does the short - range interaction ( on the scale of the healing length ) turn into a long - range interactions featuring friedel oscillations in the tonks - girardeau limit ? another issue is to assess the validity of the self - consistent harmonic approximation ( scha ) used to discuss spin @xmath240 fermions in section [ spinful ] . indeed , it is not obvious that the variational ansatz , which assumes decoupling of the charge and spin modes in presence of the impurities , is a valid starting point . in this paper , we studied static impurities . the situation becomes even more interesting if one has dynamic impurities , as in an aqd . first , we discuss the possibility of _ internal _ dynamics for the aqd . we have seen that the characteristic frequencies of vacuum modes ( excitations ) responsible for the long - range interactions between aqds are limited by @xmath55 . this means that the effective interaction potential between static impurities can be used ( in an adiabatic approximation ) for time - dependent impurity strengths @xmath334 provided that the interaction properties change slowly compared with the time scale @xmath335 . consider , for instance , the configuration with two aqds described previously , see figure [ scheme ] ( a similar scheme for two impurities in a 3d bose - einstein condensate is discussed in ref . the two - level impurity atoms can be described as isospins @xmath240 or qubits . a laser can drive transitions ( equivalent to single - qubit gate ) between the two internal levels @xmath322 and @xmath336 . in the adiabatic approximation , the aqd variables are slow and can be taken out of the integrals so that our previous treatment to calculate the interaction between two impurities applies . thus one can easily write an effective hamiltonian for the 2 aqds in the form @xmath337 where @xmath37 is the ( renormalized ) detuning , @xmath324 is the ( effective ) rabi frequency coming from the laser induced coupling @xcite and @xmath338 , @xmath339 , @xmath340 are the pauli matrices describing the isospin @xmath240 of each aqd @xmath22 . the long - range potential @xmath84 depends , as we have seen , strongly on the characteristics of the bath and on the distance between the aqds . in addition , the case of impurities with internal dynamics embedded in a spin @xmath240 fermionic liquid is of course also relevant for discussing the rkky interaction in luttinger liquids , as discussed perturbatively in @xcite . it is also possible to imagine _ external _ motion or dynamics for the impurities . the argument of adiabaticity also holds in the case where the distance between the impurities changes sufficiently slowly . this means that the casimir - like interaction could be used , for example , to create long range attractive forces in mixtures of 1d fermionic gases . however , when the external dynamics of the impurities are taken into account on the same footing as the bath dynamics other effects could reduce , if not wash out , the casimir force @xcite . this work was started in collaboration with piotr fedichev , whose substantial contributions to the ideas presented here are gratefully acknowledged . it is a pleasure also to thank prof . alan luther for helpful remarks on his unpublished work and mauro antezza , iacopo carusotto , peter kopietz , walter metzner , ins safi and peter zoller for useful discussions . laboratoire de physique des solides is a mixed research unit ( umr 8502 ) of the cnrs and the universit paris - sud @xmath341 in orsay . has been also supported by the european commission through contract ist-2001 - 38863 ( acqp ) . in this appendix , we give another derivation of the weak coupling result ( [ eq : v12idealweak ] ) using the random phase approximation ( rpa ) . the interaction between the two impurities may be written as the sum of two contributions : a direct interaction ( which is zero in the present case due to the short range nature of the impurities ) , and an indirect interaction induced via the polarization of the medium . the polarization operator in the fourier representation @xmath342 has a singularity at @xmath343 , which leads to the appearance of the long range force . specifically , we calculate the interaction energy of the two impurities using the rpa , which works well whenever only states with very few particle - hole pairs are excited by the perturbation , i.e. in the limit @xmath344 . for an ideal ( spinless ) fermi gas the rpa lagrangian can be represented as ( @xmath77 and here we use the real time formalism from the start ) : @xmath345 where @xmath346 is the rpa expression for the fermion density , @xmath347 is the energy of the electron - hole pair , and the summation over the momentum @xmath48 is limited by the conditions : @xmath348 and @xmath349 . the interaction of the aqds with the liquid can be written as @xmath350 where @xmath351 . the total rpa action @xmath352 is quadratic , and hence the fields @xmath353 can be integrated out , so that the effective interaction between the impurities is given by @xmath354 where the polarization operator is : @xmath355 substituting @xmath356 , performing the integration by transforming the integral along the real @xmath357 axis into the integrals along the branch cut corresponding to the singularity of the @xmath358-function , and removing the self energies of the separated impurities ( renormalizing the interaction ) , we find @xmath359 in agreement with eq . ( [ eq : v12idealweak ] ) . the present paper discusses the interaction energy between two impurities in an atomic quantum wire ( i.e. , a 1d quantum liquid made of atoms ) . in this appendix , we wish to make contact with related subjects in the field of mesoscopic conductors ( or solid - state quantum wires ) : namely coulomb blockade and resonant tunneling of electrons in a quantum wire with two tunnel junctions or barriers ( see , e.g. , @xcite ) . for simplicity , we only consider the case of spinless fermions . we first consider an atomic quantum wire with two impurities . when the impurities are strong ( or very far apart ) , the atomic quantum wire is cut in three disconnected pieces and we may picture the low energy behavior of the system as the following structure : left wire / impurity / island(or central wire)/impurity / right wire . the casimir effect occurring in this system is directly related to the energy cost of transferring a supplementary particle from one of the wires to the island . in our case ( neutral atoms interacting via a short - range potential ) , this energy cost is a finite size energy @xcite equal to : @xmath360 with the help of the approximate relation @xmath361 @xcite , this energy cost can be seen as the sum of two contributions : a kinetic energy cost @xmath362 and an interaction energy cost , resulting from the local interaction with the other particles on the island , @xmath363 , where @xmath364 is the forward scattering coupling constant ( in the standard notation of the g - ology , @xmath365 , see @xcite , e.g. ) . note that the finite size energy is of the order of the zero - point kinetic energy of a phonon on the island @xmath366 . in the case of electrons in a solid - state quantum wire with two tunnel junctions ( corresponding to the structure : left electrode / barrier / island / barrier / right electrode ) , the coulomb blockade is due to the energy cost of transferring a single electron from an electrode to the island @xcite . however , this energy cost is not only due to the finite size energy @xmath367 ( with @xmath368 , where @xmath369 is the electron charge and @xmath370 is an appropriate dielectric constant ) but also gets an additional contribution from the charging energy @xmath371 of the capacitors ( i.e. the two tunnel junctions ) , where @xmath372 is the sum of the capacitance of each tunnel junction @xcite . the total energy cost is equal to @xmath373 . the charging energy is due to the long - range part of the interaction between electrons and therefore does not arise in the case of cold atoms interacting via a short - range potential . another subject of comparison between the atomic quantum wire and the solid - state quantum wire is the possibility of tunneling resonances across the double barrier structure @xcite ( see @xcite for review ) . indeed , in an atomic quantum wire with two impurities , when @xmath374 @xcite a tunneling resonance occur : the energy cost to add a particle on the island vanishes and particles can therefore tunnel trough the impurities . the liquid is no more cut into disconnected pieces and we do not expect a casimir effect to occur . tunneling resonances do not appear in our calculations because they are not captured by the self - consistent harmonic approximation we used , as discussed in ref . @xcite for example . however such resonances are infinitely sharp at zero temperature @xcite and therefore they do not play a major role and should be easy to avoid experimentally . the ideal bose gas ( @xmath288 ) is a singular case : the healing length @xmath307 diverges and boundaries are therefore felt over macroscopic distances . here using quantum hydrodynamics makes no sense , but of course one can exactly solve the problem from first principles ( schrdinger equation ) . at zero temperature , all bosons are in the same single - particle wavefunction @xmath375 ( the corresponding many - body wavefunction is just a product or hartree state ) , which is the ground state of the single - particle schrdinger equation @xmath376\psi_0(x)= \epsilon_0\psi_0(x),\ ] ] where @xmath377 is the impurity coupling constant . we assume that the particles are on a ring of length @xmath9 . the ground state energy @xmath290 for @xmath8 bosons is given by @xmath378 . for @xmath379 , it is bounded as follows @xmath380 which is a direct consequence of the schrdinger equation . in the following we will show that both bounds are going to zero in the thermodynamic limit , implying that @xmath381 for all @xmath382 . therefore , the interaction energy @xmath383 vanishes whatever the distance between the impurities . on the one hand , when the impurity strength is zero @xmath384 , the ground state of the single - particle schrdinger equation with periodic boundary conditions is just the constant wavefunction , which has zero energy @xmath385 . on the other hand , when @xmath386 , the wavefunction has to vanish on the location of the impurities , implying some bending of the wavefunction and a corresponding cost in kinetic energy . the ground state wavefunction is @xmath387 where @xmath388 is the maximum value of the wavefunction and the energy is : @xmath389 this quantity vanishes in the thermodynamic limit ( @xmath390 at fixed density @xmath391 ) such that @xmath392 . therefore @xmath393 in the thermodynamic limit , for all @xmath6 such that @xmath394 . of course , this conclusion does not hold for an ideal fermi gas , because fermions have to occupy different single - particle states ( following the pauli principle ) and therefore the average energy per particle does not vanish in the thermodynamic limit . in conclusion , there is no interaction energy between two impurities in an ideal bose gas , provided that the distance @xmath6 is much smaller than the ring size @xmath9 , which is always satisfied in the thermodynamic limit . m. abramowitz , and i. a. stegun eds . , _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ ( dover , new york , 1972 ) , 27.7 ; see also http://mathworld.wolfram.com/dilogarithm.html r. a. ferrell , and a. luther , unpublished ( 1967 ) . the argument is based on calculating the energy - momentum tensor half way between the impurities , where it is determined by the asymptotic form of the scattering states . we are grateful to professor a. luther for a discussion on this point . this is true when the impurities are weak and a perturbative rg applies , but is hard to justify in the opposite limit . we also neglect higher harmonics , which again relies on the perturbative approach based on a bosonized theory , as is used here , see @xcite . r. egger , and h. grabert , phys . lett , * 75 * , 3505 ( 1995 ) ; in _ quantum transport in semiconductor submicron structures , nato advanced study institute , series e : applied sciences , vol . 326 _ , edited by b. kramer ( kluwer , dordrecht , 1996 ) . m. abramowitz , and i. a. stegun eds . , _ handbook of mathematical functions with formulas , graphs , and mathematical tables _ ( dover , new york , 1972 ) , 15 ; see also http://mathworld.wolfram.com/hypergeometricfunction.html such a finite size energy was first discussed by haldane @xcite in a slightly different context . he noticed that for a finite size luttinger liquid , phonons are not the only type of low - energy excitations , but that there are also two so - called `` zero modes '' . one of the zero modes corresponds to adding a particle to the system without adding current . for a system of size @xmath6 , the corresponding energy cost is @xmath395 ( where @xmath396 is a characteristic velocity depending on the compressibility of the system , which was introduced in ref . this is precisely the energy cost of transferring a particle from one of the wires to the island , see eq . ( [ finitesizeenergy ] ) .

cluster automorphisms have been shown to have links to the mapping class groups of surfaces , maximal green sequences and to exchange graph automorphisms for skew - symmetric cluster algebras . in this paper we generalise these results to the skew - symmetrizable case by introducing a marking on the exchange graph . many skew - symmetrizable matrices unfold to skew - symmetric matrices and we consider how cluster automorphisms behave under this unfolding with applications to coverings of orbifolds by surfaces . * keywords : * cluster algebra ; quiver mutation ; cluster automorphism ; exchange graph ; mapping class group .

1602.05494

cluster algebras were introduced by fomin and zelevinsky in @xcite , and have since found applications across many types of mathematics . these are commutative subalgebras of @xmath0 generated by rational functions constructed using a certain combinatorial procedure starting from an initial seed which produces that seed s mutation class . in the same paper fomin and zelevinsky defined the exchange graph of a cluster algebra to better visualise the combinatorics of the mutation class . these graphs proved a useful tool in their classification of finite - type cluster algebras in @xcite where these algebras were shown to correspond to dynkin diagrams . cluster algebras were shown to be closely related to triangulations of surfaces by fomin , shapiro and thurston in @xcite , where a quiver is constructed from a given triangulation and quiver mutations correspond to flipping an edge in the triangulation . these quivers from surfaces play an important role in the classification of mutation - finite quivers , given by felikson , shapiro and tumarkin in @xcite , as all such quivers are mutation - finite and there are only 11 other exceptional mutation classes . in a similar fashion triangulations of orbifolds with orbifold points of order 2 were shown to correspond to mutation - finite diagrams by felikson , shapiro and tumarkin in @xcite . unfoldings of diagrams , introduced by felikson , shapiro and tumarkin in @xcite , then correspond to coverings of the orbifold by a triangulated surface , as shown in ( * ? ? ? * section 12 ) . cluster automorphisms were introduced by assem , schiffler and shramchenko in @xcite , for cluster algebras generated from quivers , as automorphisms of the cluster algebra taking clusters to clusters and acting as either the identity or the opposite function on quivers . these ideas were extended to cluster algebras generated from certain skew - symmetrizable matrices by chang and zhu in @xcite . the group of cluster automorphisms of a cluster algebra arising from the triangulation of a surface was shown to be isomorphic to the mapping class group of this surface by brstle and qiu in @xcite . in their paper on labelled seeds and global mutations @xcite , king and pressland showed that cluster automorphisms arise naturally when mutation classes are considered as orbits of labelled seeds under the action of a global mutation group @xmath1 . the group of cluster automorphisms is a subgroup of the automorphisms of these mutation classes , @xmath2 , which commute with this group action , and in fact for mutation - finite quivers these groups are isomorphic . we use the links between automorphisms of the exchange graph and the labelled exchange graph to prove that this group @xmath2 is isomorphic to the group of exchange graph automorphisms : for a labelled mutation class @xmath3 with mutation class @xmath4 and exchange graph @xmath5 @xmath6 therefore for mutation - finite quivers , such as those from triangulations of a surface , exchange graph automorphisms are cluster automorphisms . for a cluster algebra @xmath7 constructed from a mutation - finite quiver with exchange graph @xmath8 @xmath9 this result was proved in a different way by chang and zhu in @xcite who also proved an extension of this to skew - symmetrizable matrices of type @xmath10 and @xmath11 for @xmath12 . however for other skew - symmetrizable matrices it is not true that exchange graph automorphisms are cluster automorphisms . it can be shown that the group of cluster automorphisms is isomorphic to a subgroup of the group of exchange graph automorphisms but in general there exist graph automorphisms which do not correspond to cluster automorphisms . in order to generalise these results we introduce a marking on the exchange graph in such a way that any automorphism which fixes these markings does in fact correspond to a cluster automorphism . let @xmath13 be a seed where @xmath14 is a mutation - finite skew - symmetrizable matrix with cluster algebra @xmath7 and marked exchange graph @xmath15 then @xmath16 therefore the cluster automorphisms of any cluster algebra generated by mutation - finite skew - symmetrizable matrices can be studied using just the combinatorial properties of its marked exchange graph . a skew - symmetrizable matrix associated to a good orbifold with order 2 orbifold points can be unfolded to a skew - symmetric matrix associated to a surface which covers the orbifold . in this case we show that automorphisms of the marked exchange graph induce automorphisms of the unfolded exchange graph . given a skew - symmetrizable matrix @xmath14 which unfolds to a matrix @xmath17 , with corresponding marked exchange graphs @xmath18 and @xmath19 , @xmath20 we finish the paper with a conjecture generalising a result of brstle and qiu linking the tagged mapping class group of a surface with the cluster automorphisms of the corresponding surface cluster algebra . for a cluster algebra @xmath7 arising from the triangulation of an orbifold @xmath21 @xmath22 the structure of the paper is as follows : section [ sec : mut ] gives basic definitions of cluster algebras and mutations while section [ sec : exch ] looks at the exchange graph of a cluster algebra and includes proofs linking graph automorphisms and mutation class automorphisms . section [ sec : clauts ] recalls the definition of cluster automorphisms and various known results linking these to mutation class automorphisms and exchange graph automorphisms . the section ends by explaining how a maximal green sequence of an acyclic quiver can be used to construct a cluster automorphism . in section [ sec : exchauts ] we introduce the marked exchange graph which enables us to extend these results to cluster algebras from skew - symmetrizable matrices . we show that graph automorphisms fixing the marking are in one - to - one correspondence with cluster automorphisms . in section [ sec : unf ] we consider unfoldings of skew - symmetrizable matrices and show how the cluster automorphisms of a skew - symmetrizable cluster algebra induce cluster automorphisms of its unfolded cluster algebra . section [ sec : mcg ] looks at these ideas when the skew - symmetrizable cluster algebra is constructed from an orbifold and its unfolding gives a surface cluster algebra . a _ skew - symmetric _ matrix is a matrix @xmath23 such that @xmath24 . a _ skew - symmetrizable _ matrix is a matrix @xmath14 such that there exists some diagonal integer matrix @xmath25 with positive diagonal entries for which @xmath26 is a skew - symmetric matrix . such a matrix with the smallest entries is called the _ symmetrizing matrix _ of @xmath14 . a _ quiver _ is an oriented graph possibly with multiple arrows between two vertices and in this paper we always assume that it is restricted to having no loops or 2-cycles . if @xmath17 is a quiver , then its _ opposite _ @xmath27 is the quiver constructed by reversing the direction of all arrows in @xmath17 . the restrictions on the definition of a quiver ensure that quivers are in one - to - one correspondence with skew - symmetric matrices . a given skew - symmetric matrix @xmath28 defines a quiver with @xmath29 vertices and @xmath30 arrows from the @xmath31-th vertex to the @xmath32-th vertex if @xmath33 . a _ diagram _ is a weighted oriented graph which does not have multiple arrows between any two vertices , in addition to having no loops or 2-cycles , where the weights on the edges are positive integers . similarly to quivers , if @xmath34 is a diagram then its _ opposite _ @xmath35 is constructed by reversing all arrows in @xmath34 . unlike with quivers , there is no one - to - one correspondence between diagrams and skew - symmetrizable matrices . given a skew - symmetrizable matrix @xmath36 we can construct a diagram with @xmath29 vertices and an arrow from the @xmath31-th vertex to the @xmath32-th vertex with weight @xmath37 if @xmath33 . usually weights of @xmath38 are omitted and just shown as an unweighted arrow . however a diagram only corresponds to a matrix if the product of weights along any chordless cycle is a perfect square and in this case may correspond to multiple matrices . throughout this paper we assume that all quivers and diagrams are connected . the results can be easily extended to disconnected diagrams , however care must be taken as different connected components could have their arrows reversed while other components do not , so the idea of an opposite diagram is less clear . let @xmath39 . a _ cluster _ is a set of algebraically independent elements of @xmath40 , while a _ labelled cluster _ is a cluster with some ordering of its elements . the individual elements in a cluster are called _ cluster variables_. a _ labelled seed _ is a pair @xmath13 where @xmath14 is a skew - symmetrizable matrix and @xmath41 is a labelled cluster . each cluster variable in the cluster can be thought of as being attached to one of the matrix rows , or equivalently attached to one of the vertices of the corresponding quiver or diagram . seed _ is a class of labelled seeds which differ only by permutations . throughout this paper we assume that the matrix in a seed is uniquely determined by its cluster . this has been proved for all cluster algebras of geometric type or generated from a non - degenerate matrix by gekhtman , shapiro and vainshtein in @xcite . in this case denote the matrix for a given cluster @xmath41 by @xmath42 . given a labelled seed @xmath43 , where @xmath44 and @xmath45 , then the _ mutation _ @xmath46 acts on @xmath47 to give @xmath48 where @xmath49 and @xmath50 given by @xmath51 @xmath52 it is sometimes convenient to consider the local mutation @xmath53 of a seed @xmath13 corresponding to the mutation at the vertex associated to the cluster variable @xmath54 . these local mutations act as functions on seeds , whereas global mutations act on labelled seeds . permutations act on a labelled seed @xmath55 , @xmath56 , @xmath45 in the expected way taking the @xmath31-th vertex to the @xmath57-th vertex and the @xmath31-th cluster variable to the @xmath57-th cluster variable . therefore @xmath58 where @xmath59 and @xmath60 , @xmath61 . given a 3 vertex seed @xmath62 as in figure [ fig : permex ] and permutation @xmath63 then @xmath64 maps the first vertex and cluster variable to the third , second to first and third to second . therefore @xmath65 , @xmath66 and @xmath67 . similarly @xmath68 and @xmath69 . @xmath70 the _ global mutation group _ for seeds of rank @xmath29 is given by @xmath71 where the @xmath72 are mutations and @xmath73 for @xmath74 . the _ labelled mutation class _ @xmath3 of a labelled seed @xmath62 is the orbit of @xmath13 under the action of @xmath1 . the quotient by the symmetric group action gives the _ mutation class _ @xmath75two seeds in the same mutation class are said to be _ mutation - equivalent_. the _ cluster algebra _ @xmath76 is the subalgebra of @xmath40 generated by all cluster variables occurring in the seeds in @xmath77 . a cluster algebra is said to be of _ finite type _ if there are a finite number of generating cluster variables in the mutation class , otherwise it is of _ infinite type_. if there are a finite number of distinct matrices in the seeds of @xmath77 , then the cluster algebra and all the matrices are said to be _ mutation - finite _ or of _ finite mutation type _ , otherwise it is _ mutation - infinite _ or of _ infinite mutation type_. the mutation class automorphism group @xmath78 is the group of bijections @xmath79 which commute with the action of @xmath1 , so for all @xmath80 , @xmath81 and @xmath82 @xmath83 fomin and zelevinsky in @xcite developed the idea of the exchange graph of a cluster algebra to better visualise the relations in a mutation class . these were also an important tool in their classification of finite type cluster algebras in @xcite . the _ exchange graph _ @xmath5 of a mutation class @xmath77 is constructed with vertices for each seed in @xmath77 and an edge between two seeds @xmath47 and @xmath84 if and only if there is a single local mutation @xmath85 such that @xmath86 . the _ labelled exchange graph _ @xmath87 of a labelled mutation class @xmath3 is constructed with a vertex for each labelled seed in @xmath3 and an edge labelled @xmath31 between two labelled seeds @xmath47 and @xmath84 if and only if @xmath88 ( and conversely @xmath89 ) . . ] . ] the exchange graph for the cluster algebra of type @xmath90 is the well known pentagon , as seen in figure [ fig : a2-exch ] . the labelled exchange graph is a decagon shown in figure [ fig : a2-lexch ] , with the permutation acting by taking a seed to its antipodal seed . . ] . ] the exchange graph for a cluster algebra of type @xmath91 is a hexagon , as shown in figure [ fig : b2-exch ] . the labelled exchange graph however is the disjoint union of two hexagons as shown in figure [ fig : b2-lexch ] . the permutation interchanging the cluster variables in a labelled seed gives another labelled seed which can not be obtained from the first though just mutations , so any labelled seed has a permuted counterpart in the other connected component . the _ exchange graph automorphism group _ @xmath92 is the group of permutations @xmath64 of the vertex set of the exchange graph such that there is an edge between two vertices @xmath47 and @xmath84 if and only if there is an edge between @xmath93 and @xmath94 . the _ labelled exchange graph automorphisms _ in @xmath95 must also preserve the labelling of the edges . [ thm : pullback ] for a labelled mutation class @xmath3 with quotient @xmath77 and corresponding exchange graphs @xmath87 and @xmath5 , then @xmath96 to show this we construct a unique @xmath97 for each @xmath98 . let @xmath99 denote the cluster of a seed @xmath84 . choose a seed @xmath47 in @xmath87 , then for each @xmath100 there is a vertex @xmath101 with a corresponding edge @xmath102 labelled @xmath31 in the labelled exchange graph . the cluster @xmath103 then differs from the cluster @xmath104 in just the @xmath31-th cluster variable . under the quotient by the symmetric group action the labelled seed @xmath47 gets mapped to a seed @xmath105 $ ] and in the exchange graph @xmath5 there are edges @xmath105 - [ v^i]$ ] for each @xmath106 . each unordered cluster @xmath107 $ ] differs from the unordered cluster @xmath108 $ ] in a single variable , just as the corresponding labelled clusters do . the exchange graph automorphism @xmath109 maps @xmath105 $ ] to some seed @xmath110 $ ] and preserves all edges in the graph , so @xmath110 $ ] is connected to @xmath111 $ ] for each @xmath31 . therefore each @xmath111 $ ] is a single mutation from @xmath110 $ ] , and so the unordered cluster @xmath112 ) = [ \gamma_1,\dotsc,\gamma_{k_i},\dotsc,\gamma_n]$ ] differs from @xmath113 ) = [ \gamma_1,\dotsc,\gamma'_{k_i},\dotsc,\gamma_n]$ ] in a single cluster variable . set the image @xmath114 to be the seed defined by the labelled cluster @xmath115 , obtained by choosing an order of the cluster @xmath112)$ ] such that the @xmath31-th variable of @xmath116 is the corresponding @xmath117 , while all other variables are the same as for @xmath114 . this ensures that the edge between @xmath114 and @xmath118 is labelled @xmath31 . repeat this procedure with initial seed @xmath119 to get the ordering of the seeds connected to @xmath118 . continuing this construction for all seeds in @xmath87 constructs images under @xmath120 for all seeds in the labelled exchange graph . for any two seeds @xmath121 connected by an edge labelled @xmath122 in @xmath87 this construction ensures that the images @xmath123 and @xmath124 are also connected by an edge labelled @xmath122 , and so @xmath120 is indeed an automorphism of the labelled exchange graph . . ] [ ex : b2-exchaut ] consider the automorphism @xmath109 of the @xmath91 exchange graph @xmath125 shown in figure [ fig : b2-exch ] given by a clockwise rotation by angle @xmath126 . this automorphism pulls back to an automorphism @xmath120 of the labelled exchange graph @xmath127 shown in figure [ fig : b2-lexch ] . to determine the automorphism @xmath120 , choose an initial labelled seed @xmath128 where @xmath129 the automorphism @xmath109 maps the corresponding cluster @xmath130 $ ] to @xmath131 $ ] and the mutation @xmath132 takes @xmath133 to @xmath134 , whose corresponding cluster @xmath131 $ ] is mapped to @xmath135 $ ] by @xmath109 , as shown in figure [ fig : b2-exchaut ] . denote by @xmath136 the automorphism which corresponds to @xmath137 and denote the quotient by the symmetric group action as @xmath138 . then @xmath114 is a labelled seed in @xmath3 such that @xmath139 . hence the cluster variable which differs between @xmath131 $ ] and @xmath135 $ ] needs to appear in the first position of the labelled cluster of @xmath114 and so @xmath140 this shows that the rotation of @xmath125 actually corresponds to an automorphism of @xmath127 which interchanges the two components of the graph ( see figure [ fig : b2-lexch ] ) as well as rotating each component . note that this automorphism takes the diagram @xmath141 to its opposite @xmath142 , however the matrix @xmath143 is not taken to @xmath144 , but rather to @xmath145 . consider the exchange graph @xmath125 of the mutation class of type @xmath90 shown in figure [ fig : a2-exch ] , with the labelled exchange graph @xmath127 in figure [ fig : a2-lexch ] . an order 5 clockwise @xmath146 rotation @xmath109 of @xmath125 is an exchange graph automorphism and so induces an automorphism @xmath120 of @xmath127 . the cluster @xmath147 $ ] maps to @xmath148 $ ] , so the labelled cluster @xmath149 would be mapped to either @xmath150 or @xmath151 . to determine which , consider the labelled clusters adjacent to @xmath152 : @xmath153 @xmath154 the cluster @xmath155 $ ] is mapped to @xmath156 $ ] so we need to choose an ordering for @xmath157 such that @xmath158 corresponds to the same ordering of @xmath159 $ ] . these two clusters @xmath160 $ ] and @xmath161 $ ] differ by replacing @xmath162 with @xmath163 , while @xmath132 changes the cluster variable in the first position , therefore the required ordering is @xmath164 this shows that @xmath109 induces the automorphism of @xmath127 given by clockwise @xmath165 rotation , which again has order 5 . it is not true in general that @xmath166 , as @xmath87 can have a number of connected components which are identified under the quotient by the symmetric group action . any automorphism which changes a single connected component while fixing all others would therefore not project down to an automorphism of @xmath5 . for example , in the case of the cluster algebra of type @xmath91 , the labelled exchange graph automorphism given by rotating the top hexagon in figure [ fig : b2-lexch ] while fixing the bottom hexagon would not give any valid exchange graph automorphism . given @xmath167 then @xmath97 is constructed in such a way that for @xmath138 the quotient by the symmetric group action , @xmath168 a labelled seed and @xmath46 a single global mutation , @xmath169@xmath170 the inclusion @xmath171 is a homomorphism , that is @xmath172 for any exchange graph automorphisms @xmath173 . choose a labelled seed @xmath174 then @xmath175 this shows that the labelled seeds @xmath176 and @xmath177 are the same up to permutation , however for any @xmath178 @xmath179 so after mutation in the @xmath122-th vertex @xmath176 and @xmath177 are the still same up to permutation . the only way that the @xmath122-th mutation affects two labelled seeds in the same way is if the labelled seeds are in fact equal and not permutations of one another , so @xmath180 and therefore @xmath172 . [ prop : deltaperm ] let @xmath98 with pullback @xmath181 , then for any labelled seed @xmath47 and any permutation @xmath64 @xmath182 therefore although it looks like the construction of @xmath120 from @xmath109 depends on the initial choice of ordering of @xmath47 , any other ordering just gives a permutation of @xmath120 . in @xmath87 there are edges @xmath183 for each mutation @xmath72 , applying @xmath64 gives edges @xmath184 for each @xmath185 . when projected @xmath112 ) = \textbf{x}(\phi[u \cdot \sigma])$ ] and @xmath113 ) = \textbf{x}(\phi[v^i \cdot \sigma])$ ] for each @xmath31 . the clusters @xmath112 ) = [ a , \dotsc , k_i , \dotsc]$ ] and @xmath113 ) = [ a , \dotsc , k'_i , \dotsc]$ ] differ in a single cluster variable @xmath186 to @xmath187 . in the construction of @xmath114 we specified an ordering @xmath188 on @xmath112)$ ] such that the @xmath31-th variable @xmath189))_i = k_i$ ] for each @xmath31 . to construct @xmath190 we need an ordering @xmath191 such that the @xmath57-th variable @xmath192))_{\sigma(i ) } = k_i$ ] so that the position of the variable which changes matches the label on the edge in @xmath127 . therefore @xmath193 ) ) = \sigma(\rho_{u}(\textbf{x}(\phi[u ] ) ) ) = \sigma(\textbf{x}(\phi^\delta(u ) ) ) = \textbf{x}(\phi^\delta(u ) \cdot \sigma)$ ] . so far in this section we have proved properties of automorphisms of the exchange graph of a cluster algebra . in the remainder of this paper we use these results to compare these exchange graph automorphisms to other automorphisms related to the cluster algebra . [ thm : mnexch ] for a labelled mutation class @xmath3 with mutation class @xmath4 and exchange graph @xmath5 @xmath6 let @xmath194 and let @xmath195 be the transformation of @xmath5 given by @xmath196 ) = [ \phi(u)]$ ] . the automorphism @xmath109 commutes with permutations so the choice of order of @xmath47 does not matter , because for any other choice of order @xmath197 there is some permutation @xmath64 such that @xmath198 and then @xmath199 = [ \phi(u \cdot \sigma ) ] = [ \phi(u ) \cdot \sigma ] = [ \phi(u)]$ ] . for any two seeds @xmath47 and @xmath200 related by a single mutation @xmath85 there is an edge @xmath105 - [ v]$ ] in @xmath5 . then @xmath201 ) = \psi([u \cdot \mu ] ) = [ \phi(u \cdot \mu ) ] = [ \phi(u ) \cdot \mu ] = \tilde{\mu}[\phi(u ) ] = \tilde{\mu}\psi([u])$ ] where @xmath202 is the single local mutation on @xmath105 $ ] corresponding to the global mutation @xmath85 on @xmath47 . hence there is an edge @xmath203 - \psi[v]$ ] in @xmath5 , so @xmath204 and @xmath205 . to show the converse , let @xmath206 which pulls back to @xmath207 by theorem [ thm : pullback ] . let @xmath208 be the map given by @xmath209 . any element of @xmath1 can be written as a product of mutations and permutations , so to prove @xmath194 it suffices to show that @xmath109 commutes with any permutation and any mutation . let @xmath64 be a permutation , then by proposition [ prop : deltaperm ] @xmath210 let @xmath47 and @xmath200 be two labelled seeds related by a single mutation , then @xmath211 is an automorphism of @xmath87 , so @xmath212 cluster automorphisms were introduced by assem , schiffler and shramchenko in @xcite . in their paper the authors computed some particular examples of automorphism groups and drew links between automorphisms of the cluster algebra of a surface and the mapping class group of that surface . this correspondence was later proved by brstle and qiu in @xcite for all surfaces except a select few , as discussed in section [ sec : mcg ] . [ def : claut ] a @xmath40-automorphism @xmath213 is a _ cluster automorphism _ of @xmath76 if there exists a seed @xmath13 in @xmath77 such that 1 . @xmath214 is a cluster . 2 . for every @xmath215 we have @xmath216 . cluster automorphisms were originally only defined for skew - symmetric matrices and hence quivers , but the same definitions and some results apply to skew - symmetrizable matrices as well . the cluster automorphism groups in this setting were first studied by chang and zhu in @xcite and @xcite . recall that throughout this paper we assume that the cluster @xmath41 of a seed uniquely determines the seed s matrix , and in this case the matrix is denoted @xmath42 . [ lem : clautmatrix ] if @xmath213 is a @xmath40-automorphism , then @xmath213 is a cluster automorphism if and only if there exists a seed @xmath62 such that @xmath214 is a cluster and @xmath217 or @xmath144 . the definition of a cluster automorphism only requires that there exists a single seed such that the image is a seed and the automorphism is compatible with mutations of that seed , however the compatibility with mutations allows these properties to be extended to all seeds in the cluster algebra . let @xmath213 be a cluster automorphism of a cluster algebra @xmath7 , then @xmath213 satisfies the conditions in definition [ def : claut ] and lemma [ lem : clautmatrix ] for every seed in @xmath7 . this therefore gives two ways of thinking of cluster automorphisms as either automorphisms taking clusters to clusters which are compatible with mutations or as automorphisms which fix exchange matrices ( up to multiplication by -1 ) . a cluster automorphism which fixes exchange matrices is called a _ direct _ cluster automorphism , whereas those which send an exchange matrix @xmath14 to @xmath144 are called _ inverse _ cluster automorphisms . cluster automorphisms form a group , so let @xmath218 denote the group of all cluster automorphisms of @xmath7 , and @xmath219 be the subgroup of direct cluster automorphisms . let @xmath7 be a cluster algebra generated by an exchange matrix @xmath14 . if @xmath14 is mutation - equivalent to @xmath144 then @xmath220 is a normal subgroup of @xmath218 with index 2 , otherwise @xmath221 . cluster automorphisms arise naturally in the labelled seed and global mutation setting introduced by king and pressland , with the following correspondence : [ thm : mutfinquiver ] if @xmath77 is the mutation class of a seed @xmath222 where @xmath17 is a skew - symmetric mutation - finite matrix then @xmath223 combining theorem [ thm : mutfinquiver ] with theorem [ thm : mnexch ] gives the following : [ cor : mfquiv - exchaut ] for a cluster algebra @xmath7 constructed from a mutation - finite quiver with exchange graph @xmath8 @xmath9 chang and zhu provide an alternative proof of this in @xcite and extend the result to certain finite type skew - symmetrizable matrices : [ thm : bncn ] if @xmath77 is the mutation class of a seed @xmath13 where @xmath14 is a skew - symmetrizable matrix of dynkin type @xmath10 or @xmath11 for @xmath12 then @xmath224 maximal green sequences are certain sequences of mutations of a given quiver . first studied by keller in @xcite in relation to quantum dilogarithms they have subsequently been used to study bps states in theoretical physics ( see for example @xcite ) . in their paper on maximal green sequences , brstle , dupont and protin proved that any maximal green sequence for some quiver @xmath17 takes it to a quiver which is isomorphic to @xmath17 ( * ? ? ? * proposition 2.10 ) . hence this sequence of mutations will give an element of the mutation group @xmath225 which takes a cluster to a cluster and a quiver @xmath17 to an isomorphic quiver @xmath226 , with some permutation @xmath64 which acts on the vertices of @xmath226 to give @xmath17 . then @xmath227 fixes the quiver and therefore induces a cluster automorphism . given a quiver @xmath17 , its _ framed quiver _ @xmath228 ( respectively _ coframed _ quiver @xmath229 ) is the quiver constructed from @xmath17 by adding an additional vertex @xmath230 and an additional arrow @xmath231 ( resp . @xmath232 ) for each vertex @xmath31 of @xmath17 . these additional vertices are considered frozen vertices of the ( co)framed quiver . for a quiver @xmath17 call this set of frozen vertices of the quiver @xmath233 . given a quiver @xmath17 with framed quiver @xmath228 , a non - frozen vertex @xmath31 of a quiver @xmath34 in the mutation class of @xmath228 is called _ green _ ( resp . _ red _ ) if for each @xmath234 there is no arrow @xmath235 ( resp . no arrow @xmath236 ) in the quiver @xmath34 . every ( non - frozen ) vertex in a quiver of the mutation class of @xmath228 is either green or red ( * ? ? ? * theorem 2.6 ) . a maximal green sequence is then a sequence of mutations at green vertices which continues until every non - frozen vertex is red . the quiver of type @xmath90 has framed and coframed quivers as shown in figure [ fig : framed - a2 ] . this quiver has two maximal green sequences given by @xmath237 and @xmath238 which are illustrated in figure [ fig : green - seq - a2 ] . ( left ) , with its framed quiver ( center ) and coframed quiver ( right ) . green vertices are shown as circles , red vertices as crosses and frozen vertices as plusses . ] starting with its framed quiver . the top green sequence is @xmath237 and the bottom is @xmath239 . the two resulting quivers are both isomorphic to the coframed quiver of the quiver of type @xmath90 . green vertices are shown as circles , red vertices as crosses and frozen vertices as plusses . ] if the initial labelled seed is @xmath240 with cluster @xmath241 , then the resulting cluster after these green sequences induces a cluster automorphism as shown below . the sequence @xmath238 does not give the same quiver , but after the permutation @xmath242 it does : @xmath243these both give the same cluster automorphism @xmath244 and @xmath245 . theorems [ thm : mutfinquiver ] and [ thm : bncn ] show that cluster automorphisms are linked to the automorphisms of the exchange graph for mutation - finite skew - symmetric matrices as well as a specific family of skew - symmetrizable matrices . however , in general the exchange graph automorphism group for any mutation - finite skew - symmetrizable matrix is larger than the cluster automorphism group . an example of this would be the exchange graph automorphism of the mutation class of @xmath91 considered in example [ ex : b2-exchaut ] . this graph automorphism does not correspond to a cluster automorphism as the initial matrix @xmath14 is sent to @xmath246 . in this section we aim to generalise the results of the previous section to the skew - symmetrizable case . to do this we introduce additional structure on the exchange graph , which defines a marked exchange graph . this extra structure ensures that any graph automorphism fixing this structure corresponds to a cluster automorphism . in this way the study of cluster automorphisms can be reduced to the combinatorial study of graph automorphisms . let @xmath14 be a skew - symmetrizable matrix , with symmetrizing matrix @xmath25 . if @xmath72 is any mutation , then @xmath25 is also the symmetrizing matrix for @xmath247 . similarly for any permutation @xmath64 the permuted matrix @xmath248 is the symmetrizing matrix for @xmath249 . the _ marked labelled exchange graph _ of a mutation class generated by @xmath128 where @xmath14 is a skew - symmetrizable matrix with symmetrizing matrix @xmath250 is the labelled exchange graph with an additional marking on each edge . each edge corresponds to a global mutation @xmath72 for some @xmath31 , so mark that edge with the symmetrizing entry @xmath251 . if a permutation @xmath64 acts on @xmath47 to give a labelled seed in a different component of @xmath87 , then mark the @xmath31-th edges with @xmath252 , where @xmath253 . in the exchange graph @xmath125 each edge no longer corresponds to a global mutation @xmath72 , but rather to a local mutation @xmath53 at a specific cluster variable @xmath254 in a cluster @xmath41 . for a permutation @xmath64 and permuted seed @xmath255 , then the edge @xmath185 adjacent to this seed corresponds to the local mutation @xmath256 } = \mu_{\beta_i , [ \textbf{x}]}$ ] as @xmath257 and @xmath258 = [ \textbf{x}]$ ] . this edge @xmath185 is marked with @xmath259 and hence in the quotient the edge @xmath260 has a consistent marking , so the following is well - defined . let @xmath261 be the _ marked exchange graph _ of a mutation class @xmath77 given by taking the quotient of the marked labelled exchange graph with respect to the symmetric group action . alternatively let @xmath14 be a skew - symmetrizable matrix , with symmetrizing matrix @xmath25 and let @xmath34 be the diagram corresponding to @xmath14 so each row in @xmath14 represents a vertex in @xmath34 . each diagonal entry in @xmath25 can be thought of as being attached to that row s vertex of @xmath34 , and the edge in @xmath262 representing mutation in that vertex should be marked with this diagonal entry . . dotted edges correspond to a symmetrizing entry of @xmath38 , while dashed edges correspond to @xmath263 . only denominators are shown in the cluster variables with a bar above each , unless the cluster variable is one of @xmath264 or @xmath265 where the variable is shown with a bar underneath . ] the marked exchange graph of the cluster algebra of type @xmath266 is shown in figure [ fig : b3-marked ] . the cluster variables are not written out in full , rather only the denominators are shown with a bar above except for the initial cluster variables @xmath267 , @xmath268 and @xmath265 which are shown with a bar underneath . each vertex is adjacent to two dotted edges and one dashed edge . choosing a matrix in the mutation class , the symmetrizing matrix is @xmath269 : @xmath270 the dotted edges correspond to mutations in the vertices with symmetrizing entry @xmath38 , while the dashed edge corresponds to the mutation in the vertex with symmetrizing entry @xmath263 . in this case , any automorphism of the unmarked exchange graph sends dashed edges to dashed edges , so automatically preserves the markings and hence @xmath271 . . dotted edges correspond to mutations in a vertex with symmetrizer @xmath38 while dashed edges correspond to symmetrizer @xmath263 . ] the marked exchange graph of the cluster algebra of type @xmath91 is shown in figure [ fig : b2-marked ] , where dotted edges correspond to mutations in vertices with symmetrizer @xmath38 and dashed edges correspond to symmetrizer @xmath263 . the initial matrix for the cluster @xmath130 $ ] was chosen to be @xmath272 with symmetrizing matrix @xmath273 . the automorphism considered in example [ ex : b2-exchaut ] , given by a rotation of angle @xmath126 , does not fix the markings in the graph , so is not an automorphism of the marked graph . for any skew - symmetric matrix the symmetrizing matrix is the identity , so all markings would be the same and @xmath274 . for a cluster algebra of dynkin type @xmath10 or @xmath11 , for @xmath12 , the marking on the exchange graph does not limit the number of automorphisms , so @xmath275 . this follows from theorem 3.7 in chang and zhu s paper @xcite linking exchange graph automorphisms and cluster automorphisms . let @xmath125 be an exchange graph of a seed @xmath276 with vertices labelled @xmath277 . for a subset of vertices @xmath278 the _ frozenisation _ of @xmath47 with respect to @xmath278 is the mutation class constructed by freezing all vertices in @xmath278 . it is often more convenient to consider the _ cofrozenisation _ of @xmath47 with respect to @xmath278 , denoted @xmath279 , which is constructed by freezing all vertices in @xmath47 except those in @xmath278 . this is then a frozenisation of @xmath47 with respect to @xmath280 . a _ geodesic loop _ @xmath281 is the exchange graph of a cofrozenisation @xmath282 which leaves only two vertices @xmath283 and @xmath284 unfrozen . a loop is then either a polygon with @xmath285 or @xmath286 sides or an infinite line , which embeds into the exchange graph of the mutation class of @xmath47 . the _ distance _ between a geodesic loop @xmath287 and any vertex @xmath84 in @xmath125 is the ( possibly zero ) minimum number of edges in @xmath125 between @xmath84 and any vertex in @xmath287 . the _ length _ of a geodesic loop @xmath288 is the number of edges in the loop . geodesic loops as subgraphs of a larger exchange graph give rise to the following sets , which encode the information about a given seed represented by a vertex of the exchange graph . the following construction is a slight notational variation of the one given by chang and zhu in definition 3.1 of @xcite . let @xmath47 be a seed of rank @xmath29 in an exchange graph , then define @xmath289 to be the set of @xmath290 numbers given by the length of all geodesic loops distance @xmath291 from @xmath47 . similarly define @xmath292 to be the set of @xmath293 numbers given by the lengths of all geodesic loops distance @xmath38 from @xmath47 . an exchange graph automorphism @xmath137 induces an automorphism @xmath136 and in this way @xmath109 induces a map @xmath294 which takes cluster variables in a seed @xmath47 to variables in @xmath295 . a geodesic loop @xmath296 in an exchange graph @xmath125 must get mapped to another geodesic loop of the same length by any exchange graph automorphism , however it is not clear that the image of @xmath296 will be generated by the cofrozenisation @xmath297 rather than another cofrozenisation with two different unfrozen vertices in @xmath47 . the following lemma explains that this must always be the case . [ lem : fixfrozexch ] let @xmath47 be a seed in a cluster algebra @xmath7 and @xmath298 . for any two vertices @xmath283 and @xmath284 the geodesic loop @xmath299 is isomorphic to its image @xmath300 . choose some ordering on @xmath47 so that the vertices @xmath301 and @xmath302 are indexed by @xmath31 and @xmath32 respectively , then the length of the geodesic loop specifies a relation @xmath303 . for example if the loop has length @xmath304 , then @xmath305 , whereas if the length is @xmath306 then @xmath307 . the exchange graph automorphism @xmath109 corresponds to some @xmath308 which commutes with the action of @xmath1 . hence @xmath309 so the geodesic loop @xmath300 has the same length as the geodesic loop @xmath299 , and hence the two loops are isomorphic . exchange graph automorphisms preserve the combinatorial structure around a seed . as these automorphisms are compatible with mutations the above result could be extended to the exchange graphs of cofrozenisations with any number of unfrozen vertices . [ lem : npreserved ] if @xmath137 is an exchange graph automorphism , with @xmath47 a seed and @xmath310 its image , then @xmath311 and @xmath312 . [ lem : ndefinesb ] given a mutation - finite diagram with at least 3 vertices , the exchange graph of a frozenisation leaving just two vertices unfrozen determines the weight on the arrow between the two unfrozen vertices . the exchange graph of the frozenisation leaving just two vertices @xmath283 and @xmath284 unfrozen is a geodesic loop @xmath313 with length @xmath314 . if @xmath315 then the vertices have no arrow between them , while if @xmath316 there is a single unweighted arrow . if @xmath317 then there is an arrow weighted @xmath263 and @xmath318 shows there is an arrow weighted @xmath319 . the highest edge weight in a mutation - finite diagram ( with more than 2 vertices ) is @xmath320 , so @xmath321 implies that there is an arrow weighted @xmath320 . for any @xmath263-vertex diagram @xmath14 , an edge weight of @xmath320 or more will always give @xmath321 , so the exchange graph can not determine this weight . however the only diagrams mutation - equivalent to @xmath14 are @xmath14 and @xmath322 , so all diagrams in the same mutation class have the same edge weight . [ lem : unoriented ] an exchange graph automorphism @xmath137 takes a seed @xmath276 to another seed @xmath323 where the unoriented diagram of @xmath324 is the same as the unoriented diagram of @xmath14 . fix any two vertices @xmath325 and @xmath326 in @xmath47 . under @xmath294 these vertices are mapped to corresponding vertices @xmath327 and @xmath328 in @xmath84 . the weight on ( or absence of ) the arrow between @xmath325 and @xmath326 determines the exchange graph @xmath329 of the cofrozenisation @xmath330 . by lemma [ lem : fixfrozexch ] , @xmath329 is isomorphic to the exchange graph @xmath331 of the cofrozenisation @xmath332 . hence this exchange graph determines the arrow between @xmath333 and @xmath334 by lemmas [ lem : npreserved ] and [ lem : ndefinesb ] , which necessarily must be the same as that between @xmath325 and @xmath326 . this shows that the unoriented diagrams of two seeds related by an exchange graph automorphism must be the same . to see how exchange graphs automorphisms affect the orientations of the arrows we need to consider frozenisations with three unfrozen vertices . [ lem:3vert - n ] for any seed @xmath276 with 3 vertices in an exchange graph of a mutation - finite skew - symmetrizable diagram , the diagram of @xmath14 is determined by the sets @xmath289 and @xmath292 , up to reversing all arrows . the unoriented diagram of @xmath14 is determined by @xmath335 , where each @xmath336 determines a weighted arrow , or absence of arrow , between two vertices . the orientation of @xmath14 ( up to reversing all arrows ) is given by @xmath292 as shown in tables [ tab : diconnn ] , [ tab : symmetricn ] and [ tab : symmetrizn ] , where all mutation - finite 3-vertex diagrams are illustrated along with their defining sets @xmath337 and @xmath338 . hence the pair @xmath339 defines a unique diagram , up to reversing all arrows . in the case @xmath340 the diagram is of the form : where the weight satisfies @xmath341 and so the diagram is not uniquely determined . however if @xmath342 then the resulting diagram will never appear as a subdiagram of any larger mutation - finite diagram . this is precisely the setup used in the proofs below and so @xmath340 is always assumed to correspond to a diagram of the form : [ prop : exchdiagram ] let @xmath137 be an exchange graph automorphism and @xmath128 a seed where @xmath14 is a mutation - finite skew - symmetrizable matrix with corresponding connected diagram @xmath34 . in the image @xmath343 , the diagram @xmath344 corresponding to the matrix @xmath324 is either @xmath34 or @xmath35 . choose any 3 vertices @xmath345 in @xmath47 , then by lemma [ lem : fixfrozexch ] there is an isomorphism @xmath346 and @xmath347 , @xmath348 . therefore by lemma [ lem:3vert - n ] the subdiagram @xmath349 of @xmath34 consisting just of the arrows between @xmath350 and @xmath351 is the same as the subdiagram @xmath352 of @xmath344 consisting of the arrows between @xmath353 and @xmath354 , up to reversing all arrows . choose a fourth vertex @xmath355 and consider the 3-vertex subdiagram @xmath356 on the vertex set @xmath357 . by the same reasoning as above the image @xmath358 must be the same , but possibly with all arrows reversed . however both @xmath352 and @xmath359 share the edge between vertices @xmath360 and @xmath354 , so if @xmath361 then @xmath362 whereas if @xmath363 then @xmath364 . as @xmath34 is connected , by successively choosing different vertices , the whole diagram @xmath344 must either be the same as @xmath34 or @xmath35 . this shows that any exchange graph automorphism takes clusters to clusters and a diagram to itself or its opposite . however this is not enough to show that these automorphisms are cluster automorphisms , as this requires the matrix @xmath14 of the diagram to be sent to @xmath365 . for this we require the markings on the exchange graph . [ prop : mexchmatrix ] given a marked exchange graph automorphism @xmath366 and a seed @xmath367 with image @xmath368 , then the matrix @xmath369 or @xmath144 . let @xmath34 be the diagram associated to @xmath14 , and let @xmath344 be the diagram associated to @xmath324 . let @xmath370 be the symmetrizing matrix for @xmath14 , each vertex @xmath371 in @xmath47 has a symmetrizing multiplier , which marked exchange graph automorphisms preserve , so each vertex @xmath372 in @xmath295 has the same symmetrizing multiplier as @xmath371 and @xmath373 . by proposition [ prop : exchdiagram ] , @xmath344 is the same as @xmath34 or @xmath35 with symmetrizing matrix @xmath374 which defines the skew - symmetrizable matrix @xmath369 or @xmath144 . these results ensure that a marked exchange graph automorphism fixes matrices in seeds and so correspond to cluster automorphisms . in this way we generalise corollary [ cor : mfquiv - exchaut ] to all mutation - finite skew - symmetrizable matrices . [ thm : mexchaut ] for a seed @xmath13 where @xmath14 is a mutation - finite skew - symmetrizable matrix with mutation class @xmath77 , cluster algebra @xmath375 and marked exchange graph @xmath15 then @xmath16 a cluster automorphism @xmath376 satisfies the following properties : * @xmath214 is a cluster * @xmath213 is compatible with mutations * @xmath377 or @xmath144 for all seeds @xmath378 in the mutation class @xmath77 . such an automorphism induces an automorphism of the exchange graph , and as @xmath213 sends a matrix @xmath14 to @xmath365 it also fixes the symmetrizing matrix so fixes the marking on the exchange graph . therefore @xmath379 and @xmath380 . to show that @xmath381 let @xmath13 be a labelled seed , with mutation class @xmath3 and quotient @xmath77 . if @xmath382 , then by theorem [ thm : pullback ] this pulls back to an automorphism @xmath97 . then the image @xmath383 where @xmath384 gives an automorphism @xmath385 defined by @xmath386 this @xmath213 then corresponds to @xmath109 , so @xmath214 is a cluster and it remains to show that @xmath387 , however this follows from proposition [ prop : mexchmatrix ] so @xmath388 . many skew - symmetrizable matrices @xmath14 have unfoldings to skew - symmetric matrices @xmath389 , which extend to seeds , where a given seed in @xmath390 unfolds to a seed in @xmath391 . the corresponding exchange graphs are related , with the marked exchange graph @xmath18 embedding into the exchange graph @xmath392 provided edges marked in certain ways split into multiple edges . given a skew - symmetrizable @xmath393 matrix @xmath45 with symmetrizing matrix @xmath250 , let @xmath394 and partition the set @xmath395 into @xmath29 disjoint consecutive index sets @xmath396 such that @xmath397 for all @xmath32 . construct a skew - symmetric @xmath398 matrix @xmath389 where : 1 . [ enum : unfdefa ] the sum of entries in each column of each @xmath399 block equals @xmath30 . [ enum : unfdefb ] if @xmath33 then all entries in the @xmath399 block are non - negative . [ enum : unfdefc ] all entries in each @xmath400 block are zero . given @xmath401 and any @xmath402 the corresponding mutations @xmath403 and @xmath46 commute . the @xmath31-th _ composite mutation _ @xmath404 of @xmath389 is given by @xmath405 the matrix @xmath389 is the _ unfolding _ of @xmath14 if the matrix @xmath406 satisfies the conditions [ enum : unfdefa ] and [ enum : unfdefb ] above with respect to the matrix @xmath407 for any sequence of mutations @xmath408 with corresponding composite mutations @xmath409 . a labelled seed @xmath410 , b { \aftergroup\egroup\originalright})$ ] , with skew - symmetrizable matrix @xmath14 , unfolds in the same way to @xmath411 , c { \aftergroup\egroup\originalright})$ ] where @xmath389 is the unfolding of @xmath14 . the @xmath32-th row in @xmath14 corresponds to the cluster variable @xmath412 and this row unfolds to @xmath413 rows in @xmath389 , hence @xmath412 unfolds to @xmath413 cluster variables @xmath414 . a diagram has a finite number of distinct matrix representations , each of which may give different unfoldings , or may not admit any unfolding . almost all mutation - finite matrices have an unfolding . given a permutation @xmath74 of the initial seed , construct the _ composite permutation _ @xmath415 to be the permutation given by : [ thm : autsintounf ] given a skew - symmetrizable matrix @xmath14 which unfolds to a matrix @xmath389 , with corresponding marked exchange graphs @xmath18 and @xmath416 , then @xmath417 choose an initial @xmath393 labelled seed @xmath418 , b { \aftergroup\egroup\originalright})$ ] which unfolds to the @xmath398 labelled seed @xmath411 , c { \aftergroup\egroup\originalright})$ ] with index sets @xmath419 for @xmath420 and @xmath421 . let @xmath422 be an exchange graph automorphism , then @xmath109 corresponds to both a cluster automorphism @xmath423 of the cluster algebra @xmath424 , constructed from the initial seed @xmath47 , and to a mutation class automorphism @xmath425 . this mutation class automorphism in turn corresponds to an element of @xmath1 , so there is a sequence of @xmath426 mutations @xmath408 and a permutation @xmath64 such that @xmath427 all such automorphisms are constructed to have the same action on the initial seed @xmath47 , so @xmath428 , \pm b { \aftergroup\egroup\originalright } ) = { \mathopen{}\mathclose\bgroup\originalleft } ( { \mathopen{}\mathclose\bgroup\originalleft } [ f(\beta_i ) { \aftergroup\egroup\originalright } ] , \pm b { \aftergroup\egroup\originalright } ) = \phi_m(u ) = u \cdot { \mathopen{}\mathclose\bgroup\originalleft } ( \mu_{k_1 } \mu_{k_2 } \dotsb \mu_{k_r } \sigma { \aftergroup\egroup\originalright}).\ ] ] in the unfolding , each mutation @xmath408 corresponds to the composite mutation @xmath409 and the permutation @xmath64 corresponds to the composite permutation @xmath429 , so the following commutes : the automorphism @xmath109 corresponds to a cluster automorphism , so the matrix of the image of @xmath47 is @xmath365 . the seed @xmath430,\pm b { \aftergroup\egroup\originalright})$ ] unfolds to @xmath431 , \pm c { \aftergroup\egroup\originalright})$ ] and hence @xmath432 acts on @xmath411 , c { \aftergroup\egroup\originalright})$ ] to give a seed with the same matrix up to sign , so corresponds to a cluster automorphism of the cluster algebra constructed with @xmath411 , c { \aftergroup\egroup\originalright})$ ] as the initial seed , and hence to an automorphism of the exchange graph @xmath392 . by theorem [ thm : mexchaut ] the marked exchange graph automorphisms correspond to cluster automorphisms , so for a skew - symmetrizable matrix @xmath14 which unfolds to @xmath389 and with corresponding cluster algebras @xmath424 and @xmath433 , theorem [ thm : autsintounf ] implies @xmath434 . the dotted and dashed edges show how the marked exchange graph of type @xmath91 shown in figure [ fig : b2-marked ] unfolds . a dashed edge in figure [ fig : b2-marked ] corresponds to the composite mutation denoted by a consecutive pair of dashed edges in this figure . ] the matrix @xmath14 representing the dynkin diagram of type @xmath91 @xmath435 the matrix representing a quiver of dynkin type @xmath436 . the symmetrizing matrix of @xmath14 is given by @xmath437 so the @xmath91 marked exchange graph shown in figure [ fig : b2-marked ] embeds into the exchange graph of type @xmath436 shown in figure [ fig : a3-exch ] . the dashed edges in figure [ fig : b2-marked ] correspond to the pairs of dashed edges representing composite mutations in figure [ fig : a3-exch ] . dotted edges in figure [ fig : b2-marked ] correspond to single dotted edges in figure [ fig : a3-exch ] . the seed @xmath438 , b { \aftergroup\egroup\originalright})$ ] unfolds to the seed @xmath439 , c { \aftergroup\egroup\originalright})$ ] and the cluster variables of these two seeds are related with @xmath440 as the symmetrizing matrix @xmath273 ensures that @xmath162 unfolds to two cluster variables . the automorphism @xmath441 given by rotation by @xmath442 takes the seed @xmath130 $ ] to @xmath443 $ ] and corresponds to the cluster automorphism @xmath444 given by @xmath445 this automorphism induces an automorphism of the exchange graph of @xmath436 given by a rotation along the embedded @xmath446 fixing the seeds with cyclic quivers and takes @xmath447 $ ] to @xmath448 $ ] which corresponds to the cluster automorphism @xmath449 given by @xmath450 however the automorphism could also correspond to the cluster automorphism @xmath451 where @xmath452 there is a single non - identity @xmath453 exchange graph automorphism which fixes the embedded @xmath446 , given by a reflection in the circle of the embedded subgraph and interchanging the two seeds with cyclic quivers . this then corresponds to the cluster automorphism @xmath454 given by @xmath455 such that @xmath456 . theorem [ thm : autsintounf ] shows that cluster automorphisms of @xmath424 commute with unfolding the seeds , so a direct cluster automorphism @xmath457 preserves the exchange matrix @xmath14 , which when unfolded to @xmath458 must also preserve the exchange matrix @xmath389 and so is also a direct cluster automorphism . [ cor : direct_subgroup ] @xmath459 . in their paper introducing cluster automorphisms @xcite assem , schiffler and shramchenko introduced the tagged mapping class group for surfaces with punctures . this group has been shown to coincide with the group of direct cluster automorphisms of the surface s corresponding cluster algebra . given a surface with marked points @xmath460 the _ mapping class group _ of the surface is given by @xmath461 here @xmath462 is the group of orientation - preserving homeomorphisms from @xmath349 to itself which sends the set @xmath463 to itself , but does not necessarily fix @xmath463 nor the boundary of @xmath349 pointwise , and @xmath464 is the subgroup of homeomorphisms which are isotopic to the identity such that the isotopy fixes @xmath463 pointwise . the cluster structure given by triangulations of a surface with marked points was first studied by fomin , shapiro and thurston in @xcite , where they show that flips of arcs in a triangulation coincide with mutations . however such a triangulation could contain self - folded triangles , and therefore arcs that can not be flipped ; to get around this problem , the authors introduced taggings on the arcs . a _ tagged arc _ is an arc which does not cut out a once - punctured monogon , where the enpoints are tagged either _ plain _ or _ notched _ , such that any endpoints on @xmath465 are tagged plain and if the endpoints of an arc coincide then they must be tagged the same . two tagged arcs are _ compatible _ if either their underlying arcs are the same and then at least one endpoint must be tagged in the same way , or the underlying arcs are not equal but are compatible . in this case , if they share an endpoint , the arcs must be tagged in the same way at that endpoint . a tagged triangulation is a maximal collection of compatible tagged arcs and a tagged flip is then defined in the same way as for triangulations , where a tagged arc is replaced with the unique other compatible tagged arc and these flips again correspond to mutations . see ( * ? ? ? * section 7 ) or ( * ? ? ? * section 4 ) for more details . the _ tagged mapping class group _ of a surface @xmath460 with @xmath466 punctures is the semidirect product of the standard mapping class group of the surface with @xmath467 , @xmath468 where the elements of @xmath469 act as diffeomorphisms on the surface and elements of @xmath467 switch or preserve the tags on the tagged triangulation at each puncture . they showed that for discs and annuli without punctures as well as for certain discs with 1 or 2 punctures then the tagged mapping class group is isomorphic to the group of direct cluster automorphisms of the corresponding cluster algebra . the authors conjectured that this would be the case for almost all surfaces with marked points . brstle and qiu proved that this conjecture is true in @xcite : 1 . a sphere with @xmath476 marked points ; 2 . an unpunctured disc with @xmath477 marked points on the boundary ; 3 . a disc with a single puncture and one marked point on the boundary ; 4 . a once - punctured disc with 2 or 4 marked points on the boundary ; 5 . a twice - punctured disc with 2 marked points on the boundary , diagrams correspond to triangulations of orbifolds in the same way that quivers correspond to triangulations of surfaces . a covering of the orbifold by a surface corresponds to an unfolding of the diagram to a quiver , in such a way that composite mutations of the quiver correspond to triangle flips in the triangulation of the surface , as discussed in @xcite . in their paper on the growth rate of cluster algebras , felikson , shapiro , thomas and tumarkin @xcite defined the mapping class group of a cluster algebra @xmath479 to be the elements of @xmath1 which fix the initial exchange matrix up to a quotient by those elements of @xmath1 which fix the initial seed . elements of this group would then fix the initial exchange matrix and map the initial cluster to some other cluster in the mutation class , and hence would induce a direct cluster automorphism . fix a marked orbifold @xmath21 with @xmath480 punctures . in ( * remark 4.15 ) the cluster mapping class group is argued to either contain the orbifold s mapping class group as a proper normal subgroup with quotient @xmath481 ( when @xmath482 , or when @xmath483 and the boundary non - empty ) or be isomorphic to the orbifold mapping class group ( when @xmath484 , or when @xmath485 and the boundary is empty ) . this orbifold has no punctures , so the tagged mapping class group is equal to the mapping class group . any element of the mapping class group must fix the orbifold point and permute the four boundary marked points . the only such permutations are rotations around the boundary , as any reflection would not preserve the orientation , hence the mapping class group is isomorphic to @xmath488 generated by a rotation by angle @xmath489 . this orbifold corresponds to the cluster algebra of dynkin type @xmath266 , which can be generated by the diagram in figure [ fig : mcg - b3 ] . the cluster automorphism group of @xmath490 is the dihedral group with 8 elements : @xmath491 where @xmath488 is generated by the automorphism given by the action of @xmath492 on the initial cluster and @xmath486 by @xmath493 . this can be seen as the automorphisms of the marked exchange graph shown in figure [ fig : b3-marked ] where the 4 squares are permuted while fixing the markings .

we employ a number of statistical measures to characterize neural discharge activity in cat retinal ganglion cells ( rgcs ) and in their target lateral geniculate nucleus ( lgn ) neurons under various stimulus conditions , and we develop a new measure to examine correlations in fractal activity between spike - train pairs . in the absence of stimulation ( i.e. , in the dark ) , rgc and lgn discharges exhibit similar properties . the presentation of a constant , uniform luminance to the eye reduces the fractal fluctuations in the rgc maintained discharge but enhances them in the target lgn discharge , so that neural activity in the pair no longer mirror each other . a drifting - grating stimulus yields rgc and lgn driven spike trains similar in character to those observed in the maintained discharge , with two notable distinctions : action potentials are reorganized along the time axis so that they occur only during certain phases of the stimulus waveform , and fractal activity is suppressed . under both uniform - luminance and drifting - grating stimulus conditions ( but not in the dark ) , the discharges of pairs of lgn cells are highly correlated over long time scales ; in contrast discharges of rgcs are nearly uncorrelated with each other . this indicates that action - potential activity at the lgn is subject to a common fractal modulation to which the rgcs are not subjected .

physics9910025

the sequence of action potentials recorded from cat retinal ganglion cells ( rgcs ) and lateral - geniculate - nucleus ( lgn ) cells is always irregular . this is true whether the retina is in the dark @xcite , or whether it is adapted to a stimulus of fixed luminance @xcite . it is also true for time - varying visual stimuli such as drifting gratings . with few exceptions , the statistical properties of these spike trains have been investigated from the point - of - view of the interevent - interval histogram @xcite , which provides a measure of the relative frequency of intervals of different durations . the mathematical model most widely used to describe the interevent - interval histogram under all of these stimulus conditions derives from the gamma renewal process @xcite , though point processes incorporating refractoriness have also been investigated @xcite . however , there are properties of a sequence of action potentials , such as long - duration correlation or memory , that can not generally be inferred from measures that reset at short times such as the interevent - interval histogram @xcite . the ability to uncover features such as these demands the use of measures such as the allan factor , the periodogram , or rescaled range analysis ( r / s ) , which can extend over time ( or frequency ) scales that span many events . rgc and lgn spike trains exhibit variability and correlation properties over a broad range of time scales , and the analysis of these discharges reveals that the spike rates exhibit fractal properties . fractals are objects which possess a form of self - similarity : parts of the whole can be made to fit to the whole by shifting and stretching . the hallmark of fractal behavior is power - law dependence in one or more statistical measures , over a substantial range of the time or frequency scale at which the measurement is conducted @xcite . fractal behavior represents a form of memory because the occurrence of an event at a particular time increases the likelihood of another event occurring at some time later , with this likelihood decaying in power - law fashion . fractal signals are also said to be self - similar or self - affine . this fractal behavior is most readily illustrated by plotting the estimated firing rate of a sequence of action potentials for a range of averaging times . this is illustrated in fig . 1a for the maintained discharge of a cat rgc . the rate estimates are formed by dividing the number of spikes in successive counting windows of duration @xmath0 by the counting time @xmath0 . the rate estimates of the shuffled ( randomly reordered ) version of the data are presented in fig . this surrogate data set maintains the same relative frequency of interevent - interval durations as the original data , but destroys any long - term correlations ( and therefore fractal behavior ) arising from other sources , such as the relative ordering of the intervals . comparing figs . 1a and b , it is apparent that the magnitude of the rate fluctuations decreases more slowly with increasing counting time for the original data than for the shuffled version . fractal processes exhibit slow power - law convergence : the standard deviation of the rate decreases more slowly than @xmath1 as the averaging time increases . nonfractal signals , such as the shuffled rgc spike train , on the other hand , exhibit fluctuations that decrease precisely as @xmath1 . the data presented in fig . 1 are typical of all rgc and lgn spike trains . the statistical behavior of a neural spike train can be studied by replacing the complex waveforms of each individual electrically recorded action potential ( fig . 2 , top ) by a single point event corresponding to the time of the peak ( or other designator ) of the action potential ( fig . 2 , middle ) . in mathematical terms , the neural spike train is then viewed as an unmarked point process . this simplification greatly reduces the computational complexity of the problem and permits use of the substantial methodology previously developed for stochastic point processes @xcite . the occurrence of a neural spike at time @xmath2 is therefore simply represented by an impulse @xmath3 at that time , so that the sequence of action potentials is represented by @xmath4 a realization of a point process is specified by the set of occurrence times of the events , or equivalently , of the times @xmath5 between adjacent events , where @xmath6 . a single realization of the data is generally all that is available to the observer , so that the identification of the point process , and elucidation of the mechanisms that underlie it , must be gleaned from this one realization . one way in which the information in an experimental sequence of events can be made more digestible is to reduce the data into a statistic that emphasizes a particular aspect of the data , at the expense of other features . these statistics fall into two broad classes which have their origins , respectively , in the sequence of interevent intervals @xmath5 illustrated at the lower left of fig . 2 , or in the sequence of counts @xmath7 shown at the lower right of fig . 2 . the homogeneous poisson point process , which is the simplest of all stochastic point processes , is described by a single parameter , the rate @xmath8 . this point process is memoryless : the occurrence of an event at any time @xmath9 is independent of the presence ( or absence ) of events at other times @xmath10 . because of this property , both the intervals @xmath5 and counts @xmath7 form sequences of independent , identically distributed ( iid ) random variables . the homogeneous poisson point process is therefore completely characterized by the interevent - interval distribution ( which is exponential ) or the event - number distribution ( which is poisson ) together with the iid property . this process serves as a benchmark against which other point processes are measured ; it therefore plays the role that the white gaussian process enjoys in the realm of continuous - time stochastic processes . a related point process is the nonparalyzable fixed - dead - time - modified poisson point process , a close cousin of the homogeneous poisson point process that differs only by the imposition of a dead - time ( refractory ) interval after the occurrence of each event , during which other events are prohibited from occurring @xcite . another cousin is the gamma-@xmath11 renewal process which , for integer @xmath11 , is generated from an homogeneous poisson point process by permitting every @xmath11th event to survive while deleting all intermediate events @xcite . both the dead - time - modified poisson point process and the gamma renewal process require two parameters for their description . all the examples of point process presented above belong to the class of renewal point processes , which will be defined in sec . [ iihdef ] . however , spike trains in the visual system can not be adequately described by renewal point processes ; rather , nonrenewal processes are required @xcite . of particular interest are fractal - rate stochastic point processes , in which one or more statistics exhibit power - law behavior in time or frequency @xcite . one feature of such processes is the relatively slow power - law convergence of the rate standard deviation , as illustrated in fig we have previously shown that a fractal , doubly stochastic point process that imparts multiscale fluctuations to the gamma-@xmath11 renewal process provides a reasonable description of the rgc and lgn maintained discharges @xcite . two statistical measures are often used to characterize the discrete - time stochastic process @xmath5 illustrated in the lower left corner of fig . 2 . these are the interevent - interval histogram ( iih ) and rescaled range analysis ( r / s ) . the interevent - interval histogram ( often referred to as the interspike - interval histogram or isih in the physiology literature ) displays the relative frequency of occurrence @xmath12 of an interval of size @xmath13 ; it is an estimate of the probability density function of interevent - interval magnitude ( see fig . 2 , lower left ) . it is , perhaps , the most commonly used of all statistical measures of point processes in the life sciences . the interevent - interval histogram provides information about the underlying process over time scales that are of the order of the interevent intervals . its construction involves the loss of interval ordering , and therefore dependencies among intervals ; a reordering of the sequence does not alter the interevent - interval histogram since the order plays no role in the relative frequency of occurrence . some point processes exhibit no dependencies among their interevent intervals at the outset , in which case the sequence of interevent intervals forms a sequence of iid random variables and the point process is completely specified by its interevent - interval histogram . such a process is called a renewal process , a definition motivated by the replacement of failed parts ( such as light bulbs ) , each replacement of which forms a renewal of the point process . the homogeneous poisson point process , dead - time - modified poisson point process , and gamma renewal process are all renewal processes , but experimental rgc and lgn spike trains are not . rescaled range ( r / s ) analysis provides information about correlations among blocks of interevent intervals . for a block of @xmath14 interevent intervals , the difference between each interval and the mean interevent interval is obtained and successively added to a cumulative sum . the normalized range @xmath15 is the difference between the maximum and minimum values that the cumulative sum attains , divided by the standard deviation of the interval size . @xmath15 is plotted against @xmath14 . information about the nature and the degree of correlation in the process is obtained by fitting @xmath15 to the function @xmath16 , where @xmath17 is the so - called hurst exponent @xcite . for @xmath18 positive correlation exists among the intervals , whereas @xmath19 indicates the presence of negative correlation ; @xmath20 obtains for intervals with no correlation . renewal processes yield @xmath20 . for negatively correlated intervals , an interval that is larger than the mean tends , on average , to be preceded or followed by one smaller than the mean . this widely used measure is generally assumed to be well suited to processes that exhibit long - term correlation or have a large variance @xcite , but it appears not to be very robust since it exhibits large systematic errors and highly variable estimates of the hurst coefficient for some fractal sequences @xcite . nevertheless , it provides a useful indication of correlation in a point process arising from the ordering of the interevent intervals alone . it is advantageous to study some characteristics of a point process in terms of the sequence of event numbers ( counts ) @xmath7 rather than via the sequence of intervals @xmath5 . figure 2 illustrates how the sequence is obtained . the time axis is divided into equally spaced , contiguous time windows ( center ) , each of duration @xmath0 sec , and the ( integer ) number of events in the @xmath21th window is counted and denoted @xmath22 . this sequence @xmath7 forms a random counting process of nonnegative integers ( lower right ) . closely related to the sequence of counts is the sequence of rates ( events / sec ) @xmath23 , which is obtained by dividing each count @xmath22 by the counting time @xmath0 . this is the measure used in fig . 1 . we describe several statistical measures useful for characterizing the counting process @xmath7 : the fano factor , the allan factor , and the event - number - based power spectral density estimate ( periodogram ) . the fano factor is defined as the event - number variance divided by the event - number mean , which is a function of the counting time @xmath0 : @xmath24}}{\displaystyle { { \rm e}\left[z_n(t)\right]}}}.\ ] ] this quantity provides an abbreviated way of describing correlation in a sequence of events . it indicates the degree of event clustering or anticlustering in a point process relative to the benchmark homogeneous poisson point process , for which @xmath25 for all @xmath0 . the fano factor must approach unity at sufficiently small values of the counting time @xmath0 for any regular point process @xcite . in general , a fano factor less than unity indicates that a point process is more orderly than the homogeneous poisson point process at the particular time scale @xmath0 , whereas an excess over unity indicates increased clustering at the given time scale . this measure is sometimes called the index of dispersion ; it was first used by fano in 1947 @xcite for characterizing the statistical fluctuations of the number of ions generated by individual fast charged particles . for a fractal - rate stochastic point process the fano factor assumes the power - law form @xmath26 ( @xmath27 ) for large @xmath0 . the parameter @xmath28 is defined as an estimate of the fractal exponent ( or scaling exponent ) @xmath29 of the point - process rate . though the fano factor can detect the presence of self - similarity even when it can not be discerned in a visual representation of a sequence of events , mathematical constraints prevent it from increasing with counting time faster than @xmath30 @xcite . it therefore proves to be unsuitable as a measure for fractal exponents @xmath31 ; it also suffers from bias for finite - length data sets @xcite . for these reasons we employ other count - based measures . the reliable estimation of a fractal exponent that may assume a value greater than unity requires the use of a measure whose increase is not constrained as it is for the fano factor , and which remains free of bias . in this section we present a measure we first defined in 1996 @xcite , and called the allan factor . the allan factor is the ratio of the event - number allan variance to twice the mean : @xmath32 ^ 2\right\}}}{\displaystyle { 2 { \rm e}\left[z_n(t)\right]}}}.\ ] ] the allan variance was first introduced in connection with the stability of atomic - based clocks @xcite . it is defined in terms of the variability of differences of successive counts ; as such it is a measure based on the haar wavelet . because the allan factor functions as a derivative , it has the salutary effect of mitigating linear against nonstationarities . more complex wavelet allan factors can be constructed to eliminate polynomial trends @xcite . like the fano factor , the allan factor is also a useful measure of the degree of event clustering ( or anticlustering ) in a point process relative to the benchmark homogeneous poisson point process , for which @xmath33 for all @xmath0 . in fact , for any point process , the allan factor is simply related to the fano factor by @xmath34 so that , in general , both quantities vary with the counting time @xmath0 . in particular , for a regular point process the allan factor also approaches unity as @xmath0 approaches zero . for a fractal - rate stochastic point process and sufficiently large @xmath0 , the allan factor exhibits a power - law dependence that varies with the counting time @xmath0 as @xmath35 ( @xmath36 ) ; it can rise as fast as @xmath37 and can therefore be used to estimate fractal exponents over the expanded range @xmath36 . fourier - transform methods provide another avenue for quantifying correlation in a point process . the periodogram is an estimate of the power spectral density of a point process , revealing how the power is concentrated across frequency . the count - based periodogram is obtained by dividing a data set into contiguous segments of equal length @xmath38 . within each segment , a discrete - index sequence @xmath39 is formed by further dividing @xmath38 into @xmath40 equal bins , and then counting the number of events within each bin . a periodogram is then formed for each of the segments according to @xmath41 where @xmath42 is the discrete fourier transform of the sequence @xmath39 and @xmath40 is the length of the transform . all of the segment periodograms are averaged together to form the final averaged periodogram @xmath43 , which estimates the power spectral density in the frequency range from @xmath44 to @xmath45 hz . the periodogram @xmath43 can also be smoothed by using a suitable windowing function @xcite . the count - based periodogram , as opposed to the interval - based periodogram ( formed by fourier transforming the interevent intervals directly ) , provides direct undistorted information about the time correlation of the underlying point process because the count index increases by unity every @xmath46 seconds , in proportion to the real time of the point process . in the special case when the bin width @xmath47 is short in comparison with most interevent intervals @xmath13 , the count - based periodogram essentially reduces to the periodogram of the point process itself , since the bins reproduce the original point process to a good approximation . for a fractal - rate stochastic point process , the periodogram exhibits a power - law dependence that varies with the frequency @xmath48 as @xmath49 ; unlike the fano and allan factor exponents , however , @xmath50 can assume any value . thus in theory the periodogram can be used to estimate any value of fractal exponent , although in practice fractal exponents @xmath29 rarely exceed a value of @xmath51 . compared with estimated based on the allan factor , periodogram - based estimates of the fractal exponent @xmath50 suffer from increased bias and variance @xcite . other methods also exist for investigating the spectrum of a point process , some of which highlight fluctuations about the mean rate @xcite . for a fractal - rate stochastic point process with @xmath52 , the theoretical fano factor , allan factor , and periodogram curves all follow power - law forms with respect to their arguments , and in fact we obtain @xmath53 . for @xmath54 , the theoretical fano factor curves saturate , but the relation @xmath55 still obtains . the fractal exponent @xmath29 is ambiguously related to the hurst exponent @xmath17 , since some authors have used the quantity @xmath17 to index fractal gaussian noise whereas others have used the same value of @xmath17 to index the integral of fractal gaussian noise ( which is fractional brownian motion ) . the relationship between the quantities is @xmath56 for fractal gaussian noise and @xmath57 for fractal brownian motion . in the context of this paper , the former relationship holds , and we can define another estimate of the fractal exponent , @xmath58 , where @xmath59 is the estimate of the hurst exponent @xmath17 obtained from the data at hand . in general , @xmath60 depends on the theoretical value of @xmath29 , as well as on the probability distribution of the interevent intervals . the distributions of the data analyzed in this paper , however , prove simple enough so that the approximate theoretical relation @xmath61 will hold in the case of large amounts of data . second - order methods prove useful in revealing correlations between sequences of events , which indicate how information is shared between pairs of spike trains . such methods may not detect subtle forms of interdependence to which information - theoretic approaches are sensitive @xcite , but the latter methods suffer from limitations due to the finite size of the data sets used . we consider two second - order methods here : the normalized wavelet cross - correlation function ( nwccf ) and the cross periodogram . we define the normalized wavelet cross - correlation function @xmath62 as a generalization of the allan factor ( see sec . [ afdef ] ) . it is a haar - wavelet - based version of the correlation function and is therefore insensitive to linear trends . it can be readily generalized by using other wavelets and can thereby be rendered insensitive to polynomial trends . to compute the normalized wavelet cross - correlation function at a particular counting time @xmath0 , the two spike trains first are divided into contiguous counting windows @xmath0 . the number of spikes @xmath63 falling within the @xmath21th window is registered for all indices @xmath21 corresponding to windows lying entirely within the first spike - train data set , much as in the procedure to estimate the allan factor . this process is repeated for the second spike train , yielding @xmath64 . the difference between the count numbers in a given window in the first spike train @xmath65 and the one after it @xmath66 is then computed for all @xmath21 , with a similar procedure followed for the second spike train . paralleling the definition of the allan factor , the normalized wavelet cross - correlation function is defined as : @xmath67 \left[z_{2,n}(t ) - z_{2,n+1}(t)\right ] \right\}}}{\displaystyle { 2 \left\{{\rm e}\left[z_{1,n}(t)\right ] { \rm e}\left[z_{2,n}(t)\right]\right\}^{1/2}}}}.\ ] ] the normalization has two salutary properties : 1 ) it is symmetric in the two spike trains , and 2 ) when the same homogeneous poisson point process is used for both spike trains the normalized wavelet cross - correlation function assumes a value of unity for all counting times @xmath0 , again in analogy with the allan factor . to determine the significance of a particular value for the normalized wavelet cross - correlation function , we make use of two surrogate data sets : a shuffled version of the original data sets ( same interevent intervals but in a random order ) , and homogeneous poisson point processes with the same mean rate . comparison between the value of the normalized wavelet cross - correlation function obtained from the data at a particular counting time @xmath0 on the one hand , and from the surrogates at that time @xmath0 on the other hand , indicates the significance of that particular value . the cross periodogram @xcite is a generalization of the periodogram for individual spike trains ( see sec . [ pgdef ] ) , in much the same manner as the normalized wavelet cross - correlation function derives from the allan factor . two data sets are divided into contiguous segments of equal length @xmath68 , with discrete - index sequences @xmath69 and @xmath70 formed by further dividing each segment of both data sets into @xmath40 equal bins , and then counting the number of events within each bin . with the @xmath40-point discrete fourier transform of the sequence @xmath69 denoted by @xmath71 ( and similarly for the second sequence ) , we define the segment cross periodograms as @xmath72 = \frac{1}{m } { \rm re } \left [ \widetilde{w_1}^*(f ) \widetilde{w_2}(f ) \right],\ ] ] where @xmath73 represents complex conjugation and @xmath74 represents the real part of the argument . as with the ordinary periodogram , all of the segment cross periodograms are averaged together to form the final averaged cross periodogram , @xmath75 , and the result can be smoothed . this form is chosen to be symmetric in the two spike trains , and to yield a real ( although possibly negative ) result . in the case of independent spike trains , the expected value of the cross periodogram is zero . we again employ the same two surrogate data sets ( shuffled and poisson ) to provide significance information about cross - periodogram values for actual data sets . the cross periodogram and normalized wavelet cross - correlation function will have different immunity to nonstationarities and will exhibit different bias - variance tradeoffs , much as their single - dimensional counterparts do @xcite . we have carried out a series of experiments to determine the statistical characteristics of the dark , maintained , and driven neural discharge in cat rgc and lgn cells . using the analysis techniques presented in sec . [ sec : analysis ] , we compare and contrast the neural activity for these three different stimulus modalities , devoting particular attention to their fractal features . the results we present all derive from on - center x - type cells . the experimental methods are similar to those used by kaplan and shapley @xcite and teich _ et al . experiments were carried out on adult cats . anesthesia was induced by intramuscular injection of xylazine ( rompun 2 mg / kg ) , followed 10 minutes later by intramuscular injection of ketamine hcl ( ketaset 10 mg / kg ) . anesthesia was maintained during surgery with intravenous injections of thiamylal ( surital 2.5% ) or thiopental ( pentothal 2.5% ) . during recording , anesthesia was maintained with pentothal ( 2.5% , 26 ( mg / kg)/hr ) . the local anesthetic novocain was administered , as required , during the surgical procedures . penicillin ( 750,000 units intramuscular ) was also administered to prevent infection , as was dexamethasone ( decadron , 6 mg intravenous ) to forestall cerebral edema . muscular paralysis was induced and maintained with gallium triethiodide ( flaxedil , 515 ( mg / kg)/hr ) or vecuronium bromide ( norcuron , 0.25 ( mg / kg)/hr ) . infusions of ringer s saline with 5% dextrose at 34 ( ml / kg)/hr were also administered . the two femoral veins and a femoral artery were cannulated for intravenous drug infusions . heart rate and blood pressure , along with expired co@xmath76 , were continuously monitored and maintained in physiological ranges . for male cats , the bladder was also cannulated to monitor fluid outflow . core body temperature was maintained at 37.5@xmath77 c throughout the experiment by wrapping the animal s torso in a dc heating pad controlled by feedback from a subscapular temperature probe . the cat s head was fixed in a stereotaxic apparatus . the trachea was cannulated to allow for artificial respiration . to minimize respiratory artifacts , the animal s body was suspended from a vertebral clamp and a pneumothorax was performed when needed . eyedrops of 10% phenylephrine hydrochloride ( neo - synephrine ) and 1% atropine were applied to dilate the pupils and retract the nictitating membranes . gas - permeable hard contact lenses protected the corneas from drying . artificial pupils of 3-mm diameter were placed in front of the contact lenses to maintain fixed retinal illumination . the optical quality of the animal s eyes was regularly examined by ophthalmoscopy . the optic discs were mapped onto a tangent screen , by back - projection , for use as a positional reference . the animal viewed a crt screen ( tektronix 608 , 270 frames / sec ; or conrac , 135 frames / sec ) that , depending on the stimulus condition , was either dark , uniformly illuminated with a fixed luminance level , or displayed a moving grating . a craniotomy was performed over the lgn ( center located 6.5 mm anterior to the earbars and 9 mm lateral to the midline of the skull ) , and the dura mater was resected . a tungsten - in - glass microelectrode ( 510-@xmath78 m tip length ) @xcite was lowered until spikes from a single lgn neuron were isolated . the microelectrode simultaneously recorded rgc activity , in the form of s potentials , and lgn spikes , with a timing accuracy of 0.1 msec . the output was amplified and monitored using conventional techniques . a cell was classified as y - type if it exhibited strong frequency doubling in response to contrast - reversing high - spatial - frequency gratings , and x - type otherwise @xcite . the experimental protocol was approved by the animal care and use committee of rockefeller university , and was in accord with the national institutes of health guidelines for the use of higher mammals in neuroscience experiments . results for simultaneously recorded rgc and target lgn spike trains of 4000-sec duration are presented in fig . 3 , when the retina is thoroughly adapted to the dark ( this is referred to as the `` dark discharge '' ) . the normalized rate functions ( a ) for both the rgc ( solid curve ) and lgn ( dashed curve ) recordings exhibit large fluctuations over the course of the recording ; each window corresponds to a counting time of @xmath79 sec . such large , slow fluctuations often indicate fractal rates @xcite . the two recordings bear a substantial resemblance to each other , suggesting that the fractal components of the rate fluctuations either have a common origin or pass from one of the cells to the other . the normalized interevent - interval histogram ( b ) of the rgc data follows a straight - line trend on a semi - logarithmic plot , indicating that the interevent - interval probability density function is close to an exponential form . the lgn data , however , yields a nonmonotonic ( bimodal ) interevent - interval histogram . this distribution favors longer and shorter intervals at the expense of those near half the mean interval , reflecting clustering in the event occurrences over the short term . various kinds of unusual clustering behavior have been previously observed in lgn discharges @xcite . r / s plots ( c ) for both the rgc and lgn recordings follow the @xmath80 line for sums less than 1000 intervals , but rise sharply thereafter in a roughly power - law fashion as @xmath81 , suggesting that the neural firing pattern exhibits fractal activity for times greater than about 1000 intervals ( about 120 sec for these two recordings ) . both smoothed periodograms ( d ) decay with frequency as @xmath82 for small frequencies , and the allan factors ( e ) increase with time as @xmath83 for large counting times , confirming the fractal behavior . the 0.3-hz component evident in the periodograms of both recordings is an artifact of the artificial respiration ; it does not affect the fractal analysis . as shown in table 1 , the fractal exponents calculated from the various measures bear rough similarity to each other , as expected @xcite ; further , the onset times also agree reasonably well , being in the neighborhood of 100 sec . the coherence among these statistics leaves little doubt that these rgc and lgn recordings exhibit fractal features with estimated fractal exponents of @xmath84 and @xmath85 ( mean @xmath86 standard deviation of the three estimated exponents ) , respectively . moreover , the close numerical agreement of the rgc and lgn estimated fractal exponents suggests a close connection between the fractal activity in the two spike trains under dark conditions @xcite . curves such as those presented in fig . 3 are readily simulated by using a fractal - rate stochastic point process , as described in @xcite . with the exception of the interevent - interval distribution , it is apparent from fig . 3 that the statistical properties of the dark discharges generated by the rgc and its target lgn cell prove to be remarkably similar . figure 4 presents analogous statistical results for simultaneously recorded maintained - dis/-charge rgc and target - lgn spike trains of 7000-sec duration when the stimulus presented by the crt screen was a 50 cd / m@xmath87 uniform luminance . the cell pair from which these recordings were obtained is different from the pair whose statistics are shown in fig . as is evident from table 1 , the imposition of a stimulus increases the rgc firing rate , though not that of the lgn . in contrast to the results for the dark discharge , the rgc and lgn action - potential sequences differ from each other in significant ways under maintained - discharge conditions . we previously investigated some of these statistical measures , and their roles in revealing fractal features , for maintained discharge @xcite . the rate fluctuations ( a ) of the rgc and the lgn no longer resemble each other . at these counting times , the normalized rgc rate fluctuations are suppressed , whereas those of the lgn are enhanced , relative to the dark discharge shown in fig . 3 . significant long - duration fluctuations are apparently imparted to the rgc s - potential sequence at the lgn , through the process of selective clustered passage @xcite . spike clustering is also imparted at the lgn over short time scales ; the rgc maintained discharge exhibits a coefficient of variation ( cv ) much less than unity , whereas that of the lgn significantly exceeds unity ( see table 1 ) . the normalized interevent - interval histogram ( b ) of the rgc data resembles that of a dead - time - modified poisson point process ( fit not shown ) , consistent with the presence of relative refractoriness which becomes more important at higher rates @xcite . dead - time effects in the lgn are secondary to the clustering that it imparts to the rgc s - potentials , in part because of its lower rate . the r / s ( c ) , periodogram ( d ) , and allan factor ( e ) plots yield results that are consistent with , but different from , those revealed by the dark discharge shown in fig . 3 . although both the rgc and lgn recordings exhibit evidence of fractal behavior , the two spike trains now behave quite differently in the presence of a steady - luminance stimulus . for the rgc recording , all three measures are consistent with a fractal onset time of about 1 sec , and a relatively small fractal exponent ( @xmath88 ) . for the lgn , the fractal behavior again appears in all three statistics , but begins at a larger onset time ( roughly 20 sec ) and exhibits a larger fractal exponent ( @xmath89 ) . again , all measures presented in fig . 4 are well described by a pair of fractal - rate stochastic point processes @xcite . figure 5 presents these same statistical measures for simultaneously recorded 7000-sec duration rgc and lgn spike trains in response to a sinusoidal stimulus ( drifting grating ) at 4.2 hz frequency , 40% contrast , and 50 cd / m@xmath87 mean luminance . the rgc / lgn cell pair from which these recordings were obtained is the same as the pair illustrated in fig . 4 . the results for this stimulus resemble those for the maintained discharge , but with added sinusoidal components associated with the restricted phases of the stimulus during which action potentials occur . using terminology from auditory neurophysiology , these spikes are said to be `` phase locked '' to the periodicity provided by the drifting - grating stimulus . the firing rate is greater than that observed with a steady - luminance stimulus , particularly for the lgn ( see table 1 ) . again , the rgc and lgn spike trains exhibit different behavior . the rate fluctuations ( a ) of the lgn still exceeds those of the rgc , but not to as great an extent as in fig . 4 . both action - potential sequences exhibit normalized interevent - interval histograms ( b ) with multiple maxima , but the form of the histogram is now dominated by the modulation imposed by the oscillatory stimulus . over long times and small frequencies , the r / s ( c ) , periodogram ( d ) , and allan factor ( e ) plots again yield results in rough agreement with each other , and also with the results presented in fig . the most obvious differences arise from the phase locking induced by the sinusoidal stimulus , which appears directly in the periodogram as a large spike at 4.2 hz , and in the allan factor as local minima near multiples of @xmath90 sec . the rgc results prove consistent with a fractal onset time of about 3 sec , and a relatively small fractal exponent ( @xmath91 ) , whereas for the lgn the onset time is about 20 sec and the fractal exponent is @xmath92 . for both spike trains fractal behavior persists in the presence of the oscillatory stimulus , though its magnitude is slightly attenuated . we previously examined information exchange among pairs of rgc and lgn spike trains using information - theoretic measures @xcite . while these approaches are very general , finite data length renders them incapable of revealing relationships between spike trains over time scales longer than about 1 sec . we now proceed to investigate various rgc and lgn spike - train pairs in terms of the correlation measures for pairs of point processes developed in sec . [ sec : corrmeas ] . pairs of rgc discharges are only weakly correlated over long counting times . this is readily illustrated in terms of normalized rate functions such as those presented in fig . 6a , in which the rate functions of two rgcs are computed over a counting time @xmath79 sec . calculation of the correlation coefficient ( @xmath93 ) shows that the fluctuations are only mildly correlated . unexpectedly , however , significant correlation turns out to be present in pairs of lgn discharges over long counting times . this is evident in fig . 6b , where the correlation coefficient @xmath94 ( @xmath95 ) for the rates of two lgn discharges computed over the same counting time @xmath79 sec . for shorter counting times , there is little cross correlation for either pairs of rgc or of lgn spike trains ( not shown ) . however , strong correlations are present in the spike rates of an rgc and its target lgn cell as long as the rate is computed over times shorter than 15 sec for this particular cell pair . the cross correlation can be quantified at all time and frequency scales by the normalized wavelet cross - correlation function ( see sec . [ sec : nwccf ] ) and the cross periodogram ( see sec . [ sec : cpg ] ) , respectively . figure 6c shows the normalized wavelet cross - correlation function , as a function of the duration of the counting window , between an rgc / lgn spike - train pair recorded under maintained - discharge conditions , as well as for two surrogate data sets ( shuffled and poisson ) . for this spike - train pair , it is evident that significant correlation exists over time scales less than 15 seconds . the constant magnitude of the normalized wavelet cross - correlation function for @xmath96 sec is likely associated with the selective transmission properties of the lgn @xcite . figure 6d presents the normalized wavelet cross - correlation function for the same rgc / lgn spike - train pair shown in fig . 6c ( solid curve ) , together with that between two rgc action - potential sequences ( long - dashed curve ) , and between their two associated lgn spike trains ( short - dashed curve ) . also shown is a dotted line representing the aggregate behavior of the normalized wavelet cross - correlation function absolute magnitude for all surrogate data sets , which resemble each other . while the two rgc spike trains exhibit a normalized wavelet cross - correlation function value which remains below 7 , the two lgn action - potential sequences yield a curve that steadily grows with increasing counting window @xmath0 , attaining a value in excess of 1000 . indeed , a logarithmic scale was chosen for the ordinate to facilitate the display of this wide range of values . it is of interest to note that the lgn / lgn curve begins its steep ascent just as the rgc / lgn curve abruptly descends . further , the normalized wavelet cross - correlation function between the two lgn recordings closely follows a power - law form , indicating that the two lgn action - potential rates are co - fractal . one possible origin of this phenomenon is a fractal form of correlated modulation of the random - transmission processes in the lgn that results in the two lgn spike trains . some evidence exists that global modulation of the lgn might originate in the parabrachial nucleus of the brain stem ; the results presented here are consistent with such a conclusion . analogous results for the cross - periodograms , which are shown in figs . 6e and f , provide results that corroborate , but are not as definitive as , those obtained with the normalized wavelet cross - correlation function . the behavior of the normalized wavelet cross - correlation functions for pairs of driven spike trains , shown in fig . 7 , closely follow those for pairs of maintained discharges , shown in fig . 6 , except for the presence of structure at the stimulus period imposed by the drifting grating . the presence of a stimulus alters the manner in which spike trains in the visual system exhibit fractal behavior . in the absence of a stimulus , rgc and lgn dark discharges display similar fractal activity ( see fig . the normalized rate functions of the two recordings , when computed for long counting times , follow similar paths . the r / s , allan factor , and periodogram quantify this relationship , and these three measures yield values of the fractal exponents for the two spike trains that correspond reasonably well ( see table 1 ) . the normalized interevent - interval histogram , a measure which operates only over relatively short time scales , shows a significant difference between the rgc and lgn responses . such short - time behavior , however , does not affect the fractal activity , which manifests itself largely over longer time scales . the presence of a stimulus , either a constant luminance ( fig . 4 ) , or a drifting grating ( fig . 5 ) , causes the close linkage between the statistical character of the rgc and lgn discharges over long times to dissipate . the normalized rate functions of the lgn spike trains display large fluctuations about their mean , especially for the maintained discharge , while the rgc rate functions exhibit much smaller fluctuations that are minimally correlated with those of the lgn . again , the r / s , allan factor , and periodogram quantify this difference , indicating that fractal activity in the rgc consistently exhibits a smaller fractal exponent ( see also table 1 ) , and also a smaller fractal onset time ( higher onset frequency ) . both the r / s and allan - factor measures indicate that the lgn exhibits more fluctuations than the rgc at all scales ; the periodogram does not , apparently because it is the only one of the three constructed without normalization . in the driven case ( fig . 5 ) , the oscillatory nature of the stimulus phase - locks the rgc and lgn spike trains to each other at shorter time scales . the periodogram displays a peak at 4.2 hz , and the allan factor exhibits minima at multiples of @xmath90 sec , for both action - potential sequences . the normalized interevent - interval histogram also suggests a relationship between the two recordings mediated by the time - varying stimulus ; both rgc and lgn histograms achieve a number of maxima . although obscured by the normalization , the peaks do indeed coincide for an unnormalized plot ( not shown ) . in the presence of a stimulus , rgcs are not correlated with their target lgn cells over the long time scales at which fractal behavior becomes most important , but significant correlation does indeed exist between pairs of lgn spike trains for both the maintained and driven discharges ( see figs . 6 and 7 , respectively ) . these pairs of lgn discharges , exhibiting linked fractal behavior , may be called co - fractal . the normalized wavelet cross - correlation function and cross periodogram plots between rgc 1 and lgn 1 remain significantly above the surrogates for small times ( figs . 6c and 6e ) . the results for the two rgcs suggest some degree of co - fractal behavior , but no significant correlation over short time scales for the maintained discharge ( figs . 6d and 6f ) . since the two corresponding rgc spike trains do not appear co - fractal nearly to the degree shown by the lgn recordings , the co - fractal component must be imparted at the lgn itself . this suggests that the lgn discharges may experience a common fractal modulation , perhaps provided from the parabrachial nucleus in the brain stem , which engenders co - fractal behavior in the lgn spike trains . although similar data for the dark discharge are not available , the tight linkage between rgc and lgn firing patterns in that case ( fig . 3 ) suggests that a common fractal modulation may not be present in the absence of a stimulus , and therefore that discharges from nearby lgn cells would in fact not be co - fractal ; this remains to be experimentally demonstrated . correlations in the spike trains of relatively distant pairs of cat lgn cells have been previously observed in the short term for drifting - grating stimuli @xcite ; these correlations have been ascribed to low - threshold calcium channels and dual excitatory / inhibitory action in the corticogeniculate pathway @xcite . in the context of information transmission , the lgn may modulate the fractal character of the spike trains according to the nature of the stimulus present . under dark conditions , with no signal to be transmitted , the lgn appears to pass the fractal character of the individual rgcs on to more central stages of visual processing , which could serve to keep them alert and responsive to all possible input time scales . if , as appears to be the case , the responses from different rgcs do not exhibit significant correlation with each other , then the lgn spike trains also will not , and the ensemble average , comprising a collection of lgn spike trains , will display only small fluctuations . in the presence of a constant stimulus , however , the lgn spike trains develop significant degrees of co - fractal behavior , so that the ensemble average will exhibit large fluctuations @xcite . such correlated fractal behavior might serve to indicate the presence of correlation at the visual input , while still maintaining fluctuations over all time scales to ready neurons in later stages of visual processing for any stimulus changes that might arrive . finally , a similar behavior obtains for a drifting - grating stimulus , but with somewhat reduced fractal fluctuations ; perhaps the stimulus itself , though fairly simple , serves to keep more central processing stages alert . fractal behavior is present in all 50 of the rgc and lgn neural spike - train pairs that we have examined , under dark , maintained - discharge , and drifting - grating stimulus conditions , provided they are of sufficient length to manifest this behavior . indeed , fractal behavior is ubiquitous in sensory systems . its presence has been observed in cat striate - cortex neural spike trains @xcite ; and in the spike train of a locust visual interneuron , the descending contralateral movement detector @xcite . it is present in the auditory system @xcite of a number of species ; primary auditory ( viii - nerve ) nerve fibers in the cat @xcite , chinchilla , and chicken @xcite all exhibit fractal behavior . it is exhibited at many biological levels , from the microscopic to the macroscopic ; examples include ion - channel behavior @xcite , neurotransmitter exocytosis at the synapse @xcite , and spike trains in rabbit somatosensory - cortex neurons @xcite and mesencephalic reticular - formation neurons @xcite . in almost all cases , the upper limit of the observed time over which fractal correlations exist is imposed by the duration of the recording . the significance of the fractal behavior is not fully understood . its presence may serve as a stimulus to keep more central stages of the sensory system alert and responsive to all possible time scales , awaiting the arrival of a time - varying stimulus whose time scale is _ a priori _ unknown . it is also possible that fractal activity in spike trains provides an advantage in terms of matching the detection system to the expected signal @xcite since natural scenes have fractal spatial and temporal noise @xcite . using a variety of statistical measures , we have shown that fractal activity in lgn spike trains remains closely correlated with that of their exciting rgc action - potential sequences under dark conditions , but not with stimuli present . the presence of a visual stimulus serves to increase long - duration fluctuations in lgn spike trains in a coordinated fashion , so that pairs of lgn spike trains exhibit co - fractal behavior largely uncorrelated with activity in their associated rgcs . such large correlations are not present in pairs of rgc spike trains . a drifting - grating stimulus yields similar results , but with fractal activity in both recordings somewhat suppressed . co - fractal behavior in lgn discharges under constant luminance and drifting - grating stimulus conditions suggests that a common fractal modulation may be imparted at the lgn in the presence of a visual stimulus . this work was supported by the u.s . office of naval research under grants n00014 - 92-j-1251 and n0014 - 93 - 12079 , by the national institute for mental health under grant mh5066 , by the national eye institute under grants ey4888 and ey11276 , and by the whitaker foundation under grant rg-96 - 0411 . e. kaplan is jules and doris stein research - to - prevent - blindness professor at mt . sinai school of medicine . 99 mastronarde , d. n. ( 1983 ) _ j. neurophysiol . _ * 49 * , 303324 . bishop , p. o. , levick , w. r. , and williams , w. o. ( 1964 ) _ j. physiol . ( london ) _ * 170 * , 598612 . kuffler , s. w. , fitzhugh , r. , and barlow , h. b. ( 1957 ) _ j. gen . physiol . _ * 40 * , 683702 . levine , m. w. , and troy , j. b. ( 1986 ) _ j. physiol . ( london ) _ * 375 * , 339359 . troy , j. b. , and robson , j. g. ( 1992 ) _ vis . neurosci . _ * 9 * , 535553 . teich , m. c. , heneghan , c. , lowen , s. b. , ozaki , t. , and kaplan , e. ( 1997 ) _ j. opt . soc . a _ * 14 * , 529546 . robson , j. g. , and troy , j. b. ( 1987 ) _ j. opt . soc . am . a _ * 4 * , 23012307 . levick , w. r. ( 1973 ) _ in _ handbook of sensory physiology , * vii/3 * , central processing of visual information , part a ( jung , r. , ed . ) , pp . 575598 , springer - verlag , new york . teich , m. c. , matin , l. , and cantor , b. ( 1978 ) _ j. opt . * 68 * , 386402 . teich , m. c. , and khanna , s. m. ( 1985 ) _ j. acoust . soc . _ * 77 * , 11101128 . thurner , s. , lowen , s. b. , heneghan , c. , feurstein , m. c. , feichtinger , h. g. , and teich , m. c. ( 1997 ) _ fractals _ * 5 * , 565595 . hurst , h. e. ( 1951 ) _ trans . eng . _ * 116 * , 770808 . feller , w. ( 1951 ) _ ann . * 22 * , 427432 . mandelbrot , b. b. ( 1983 ) the fractal geometry of nature , freeman , new york . schepers , h. e. , van beek , j. h. g. m. , and bassingthwaighte , j. b. ( 1992 ) _ ieee eng . med . biol . mag . _ * 11 * , 5771 . beran , j. ( 1994 ) statistics for long - memory processes , chapman and hall , new york . bassingthwaighte , j. b. , and raymond , g. m. ( 1994 ) _ ann . biomed . * 22 * , 432444 . fano , u. ( 1947 ) _ phys . _ * 72 * , 2629 . lowen , s. b. , and teich , m. c. ( 1996 ) _ j. acoust . soc . * 99 * , 35853591 . lowen , s. b. , and teich , m. c. ( 1995 ) _ fractals _ * 3 * , 183210 . allan , d. w. ( 1966 ) _ proc . ieee _ * 54 * , 221230 . teich , m. c. , heneghan , c. , lowen , s. b. , and turcott , r. g. ( 1996 ) _ in _ wavelets in medicine and biology ( aldroubi , a. , and unser , m. , eds . ) , pp . 383412 , crc press , boca raton , fl . abry , p. and flandrin , p ( 1996 ) _ in _ wavelets in medicine and biology ( aldroubi , a. , and unser , m. , eds . ) , pp . 413437 , crc press , boca raton , fl . oppenheim , a. v. , and schafer , r. w. ( 1975 ) digital signal processing , prentice - hall , englewood cliffs , nj . lange , g. d. , and hartline , p. h. ( 1979 ) _ biol . cybern . _ * 34 * , 3134 . lowen , s. b. , ozaki , t , kaplan , e , and teich , m. c. ( 1998 ) _ in _ computational neuroscience : trends in research , 1998 ( bower , j. m. , ed . ) , pp . 491496 , plenum , new york . tuckwell , h. c. ( 1989 ) stochastic processes in the neurosciences , society for industrial and applied mathematics , philadelphia . kaplan , e. , and shapley , r. m. ( 1982 ) _ j. physiol . ( london ) _ * 330 * , 125143 . merrill , e. g. , and ainsworth , a. ( 1972 ) _ med . * 10 * , 662672 . hochstein , s. , and shapley , r. m. ( 1976 ) _ j. physiol . ( london ) _ * 262 * , 237264 . shapley , r. m. , and hochstein , s. ( 1975 ) _ nature ( london ) _ * 256 * , 411413 . funke , k. , and wrgtter , f. ( 1997 ) _ prog . _ * 53 * , 67119 . sillito , a. m. , jones , h. e. , gerstein , g. l. , and west , d. c. ( 1994 ) _ nature ( london ) _ * 369 * , 479482 . kirkland , k. l. , and gerstein , g. l. ( 1998 ) _ vision res . _ * 38 * , 20072022 . teich , m. c. , turcott , r. g. , and siegel , r. m. ( 1996 ) _ ieee eng . med . biol . mag . _ * 15 * ( # 5 ) , 7987 . turcott , r. g. , barker , p. d. r. , and teich , m. c. ( 1995 ) _ j. statist . simul . _ * 52 * , 253271 . teich , m. c. ( 1989 ) _ ieee trans . eng . _ * 36 * , 150160 . kelly , o. e. , johnson , d. h. , delgutte , b. , and cariani , p. ( 1996 ) _ j. acoust . * 99 * , 22102220 . powers , n. l. , salvi , r. j. ( 1992 ) _ in _ abstracts of the fifteenth midwinter research meeting of the association for research in otolaryngology ( lim , d. j. , ed . ) , abstract 292 , p. 101 , association for research in otolaryngology , des moines , ia . luger , p. ( 1988 ) _ biophys . j. _ * 53 * , 877884 . millhauser , g. l. , salpeter , e. e. , and oswald , r. e. ( 1988 ) _ proc . ( usa ) _ * 85 * , 15031507 . liebovitch , l. s. , and t. i. tth ( 1990 ) _ ann . biomed . eng . _ * 18 * , 177194 . lowen , s. b. , and teich , m. c. ( 1993 ) _ in _ noise in physical systems and @xmath97 fluctuations , aip conference proceedings * 285 * ( handel , p. h. , and chung , a. l. , eds . ) , pp . 745748 , american institute of physics , new york . lowen , s. b. , cash , s. c. , poo , m .- m . , and teich , m. c. ( 1997 ) _ j. neurosci . _ * 17 * , 56665677 . wise , m. e. ( 1981 ) _ in _ statistical distributions in scientific work * 6 * ( taillie , c. e. a. , ed . ) , pp . 211231 , reidel , boston . grneis , f. , nakao , m. , mizutani , y. , yamamoto , m. , meesmann , m. , and musha , t. ( 1993 ) _ biol . cybern . _ * 68 * , 193198 . olshausen , b. a. , and field , d. j. ( 1996 ) _ network _ * 7 * , 333339 . dan , y. , atick , j. j. , and reid , r. c. ( 1996 ) _ j. neurosci . _ * 16 * , 33513362 . [ cols="<,^,>,^,^,^,^ " , ] neural - discharge statistics for cat retinal ganglion cells ( rgcs ) and their associated lateral geniculate nucleus ( lgn ) cells , under three stimulus conditions : dark discharge in the absence of stimulation ( data duration @xmath98 sec ) ; maintained discharge in response to a uniform luminance of 50 cd / m@xmath87 ( data duration @xmath99 sec ) ; and driven discharge in response to a drifting grating ( 4.2 hz frequency , 40% contrast , and 50 cd / m@xmath87 mean luminance ; data duration @xmath99 sec ) . all cells are on - center x - type . the maintained and driven data sets were recorded from the same rgc / lgn cell pair , whereas the dark discharge derived from a different cell pair . statistics , from left to right , are mean interevent interval , interevent - interval coefficient of variation ( cv = standard deviation divided by mean ) , and fractal exponents estimated by least - squares fits on doubly logarithmic plots of 1 ) the rescaled range ( r / s ) statistic for @xmath100 , which yields an estimate of the hurst exponent @xmath59 , and of @xmath60 , in turn , through the relation @xmath58 ; 2 ) the count - based periodogram for frequencies between 0.001 and 0.01 hz which yields @xmath50 ; and 3 ) of the allan factor for counting times between @xmath101 and @xmath102 where @xmath103 is the duration of the recording , which yields @xmath104 . ' '' '' * figure 1 * : rate estimates formed by dividing the number of events in successive counting windows by the counting time @xmath0 . the stimulus was a uniformly illuminated screen ( with no temporal or spatial modulation ) of luminance 50 cd / m@xmath87 . * a ) * rate estimate for a cat rgc generated using three different counting times ( @xmath105 1 , 10 , and 100 sec ) . the fluctuations in the rate estimates converge relatively slowly as the counting time is increased . this is characteristic of fractal - rate processes . the convergence properties are quantified by measures such as the allan factor and periodogram . * b ) * rate estimates from the same recording after the intervals are randomly reordered ( shuffled ) . this maintains the same relative frequency of interval sizes but destroys the original relative ordering of the intervals , and therefore any correlations or dependencies among them . for such nonfractal signals , the rate estimate converges more quickly as the counting time @xmath0 is increased . the data presented here are typical of the 50 data sets examined . * figure 2 * : a sequence of action potentials ( top ) is reduced to a set of events ( represented by arrows , middle ) that form a point process . a sequence of interevent intervals @xmath5 is formed from the times between successive events , resulting in a discrete - time , positive , real - valued stochastic process ( lower left ) . all information contained in the original point process remains in this representation , but the discrete - time axis of the sequence of interevent intervals is distorted relative to the real - time axis of the point process . the sequence of counts @xmath7 , a discrete - time , nonnegative , integer - valued stochastic process , is formed from the point process by recording the numbers of events in successive counting windows of duration @xmath0 ( lower right ) . this process of mapping the point process to the sequence @xmath7 results in a loss of information , but the amount lost can be made arbitrarily small by reducing @xmath0 . an advantage of this representation is that no distortion of the time axis occurs . * figure 3 * : statistical measures of the dark discharge from a cat on - center x - type retinal ganglion cell ( rgc ) and its associated lateral geniculate nucleus ( lgn ) cell , for data of duration @xmath98 sec . rgc results appear as solid curves , whereas lgn results are dashed . * a ) * normalized rate function constructed by counting the number of neural spikes occurring in adjacent 100-sec counting windows , and then dividing by 100 sec and by the average rate . * b ) * normalized interevent - interval histogram ( iih ) _ vs _ normalized interevent interval constructed by dividing the interevent intervals for each spike train by the mean , and then obtaining the histogram . * c ) * normalized range of sums @xmath15 _ vs _ number of interevent intervals @xmath14 ( see sec . [ rsdef ] ) . * d ) * periodogram @xmath43 _ vs _ frequency @xmath48 ( see sec . [ pgdef ] ) . * e ) * allan factor @xmath106 _ vs _ counting time @xmath0 ( see sec . [ afdef ] ) . * figure 4 * : statistical measures of the maintained discharge from a cat on - center x - type rgc and its associated lgn cell , at a steady luminance of 50 cd / m@xmath87 , for data of duration @xmath99 sec . this cell pair is different from the one illustrated in fig . 3 . the results for the rgc discharge appear as solid curves , whereas those for the lgn are presented as dashed curves . panels * a)e ) * as in fig . 3 . * figure 5 * : statistical measures of the driven discharge from a cat on - center x - type rgc and its associated lgn cell , for a drifting - grating stimulus with mean luminance 50 cd / m@xmath87 , 4.2 hz frequency , and 40% contrast , for data of duration @xmath99 sec . this cell pair is the same as the one illustrated in fig . the results for the rgc discharge appear as solid curves , whereas those for the lgn are presented as dashed curves . panels * a)e ) * as in figs . 3 and 4 . * figure 6 * : statistical measures of the maintained discharge from pairs of cat on - center x - type rgcs and their associated lgn cells , stimulated by a uniform luminance of 50 cd / m@xmath87 , for data of duration @xmath99 sec . rgc and lgn spike trains denoted `` 1 '' are those that have been presented in figs . 4 and 5 , while those denoted `` 0 '' are another simultaneously recorded pair . * a ) * normalized rate functions constructed by counting the number of neural spikes occurring in adjacent 100-sec counting windows , and then dividing by 100 sec and by the average rate , for rgc 1 and rgc 0 . note that the ordinate scale differs from that in ( a ) . * b ) * normalized rate functions for the two corresponding target lgn cells , lgn 1 and lgn 0 . * c ) * normalized wavelet cross - correlation function ( nwccf ) between the rgc 1 and lgn 1 recordings ( solid curve ) , shuffled surrogates of these two data sets ( long - dashed curve ) , and poisson surrogates ( short - dashed curve ) . unlike the allan factor @xmath106 , the normalized wavelet cross - correlation function can assume negative values and need not approach unity in certain limits . negative normalized wavelet cross - correlation function values for the data or the surrogates are not printed on this doubly logarithmic plot , nor are they printed in panel ( d ) . comparison between the value of the normalized wavelet cross - correlation function obtained from the data at a particular counting time @xmath0 on the one hand , and from the surrogates at that time @xmath0 on the other hand , indicates the significance of that particular value . * d ) * normalized wavelet cross - correlation functions between rgc 1 and lgn 1 ( solid curve , repeated from panel ( c ) , the two rgc spike trains ( long - dashed curve ) , and the two lgn spike trains ( short - dashed curve ) . also included is the aggregate behavior of both types of surrogates for all three combinations of recordings listed above ( dotted line ) . * e ) * cross periodograms of the data sets displayed in panel ( c ) . * f ) * cross periodograms of the data sets displayed in panel ( d ) . * figure 7 * : statistical measures of the driven discharge from pairs of cat on - center x - type rgcs and their associated lgn cells , stimulated by a drifting grating with a mean luminance of 50 cd / m@xmath87 , 4.2 hz frequency , and 40% contrast , for data of duration @xmath99 sec . rgc and lgn spike trains denoted `` 1 '' are recorded from the same cell pair that have been presented in figs . 46 , while those denoted `` 0 '' are recorded simultaneously from the other cell pair , that was presented in fig . panels * a)f ) * as in fig .

charge - transfer effects in photometry with wide field planetary camera 2 aboard the _ hubble space telescope _ are investigated by a comparison of wfpc2 observations with groundbased photometry for the galactic globular clusters tauri and . simple numerical formulae describing the fraction of lost signal as functions of position on the detector , stellar brightness , and the diffuse sky brightness recorded in an image are presented , and the resulting corrections are compared to those previously derived by whitmore & heyer ( 1997 , _ instrument science report wfpc2 97 - 08 _ ) . significant lost - charge effects are seen that are proportional to both the coordinate ( , the number of shifts along the parallel register during readout ) and the coordinate ( number of shifts along the serial register ) . a `` typical '' star image ( one containing photoelectrons ) near the center of a `` typical '' intermediate - length exposure ( one with a diffuse sky brightness ofe/pixel , obtained at a camera temperature of 88c ) loses approximately 2% of its electrons to charge traps during readout ; a star in the corner of the image most remote from the readout electronics loses twice that . the percentage of charge lost decreases as the star brightness or the diffuse sky brightness increases . charge losses during the brief period when wfpc2 was operated at a temperature of 76c were approximately 85% greater , but apart from that no significant change in the charge transfer losses with time during the first 3.5 years of wfpc2 s mission is evident , except possibly a weak effect for the very faintest star images . these results are quite similar to those of whitmore & heyer , which were based on a much smaller data set , but there are some differences in detail . even with the present set of corrections , additional sources of calibration uncertainty which i am unable identify or characterize with the available data probably limit the _ external _ accuracy of photometry from wfpc2 to of order 12% . = 10000 = 10000 # 1m#1 # 1ngc#1 # 1 and

astro-ph9810039

the ccds used in the wide field planetary camera 2 ( `` wfpc2 '' ) of the _ hubble space telescope _ are known to have imperfect charge - transfer efficiency ( `` cte '' ) . this property of the camera has adverse photometric consequences that have proven difficult to characterize quantitatively . the source of the problem appears to be chemical impurities in the crystal lattice of the detectors that trap some photoelectrons during exposure and readout , resulting in an underestimate of the original luminous flux impingent on the camera . these missing electrons eventually diffuse out of the traps and contribute to the background noise of that same exposure or a subsequent one . since the charge packets that represent stellar images recorded at high row numbers must be clocked across a large fraction of the chip during readout , it is to be expected that they will encounter more traps and will lose more electrons than images recorded at lower row numbers . on the other hand , in frames where the diffuse sky brightness is high such as in long exposures through broad - band filters charge traps over the entire detector area can be filled by sky - generated photoelectrons . under these circumstances , the net effect of the charge - transfer inefficiency would be a small overall lowering of the diffuse sky brightness perceived in the astronomical scene , while the apparent projection of the stellar images _ above _ the diffuse sky surface brightness could be comparatively unaffected . an early analysis suggested that in images where the sky brightness is low ( @xmath62030@xmath4e@xmath5/pixel ) a simple `` ramp '' correction proportional to the @xmath0-coordinate position of the star image , amounting to 4% per 800 rows , would approximately correct recorded fluxes for the lost electrons ; in images where the sky was bright ( @xmath7250@xmath4e@xmath5/pixel ) , no correction was needed ; and in cases of moderate sky brightness an intermediate correction was recommended ( holtzman 1995a ) . while holtzman s ramp model corrects for the first - order effects of charge loss during readout , second - order effects may be significant . for instance , one would expect any given charge trap to absorb only a limited number of electrons before their mutual repulsion effectively blocked the ingestion of any additional charge . thus , faint star images consisting of comparatively few photoelectrons should lose a larger fraction of their total charge than images of bright stars containing many photoelectrons . one would therefore expect faint stars to require a larger fractional correction than bright stars near the same row number and in the presence of the same sky brightness . on the other hand , the large electron swarm corresponding to the image of a bright star would occupy a somewhat larger volume within the three - dimensional body of the detector , and might therefore actually enounter more lattice imperfections than would be accessible to smaller electron clouds . consequently , while we might still expect that brighter stars will require smaller magnitude corrections than fainter ones , it may be that the correction will approach zero more slowly than the expected simple exponential dependence as arbitrarily bright stars are considered . in addition , while it has been demonstrated that a given star s electron swarm can lose some charge while being clocked through some number of rows along a ccd s parallel register , it is also possible that said packet may lose yet more charge while being transferred through some number of columns along the serial register . again , this second ramp correction , presumably proportional to the star s @xmath1-coordinate position in the digital image , may also depend upon the sky brightness in the scene and on the brightness of the star image itself . the numerical coefficients of the @xmath0 and @xmath1 ramp corrections may be different , due to the dissimilar morphologies of the parallel and serial registers within the semiconductor material . whitmore & heyer ( 1997 ; wh97 ) have estimated the above first- and second - order effects of the charge - transfer inefficiency in wfpc2 , based upon a special sequence of observations of the globular cluster tauri (= ) . observations were obtained at four pointings of the telescope , with deliberate translations imposed so that a particular piece of the star field could be recorded on the four different chips during the course of the sequence . since the natural coordinate systems of the four detectors are rotated by 90 with respect to each other , this allowed several hundred star images to be used for a purely differential determination of the fraction of charge lost as a function of position on the detector , without a need to know the absolute brightness of any given object . wh97 presented their results in four alternative formulations , of which the most relevant to the present discussion is given by their eqs . 1 , 2d , and 3d : @xmath8 \hbox{\it cts}_{\hbox{\scriptsize observed}}\ ] ] @xmath9 @xmath10 here , _ cts _ represents total wfpc2 data numbers belonging to the stellar image within an aperture 05 in radius , as perceived with the gain = 7@xmath4e@xmath5/dn electronics ; @xmath1 and @xmath0 represent the position of the stellar image in the natural coordinate system of the detector ( pixel ( 1,1 ) being the first one to be read through the amplifier and ( 800,800 ) being the last , with the @xmath1 coordinate varying more rapidly ) ; and @xmath11 is the perceived diffuse sky brightness in the scene , exclusive of recognized stellar images . ( i have opportunistically adopted this version from the four variants presented , because following holtzman ( 1995b ) in reducing wfpc2 data i routinely correct measured magnitudes to their half - arcsecond - aperture equivalents , and because the data files produced by my software contain an estimate of the local sky brightness around and underlying each detected object . ) certain properties of these formulae may be noted . ( 1 ) the correction that is proportional to the @xmath0 position of the star depends only on the diffuse sky brightness in the image , and not on the brightness of the star itself . in contrast , the alternative @xmath0 ramp corrections presented for magnitudes as measured through 02 synthetic apertures depend upon both sky and star brightness . this change in formulation stems from the finding by wh97 that the difference between magnitudes measured in 02 and 05 apertures itself depends upon the stellar brightness : `` we find the cte loss is slightly reduced wthen the larger aperture is used , by about 1% for the bright stars and 3% for the faint stars . '' apparently , in the data studied by wh97 this effect tended to erase the magnitude dependence of the ramp corrections in the larger aperture . however , it should be noted that a simple differencing of the formulae provided by wh97 indicates that perfect cancellation can occur only at a particular value of the sky brightness , and will not be the case in general . ( 2 ) the correction that is proportional to the @xmath1 coordinate depends only on the brightness of the star and not at all on the diffuse brightness of the sky , a result not necessarily to be predicted from the electron - trap model of the charge - transfer inefficiency . ( 3 ) the @xmath0-coordinate ramp is proportional to an exponential function of the logarithm of the sky brightness , so ( a ) the correction tends asymptotically ( but slowly ) to zero as the sky brightness increases without limit , but ( b ) as the sky brightness decreases to zero , the correction @xmath12 this implies a non - physical result for a star of any apparent magnitude detected in the presence of a very low diffuse sky brightness . ( 4 ) the @xmath1-coordinate ramp is a linear function of the logarithm of the star counts , so ( a ) the correction again diverges as the number of star counts approaches zero . however , in this case the divergence is not a serious drawback , since photometry of any stellar image containing that few photoelectrons is unlikely to be of genuine astronomical interest . conversely , ( b ) the correction is identically zero when @xmath13 , and changes sign ( , the lattice imperfections in the serial register appear to act as electron sources rather than electron sinks ) for stellar images containing more than 21,000 dn within a half - arcsecond aperture . this is approximately the amount of flux contained within a star image whose central pixel just equals the digital saturation level of the analog - to - digital converters in wfpc2 when the gain@xmath14e@xmath5/dn electronics are used . therefore , this extrapolation , too , is probably of minimal practical relevance except possibly for very bright stars observed with the high - gain electronics bay . these extrapolations do emphasize , however , that the wh97 corrections are purely empirical constructs , and are not tied directly to a self - consistent physical model of the nature of the charge - transfer inefficiency . ( 5 ) the model correction makes no allowance for the possibility of a charge loss that is independent of position , regardless of whatever dependence it may have on either stellar brightness or sky flux . this feature of the model is a direct consequence of the experimental design , where apparent flux ratios were observed for individual stars recorded at various positions on the chips , without any consideration of ( a ) flux ratios for different stars with externally known magnitude differences , or ( b ) flux ratios for images of individual stars as recorded in frames with different exposure times . charge losses that are independent of position may be expected within the lattice - flaw model : simply by virtue of being formed in the body of the silicon wafer in the first place , the stellar image may lose some charge to traps located near the place of its formation . in addition , any inefficiency inherent in the transfer of charge packets from the top of a parallel register to the serial register , or from the end of the serial register into the on - chip amplifier , would be perceived as a charge loss that is independent of the original location of the stellar image on the detector . a hint that position - independent charge losses _ might _ be significant came from preliminary analyses of wfpc2 images of the globular clusters and palomar 4 ( see w. e. harris 1997 ; stetson 1998 , 1999 ) . a comparison of short - exposure ( @xmath15 s ) images to ground - based photometry of the same objects implied photometric zero - points consistent with those found by holtzman ( 1995b ) from similarly short exposures of the standard field . in contrast , comparison of longer ( @xmath16 s ) exposures of the same fields to the same ground - based photometry implied zero points that were different by approximately 0.05 mag , in the sense that the effective quantum efficiency of the detectors was higher in the longer exposures . a subsequent recalibration of the ground - based photometry , including the addition of many more independent ground - based and hst observations of , suggested that the original estimate of 0.05 mag might in fact have been high ; the additional data suggested that an anomaly of order 0.03 mag on average might be more nearly correct . with the publication of the wh97 corrections , it is apparent that the perceived short - minus - long zero - point difference might just be a reflection of the dependence of the fractional charge loss on the total number of recorded stellar photoelectrons and on the perceived surface flux of the sky . however , it remains possible that some residual position - independent charge loss could occur . in this paper i reinvestigate the photometric consequences of the charge - transfer inefficiency in wfpc2 , based on extensive series of observations of the galactic globular clusters tauri and . in addition to testing the validity of the corrections proposed by wh97 on a much larger data set , i investigate certain aspects of the problem not considered by them . the data analyzed in this paper are all in the public domain , and were acquired from the _ hst _ science archive of the canadian astronomy data centre . the observations include 318 individual wfpc2 exposures lying within 2 arcminutes of the nominal center of the standard calibration field in the nearby galactic globular tauri ( equinox 2000.0 coordinates 1325370 , 473538 according to h. c. harris 1993 ) . to these are added 91 exposures mostly centered within a few dozen arcseconds of 738068 , + 385155 : about 1 south of , but including , the center of the far - halo globular cluster . these latter data include all of the observations discussed by w. e. harris ( 1997 ) , plus some additional exposures of the cluster that were obtained by stsci staff specifically to assist in the calibration of wfpc2 . some differences between the present data set and that of whitmore & heyer ( 1997 ) deserve particular mention . ( 1 ) the wh97 analysis relied on eight observations made in each of the standard broad - band filters ( f336w , f439w , f555w , f675w , and f814w ) , as well as four observations in f606w intended to allow a comparison with a skyflat developed during the medium deep survey project . from each of the 44 exposures , data from only one of the four wfpc2 chips were considered in their analysis . the present study concentrates on observations made in the f555w and f814w filters (: 152 and 159 exposures , respectively ; : 27 and 58 exposures ) , because these are particularly relevant for the _ key project on the extragalactic distance scale _ and for studies of certain local group star clusters and galaxies in which i am especially interested . data from all four chips of each of the 402 @xmath17 and @xmath18 exposures were considered in the analysis below . seven observations of in the intermediate - band f547 m filter were also included in the reductions , to allow the eventual estimation of the fundamental photometric zero - points for this filter for a particular project , but these data were not utilized in the analysis of the charge - transfer inefficiency to be presented below . ( 2 ) the wh97 data were all taken on 1996 june 29 , whereas the data employed here were taken from 1994 january 11 through 1997 june 26 . the data were all obtained during either 1994 may 2122 ( the 28 w. e. harris observations ) , 1996 december 21 ( 6 exposures ) , or 1997 november 1819 ( 57 exposures ) . since the observation date determines the spacecraft roll angle required to direct the solar panels toward the sun while the telescope is pointed at a particular spot on the celestial sphere , the wh97 observations were all obtained at the same telescope orientation . the observations discussed here , having been obtained at essentially all times of the year , represent several full cycles of spacecraft roll angle . the observations were also taken at several different orientations . ( 3 ) the wh97 data were all taken with wfpc2 operating at a camera temperature of 88 c , whereas 116 of the 318 observations considered here were made at the warmer temperature of 76 c used early in wfpc2 s mission ; the remaining 202 observations and all 91 exposures were taken at 88 c. ( 4 ) the wh97 data were obtained with electronics bay 4 , which gives a nominal gain of 7@xmath4e@xmath5/dn , whereas 264 of the observations analysed here were obtained through electronics bay 3 ( gain approximately 14@xmath4e@xmath5/dn ) . the rest of the and all the observations were taken at 7@xmath4e@xmath5/dn . ( 5 ) the wh97 analysis was purely differential on a star - by - star basis , considering only the flux _ ratio _ between two observations of a given star as a function of the positional difference between the two stellar images in the natural coordinate systems of the detectors . the new analysis will assume that the true standard - system @xmath17 ( johnson ) and @xmath18 ( kron - cousins ) magnitudes are known _ a priori _ from ground - based observations of the two target fields . the transformation equations relating the natural - system magnitudes to the groundbased standard system will be determined by a robust least - squares technique , assuming that the color transformation coefficients of holtzman ( 1995b ; their table 7 ) are correct . the statistical solution of these equations will yield the fundamental zero points of the four wfpc2 ccds , and the residuals of individual stars from these solutions will serve as diagnostics of the charge - transfer inefficiency . ( 6 ) wh97 estimated the mean clear - sky surface brightness for a given frame from a star - free region near the center of the field ; this value was then considered appropriate for every star in the image . the software used in the present analysis estimates a unique local sky - brightness value for every star , from the mode of brightness values in a surrounding annulus if it is aperture photometry that is being derived , or from the median brightness value at and around the position of the star in a residual image from which all known stars have been subtracted in the profile - fitting analysis . because these individual , local brightness measurements are conveniently available , they will be used in the following analyses rather than a global average sky for the frame as a whole . for the globular - cluster images used here , there is little net difference between the two methods ( typically @xmath19e@xmath5/pixel ) . both the present study and that of wh97 include a few observations obtained with internal preflashes , whose purpose was to increase the leverage for determining the dependence of the charge loss on diffuse sky brightness : wh97 had three exposures in the f555w filter with some preflash , while the present sample includes those three plus another nine preflashed exposures of . in each case , the preflash produced a diffuse surface brightness averaging approximately 175@xmath4e@xmath5/pixel . in the long exposures of the sky brightness was typically of order 250@xmath4e@xmath5/pixel , and in all other cases the background level was quite close to zero . all wfpc2 images studied here were recalibrated by the canadian astronomy data centre , using the best bias and flat - field frames available at the time the data were requested . the intensities in the frames were corrected for geometric distortion , known defective pixels were masked , and the resulting images were multiplied by 4.0 and coverted to short integers as described by stetson ( 1998 ) . instrumental magnitudes were determined via the profile - fitting technique with the computer program allframe ( stetson 1994 ) , and selected bright , isolated stars were used to compute magnitude corrections to a system defined by half - arcsecond synthetic apertures via a growth - curve analysis ( stetson 1990 ) . all transformations from the wfpc2 instrumental system to the standard ground - based @xmath20 system were performed with a version of the program ccdstd ( see stetson 1993 ) . because the edges of the pyramid mirror in wfpc2 are imaged onto the photosensitive areas of the four camera ccds , this analysis will be confined to that area of each chip which is almost certainly unvignetted ; this has been taken to be @xmath21 in the three wfc chips , and @xmath22 in pc . the ground - based johnson @xmath17 and kron - cousins @xmath18 magnitudes which will be used for calibrating wfpc2 are taken from walker ( 1994 ) for , and from my own analysis of ground - based data for . the photometry for the latter cluster is based upon 68 @xmath17 and 62 @xmath18 frames obtained during the course of 14 nights in six observing runs . the images were taken with the kitt peak 4 m and 2.1 m telescopes , the canada - france - hawaii telescope ( data obtained through the courtesy of the canadian astronomy data centre ) , and the isaac newton telescope ( data obtained through the courtesy of the isaac newton group archive ) , and have been rigorously transformed to the photometric system of landolt ( 1992 ) as part of an ongoing program that has now homogenized photometric data from over 40 individual observing runs . is a massive , rich globular cluster located about 100 kpc away , and the field is quite crowded under groundbased seeing conditions . therefore the hst imagery was examined to identify stars that appear to be minimally crowded . wherever possible , a star identified in the groundbased images was cross - identified with a star detected in the wfpc2 data . then the groundbased photometry was differentially corrected for those neighbors that were found in the hst images but were _ not _ identified and accounted for in the analysis of the groundbased data . any star that required a blending correction in excess of 0.1 mag was rejected as a potential standard . as the analysis proceeded , any star in either or for which the mean of the calibrated hst photometry differed from the mean of the groundbased data by more than 0.3 mag in either filter was likewise discarded , and the solutions were redone until no further rejections were required . stars that were rejected at this stage were walker numbers 1 , 9 , 24 , 48 , 49 , 57 , 60 , 77 , 99 , 129 , 139 , 152 , 154 , 174 , 196 , 203 , 238 , 240 , 243 , 247 , 248 , 261 , 274 , and 275 ; this left 184 stars for which the groundbased and hst photometry were concordant to within @xmath23 mag . fifteen of 366 stars in were similarly discarded . it may be argued that the inclusion of stars with such large standard errors would vitiate any delicate photometric analysis . however , this should not be a serious problem . in the analyses that follow all observations are weighted in accordance with their inferred standard errors , which are fairly well known from the photon statistics and repeatability from multiple images . then a robust version of least squares is used ( see stetson 1989 ) which permits the solution to follow the main trend of the data without being unduly influenced by outliers . inclusion of faint stars is necessary to provide the needed leverage on the magnitude dependence of the charge loss , and a more strict censoring of the data might introduce bias rather than reduce it . in addition , most stars were eventually imaged at many different locations on the various chips . a star whose photometric errors give a semblance of a large charge loss when imaged at a high value of @xmath1 or @xmath0 would produce a cancelling effect on other occasions when it appeared at low column or row numbers . as in the differential wh97 study , then , gradients in the photometric zero point across the face of the chips , and variations of those gradients with sky brightness and stellar magnitude will not be sensitive to the accuracy of the groundbased magnitudes . prior knowledge of the standard photometric magnitudes will be useful only in testing for the presence of position - independent photometric non - linearities , and those tests alone will be affected by uncertainty in the groundbased results . the present data set will be analysed in three ways : ( 1 ) assuming that charge - transfer losses are negligible ; ( 2 ) correcting the observed stellar fluxes according to the precepts of wh97 ; and ( 3 ) solving for a new numerical model of the charge loss as a function of position , stellar brightness , and sky brightness simultaneously with the determination of the photometric zero points . the results of these three analyses will be presented in 3.1 , .2 , and .3 , respectively , for the data obtained at a camera temperature of 88@xmath24c . a briefer discussion of the data obtained at the warmer temperature will be presented in 4 . the fundamental assumptions of the present analysis are : 1 . the charge - transfer inefficiency affects the gain = 7 and the gain = 14 observations identically , provided care is taken to evaluate the charge loss in units of electrons . the inefficiency is the same for f555w and f814w . the size of the effect has not evolved significantly between 1994 may and 1997 november . the charge loss may be significantly different at a camera temperature of 76 c as opposed to 88 c. wh97 did not require assumptions ( 1 ) , ( 3 ) , or ( 4 ) because all their observations were obtained on a single date with a single gain setting and camera temperature . our assumption ( 1 ) follows from the hypothesis that the charge transfer inefficiency is completely the result of physical processes within the body of the ccd , and not from subsequent effects that take place during amplification and digitization . assumptions ( 2 ) and ( 3 ) are made primarily for simplicity and ease of modifying existing software to perform the present analysis , but the work of wh97 is relevant for assessing the validity of assumption ( 2 ) , at least . while they found that the gross correction for lost charge does appear to depend upon the filter the corrections being the greatest at short and long wavelengths and being least for the f555w and f675w filters they noted , `` this is probably because the chips are more efficient at these wavelengths [ , near 555 and 675@xmath4 nm ] , hence the background is higher . the higher background appears to reduce the cte loss '' thus , when proper quantitative account is taken of the apparent dependence of charge loss on background brightness level , wh97 suggest that the photometric correction does not depend heavily on bandpass . in any case , the validity of assumptions ( 2 ) and ( 3 ) can be checked _ ex post facto _ once the average baseline effect of charge - transfer inefficiency has been calibrated and removed . as a result of assumption ( 4 ) , i have performed separate and independent analyses for the colder and warmer camera temperatures . a least - squares fit of the basic transformation equations to the images of 36,234 stars recorded in the cold - camera observations of and yielded the photometric zero points listed in table 1 . as mentioned above , the color coefficients derived by holtzman ( 1995b ) were imposed as givens . specifically , @xmath25 the first part of the table gives the zero point for each chip / filter combination based upon the full set of observations obtained with electronics bay 3 ( exclusively short exposures of the field ) , and the full set with electronics bay 4 ( a combination of short exposures of and both short and long exposures of ) . in the second part of the table , electronics bay 4 zero points are presented for the short exposures , the short exposures , and the long exposures considered separately . all of these derived zero points are tabulated in detail because they will be used below to test the validity of the various correction schemes . the calibration residuals @xmath26(magnitude ) resulting from the 16 combined solutions ( 2 gain settings @xmath27 2 filters @xmath27 4 chips , corresponding to the zero points listed in table 1a ) were all merged into a single file ; the sense of the residual is ( observed instrumental magnitude ) _ minus _ ( magnitude predicted from the standard - system indices and the mean transformation equations ) . figure 1 shows the individual residuals plotted versus ( top ) @xmath1 coordinate on the chip , ( middle ) @xmath0 coordinate on the chip , and ( bottom ) raw instrumental magnitude @xmath28 (= _ constant _ @xmath29[number of stellar photoelectrons within a radius of 05 ] ) ; for clarity , only unsaturated stars where the combined uncertainty of the ground - based standard magnitude and the wfpc2 instrumental magnitude is less than 0.05 mag have been plotted . it is apparent that there are significant residual trends with both the @xmath0 and @xmath1 coordinates , but no systematic nonlinearity with magnitude is obvious . to provide a more quantitative , albeit still crude , measure of the magnitude of these effects , a robust , weighted least - squares surface of the form @xmath30 was fitted to all 36,234 fitting residuals . the natural - system position variables are @xmath1 and @xmath0 , @xmath28 is the instrumental magnitude as defined above , and @xmath31 is the date of observation , measured in years from heliocentric julian date 2,450,000.0 . to place the amplitude of the position and magnitude trends in a comfortable set of units , i take the maximum displacement of any given well - measured star from the mean of all stars to be @xmath32 px , @xmath33 px , @xmath34 mag . from the least - squares fit i estimate the maximum photometric consequences of the neglected charge - loss corrections to be @xmath35 the positive signs of the first three coefficients , representing the linear dependence on position and magnitude , indicate that the instrumental magnitude is too large , , the star is measured too faint with respect to the average zero - point for all stars taken together , when the @xmath1 coordinate is large , the @xmath0 coordinate is large , or the star is faint ( large @xmath28 ) all in accordance with the simple predictions of the charge - loss model . the positive signs of both second derivatives demonstrate that the effect of charge loss is further compounded for faint stars at large @xmath1 and @xmath0 positions . for a star in a typical exposure , any one of these neglected corrections can cause a systematic error of no more than about 0.015 mag , which corresponds to a root - mean - square error of about 0.01 mag . a star in the wrong corner of position - magnitude space could , however , be subject to an error more than four times larger . note in particular that the result for @xmath36 implies a stretching of the instrumental magnitude scale , on average : faint stars are measured too faint compared to bright ones , by about 0.03 mag per 5 magnitudes ( an approximate figure for the effective dynamic range of wfpc2 ) . the remaining coefficient determined from the least - squares fit to the residuals is @xmath37 the positive sign of the result indicates that for fixed true - system magnitudes , the observed instrumental magnitudes are becoming larger , , fainter , as time passes . however , with an indicated rate @xmath38 mag yr@xmath39 , this secular change is not very important even over the more than 3.5 years over which the cold - camera data were obtained . in applying the wh97 corrections to the present data sets , all star and sky fluxes were converted to units of electrons , assuming gain values of identically 7.0 or 14.0@xmath4e@xmath5/dn , to facilitate combining data obtained with either of the two electronics bays . since i will be calibrating small , differential magnitude corrections directly from stellar observations , minor differences between the postulated and actual gain factors will not seriously damage the analysis . the numerical coefficients of the wh97 equations were adjusted to represent fluxes expressed in units of electrons rather than electronics bay 4 data numbers . as noted above , the logarithm of the sky brightness varies rapidly as the sky brightness approaches zero , and is undefined for any non - positive estimate of the sky flux . in fact , table 1 of wh97 indicates that seven of their 44 digital images had inferred median sky - brightness values , @xmath40 , that were negative ; a further 18 images had positive median brightness values that were less than 0.5 dn px@xmath39 , where a small inaccuracy in @xmath11 begins to lead to a significant change in @xmath41 . wh97 make no explicit mention of how images like these are to be treated in their methodology . the electron - trap model of the charge - transfer inefficiency predicts that a realistic stellar image would lose only a finite number of electrons even in the limit of a hypothetical exposure with a zero effective background ( a short exposure of bright stars through a narrow filter , for instance ) . therefore , it is apparent that some modification must be made to the formulae so that the predicted photometric correction approaches a finite constant value as the estimated background brightness reaches ( and , in the presence of noise , possibly passes through ) zero . in the present analysis i do this by simply `` softening '' the inferred sky brightness at low values : @xmath42 ^ 2\right\}^{1\over2}.\end{aligned}\ ] ] ( please note that in the present analysis , the units of _ bkg _ are electrons , not dn as in the wh97 study . ) the softening constant , 1@xmath4e@xmath5 , has been adopted arbitrarily , and has as its sole justification the fact that it is better than doing nothing . fig . 9 of wh97 indicates that the fraction of charge lost from a stellar image continues to increase down to @xmath43 \sim -1,$ ] or @xmath44e@xmath5/pixel , but diffuse sky fluxes below this have not been sampled . the imposition of this minimum inferred sky brightness means that all frames where the diffuse sky brightness is small compared to 1@xmath4e@xmath5/pixel will be subject to similar , finite , photometric corrections . henceforth in this paper , the unqualified symbol _ bkg _ will refer to @xmath45 as defined in eq . 1 . having made this one modification to the wh97 photometric corrections , i then recomputed the cold - camera zero points for the various combinations of gain factor , chip , and filter , arriving at the results shown in table 2 . i forgo plotting the residuals from these solutions , but the numerical correlations with the various independent variables are : @xmath46 thus , the wh97 corrections operate in the right direction to reduce the size of the maximum systematic error induced by the electron traps , but they do not eliminate the errors entirely , at least for the present data set : some of the charge - loss effects are slightly over - corrected , others are significantly under - corrected . most notably , in these data the amount of charge lost in the serial register ( the @xmath1 direction ) appears to be larger than estimated by wh97 , as is the dependence of the @xmath0-coordinate ramp on instrumental magnitude ( which has been set to zero in their model ) . as a result , a maximum systematic error in excess of 0.02 mag is still possible for stars in an unfortunate location in @xmath47 space in an average exposure . for internal consistency , unlike wh97 i adopt symmetric , exponential - of - a - logarithm , formulations for the charge - transfer losses in both @xmath1 and @xmath0 as functions of both star and sky flux . in doing this , i claim no superior knowledge of the detailed physics of charge loss , but rather follow wh97 s lead in postulating a simple , phenomenological , numerical model for the required photometric corrections . at the same time , in reconsidering the wh97 equations , i have tried to improve on some of the aforementioned features of their formulation . in particular , i have added a term to the formula which allows for the possibility of a charge loss which is independent of the image s position on the chip . in doing this , i have arbitrarily assumed that any charge loss which _ is _ independent of position will have precisely the same dependence on sky brightness and stellar flux as the @xmath0-dependent ramp correction . again , this assumption is arbitrary and has been made to reduce the number of independent coefficients that must be determined ; it is physically reasonable provided that the electron traps contributing to any position - independent charge loss are physically similar to those that produce the well - documented @xmath0 photometric ramp in the body of the detector , which may be different from the traps that produce the less well - understood @xmath1-dependent charge losses that probably occur in the serial register . of course , if it should happen that there are no position - independent effects , the least - squares analysis will allow the coefficient of the constant term to assume a negligibly small value . finally , rather than expressing the equations in terms of the image positions and the star and sky brightness in absolute terms , i consider expansions about `` typical '' values of these variables . this makes the various terms in the formulae more nearly independent of each other , so that the least - squares solution is less ill - determined , and also leads to derived standard errors of the coefficients that can be understood in a simple and direct way . with _ bkg _ representing the softened sky flux in e@xmath5/pixel , _ cts _ representing the stellar flux within a 05 ( radius ) aperture in e@xmath5 , and @xmath1 and @xmath0 the position of the star image in the natural coordinate system of the chip , the adopted numerical model for the photometric corrections is : @xmath48 - 1 ; \\ \hbox{\it star } & = & \log_{10}{[\hbox{\it cts } } ] - 4 ; \\ y\hbox{-cte } & = & c_1 + c_2\times\hbox{\it sky } + c_3\times\hbox{\it star } ; \\ x\hbox{-cte } & = & c_4 + c_5\times\hbox{\it sky } + c_6\times\hbox{\it star } ; \\ c & = & 1 + 0.01\times\left(c_7 + \delta y\right)\times e^{\hbox{\small $ y$-cte } } + 0.01\times \delta x\times e^{\hbox{\small $ x$-cte}}.\end{aligned}\ ] ] if @xmath28 represents the measured stellar magnitude relative to whatever zero point , then @xmath49 this model is not as formidable as it looks . @xmath50 is equivalent to @xmath51 a point near the center of the unvignetted area on each of the wfpc2 ccds while @xmath52 , @xmath53 refer to points near the edges of the photometrically reliable field . the peculiar combination of natural and base-10 logarithms and exponents used here is not entirely whimsical ; for me , at least , it facilitates performing order - of - magnitude estimates in the head . for instance , the constant @xmath54 approximately represents the fraction of charge lost ( expressed in percent ) from a star image containing 10,000 electrons near the effective center of an image that has a sky surface brightness of 10@xmath4e@xmath5/pixel . the constants @xmath55 and @xmath56 represent the degree to which such a star would lose more or less charge if measured near the edge of the field rather than near the center as a result of , respectively , the @xmath0 and @xmath1 ramps . for star images containing either 100,000 or 1,000 electrons , these corrections would differ by @xmath57 and @xmath58 percent , respectively . likewise , for sky values of 100 or @xmath59e@xmath5/pixel ( `` @xmath6 '' because of the softening ) , the corrections for a 10,000-electron star image would differ by @xmath60 and @xmath61 percent . encoding the corrections as proportional to @xmath62 $ ] and @xmath63 $ ] ensures that they will go asymptotically although not necessarily quickly to zero as @xmath64 or @xmath65 , as long as @xmath66 , @xmath67 , @xmath68 , and @xmath69 are all less than or equal to zero . at the other extreme , the softening of the sky brightness ensures that the correction approaches a constant value for any given star in the limit as the background flux becomes negligible . no corresponding numerical protection in the limit of negligibly small stellar flux is required , since the apparent magnitude of a statistically insignificant star is usually of little astronomical interest . a copy of the computer program ccdstd ( stetson 1993 ) was modified to yield numerical values and uncertainties for the seven model constants @xmath70 , as well as the zero points of the transformation from instrumental to standard magnitudes . as before , the 16 separate zero points @xmath71 and @xmath72 were computed for the four chips and the two gain settings ( eight values for @xmath71 and eight for @xmath72 ) , based on the cold ( camera temperature = 88@xmath4c ) data only . the same gain = 7@xmath4e@xmath5/dn zero points were assumed to apply for the data and both short and long exposures of . at the same time , the cold data for all four chips , both filters , both gain settings , and both clusters were used to determine a single , unique set of values @xmath70 . the numerical results of this exercise were the zero points listed in table 3a and @xmath73 all of these terms @xmath74 appear to be highly significant , with the sole exception of the sky dependence of the @xmath1-coordinate ramp ( @xmath68 ) : for a ten - fold change in the sky brightness ( , from 10@xmath4e@xmath5/pixel to either 1 or 100@xmath4e@xmath5/pixel ) the required correction changes by only 0.03@xmath750.03% at the right and left edges of the field , which abundantly justifies the neglect of this term by wh97 . however , a dependence of the @xmath0 ramp on stellar flux ( @xmath67 ) , which wh97 also omitted , appears to be strongly supported by these data . according to this model , a star image consisting of 10,000@xmath4e@xmath5 near the center of an image with a sky brightness of 10@xmath4e@xmath5/pixel loses approximately @xmath76% of its electrons ; at @xmath77 and @xmath78 it loses @xmath79% and @xmath80% yielding , in this case , a ramp of about 3.8% per 750 pixels or 4.0% per 800 pixels . in the hypothetical row @xmath81 the same star would lose @xmath82e^{+0.629 } = 0.1$]% , so a position - independent contribution to the total correction , while apparently statistically significant , is clearly not very important . decreasing the sky brightness by a factor of ten would increase the @xmath0 ramp by a factor @xmath83 , to about 7.0% per 800 pixels . for a star 2.5 magnitudes brighter , with 100,000@xmath4e@xmath5 in its image , the @xmath0 ramp would be smaller by a factor @xmath84 , yielding 2.2% per 800 pixels for 10@xmath4e@xmath5/pixel in the sky , or 3.9% per 800 pixels for a sky brightness of 1@xmath4e@xmath5/pixel . the numerical correlations between the fitting residuals and position and magnitude are : @xmath85 these are not identically equal to zero because , ( a ) these linear correlations do not have the same functional forms as the terms defined in the least - squares photometric solution , and ( b ) in the two separate robust least - squares fits one for the model parameters plus independent zero points for 16 distinct data sets , and one for the correlations between residuals and positions and magnitudes for all 16 data sets considered together the weighting schemes can not be perfectly reproduced : the degrees of freedom are different . in particular , in fitting the correction model no allowance was made for a separate trend of photometric error with magnitude independent of position on the chip , so if such a trend is present it can not be perfectly compensated by the model . as a result , the single most significant residual correlation is @xmath86 . the sign of this correlation has changed from the uncorrected solutions , indicating that trend of increasing fractional charge loss with decreasing stellar flux has been overcorrected , on a global average over the face of the four chips . the corrected magnitude scale is now apparently compressed by 0.013 mag per 5 magnitudes , as compared to the groundbased results . i believe that this level of systematic error is small compared to other unavoidable biases endemic to photometric measurements of very faint stars ( see , , stetson & harris 1988 , v(b ) ) , and it is unlikely that a much more satisfactory calibration can be derived from the present data . certainly , in view of the crowding conditions in these fields , i am unable to guarantee that the ground - based photometry is itself linear to this degree of accuracy . additional well - established , deep , groundbased photometric sequences in other star fields targeted by hst , such as the palomar - type globular clusters and fields in the magellanic clouds , would be most welcome additions to the study of this problem . still , after application of this set of corrections to the wfpc2 data , the maximum systematic errors associated with the purely positional variables @xmath0 and @xmath1 are now well under 1% , at least averaged over the present data set . the secular trend of the photometric residuals with time that remains after applying these corrections is @xmath87 this is larger than was estimated in 3.1 presumably because in the uncorrected data the mean trend with time was partially masked by scatter resulting from the neglect of the charge - loss effects but over the 3.5 years of available wfpc2 data this still implies a maximum departure from the mean zero points of order 0.002 mag at the two ends of the time interval covered . there are several additional sanity checks which can be derived from these data . for instance , if the corrections truly are valid , then the photometric zero point for a given chip / filter combination should be independent of exposure time . we can perform this test with the short and long exposures of ; i omit the exposures from consideration here because this eliminates confusion produced by any possible differences between the ground - based photometric systems for the two clusters . the four chips and two filters provide us with eight independent measures of the ( short long ) zero - point difference ( data listed in tables 1b , 2b , 3b ) . taking the unweighted mean of the eight ( short long ) zero - point differences , i find net values of @xmath88 ( standard error of the mean ; 0.0245 standard deviation of one difference ) based on the reductions with no photometric corrections ; @xmath89 ( s.e.m . ; 0.0243 s.d . ) with the wh97 corrections ; and @xmath90 ( s.e.m . ; 0.0224 s.d . ) with the new corrections derived in 2.3 . in other words , the wh97 corrections largely eliminate the perceived difference in zero points between short and long exposures , but the chip - to - chip and filter - to - filter scatter in the differences is not much reduced by the wh97 corrections . the newly derived corrections represent a slight improvement on those of wh97 , in that they reduce the residual ( short long ) zero - point difference a bit further , and also slightly reduce the chip - to - chip and filter - to - filter scatter . repeating the experiment with weighted means ( after all , the pc1 zero points are based on roughly one - fourth as many stars as those for the wide - field chips ) we obtain a mean ( short long ) difference of @xmath91 for the uncorrected data ; @xmath92 with the wh97 corrections ; and @xmath93 with the new corrections . the gain ratio between electronics bay 3 and electronics bay 4 may differ from one ccd to the next , but we may expect that the gain will not depend upon the filter with which a given exposure was made . if this is the case , then the difference between the gain = 14 zero point and the gain = 7 zero point should be the same in @xmath17 and @xmath18 . for instance , with no photometric corrections , for chip 1 the @xmath17 zero point from the gain = 14 observations of is 1.7673 ( table 1a ) , and the gain = 7 zero point is 0.9974 ( table 1b ; again , this test considers the gain = 7 observations for alone to eliminate the effects of any residual differences between the and ground - based photometry ) , for a gain ratio of 0.7699 mag . the @xmath18-band gain ratio for the same chip is @xmath94 mag , and the difference between the @xmath17 and @xmath18 gain ratios is 0.0437 mag , rather than the expected zero . each separate calibration produces four values of this gain - ratio difference , from the four chips . the unweighted root - mean - square value of these four differences is : 0.0245 mag for the uncorrected reductions , 0.0258 mag with the wh97 corrections , and 0.0285 mag with the new corrections . for each filter , the zero points of the gain = 14 observations should be the same in the four chips , because the flat - field observations were made with the gain = 14 electronics and the flat fields were normalized over the four chips taken together so that surface brightnesses measured in the different chips would be restored to a common system . the standard deviations of the individual chips about the unweighted mean zero points for the two filters ( eight observations , six degrees of freedom ) are 0.0114 mag for the uncorrected reductions , 0.0098 mag with the wh97 corrections , and 0.0142 mag with the new corrections . the tests described in this section are quite indirect , and rely on some untested assumptions for their validity : that gain ratio is independent of wavelength , for instance . therefore these results can be no more than suggestive . there is a hint that the wh97 corrections do better than the new ones when applied to the data alone , as in the gain - ratio test and the test assuming the equality of the gain = 14 zero points . the new corrections seem to do better when the observations are added to the sample , as in the short long test here , and in the global fitting residuals discussed in 3.2 and .3 . if the self - consistency tests of this section have any meaning at all , it is that the errors in the present data set are not totally random , but rather contain unmodeled systematic components as well . for instance , the photometric zero points may vary erratically with time , or with the history of previous exposures , or with some other variable not included in the phenomenological model . if this is true , then the standard errors quoted at various places in 3 and in tables 13 which were estimated simply from the root - mean - square scatter divided by the square root of the number of fitting residuals are merely attractive fictions ; the true errors for any given selected subset of the data may be considerably larger . it appears that the photometric zero point for any given chip / filter combination in any given data set is really externally reliable only to @xmath95 mag , regardless of the standard errors listed in the tables . for instance , the dispersion of zero - point _ differences _ in the ( short@xmath4@xmath4long ) test just performed was @xmath96 mag , so the typical uncertainty of the zero - points themselves would by smaller by @xmath97 , or @xmath98 mag . similarly , the gain - ratio test involved differences of differences of zero points , with a scatter @xmath99 , implying zero points individually reliable to @xmath100 mag , while the third test , involving zero points directly , implied a typical external uncertainty in the range 0.010 0.014 mag . whitmore ( 1998 ) has conducted an analysis of the time variability of the charge - transfer inefficiency based on seven images of the standard field obtained with wfpc2 operating at the colder temperature and spanning the dates 1994 april to 1997 june . he concluded that the cte loss has increased over that period of time , but only for the faintest stars : those containing from 20 to 50 dn within a 02 aperture in very short exposures at the 14@xmath4e@xmath5/dn gain setting appeared to lose @xmath101% of their electrons when clocked out through 800 rows in the detector at the end of that period , as compared to @xmath102% at the beginning . a flux of 280 to 700@xmath4e@xmath5 within 02 corresponds , on my instrumental magnitude scale , to @xmath103 at gain = 7@xmath4e@xmath5/adu and @xmath104 at gain = 14 . accordingly , i have selected out the photometric residuals , _ after _ the application of the new corrections derived in 3.3 , for stars with predicted instrumental magnitudes fainter than those limits . for these stars considered separately , i find @xmath105 ( systematic variations of the residuals with stellar brightness are not worth investigating in a sample with this limited range of instrumental magnitude . ) for these faint stars , it seems that the new corrections overcompensate a bit for the mean rate of charge loss with @xmath1 and @xmath0 position on the chip the net trends are slightly negative , indicating that the corrected magnitudes are a little too bright at positive values of @xmath106 and @xmath107 but the maximum error is @xmath108% and the r.m.s . error averaged over the face of the chip is @xmath109mag in both @xmath1 and @xmath0 . the residuals for these very faint star images do show a systematic positive trend with time , as found by whitmore : the mean residual increases at a rate of 0.0025@xmath4mag yr@xmath39 , for a total net difference of 0.009@xmath4mag over the 3.5 years spanned by the current data set for a faint star located near @xmath110 or , equivalently , for the average of many stars scattered uniformly over the detector field . both the overcorrection of the @xmath1 and @xmath0 ramps and the apparent secular increase of the charge loss with time are much smaller than the random poisson noise for any star image containing 280700@xmath4e@xmath5 . indeed , the random observational errors for star images this faint are almost always larger than 0.05@xmath4mag when all error contributors are considered . that is why these stars are essentially absent from fig . 1 where , for reasons of clarity , only the more precise measurements have been plotted . the unmodeled systematic trends of these faint - star residuals are also much smaller than the systematic statistical effects discussed , , by stetson & harris ( 1988 ) . still , while these remaining systematic effects are unimportant on a star - by - star basis , the reader should be aware that ensemble average magnitudes for large numbers of very faint stars observed in short exposures may be subject to these additional ( relatively minor ) sources of systematic error . whitmore further states that for star images containing at least 2800@xmath4e@xmath5 within 02 ( radius ) the rate of increase of charge loss with time is negligible , even in exposures with very low background levels . it is probably safe to assume that the secular evolution of the charge losses will be doubly negligible in images that contain significant background levels , and that one generic set of charge - loss corrections will be reasonably valid over the 33.5 years spanned by existing studies . the results presented here certainly seem to bear that conclusion out . wfpc2 was operated at a camera temperature of 76 c from installation in 1993 december until 1994 april 23 . between 1994 january 11 and april 21 the tauri standard field was observed a number of times in f555w employing both electronics bays , but observations in f814w are only available for the gain = 14 electronics . determining the mean photometric zero points in the usual way , under the assumption of no charge loss , we arrive at the fitting residuals shown in figure 2 . there are comparatively few observations from the short period of time when the camera was operated at the higher temperature ( 14,521 residuals are shown in fig . 2 ) , and the data do not span the chips nearly as well as in the more extensive cold data . it is clear that there is a considerable charge - loss effect in the @xmath0 coordinate , but any systematic trends with @xmath1 and magnitude are too small to be seen at this scale . i have attempted to determine charge - loss coefficients for the warm data as i did for the cold . i quickly found that the solution for the fitting parameter @xmath111 rapidly diverged to @xmath112 , indicating that any charge loss in the serial register is too small to measure even in a statistical sense . in order to get a self - consistent solution , therefore , it was necessary to simplify the problem . accordingly , i effectively turned off the ramp correction in @xmath1 by assigning @xmath111 a value of 20 , and i also eliminated the possibility of a position - independent charge loss by setting @xmath113 identically equal to @xmath114 . the standard robust solution for the remaining three sigificant terms yielded @xmath115 plus the zero points listed in table 4 . in a frame with a mean background of 10@xmath4e@xmath5/pixel , the total amount of charge lost from a 10,000 electron star at @xmath78 ( @xmath116 ) is @xmath117% , as compared to @xmath118% for the cold data . for the warm data , the variation of the charge loss with background level appears marginally less than for the cold data , but the dependence on the brightness of the star is about twice as strong . for , a total of 152 wfpc2 exposures in the f555w filter and 159 exposures in f814w spread over 3.5 years of observations have now been calibrated to a common photometric system . in contrast , walker s ( 1994 ) groundbased study of was based on only 12 @xmath17 and 10 @xmath18 frames obtained over the course of six nights in 1985 , using an early rca ccd . by combining the hst data with walker s it should be possible to get an improved standard sequence in , `` improved '' at least in the sense of being internally more precise even if not externally more accurate . figure 3(a ) is a ( , @xmath17 ) color - magnitude diagram for walker s stars in , while fig . 3(b ) shows the same stars , but with the present wfpc2 photometry corrected for charge - transfer inefficiency and transformed to the standard @xmath20 system using the precepts of 2.3 and 3 and the zero points of tables 3 and 4 averaged together with walker s . given the large body of hst data , it is also possible to define new standard - sequence stars in . accordingly , a list of minimally crowded detections in the standard field was derived from the output of the allframe reductions . from among these relatively isolated stars , a total of 1,318 were selected which had been measured at least 10 times in each filter , had standard errors of the mean no larger than 0.03@xmath4mag in each of @xmath17 and @xmath18 , and gave no evidence of instrinsic variability in excess of 0.03@xmath4mag , root - mean - square , averaged over the two filters . these are shown in fig . ( the figure 1,318 includes stars listed by walker ( 1994 ) ; 1,116 of them are new ones without walker , harris , or woolley identification numbers as tabulated by walker . ) figures 4(a ) and ( b ) show , respectively , the ground - based photometry for , and the -based photometry averaged together with the ground - based data for the same stars . since in i had already selected out a large sample of minimimally crowded stars to define the ground - based standard sequence in the first place , there is no need to pick out additional local standards based on the hst data alone . there is an advantage in re - deriving final sets of zero points based upon these improved , homogenized standard sequences in the two clusters . with most of the stars having been measured on more than one of the four chips during the course of the observations , as well as having been measured in both long and short exposures ( in the case of ) , with the gain = 7 and gain = 14@xmath4e@xmath5/dn electronics ( in the case of ) , and with the camera operating at both the warmer and colder temperatures ( ) , any residual systematic errors remaining in the individual data subsets will be diluted when all are averaged together . add to these considerations the fact that the total number of standards has now been increased from 534 to 1,714 while the random errors have been reduced from a median value of 0.017@xmath4mag per star to a median of 0.007@xmath4mag per star , and it follows that the derivation of new zero points based on the improved standard sequence will yield photometry that is more homogeneous across the various data sets , even though the reanalysis can not lead to photometry that is more absolutely correct on the average over all data sets . the final zero points that result from calibration of the cte - corrected data with the augmented list of standards are given in table 5 . again , the reader should note that the formal standard errors listed in the table are outrageously optimistic ; true uncertainties of order 1% or slightly more are far more realistic . only additional observations will improve this situation . application of the present cte corrections to other wfpc2 data is straightforward . provided that the reader has stellar positions referred to the natural coordinate system of each chip ( @xmath119 ) and is careful to use the known gain ratio of the observation to convert the instrumental magnitude and some estimate of the diffuse sky brightness in the image to units of electrons , then it is merely necessary to add nine lines of code , corresponding to eqs . ( 1)(9 ) , to whatever software the reader is now using . if , in addition , the reader happens to be following the key project precepts on the standardization of the raw images ( , the images have been multiplied by 4.0 so that they can be stored as short integers without significant loss of precision ; see stetson 1998 ) , then the zero points of table 5 can be used to convert the corrected half - arcsecond aperture magnitudes to the standard system . if the reader does not use the key project precepts , stetson ( 1998 ) list the steps necessary to relate these zero points to those of holtzman ( 1995b ) . the external reliability of photometry on this system is still probably of order 0.01@xmath4mag , since the present zero points represent a compromise among gain 7 and gain 14 , warm- and cold - camera , and short- and long - exposure data sets , each of which has zero points individually uncertain at the 0.010.02@xmath4mag level . internally , however , the various chips should yield photometry consistent at a level somewhat better than this . i am extremely grateful to dr . jeremy mould and the entire staff of mount stromlo and siding spring observatories for the financial support and hospitality during the time when much of this work was being done . additional financial support from the jet propulsion laboratory of the california institute of technology is also very much appreciated . rcccc & cold , gain = 14 , @xmath17 & 1.7673 @xmath75 0.0029 & 1.7515 @xmath75 0.0010 & 1.7824 @xmath75 0.0008 & 1.7593 @xmath75 0.0012 cold , gain = 14 , @xmath18 & 2.6667 0.0021 & 2.6534 0.0007 & 2.6695 0.0010 & 2.6513 0.0010 cold , gain = 7 , @xmath17 & 0.9957 0.0023 & 0.9860 0.0016 & 0.9922 0.0015 & 0.9897 0.0016 cold , gain = 7 , @xmath18 & 1.9728 0.0021 & 1.8963 0.0012 & 1.9240 0.0012 & 1.9473 0.0011 & @xmath120 cen , @xmath17 & 0.9974 0.0030 & 1.0041 0.0024 & 1.0056 0.0029 & 0.9965 0.0024 @xmath120 cen , @xmath18 & 1.9405 0.0031 & 1.9029 0.0017 & 1.9035 0.0025 & 1.9134 0.0016 , short , @xmath17 & 0.9979 0.0044 & 0.9709 0.0026 & 1.0050 0.0020 & 0.9899 0.0023 , short , @xmath18 & 1.9981 0.0026 & 1.8929 0.0015 & 1.9367 0.0014 & 1.9594 0.0013 , long , @xmath17 & 0.9938 0.0061 & 0.9874 0.0032 & 0.9591 0.0031 & 0.9743 0.0037 , long , @xmath18 & 1.9468 0.0039 & 1.8826 0.0038 & 1.8900 0.0023 & 1.9224 0.0023 rcccc & cold , gain = 14 , @xmath17 & 1.7264 @xmath75 0.0026 & 1.7254 @xmath75 0.0009 & 1.7517 @xmath75 0.0008 & 1.7326 @xmath75 0.0014 cold , gain = 14 , @xmath18 & 2.6217 0.0021 & 2.6263 0.0007 & 2.6355 0.0009 & 2.6219 0.0011 cold , gain = 7 , @xmath17 & 0.9610 0.0023 & 0.9594 0.0016 & 0.9666 0.0014 & 0.9649 0.0016 cold , gain = 7 , @xmath18 & 1.9345 0.0018 & 1.8640 0.0010 & 1.8942 0.0010 & 1.9195 0.0010 & @xmath120 cen , @xmath17 & 0.9571 0.0032 & 0.9765 0.0023 & 0.9777 0.0028 & 0.9680 0.0026 @xmath120 cen , @xmath18 & 1.8967 0.0033 & 1.8715 0.0014 & 1.8744 0.0026 & 1.8795 0.0019 , short , @xmath17 & 0.9584 0.0038 & 0.9379 0.0023 & 0.9744 0.0018 & 0.9617 0.0023 , short , @xmath18 & 1.9539 0.0021 & 1.8554 0.0012 & 1.9024 0.0012 & 1.9285 0.0011 , long , @xmath17 & 0.9734 0.0058 & 0.9712 0.0032 & 0.9433 0.0031 & 0.9672 0.0037 , long , @xmath18 & 1.9267 0.0036 & 1.8668 0.0038 & 1.8729 0.0024 & 1.9128 0.0023 rcccc & cold , gain = 14 , @xmath17 & 1.7168 @xmath75 0.0025 & 1.7300 @xmath75 0.0009 & 1.7556 @xmath75 0.0008 & 1.7365 @xmath75 0.0015 cold , gain = 14 , @xmath18 & 2.6096 0.0022 & 2.6315 0.0007 & 2.6368 0.0008 & 2.6275 0.0012 cold , gain = 7 , @xmath17 & 0.9601 0.0026 & 0.9678 0.0016 & 0.9750 0.0014 & 0.9732 0.0016 cold , gain = 7 , @xmath18 & 1.9343 0.0019 & 1.8720 0.0010 & 1.8991 0.0009 & 1.9252 0.0009 & @xmath120 cen , @xmath17 & 0.9478 0.0039 & 0.9850 0.0024 & 0.9861 0.0028 & 0.9760 0.0029 @xmath120 cen , @xmath18 & 1.8891 0.0036 & 1.8776 0.0013 & 1.8844 0.0024 & 1.8897 0.0022 , short , @xmath17 & 0.9621 0.0039 & 0.9466 0.0023 & 0.9824 0.0017 & 0.9709 0.0022 , short , @xmath18 & 1.9542 0.0019 & 1.8646 0.0012 & 1.9059 0.0011 & 1.9337 0.0010 , long , @xmath17 & 0.9818 0.0062 & 0.9791 0.0032 & 0.9525 0.0030 & 0.9720 0.0036 , long , @xmath18 & 1.9359 0.0041 & 1.8743 0.0038 & 1.8815 0.0023 & 1.9187 0.0023 rcccc warm , gain = 14 , @xmath17 & 1.7805 @xmath75 0.0019 & 1.7381 @xmath75 0.0011 & 1.7682 @xmath75 0.0008 & 1.7612 @xmath75 0.0009 warm , gain = 14 , @xmath18 & 2.5957 0.0037 & 2.5852 0.0010 & 2.6099 0.0008 & 2.6244 0.0011 warm , gain = 7 , @xmath17 & 1.0656 0.0047 & 0.9833 0.0043 & 1.0298 0.0025 & 1.0311 0.0037 rcccc cold , gain = 14 , @xmath17 & 1.7222 @xmath75 0.0010 & 1.7282 @xmath75 0.0005 & 1.7475 @xmath75 0.0004 & 1.7285 @xmath75 0.0005 cold , gain = 14 , @xmath18 & 2.6123 0.0008 & 2.6313 0.0003 & 2.6355 0.0004 & 2.6353 0.0003 cold , gain = 7 , @xmath17 & 0.9743 0.0012 & 0.9767 0.0005 & 0.9820 0.0004 & 0.9830 0.0005 cold , gain = 7 , @xmath18 & 1.9088 0.0012 & 1.8792 0.0004 & 1.8957 0.0004 & 1.9140 0.0004 warm , gain = 14 , @xmath17 & 1.7836 0.0007 & 1.7441 0.0004 & 1.7683 0.0004 & 1.7602 0.0005 warm , gain = 14 , @xmath18 & 2.5923 0.0013 & 2.5941 0.0004 & 2.6122 0.0004 & 2.6159 0.0005 warm , gain = 7 , @xmath17 & 1.0643 0.0034 & 0.9997 0.0016 & 1.0356 0.0017 & 1.0264 0.0024

we have undertaken an imaging survey of 34 nearby galaxies in far ultraviolet ( fuv , ) and optical ( ) passbands to characterize galaxy morphology as a function of wavelength . this sample , which includes a range of classical hubble types from elliptical to irregular with emphasis on spirals at low inclination angle , provides a valuable database for comparison with images of high galaxies whose fuv light is redshifted into the optical and near infrared bands . ultraviolet data are from the uit _ astro2 _ mission . we present images and surface brightness profiles for each galaxy , and we discuss the wavelength dependence of morphology for different hubble types in the context of understanding high objects . in general , the dominance of young stars in the fuv produces the patchy appearance of a morphological type later than that inferred from optical images . prominent rings and circumnuclear star formation regions are clearly evident in fuv images of spirals , while bulges , bars , and old , red stellar disks are faint to invisible at these short wavelengths . however , the magnitude of the change in apparent morphology ranges from dramatic in early type spirals with prominent optical bulges to slight in late type spirals and irregulars , in which young stars dominate both the uv and optical emission . starburst galaxies with centrally concentrated , symmetric bursts display an apparent `` e / s0 '' structure in the fuv , while starbursts associated with rings or mergers produce a peculiar morphology . we briefly discuss the inadequacy of the optically defined hubble sequence to describe fuv galaxy images and estimate morphologicalcorrections , and we suggest some directions for future research with this dataset .

astro-ph0002111

an understanding of the ultraviolet ( uv ) properties of local galaxies is essential for interpreting images of high redshift systems and especially critical in searches for morphological evolution in the galaxy population . at redshifts of @xmath4 , the rest frame uv light of galaxies is shifted into the optical ( 1500@xmath5 observed in @xmath6band at @xmath7 ) and near infrared ( 1500@xmath5 observed in @xmath8band at @xmath9 ) regions of the spectrum that are easily observable from the ground . although one operational approach to overcoming this `` band shifting '' problem is to observe in the infrared ( _ e.g. _ @xcite ) , the rate of data acquisition for galaxy images is generally highest at optical wavelengths where detectors are larger and the sky background is lower than in the infrared . under these circumstances , the most direct method to compare present day galaxies to those in the early universe is to build a large sample of local uv galaxy data . recent deep surveys suggest that galaxies with peculiar morphology are more prevalent at high redshift ( _ e.g. _ @xcite ; @xcite ) . however , it is difficult to quantify the influence of band shifting on observed differences between the optical characteristics of nearby galaxies and the rest frame uv appearance of distant ones because uv and optical morphology are not often well coupled ( @xcite ; marcum et al . 1997 , 2000 ) . restframe uv images of galaxies often suggest a later hubble type than the corresponding optical appearance : bulges are less prominent , galaxy light appears more patchy , and in extreme cases the different star forming regions in a single galaxy may appear as separate systems in the uv ( @xcite ; @xcite ) . on the other hand , recent deep nicmos imaging of distant galaxies at longer rest wavelengths suggests that some of the morphological peculiarities are still visible in the optical and thus are intrinsic in the galaxies structure ( _ e.g. _ @xcite ; @xcite ) . although methods exist to estimate uv morphology by extrapolation from optical data ( @xcite ; @xcite ) , they rely on the use of global galaxy spectral energy distributions that may not accurately represent localized conditions , especially in dusty or starbursting regions . for example , donas , milliard , & laget ( 1995 ) find a variation among uv@xmath10 colors in galaxies with identical @xmath11 . the availability of fuv images for bright , well resolved galaxies of a range of hubble types would facilitate a more direct comparison between local and distant galaxy populations . the appearance of galaxies in the uv is determined primarily by emission from hot stars and by the distribution of dust , with a contribution in the central regions from an agn if one is present and unobscured . light in the @xmath121500@xmath5 far ultraviolet ( fuv ) band of the ultraviolet imaging telescope ( uit , @xcite ) originates in young objects of spectral types o and b ( fanelli et al . 1997 ) and , if old populations are present , in certain types of low mass stars in late stages of evolution ( the `` uvx '' population , see @xcite for a review ) . thus this region of the spectrum can probe timescales of young stellar populations between that studied via @xmath13 emission ( only o stars , @xmath12 5 myr ) and that evident in the optical colors ( which trace the age of a stellar population on timescales of a few gyr ; oconnell 1997b ) . opacity due to dust in the fuv is a factor of @xmath122.5 higher than in the optical @xmath6 band , assuming a galactic extinction law ( @xcite ) , and even larger for small magellanic cloud type dust ( @xcite ) . the young , massive stars that provide the bulk of the uv emission also tend to have small scale heights , less than or similar to that of the dust , while older stars that dominate at optical wavelengths and have scale heights greater than that of the dust layer ( @xcite ) . the degree to which dust extinguishes the uv light of highly inclined disk galaxies is unclear . it likely depends strongly on the relative geometry of star forming regions and dust concentrations , a phenomenon recently noted in optical / nir studies of nearby galaxies as well as in theoretical studies of systems with a mixture of stars and dust ( @xcite ; gordon et al . 1997 ; @xcite ) . for instance , ngc 4631 is quite bright in the fuv and blue in fuv optical colors , and it does not show evidence for strong attenuation ( smith et al . 1997 , 2000 ; @xcite ) . however , ngc 891 is only barely detected ( marcum et al 1997 , 2000 ) , a phenomenon that smith et al . suggest is due to extinction effects . uv morphology may also be dramatically influenced by specific features such as dust lanes , as noted in cen a by oconnell ( 1997a ) . scattering by dust grains is quite efficient in the fuv ( @xcite ) and may contribute significantly to the emission near h ii regions and in spiral arms ( @xcite ; @xcite ) . prior to the space shuttle borne uit _ missions , uv imaging of galaxies was limited to low resolution and/or small fields of view ( see @xcite for a detailed review ) . uit prototype rocket borne experiments in the 1980 s provided images of five nearby galaxies with a resolution of @xmath1215@xmath14 and ten more with a resolution of 100@xmath14 ( @xcite ; @xcite ) . the foca balloon experiment yielded images of similar resolution for six nearby galaxies ( @xcite ) . recently , higher resolution uv images of 110 galaxy nuclei ( @xcite ) and nine starburst galaxies ( @xcite ) have been obtained at 2200@xmath5 using the hubble space telescope s faint object camera with a very small ( 22 ) field of view . a fuv ( @xmath121500@xmath5 ) survey of galaxies at redshifts of @xmath12 0.10.2 is also underway using the space telescope imaging spectrograph , again with a small ( 25 ) field of view ( @xcite ) . with a 40@xmath15 field of view and @xmath123@xmath14 resolution , uit was optimally suited to study the morphology of nearby galaxies with good resolution . the usefulness of uv and optical images and radial profiles to study morphology variations with wavelength was demonstrated by @xcite with uit _ data for m74 , by @xcite for m51 , and by @xcite for m81 . in this and a companion paper ( marcum et al . 2000 ) , we present the results of an uv optical imaging survey of nearby bright galaxies . the uv images were obtained by the uit during the _ astro1 _ and _ astro2 _ spacelab missions in 1990 and 1995 respectively . ground based optical ccd images of the sample galaxies were obtained with pixel scales and fields of view approximately comparable to those of the uit images . the _ astro1 _ data presented in marcum et al . include both fuv ( @xmath121500@xmath5 ) and nuv(@xmath122500@xmath5 ) data as well as optical images for 43 galaxies . this paper includes fuv and optical data for 34 galaxies observed during _ astro2 _ with hubble types specifically chosen to span the range e to irr , including some interacting / merging systems . there is some emphasis in the sample selection on face on spirals in which spiral arm , bar , and ring morphologies are easily observed . in section 2 we discuss the sample selection , describe the uit and optical observations , and explain the basic data reduction . section 3 contains details of the registration of fuv and optical images and the extraction of surface brightness and color profiles . a qualitative comparison of uv and optical morphology , based on the images and profiles , is presented in section 4 . in section 5 we provide a brief summary and discuss avenues of future investigation , including studies of the morphological @xmath3correction and analysis of the star formation histories of individual galaxies . in a subsequent paper ( @xcite ) , we will investigate the quantitative concentration and asymmetry indices for these galaxies and consider the implications of these results for the study of high redshift galaxies . our sample consists of 34 galaxies observed with uit during the _ astro2 _ mission for which we have also obtained ground based optical images . many of the galaxies were selected as part of a uit guest observer program specifically to investigate the uv morphology in prototypes of different hubble types . others were observed during uit studies of starburst galaxies ( _ e.g. _ @xcite ) , early type galaxies ( @xcite ) , agns ( fanelli et al . 1997a ) , or individually for selected purposes . it is important to remember that while this sample does contain galaxies with a range of morphologies , it was not chosen to statistically represent the relative distribution of hubble types in the local population . the galaxies have types e to irr ( @xmath16 = 5.0 + 10.0 ) , and are located at distances of @xmath17 mpc ( @xmath18km / s / mpc ) . late type systems in which star formation produces prominent morphological features in the fuv are emphasized . three sets of interacting pairs are included : ngc 3226/7 , ngc 4038/9 , and ngc 5194/5 . table [ galxdat ] gives basic properties of the sample galaxies . the sample galaxies were observed in the fuv with uit , a 38 cm . chretien telescope mounted on the _ astro _ payload . astro2 _ mission during which these data were obtained was flown on the space shuttle endeavour on 218 march 1995 . details of the telescope and instrumentation , as well as specific information about pipeline data processing of the uv images , can be found in stecher et al . ( 1997 ) and will be briefly summarized here . only the fuv ( @xmath191500 ) camera was operational during the _ astro2 _ mission . two fuv filters were used : the b1 filter with an effective wavelength of 1520@xmath20 and width 354@xmath20 , and the slightly narrower b5 filter with effective wavelength of 1615@xmath20 and width 225@xmath20 , which was used during daylight observations to exclude dayglow emission lines ( @xcite ; @xcite ) . based on observations of early type galaxies ( @xcite ) , there is no evidence for a `` red leak '' in the system . the camera consisted of a two stage image intensifier coupled to kodak iia0 film on which the images were recorded . the photographic film was digitized with a 20@xmath21 m square aperture and 10@xmath21 m sample spacing , then binned to a 20@xmath21 m spacing ( @xcite ) . in the pipeline data reductions , background `` fog '' from the photographic film was subtracted , and the images were linearized , flatfielded , and calibrated . the final output images have a scale of 1.136/pixel and a typical point spread function of fwhm @xmath12 3@xmath14 in the central 16@xmath15 of the frame , in which nearly all of the galaxies studied here are imaged ( @xcite ) . uit images with enough stars to calculate astrometric solutions were corrected for distortion ( @xcite ; @xcite ) ; these data products were used where available . however , we note that even on uncorrected images , the uncertainty due to distortion in the central part of the frame ( where most of our galaxy images are located ) is less than or equal to the psf fwhm . after visual inspection of all available fuv images for each galaxy , we selected the ones with the best signal to noise and image quality for further analysis . in most cases this is the image with the longest exposure time , unless that image is noted in the uit log to suffer pointing problems or other defects . in such cases , the longest exposure time image of good quality was used . the uv bright center of ngc 4151 is saturated in the uit _ astro2 _ images ; for photometry of this galaxy , see @xcite . we select a long exposure uit image of this galaxy to show the faint spiral arms as well as the active nucleus . the uit exposure time of the image selected for each galaxy is given in table [ galxd2 ] . many of the fuv images required no further processing beyond the uit pipeline , but some have a noticeable `` stripe '' artifact through the center of the frame ( @xcite ) . if the entire galaxy was located within the stripe , no correction was made and care was taken to determine the sky value within the stripe as well . for galaxies located partly within and partly outside the stripe , a quadratic surface was fit to the frame after masking out the central region ( where the galaxy lies ) and the frame edges . this surface was then subtracted before the sky background and surface brightness profile were measured . the fuv sky background is typically low ( see @xcite for a discussion of the uv background ) , ranging from undetectable ( with a 1@xmath22 limit of @xmath1225 mag arcsec@xmath23 ) up to 22 mag arcsec@xmath23 for a few images acquired during daylight . broad band optical images of the sample galaxies were obtained over the past several years with the las campanas observatory 2.5 m du pont telescope , the ctio 1.5 m telescope , and the palomar 1.5 m and 5 m telescopes . all of the galaxies presented here were imaged in at least one of the @xmath24 bands ; many have multi color photometry . observations at las campanas were carried out using the 2048@xmath252048 tek 5 ccd with a scale of 0.26/pixel . the ctio observations utilized the tek 2048@xmath252048 ccd with a scale of 0.43/pixel . data from the palomar 1.5 m telescope were taken using the 2048@xmath252048 ccd13 or ccd16 with a scale of 0.37/pixel ; some of these were binned to 1024@xmath251024 with a scale of 0.74/pixel . observations with the palomar 5 m telescope used the cosmic camera with a 2048@xmath252048 ccd and a scale of 0.28/pixel ; again some images were binned to a scale of 0.55/pixel . table [ galxd2 ] shows the telescope , filters , and exposure times for each galaxy . data reduction for the optical images was carried out using the iraf ( @xcite ) and vista ( @xcite ) packages . the images were bias subtracted and flatfielded with twilight or dome flats for each filter . an apparent gradient in the background of some of the @xmath6 , @xmath26 , and @xmath27 frames taken at palomar remained after these steps . a planar surface was fit to these frames after masking out the galaxy , then subtracted before further processing . cosmic rays were identified with the iraf `` cosmicrays '' package and were replaced with an average of surrounding pixel values . if more than one data frame was available for a galaxy in a specific filter , the frames were registered using the coordinates of several bright stars , scaled to the same exposure time , and a zeropoint offset was applied to match the background levels . the frames were then averaged ( for two ) or median combined ( for three or more ) . some of the images are saturated at the galaxy centers ; no correction is applied but the saturated pixels were flagged for future consideration . sky levels were determined in boxes away from the galaxy . in some cases ( flagged in table [ galxd2 ] ) the galaxy nearly fills the frame and thus the background level is somewhat uncertain . on images in which several regions of sky away from the galaxy can be measured , the typical r.m.s . scatter in sky levels across the frame is less than 2% of the sky value . calibrations were applied to each image to obtain the approximate relative alignment of the multicolor light profiles presented in section 3 . the uit data were calibrated using standard stars measured by the international ultraviolet explorer ( @xmath28 ) . the resulting flux calibration is estimated to be accurate to @xmath29 ( @xcite ) ; this value includes the uncertainty in the final @xmath28 calibration . magnitudes are on the monochromatic system : @xmath30 where the flux is in units of erg cm@xmath23s@xmath32 @xmath33 . calibration data for each image are stored in the image headers of the uit data products . because many of the optical data were obtained under non photometric conditions , the frames were calibrated via comparison to published aperture photometry for each galaxy . we used the catalog of @xcite , excluding any data whose zeropoint was found by these authors to be systematically offset by more than 0.05 mag . from that galaxy s curve of growth . we calculated the difference between a synthetic aperture magnitude measured on our galaxy image and the published value for each of the tabulated apertures that fit on our frame . after examining the residuals as a function of ( @xmath34 ) or ( @xmath35 ) color for those galaxies with sufficient published data over a range of colors , we found no strong evidence for color terms in the calibration . therefore , we simply used the mean difference between the measured and published magnitudes as a calibration constant . for images in which the galaxy center was saturated , we compared the observed and published magnitudes in annuli between the smallest published aperture that enclosed all the saturated pixels and subsequent apertures . the typical scatter around this mean difference ranges from 0.020.15 mag . , which is not unexpected for a mix of photometric and non photometric data from several different instruments . the uncertainty in optical aperture magnitudes due to sky and instrument noise is typically @xmath36 . if no catalog data were available for a particular galaxy / filter , we substituted an average of the calibration constants determined in that filter for other galaxies observed during the same night . as all but one of these cases involved calibration of @xmath26 or @xmath27band data , airmass differences between the galaxies that were bootstrapped and those for which calibrations were determined from the literature are likely to contribute only small errors . galaxies for which this procedure was necessary are flagged in table [ galxd2 ] ; their calibrations are correspondingly less certain than the others . overall , combining the noise and calibration errors , we estimate an uncertainty in the optical magnitudes of @xmath37 , with the exact value for each galaxy depending on the quality of the observations and the availability of published data . in order to facilitate the direct comparison of galaxy morphology at uv and optical wavelengths , we transformed the coordinates of the optical data onto the system of the uit data and smoothed all data for a particular galaxy to the same resolution . because the fuv images contain few or no stars , the coordinates of uv bright star forming regions were used to determine a linear transformation ( rotation , scaling , and translation ) between the shortest wavelength optical image and the uit data . after the transformation was applied to that optical image , the centroids of several bright stars were used to align additional optical data . finally , we measured the fwhm of the point spread function ( psf ) on each optical image and smoothed the frame to the uit resolution of 3 . the psf fwhm for two galaxies , m51 and ngc 925 , were slightly larger than the uit fwhm ( 4@xmath14 and 4.5@xmath14 respectively ) ; in these cases the uit images were smoothed to the optical resolution . the final images have the uit scale of 1.136/pixel , implying physical scales ranging from @xmath38 pc / pixel for our sample galaxies . foreground stars and any remaining artifacts were masked out on the optical and fuv data before producing the images and surface brightness profiles presented in section 3 . for the optical data , we identified stars on the longest wavelength image and created a mask that was used to interpolate over those pixels on all of the optical images . most of the sample galaxies lie at high galactic latitude , so few foreground bright stars were present . however , in the case of ngc 2403 , badly saturated bright stars in the @xmath26 and @xmath27bands necessitated masking large areas of those images . for this galaxy , we created a separate mask for the @xmath39 images , which did not require such extensive corrections . in a few cases it was difficult to determine whether bright spots within the galaxy were foreground stars or hii regions ; here we carefully compared the optical and fuv images and masked only those spots that were not visible in the fuv . some galaxies also required a fuv mask to interpolate over small , bright point like or streak artifacts ( mostly due to cosmic rays ) that remained after prior data processing . in figures [ ims](@xmath40 ) , we present multiwavelength images for the sample galaxies . the sky backgrounds have been removed , foreground stars have been masked as described above . for display purposes only , image pixel values have been converted to `` calibrated '' flux units such that : @xmath41 the images for a given galaxy are displayed using the same range of calibrated brightness for each filter , so that pixels with the same brightness ( flux/@xmath5 ) have a constant gray level on each image . this mode of presentation shows the relative limiting surface brightnesses of the data at different wavelengths . to conserve space , we do not show every optical image , as those from adjacent optical filters often look nearly identical . it is quite difficult to simultaneously show structure in the bright centers of galaxies and faint outlying features , so we present additional images of some galaxies with different scales or normalizations in figures [ vfigsf ] and [ vfigearly ] . the four fuv bright galaxies shown in figure [ vfigsf ] are displayed with a much larger scale to show the central regions . in figure [ vfigearly ] , we show several galaxies with red fuv optical colors , mainly early type galaxies or those with a prominent optical bulge . here we show the optical images with a much larger stretch than the fuv data , so that the central structure of the optical image can be compared to the fuv . we also plan to make our images available to the community in digital form via the nasa / ipac extragalactic database ( ned ) ; they can then be renormalized as appropriate for any application . we extract surface brightness profiles for each galaxy by azimuthally averaging around concentric ellipses . the ellipse center , ellipticity , and position angle ( p.a . ) for each galaxy were determined by running an ellipse fitting package in vista ( based on the method of @xcite ) on the longest available wavelength image , usually @xmath26 or @xmath27 . the centroid was determined in a small box around the visual center of the galaxy , taking care to use an image in which the center was not saturated . the ellipticity and p.a . are determined from the outer ellipses , where their values were typically stable over a large range in radii . these ellipse parameters were used to obtain surface brightness profiles at all wavelengths , in order to avoid problems with determining centroids and fitting ellipses on the often asymmetric and irregular fuv images . for one galaxy , ngc 3115 , we were unable to align the optical and fuv images due to a lack of bright features or foreground stars in the fuv . we determined a separate fuv ellipticity and p.a . for this galaxy using the ellipse fitting routine . the ellipticity and p.a . for each galaxy , which are given in table [ galxdat ] , are quite similar to the @xmath42 ( @xcite ) values ( @xmath43 and @xmath44 ) . the surface brightness profiles derived for each sample galaxy are shown in figure [ sbprofs](@xmath45 ) , with data from the various filters denoted by different symbols . images in which the galaxy center was saturated are identifiable by their flat profiles in the central regions ; this is the case for several optical images and the inner @xmath1210@xmath14 of the fuv data for ngc 4151 . profiles for a subset of our sample were compared to those extracted by fanelli et al . ( 1997b ) from the same data ; no significant differences are noted . the profiles have not been corrected for foreground galactic extinction . however , the foreground @xmath46 values ( given in table 1 ) are generally small , so even fuv extinctions of @xmath47 amount to @xmath48mag and would have little effect on the plots of figure [ sbprofs ] . three systems in our sample are interacting pairs for which it is difficult to separate the light profiles of the two galaxies : ngc 4038/9 , ngc 3226/7 , and ngc 5194/5 ( m51a , b ) . ( ngc 4647 and ngc 4649 are a close pair that may be undergoing a mild interaction , but their fuv isophotes do not overlap in our images . ) for ngc 5194/5 , the companion galaxy ngc 5195 is undetected in the fuv image and thus is not analyzed separately . the ellipse parameters for ngc 5194 are determined from a region unaffected by the companion on the optical images . many of the outer ellipses cut through the companion , creating a slight `` bump '' in the outer part of the optical surface brightness profiles . the outer ellipses for ngc 3227 encompass all of the fuv light and a significant fraction of the optical light of its companion ngc 3226 . within the available software packages , it is impossible to avoid this situation without either severely truncating the ngc 3227 profile or changing the ellipticity so much that it no longer accurately reflects the shape of ngc 3227 . however , the ngc 3226 profile of the region with detectable fuv emission is relatively unaffected . the merger system ngc 4038/9 is treated as one galaxy , as the fuv image does not suggest a simple division into two separate disks . we created an image of this system consisting of only pixels with flux greater than 1.5@xmath49 above the background that also have at least 4 adjacent pixels above the threshold . ( `` adjacent '' pixels are defined as those that share either an edge or corner with the pixel in question ; thus each pixel has 8 adjacent neighbors . ) this image is approximately circular , so we defined an aperture center at the geometric center of the fuv image and used circular annuli to measure the light profiles . color profiles are produced by subtracting the surface brightness profiles ( in mag arcsec@xmath23 ) of different filters . the use of a single set of ellipse parameters for all images of a particular galaxy ensures that the light profiles at each wavelength sample the same physical region and thus can be compared in this simple manner . color profiles for many of the sample galaxies will be presented and discussed in the next section . in this section we present a qualitative discussion of the differences in apparent morphology between fuv and optical images . the morphological types included in each subsection below are approximate , stemming in part from the fact that several galaxies have characteristics associated with more than one `` type '' . rather than discussing each galaxy individually ( as is done in marcum et al . 2000 ) , we summarize the changes in apparent morphology with wavelength for various hubble types . ( interested readers can see the results for each individual galaxy from the data presented in figure [ ims ] . ) throughout this section , we also consider implications of our analysis for the study of high redshift galaxies that are observed in their rest it is important to remember that this level of discussion assumes no evolution of the star formation rate : we simply consider what galaxies _ identical _ to local ellipticals , spirals , and irregulars would look like at high redshift . galaxies undergoing their initial starburst or a subsequent period of increased star formation activity due to mergers , etc . would , of course , look different from the quiescent ellipticals and `` normal '' spirals in this sample and may more closely resemble star forming irregular or peculiar systems . the elliptical and s0 galaxies in our sample feature smooth , compact fuv emission centered on the nuclear regions ( see also the detailed analysis of uit data for e / s0 galaxies in ohl et al . 1998 ) . in galaxies of this type , fuv emission is produced not by young stars but by the hot , evolved population responsible for the spectrosopic uv upturn ( @xcite ) . the fuv light profiles of these systems ngc 3379 , ngc 3384 , ngc 4649 , ngc 3226 , and ngc 3115 drop off much more rapidly than the optical light , and the colors throughout are quite red ( fuv@xmath26 @xmath12 25 ) . light radii for these galaxies are @xmath122035% of their @xmath50band values tabulated in the @xmath42 ( we can not compare directly to our optical data because most of the e / so optical images suffer saturation in the center ) . the only exception to this pattern is ngc 1510 , the interacting companion of ngc 1512 . the fuv emission in this galaxy is comparable to the optical light in both brightness and physical extent and is likely due to tidally induced star formation rather than the evolved population . although it is classified in the @xmath42 as a peculiar s0 galaxy , @xcite classify ngc 1510 as an amorphous galaxy , and it is also noted to be a star forming blue compact dwarf ( bcd ) in @xcite and an hii galaxy in ned . ( the other s0/amorphous galaxy in our sample , ngc 5195 , is also interacting with companion galaxy ngc 5194 but is not detected in the fuv . ) it is difficult to predict the appearance of high redshift e / s0 galaxies in their rest frame fuv because the present day uv emission mechanism , that of evolved stars , is likely to be weak to nonexistent at redshifts approaching @xmath51 ( brown et al . 1998 , 2000 ) . indeed , the uv upturn phenomenon is a potentially sensitive age indicator for evolved stellar populations , and thus its evolution with redshift could help identify the formation epoch of early type galaxies ( @xcite ) . at high redshifts ellipticals may be undergoing an initial starburst and/or the uv upturn population will not yet have had time to evolve . in this case , the progenitors of present day elliptical and s0 galaxies may look like local starbursts or mergers depending on their formation mechanism and the timing of the observations . however , if some high redshift e / s0 s were quiescent and identical to local early type galaxies ( _ i.e. _ assuming a very high formation redshift ) , they would still have the smooth and symmetric light distributions associated with early types . their fuv emission would be much more compact than the optical light , which would lead to difficulties in detection and confusion in estimating their typical size . galaxies of hubble type sa sbc tend to appear as later type spirals in the fuv than at optical wavelengths , primarily due to fading of the red bulge and bar in uv images . a significant fraction of the fuv light in these systems comes from star forming regions in the spiral arms , which are visible in the images of figure [ ims ] and appear as `` bumps '' in the fuv surface brightness profiles of figure [ sbprofs ] ( see also @xcite ) . some early type spirals , such as m63 and ngc 1672 , also appear more patchy or fragmented in the fuv because old stars in the underlying smooth disk do not produce much light at very short wavelengths . these results are in agreement with earlier work by oconnell ( 1997a ) and oconnell & marcum ( 1996 ) for smaller samples of galaxies . the dramatic fading of light in the central regions ( also noted by @xcite and @xcite ) produces a global pattern in which the central fuv optical colors are much redder than the outer regions , as shown in figure [ ecolpr ] for m 51 , ngc 1566 , and m 63 . two notable exceptions to this pattern are ngc 4151 , in which the seyfert 1 nucleus dominates the central fuv emission and is much brighter than the weak spiral arms , and ngc 1068 , in which the fuv bright seyfert 2 nucleus and surrounding star formation combine to produce a very blue center . the uv morphologies and luminosities of the agn and star forming components in these two galaxies are discussed in more detail by @xcite . many early type spirals also show evidence of star forming inner rings or circumnuclear star formation , which often dominate the fuv images and light profiles . images of these features , which are present in ngc 1097 , ngc 1512 , ngc 1672 , ngc 3310 , ngc 3351 , and ngc 4736 , are shown in figure [ sfring ] , and the associated color profiles are plotted in figure [ ringcolpr ] . the star forming rings appear as spikes of fuv brightness or very blue color in the radial profiles of figures [ sbprofs ] and [ ringcolpr ] . inner and circumnuclear rings are thought to be orchestrated by dynamical resonances , which are often associated with central bars ( _ e.g. _ @xcite ; @xcite ) . the fuv properties of some specific ringed galaxies are discussed by waller et al . ( 1997c , 2000 ) for ngc 4736 , smith et al . ( 1996 ) for ngc 3310 , and marcum et al . ( 2000 ) . galaxies classified from their optical images as early type spirals may present a variety of different appearances when viewed at high redshift , in their rest frame fuv . optically barred galaxies such as ngc 1097 and ngc 1365 will appear unbarred in the ultraviolet . isolated galaxies such as ngc 1672 that appear highly fragmented in the fuv may be mistaken for an ongoing merger . images of distant systems may recover only the fuv bright star forming rings or circumnuclear star formation , yielding little evidence of the underlying regular structure seen in the galaxy s optical light ( see also waller et al . on the other hand , galaxies such as ngc 4151 appear to have an earlier morphological type in the fuv due to the prominence of highly symmetric and strongly peaked emission from the active nucleus . finally , the appearance of a galaxy undergoing a global starburst , such as ngc 3310 , does not differ a great deal between the fuv and optical because the light at both wavelengths is dominated by young stars . we have found that all of these effects can vary in magnitude , such that some galaxies would be assigned nearly the same hubble type in the optical and fuv , while others would have very different classifications . overall , most optically classified early type spirals present the appearance of a later type in the fuv , but a small fraction may appear to have identical or earlier fuv types in the presence of starburst activity or an agn . galaxies that are optically classified as late type spirals generally do not show as dramatic a difference between their fuv and optical morphologies as do early type spirals . although they often appear somewhat more patchy in the fuv , there is no prominent optical bulge whose absence at short wavelengths is noticeable . star formation occurs over a large fraction of the galaxy , producing both fuv and optical light in the same physical regions . a number of these galaxies , including ngc 4449 , ngc 925 , ngc 1313 , and ic 2574 , show color profiles that have no definite trend with radius . these profiles , examples of which are displayed in figure [ flatcolpr ] , are either flat or bumpy depending on the relative prominence of star forming regions against the background disk . figure [ bluecolpr ] shows color profiles for the central starburst galaxies m83 , ngc 4214 , and ngc 5253 , which are bluest in their centers . ngc 4214 and ngc 5253 have global color gradients in the opposite direction of that observed in early type spirals due to the presence of a blue starburst superposed on a more extended old stellar distribution . this `` starburst core '' morphology has been previously noted in fuv optical image comparisons of ngc 4214 ( fanelli et al . 1997c ) and ngc 5253 ( kinney et al . 1993 ) . a central starburst combined with bright star forming regions in the outer disk can also produce a flat color profile , which is the case for ngc 2903 ( also shown in figure [ bluecolpr ] ) . we find that the morphological @xmath3correction for late type galaxies is expected to be highly dependent on whether or not they host starburst activity . as evident from their similar fuv and optical morphology , non starbursting late type galaxies ( dominated by ongoing disk star formation ) viewed at high redshift would not generally appear to have a very different type from their local counterparts . increased patchiness in the fuv ( if it were still detectable at large distances ) might suggest a slightly later type than the optical appearance . however , galaxies with central starbursts like those seen in ngc 4214 and ngc 5253 might present compact cores surrounded by diffuse light , similar to some objects seen on deep hst images ( giavalisco , steidel , & duccio macchetto 1996a ) . the symmetric central burst in ngc 5253 might even appear to be an e or s0 galaxy in the absence of the old stars . if starburst systems are more prevalent at high redshift , their strong influence on the rest frame uv morphology must be considered when studying the evolution of the galaxy population . as might be expected , the morphological @xmath3correction for interacting galaxies depends strongly on the details of the interaction itself . three objects in our sample are currently undergoing interactions of various degrees : ngc 4038/9 ( the antennae ) , ngc 3226/7 , and ngc 5194/5 ( m51 ) . the merger of two disk galaxies in ngc 4038/9 has produced a significant amount of massive star formation , concentrated mainly between the two disks and around the edge of the ngc 4038 disk . thus the fuv image shows a partial outline of the merger system , and , without the corresponding optical data , suggests the appearance of a single , drastically perturbed galaxy . however , the old populations visible in long wavelength optical images clearly delineate the inner disks and tidal tails that identify ngc 4038/9 as an ongoing disk disk merger . in spite of the dramatic difference between its fuv and optical appearances , ngc 4038/9 is clearly peculiar at both wavelengths and thus would not be misclassified as any sort of `` normal '' hubble type galaxy when viewed at high redshift . both galaxies in the ngc 3226/7 interacting pair present fuv morphologies consistent with what would be expected from their hubble types : the elliptical ngc 3226 appears highly concentrated and the sa ngc 3227 shows emission from the center and a few faint disk star formation regions . in this case , the influence of the interaction in determining the fuv morphology is confined to its possible role in driving the seyfert nucleus and circumnuclear star formation in ngc 3227 ( @xcite ; @xcite ) . ngc 5195 is not detected in the fuv , likely due to obscuration by dust from a foreground spiral arm in ngc 5194 as well as possible internal extinction ( see @xcite . @xcite also suspect that dust is responsible for their failure to detect ngc 5195 at 2200@xmath5 . ) observed at high redshift in its rest frame uv , the m51 double galaxy system would appear to be an isolated spiral galaxy . although the examples discussed above demonstrate the difficulty of relating uv and optical morphology for interacting systems , interactions that are significant enough to severely distort the galaxies involved are often associated with enhanced star formation rates that may identify the system as `` abnormal '' in the fuv as well ( @xcite and references therein ) . throughout this section we have discussed fuv morphology mainly in terms of the current star formation , but , as suggested by the example of m51 , one can not neglect the possibility that extinction also influences the uv appearance of galaxies . it is difficult to ascertain whether the `` patchiness '' or fragmented appearance of many galaxies in the fuv is due to internal extinction or to the pattern of massive star formation . two of the most highly inclined ( @xmath52 ) galaxies in our sample , ngc 2841 and ngc 2903 , have asymmetric fuv emission with the bright side corresponding to what appears to be the unobscured side of the optical disk . on the other hand , ngc 2403 and ic 2574 also have @xmath53 but do not appear to suffer heavy extinction on one side of the disk . these results provide further confirmation of the variations in fuv brightness of edge on galaxies noted by smith et al . ( 1997 , 2000 ) for ngc 4631 ( fuv bright ) and ngc 891 ( undetected ) . an extreme case of extinction effects was discussed above for ngc 5195 ( m51b ) , which is completely obscured in the fuv by the foreground spiral arm of ngc 5194 ( m51a ) . as noted in the introduction , the role of dust in determining the observed fuv morphology depends strongly on the dust star geometry of individual galaxies and is thus difficult to predict from optical images alone . we have presented fuv and optical images and surface brightness profiles for 34 galaxies observed with the uit and compared the apparent morphology at different wavelengths in the context of interpreting images of high redshift objects . the uit data represent the most detailed set of fuv images obtained to date for large , nearby galaxies that are well resolved . day elliptical and s0 galaxies appear smooth and symmetric in fuv as well as in optical images , but they are much more compact in the fuv and may be faint and/or unresolved at high redshift unless they are actively forming stars at that particular look back time . the majority of spiral galaxies in our sample appear to have a later hubble type in the fuv than at optical wavelengths due to increased patchiness in combination with the fading of light from the bulge and/or bar populations . this effect is particularly dramatic for the early type spirals , in which bulges and often bars are prominent optical features . the optical and fuv morphologies of late type spirals and irregular galaxies do not differ as much as the earlier types because young stars dominate the light in both spectral regimes . some galaxies appear highly fragmented in the fuv images in spite of a regular optical morphology , making it hard to determine from uv data alone whether they are single systems or multiple mergers . even assuming their current rates and patterns of star formation , many spirals viewed at high redshift would be assigned later hubble types than their local counterparts due to the bandshifting effects . central or circumnuclear starbursts and star forming rings , and bright spiral arms dominate the fuv light and fuv optical color profiles of many galaxies . depending on the relative geometry of dust and young stars , extinction can play a varying role ( small to dominant ) in determining the apparent fuv morphology . from the galaxy sample presented here , it is apparent that understanding the correlation between fuv and optical morphology depends on one s knowledge of not only the global optical hubble type but also of smaller scale features , mostly those due to recent star formation patterns . the fragmented appearance of many of the fuv images and the prominence of young stars in structures such as rings or fragments of spiral arms produces fuv morphologies that do not readily fit into the traditional bins of the hubble sequence . in order to characterize rest frame uv galaxy populations at both low and high redshift , it will be necessary to find new ways of describing the morphology that do not by default cause the majority of galaxies to fall into the `` peculiar '' or other catch all bins . it would also be desirable to have an objective classification scheme to express the similarities or differences between local and distant galaxies in quantitative terms . however , uv data pose problems for many automated classification schemes because galaxy centers are often ill defined and the patchy nature complicates the selection of an aperture that encloses a single system . future work in this area will use uit data to identify new features or indices that better describe the uv morphology of galaxies . the dataset presented in this paper will be valuable for studying the morphological @xmath3correction that must be taken into account when interpreting deep optical images such as the _ hubble deep field _ ( hdf , @xcite ) . the dependence of fuv morphology on features that are considered minor or secondary details in the optical highlights the difficulty of studying evolution in the galaxy population by comparing rest frame uv observations of high redshift galaxies to the optical properties of local samples . the comparison between local and distant uv galaxy images is much more direct than relying on estimates of the uv morphology extrapolated from optical light and colors ( _ e.g. _ abraham et al . 1997 ) . indeed , a small sample of uit galaxy images from the _ astro1 _ mission have been used to simulate the appearance of high redshift counterparts of nearby bright galaxies ( giavalisco et al . in a subsequent paper ( @xcite ) , we simulate the cosmological distance effects of surface brightness dimming and loss of spatial resolution on the fuv images of our larger _ astro2 _ galaxy sample . these simulations provide examples of how local galaxies would appear in the hdf at redshifts of @xmath54 . in that paper , we also attempt to quantify the effects of bandshifting and distance on morphology using simple parameters such as the central concentration and asymmetry of galaxies . one of the primary goals of studying morphology is to infer the past or ongoing physical processes that have shaped galaxies . in combination with current knowledge about the dynamical states of nearby galaxies , our multi wavelength dataset and future uv imaging data from the galex sky survey ( @xcite ) could be used to relate uv morphology to physical structures or properties . for example , waller et al . ( 1997b , 2000 ) find that uv bright starburst rings likely result from dynamical resonances with a bar component . thus ring features may predict the presence of an underlying disk and bar system even in uv images that appear to be fragmented or otherwise dynamically disorganized . it would also be of interest to further investigate the suggestion of waller et al . ( 1997b ) that optical hubble type may be correlated with fuv@xmath6 color ( although rest frame @xmath6band data will not necessarily be available for high redshift galaxies ) . if relations between uv morphology and physical properties can be found in local galaxies _ and _ shown to extend to large look back times , they could provide a valuable tool for studying the distant universe through rest frame uv imaging . the authors gratefully acknowledge the help of r. bernstein , j. parker , r. phelps , and n. silbermann for obtaining some of the optical imaging data presented here . we thank o. pevunova for assistance with preliminary data reduction . we are also grateful to the anonymous referee for many constructive suggestions . funding for the uit project was provided through the spacelab office at nasa headquarters under project number 440 - 51 . wlf and bfm acknowledge support from the astro2 guest investigator program through grant number nag81051 . some of the data presented here were obtained at ctio , which is operated by aura as part of noao under a cooperative agreement with the national science foundation . this research has made use of the nasa / ipac extragalactic database ( ned ) , which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . abraham , r. 1998 in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 195 abraham , r. g. , freedman , w. l. , & madore , b. f. 1997 , in _ hst _ and the high redshift universe , ed . n. r. tanvir , a. aragon salamanca , & j. v. wall ( singapore : world scientific ) , 57 abraham , r. g. , tanvir , n. r. , santiago , b. x. , ellis , r. s. , glazebrook , k. g. , & van den bergh , s. 1996 , , 279 , l47 barth , a. j. , ho , l. c. , filippenko , a. v. , & sargent , w. l. w. 1998 , , 496 , 133 blecha , a. , golay , m. , huguenin , d. , reichen , d. , & bersier , d. 1990 , , 233 , l9 bohlin , r. c. , et al . 1991 , , 368 , 12 brinchmann , j. , et al . 1998 , , 499 , 112 brosch , n. 1999 , experimental astronomy , 9 , 119 brown , t. m. , bowers , c. w. , kimble , r. a. , & ferguson , h. c. 2000 , , 529 , l89 brown , t. m. , ferguson , h. c. , deharveng , j.m . , & jedrzejewski , r. i. 1998 , , 508 , l139 bunker , a. j. 1999 , to appear in the ociw workshop on photometric redshifts , ed . weymann , r. , storrie lombardi , l. , sawicki , m. , & brunner , r. ( _ astro - ph _ 9907196 ) burstein , d. , & heiles , c. 1984 , , 54 , 33 buta , r. , & combes , f. 1996 , fund . cosmic phys . , 17 , 95 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 cornett , r. h. , et al . 1994 , , 426 , 553 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , buta , r. j. , paturel , g. , & fouque , p. 1991 , third referencecatalogue of bright galaxies , ( new york : springer verlag ) ( _ rc3 _ ) donas , j. , milliard , b. , & laget , m. 1995 , , 303 , 661 fanelli , m. n. , collins , n. , bohlin , r. c. , neff , s. g. , oconnell , r. w. , roberts , m. s. , smith , a. m. , & stecher , t. p. 1997a , , 114 , 575 fanelli , m. n. , et al . 1997b , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 83 fanelli , m. n. , et al . 1997c , , 481 , 735 gardner , j. p. , heap , s. r. , malumuth , e. m. , hill , r. s. , & smith , e. p. 1997 , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 429 giavalisco , m. , livio , m. , bohlin , r. c. , duccio macchetto , f. , & stecher , t. p. 1996b , , 112 , 369 giavalisco , m. , steidel , c. c. , & duccio macchetto , f. 1996 , , 470 , 189 gonzalez delgado , r. m. , & perez , e. 1997 , , 284 , 931 gordon , k. d. , calzetti , d. , & witt , a. n. 1997 , , 487 , 625 greason , m. r. , offenberg , j. d. , cornett , r. h. , hill , r. s. , & stecher , t. p. 1994 , , 106 , 1151 hill , j. k. , et al . 1997 , , 477 , 673 hill , j. k. , bohlin , r. c. , & stecher , t. p. 1984 , , 277 , 542 hill , r. s. , et al . 1998 , , 507 , 179 huizinga , j. e. 1994 , phdt , univ . groningen keel , w. c. 1996 , , 111 696 kennicutt , r. c. 1998 , , 36 , 189 kent , s. m. 1983 , , 266 , 562 kinney , a. l. , bohlin , r. c. , calzetti , d. , panagia , n. , & wyse , r. f. g. 1993 , , 86 , 5 kuchinski , l. e. , madore , b. f. , freedman , w. l. , & trewhella , m. 2000 , in preparation kuchinski , l. e. , terndrup , d. m. , gordon , k. d. , & witt , a. n. 1998 , , 115 , 1438 maoz , d. , filippenko , a. v. , ho , l. c. , macchetto , f. d. , rix , h.w . , & schneider , d. p. 1996 , , 107 , 215 marcum , p. m. , et al . 1997 , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 88 marcum , p. m. , et al . 2000 , in preparation martin , c. , et al . 1997 , baas , 191 , 63.04 martin , p. , & roy , j.r . 1994 , , 424 , 599 meurer , g. r. , heckman , t. m. , leitherer , c. , kinney , a. , robert , c. , & garnett , d. r. 1995 , , 110 , 2665 mihalas , d. & binney , j. 1981 , galactic astronomy ( san francisco : w. h. freeman ) oconnell , r. w. 1999 , to appear in , 1999 oconnell , r. w. 1997a , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 11 oconnell , r. w. 1997b , in star formation near and far , ed . s. s. holt & l. g. mundy ( new york : aip ) , 491 oconnell , r. w. , & marcum , p. m. 1996 , in hst and the high redshift universe , ed . n. r. tanvir , a. aragon salamanca , & j. v. wall ( singapore : world scientific ) ohl , r. g. , et al . 1998 , , 505 , l11 prugniel , p. , & heraudeau , p. 1998 , , 128,299 reichen , m. , kaufman , m. , blecha , a. , golay , m. , & huguenin , d. 1994 , , 106 , 523 roberts , m. s. , & haynes , m. p. 1994 , , 32 , 115 sandage , a. 1961 , the hubble atlas of galaxies ( washington : carnegie institution of washington ) sandage , a. , & brucato , r. 1979 , , 84 , 472 smith , a. m. , et al . 2000 , , submitted smith , a. m. , collins , n. r. , waller , w. h. , fanelli , m. n. , stecher , t. p. , & the uit team 1997 , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 439 smith , d. a. , et al . 1996 , , 473 , l21 stecher , t. p. , bohlin , r. c. , hill , j. k. , & jura , m. a. 1982 , , 255 , l99 stecher , t. p. , et al . 1997 , , 109 , 58 , 4 storchi bergmann , t. , wilson , a. s. , & baldwin , j. a. 1996 , , 472 , 83 stover , r. j. 1988 , in instrumentation for ground based optical astronomy , ed . l. b. robinson ( new york : springer ) , 443 teplitz , h. i. , gardner , j. p. , malumuth , e. m. , & heap , s. r. 1998 , , 507 , l17 tody , d. 1986 , spie , 627 , 733 tully , r. b. 1988 , catalog of nearby galaxies , ( cambridge : cambridge up ) waller , w. h. , et al . 2000 , , submitted waller , w. h. , et al . 1997a , , 481 , 169 waller , w. h. , fanelli , m. n. , collins , n. r. , cornett , r. h. , offenberg , j. , marcum , p. m. , stecher , t. p. , & the uit science team 1997b , in the ultraviolet universe at low and high redshift , ed . w. h. waller et al . ( woodbury , ny : aip ) , 39 waller , w. h. , stecher , t. p. , & the uit science team 1997c , in star formation near and far , ed . s. s. holt & l. g. mundy ( new york : aip ) , 287 waller , w. h. , et al . 1995 , , 110 , 1255 williams , r. e. , et al . 1996 , , 112 . 1335 witt , a. n. , petersohn , j. k. , bohlin , r. c. , oconnell , r. w. , roberts , m. s. , smith , a. m. , & stecher , t. p. 1992 , , 395 , l5 yi , s. , lee , y.w . , woo , j.h . , park , j.h . , demarque , p. , & oemler , a. 1999 , , 513 , 128 lllccrrrr ngc 925 & 02 27 16.8 & + 33 34 41 & sab(s)d & & 9.4 & 0.25 & 0.40 & 115 ngc 1068 ( m 77 ) & 02 42 40.2 & @xmath5500 00 48 & rsa(rs)b & sy 2 & 14.4 & 0.05 & 0.20 & 84 ngc 1097 & 02 46 18.9 & @xmath5530 16 21 & sb(s)b & sy 1 & 14.5 & 0.07 & 0.32 & 140 ngc 1313 & 03 18 15.5 & @xmath5566 29 51 & sb(s)d & & 3.7 & 0.03 & 0.20 & 40 ngc 1365 & 03 33 36.6 & @xmath5536 08 17 & sb(s)b & sy 1.8 & 16.9 & 0.00 & 0.45 & 32 ngc 1510 & 04 03 32.6 & @xmath5543 24 01 & s0,pec & h ii & 10.3 & 0.00 & 0.12 & 115 ngc 1512 & 04 03 54.6 & @xmath5543 21 03 & sb(r)a & & 9.5 & 0.00 & 0.36 & 56 ngc 1566 & 04 20 00.4 & @xmath5554 56 18 & sab(s)bc & sy 1 & 13.4 & 0.00 & 0.21 & 40 ngc 1672 & 04 45 42.2 & @xmath5559 14 57 & sb(s)b & sy 2 & 14.5 & 0.00 & 0.13 & 161 ngc 2403 & 07 36 54.5 & + 65 35 58 & sab(s)cd & & 4.2 & 0.16 & 0.44 & 115 ngc 2841 & 09 22 01.8 & + 50 58 31 & sa(r)b & liner , sy 1 & 12.0 & 0.00 & 0.56 & 147 ngc 2903 & 09 32 09.7 & + 21 30 02 & sab(rs)bc & h ii & 6.3 & 0.07 & 0.53 & 24 ngc 3115 & 10 05 14.1 & @xmath5507 43 07 & s0 & & 6.7 & 0.10 & 0.66 & 43 ngc 3226 & 10 23 27.4 & + 19 53 55 & e2*,pec & liner & 23.4 & 0.02 & 0.20 & 10 ngc 3227 & 10 23 31.5 & + 19 51 48 & sab(s)a , pec & sy 1.5 & 20.6 & 0.02 & 0.55 & 157 ic 2574 & 10 28 22.5 & + 68 24 39 & sab(s)m & & 2.7 & 0.06 & 0.59 & 62 ngc 3310 & 10 38 46.1 & + 53 30 08 & sab(r)bc , pec & h ii & 18.7 & 0.00 & 0.22 & 170 ngc 3351 ( m 95 ) & 10 43 58.0 & + 11 42 15 & sb(r)b & h ii & 8.1 & 0.04 & 0.32 & 17 ngc 3379 ( m 105 ) & 10 47 49.9 & + 12 34 57 & e1 & & 8.1 & 0.05 & 0.13 & 65 ngc 3384 & 10 48 17.2 & + 12 37 49 & sb0 * & & 8.1 & 0.05 & 0.50 & 51 ngc 3389 & 10 48 27.9 & + 12 32 01 & sa(s)c & & 22.5 & 0.06 & 0.55 & 108 ngc 4038 & 12 01 52.9 & @xmath5518 51 54 & sb(s)m , pec & & 25.5 & 0.05 & 0.00 & ngc 4039 & 12 01 53.9 & @xmath5518 53 06 & sa(s)m , pec & & 25.3 & 0.05 & 0.00 & ngc 4151 & 12 10 33.0 & + 39 24 28 & sab(rs)ab*,pec & sy 1.5 & 20.3 & 0.00 & 0.36 & 145 ngc 4156 & 12 10 49.7 & + 39 28 24 & sb(rs)b & liner & 90.4 & 0.00 & 0.20 & 45 ngc 4214 & 12 15 39.5 & + 36 19 39 & iab(s)m & h ii & 3.5 & 0.00 & 0.18 & 132 ngc 4449 & 12 28 11.4 & + 44 05 40 & ibm & & 3.0 & 0.00 & 0.46 & 60 ngc 4647 & 12 43 32.4 & + 11 34 56 & sab(rs)c & & 16.8 & 0.04 & 0.20 & 135 ngc 4649 ( m 60 ) & 12 43 40.3 & + 11 32 58 & e2 & & 16.8 & 0.04 & 0.17 & 130 ngc 4736 ( m 94 ) & 12 50 53.6 & + 41 07 10 & ( r)sa(r)ab & liner & 4.3 & 0.00 & 0.22 & 95 ngc 5055 ( m 63 ) & 13 15 49.3 & + 42 02 06 & sa(rs)bc & h ii , liner & 7.2 & 0.00 & 0.47 & 102 ngc 5194 ( m 51 ) & 13 29 53.3 & + 47 11 48 & sa(s)bc , pec & sy 2.5 & 7.7 & 0.00 & 0.30 & 30 ngc 5236 ( m 83 ) & 13 37 00.3 & @xmath5529 52 04 & sab(s)c & h ii & 4.7 & 0.14 & 0.10 & 80 ngc 5253 & 13 39 55.9 & @xmath5531 38 41 & pec & h ii & 3.2 & 0.19 & 0.57 & 43 ngc 5457 ( m 101 ) & 14 03 12.5 & + 54 20 55 & sab(rs)cd & & 5.4 & 0.00 & 0.00 & lccrccccccc ngc 925 & 950314 & 1591 ( b5 ) & 990207 & pal1.5 & & & & 300 & & & & 990213 & pal1.5 & 3600 & 600 & 600 & & ngc 1068 & 950307 & 753 ( b5 ) & 990214 & pal1.5 & 3600 & 300 & 300 & 200 & ngc 1097 & 950312 & 1121 ( b5 ) & 951221 & lco2.5 & 3x400 & 3x200 & 3x200 & & 3x200 ngc 1313 & 950312 & 1071 ( b5 ) & 951224 & lco2.5 & 2x400 & 2x200 & 2x200 & & 2x200 ngc 1365 & 950315 & 974 ( b5 ) & 951223 & lco2.5 & 2x400 & 2x200 & 2x200 & & 2x200 ngc 1512/10 & 950315 & 949 ( b5 ) & 951220 & lco2.5 & 3x400 & 3x200 & 3x200 & & 3x200 ngc 1566 & 950316 & 1391 ( b5 ) & 951222 & lco2.5 & 2x400 & 2x200 & 2x200 & & 2x200 ngc 1672 & 950307 & 927 ( b5 ) & 951219 & lco2.5 & 4x400 & 3x200 & 3x200 & & 3x200 ngc 2403 & 950308 & 772 ( b1 ) & 971031 & pal1.5 & 1600 & 600 & 300 & & 300 & & & 990208 & pal1.5 & & & & 60 & ngc 2841 & 950308 & 1021 ( b1 ) & 940214 & pal5.0 & & 300 & & & 120,2x60 & & & 990213 & pal1.5 & 3600 & & 300 & 200 & ngc 2903 & 950307 & 549 ( b1 ) & 951123 & pal1.5 & & 900 & 600 & & 600 & & & 990208 & pal1.5 & & & & 300 & & & & 990214 & pal1.5 & 3600 & & & & ngc 3115 & 950307 & 1081 ( b1 ) & 951226 & lco2.5 & 300 & & 300 & & 300 ngc 3226/7 & 950316 & 1271 ( b1 ) & 960218 & pal1.5 & & 450 & & & & & & 990207 & pal1.5 & & & & 300 & ic 2574 & 950316 & 624 ( b1 ) & 960414 & pal5.0 & & & 3x600 & & 5x400 & & & 990208 & pal1.5 & & & & 300 & & & & 990214 & pal1.5 & 3600 & 600 & & & ngc 3310 & 950311 & 1131 ( b1 ) & 990207 & pal1.5 & & & & 300 & & & & 990214 & pal1.5 & & & 300 & & ngc 3351 & 950306 & 881 ( b1 ) & 960118 & pal1.5 & & 600 & 300 & & 2x600,300 & & & 990208 & pal1.5 & & & & 300 & ngc 3379/84/89 & 950306 & 1301 ( b1 ) & 990208 & pal1.5 & & & & 120 & ngc 4038/9 & 950307 & 881 ( b1 ) & 951222 & lco2.5 & 1200 & & 600 & & 600 ngc 4151/6 & 950313 & 833 ( b1 ) & 990207 & pal1.5 & & & & 300 & ngc 4214 & 950313 & 1061 ( b1 ) & 960514 & pal1.5 & & 600 & 2x300 & & 300 & & & 990208 & pal1.5 & & & & 300 & ngc 4449 & 950307 & 987 ( b1 ) & 960515 & pal1.5 & & 600 & 300 & & 300 & & & 990207 & pal1.5 & & & & 300 & ngc 4647/9 & 950311 & 1301 ( b1 ) & 990207 & pal1.5 & & & & 300 & ngc 4736 & 950312 & 1041 ( b1 ) & 990208 & pal1.5 & & 2x600 & & 300 & & & & 990213 & pal1.5 & 3600 & & & & & & & 990320 & pal1.5 & & & 300 & & ngc 5055 & 950315 & 1141 ( b1 ) & 990213 & pal1.5 & 3600 & 600 & 300 & 200 & ngc 5194 & 950312 & 1101 ( b1 ) & 980423 & pal1.5 & 6x1800 & & & & & & & 990207 & pal1.5 & & 900 & 600 & 300 & 300 ngc 5236 & 950306 & 819 ( b1 ) & 960623 & ctio1.5 & 4x300 & 4x300 & & 4x300 & 4x60 ngc 5253 & 950307 & 727 ( b5 ) & 951222 & lco2.5 & & 300 & & & & & & 951223 & lco2.5 & & & 3x600 & & 2x600 ngc 5457 & 950311 & 1311 ( b1 ) & 960515 & pal1.5 & & 600 & 300 & & 300

we investigate the thermal casimir interaction of a suspended graphene described by the dirac model with a plate made of dielectric or metallic materials . the reflection coefficients on graphene expressed in terms of a temperature - dependent polarization tensor are used . we demonstrate that for a graphene with nonzero mass gap parameter the casimir free energy remains nearly constant ( and the thermal correction negligibly small ) over some temperature interval . for the interaction of graphene with metallic plate , the free energy is nearly the same , irrespective of whether the metal is nonmagnetic or magnetic and whether it is described using the drude- or plasma - model approaches . the free energy computed using the dirac model was compared with that computed using the hydrodynamic model of graphene and big differences accessible for experimental observation have been found . for dielectric and nonmagnetic metallic plates described by the drude model these differences vanish with increasing temperature ( separation ) . however , for nonmagnetic metals described by the plasma model and for magnetic metals , a severe dependence on the chosen theoretical description of graphene remains even at high temperature . in all cases the analytic asymptotic expressions for the free energy at high temperature are obtained and found in a very good agreement with the results of numerical computations .

1209.3302

it has been known that nanostructures based on carbon possess unique mechanical , electrical and optical properties.@xcite among them particular attention has been given to graphene , a two - dimensional sheet of carbon atoms arranged in a hexagonal structure with low - energy electronic excitations described by the dirac equation.@xcite at the present time suspended graphene membranes up to @xmath0 m diameter are produced.@xcite this makes possible to investigate the van der waals and casimir interaction between graphene and different material structures , such as another sheet of graphene , atoms , molecules , dielectric and metallic plates , spheres etc . respective theoretical investigations were performed using the phenomenological density - functional methods,@xcite second - order perturbation theory@xcite and , for multilayered carbon nanostructures , using the lifshitz theory.@xcite to apply the lifshitz theory , one needs the reflection coefficients on graphene over a wide frequency region . it is customary to express the reflection coefficients in terms of the frequency - dependent dielectric permittivity , a concept which is not well defined for one - atom - thick carbon nanostructures . because of this , two models for the reflection coefficients on graphene with no use of dielectric permittivity were proposed , the hydrodynamic one@xcite and the dirac one.@xcite in the framework of the hydrodynamic model , graphene is considered as an infinitesimally thin positively charged sheet , carrying a homogeneous fluid with some mass and negative charge densities . the hydrodynamic model was applied for calculation of the casimir and casimir - polder interactions.@xcite in the framework of the dirac model , it is taken into account that for energies below a few ev the dispersion relation for quasiparticles in graphene is linear with respect to the momentum , whereas it is quadratic in the hydrodynamic model . the reflection coefficients of graphene at zero and nonzero temperature were found in refs . @xcite and @xcite , respectively . some calculations of the casimir - polder graphene - atom interaction were performed at zero temperature using both the hydrodynamic and dirac models.@xcite it should be remembered , however , that both these models are only approximations . as was already mentioned , the hydrodynamic model disregards the dirac character of charge carriers in graphene . as to the dirac model , it extends the linear dispersion relation for quasiparticles to any energy , whereas this property applies only at low energies . because of this , one can conclude that in calculations of the casimir and casimir - polder forces the dirac model of graphene should be applicable at large separations between the test bodies , whereas the hydrodynamic model might work at short separation distances . there is also another approach to the application of the lifshitz theory to graphene . in this approach graphene is characterized by a spatially nonlocal dielectric permittivity depending both on the frequency and the wave vector . different authors express such a dielectric permittivity either through the polarizability of one graphene layer @xcite or in the random phase approximation.@xcite it is well known that thermal casimir effect is a subject of debate and there are different theoretical approaches to its description.@xcite for two graphene sheets in a nonretarded regime it was found@xcite that a relatively large thermal correction to the casimir force at room temperature arises at short separation distances of tens of nanometers . this conclusion was qualitatively confirmed,@xcite using the dirac model of graphene , with an alternative explanation . the reason for the origin of large thermal correction for graphene is that the contribution of all terms with nonzero matsubara frequencies at room temperature becomes small in comparison with the zero - frequency term even at short separations . it was also found@xcite that in graphene - atom casimir - polder interaction the thermal effect depends crucially on the magnitude of a mass gap parameter in the dirac model . specifically , for a nonzero gap there exists an interval of temperatures ( separations ) where the thermal correction remains small with increasing temperature ( separation ) . the possibility of large thermal correction at short separations links the dirac model of graphene to the drude model used in the literature to describe the casimir effect between real metals.@xcite because of this , it is of much interest to investigate the thermal casimir effect in the interaction of graphene with real material bodies made of different materials . in this paper , we calculate the free energy of the casimir interaction between a suspended graphene membrane described by the dirac model and dielectric ( silicon , sapphire ) or metallic ( au , ni ) plates . in so doing materials of the plate are described by realistic dielectric permittivities taking into account the interband transitions of core electrons . we demonstrate that , similar to graphene - atom interaction , the behavior of thermal correction crucially depends on the mass gap parameter @xmath1 of the dirac model . note that although the dirac - type excitations in pristine graphene are gapless , the influence of electron - electron interaction , substrates , defects of structure , and other effects leads to a nonzero mass gap.@xcite specifically , we show that larger is the magnitude of mass gap parameter , wider is the separation ( temperature ) interval , where the thermal correction remains small with increasing separation ( temperature ) . for a metallic plate interacting with graphene , we perform all calculations using both the drude- and plasma - model approaches to the dielectric permittivity of metal . in the case when graphene described by the dirac model interacts with a metallic plate ( either nonmagnetic or magnetic ) the calculation results obtained using both approaches nearly coincide and do not depend on the magnetic properties . in contrast to the case of two drude metals , the thermal correction for a graphene - metal interaction has the same sign as the interaction energy at zero temperature , i.e. , the magnitude of the casimir free energy increases with increasing temperature . note that all results obtained for a free energy in graphene - plate geometry can be reformulated as the casimir force between a material sphere and a graphene sheet using the proximity force approximation ( pfa).@xcite as was recently shown,@xcite the error arising from the use of pfa is less than the ratio of separation distance to sphere radius . in this paper we also compare the predictions of the dirac model for the thermal casimir effect with respective predictions of the hydrodynamic model and discuss the application region of each . specifically , it is shown that the hydrodynamic and dirac models of graphene lead to different results at short separations and to nearly coinciding results at large separations for the casimir free energy of graphene interacting at room temperature with a dielectric plate or with a nonmagnetic metallic plate described by the drude model . for a nonmagnetic metallic plate described by the plasma model or for a magnetic plate the predictions of the hydrodynamic and dirac models of graphene are significantly different at all separations considered and can be discriminated experimentally . at large separations the asymptotic expressions for the casimir free energy of graphene - plate interaction are derived and compared with the computational results . the paper is organized as follows . in sec . ii we present the reflection coefficients of the electromagnetic oscillations on graphene in the dirac and hydrodynamic models . section iii is devoted to the thermal casimir interaction of graphene described by the dirac model with a dielectric plate made of silicon or sapphire . similar results for graphene interacting with a metallic plate made of au and ni are presented in sec . iv . in sec . v the theoretical predictions following from the dirac and hydrodynamic models are compared . our conclusions and discussions are contained in sec . the reflection coefficients on graphene in the dirac model were expressed in terms of the components of the polarization tensor in three - dimensional space - time in the following way@xcite @xmath19 here , the dimensionless components of the polarization tensor are defined as @xmath20 and trace stands for the sum of spatial components @xmath21 and @xmath22 . at nonzero temperature the explicit expression for the polarization tensor in the dirac model with arbitrary mass gap parameter @xmath1 and chemical potential @xmath23 was found in ref . @xcite . here we consider the case of undoped graphene and put @xmath24 . then in terms of our dimensionless variables the result of ref . @xcite for the 00-component of the polarization tensor takes the following equivalent form:@xcite @xmath25^{1/2 } } + \frac{8\alpha}{{\tilde{v}}_f^2}\int_{0}^{1}dx \label{eq4 } \\ & & ~\times \left\{\vphantom{\frac{{\tilde{\delta}}^2+\zeta_l^2x(1-x)}{\left[{\tilde{\delta}}^2 + x(1-x)f(\zeta_l , y)\right]^{1/2 } } } \frac{\tau}{2\pi}\ln\left[1 + 2\cos(2\pi lx)e^{-g(\tau,\zeta_l , y ) } + e^{-2g(\tau,\zeta_l , y)}\right ] -\frac{\zeta_l}{2}(1 - 2x ) \frac{\sin(2\pi lx)}{\cosh{g(\tau,\zeta_l , y)}+ \cos(2\pi lx ) } \right . \nonumber \\ & & ~ \left . + \frac{{\tilde{\delta}}^2+\zeta_l^2x(1-x)}{\left[{\tilde{\delta}}^2 + x(1-x)f(\zeta_l , y)\right]^{1/2}}\ , \frac{\cos(2\pi lx)+e^{-g(\tau,\zeta_l , y)}}{\cosh{g(\tau,\zeta_l , y)}+ \cos(2\pi lx ) } \right\}. \nonumber\end{aligned}\ ] ] in this expression , @xmath26 in the fine - structure constant ( @xmath27 is the electron charge ) , @xmath28 is the dimensionless mass gap parameter , the dimensionless fermi velocity is @xmath29 , and the dimensionless variable @xmath30 is defined as @xmath31 . equation ( [ eq4 ] ) also contains two dimensionless functions defined by @xmath32^{1/2}. \nonumber\end{aligned}\ ] ] the result of ref . @xcite for the trace of the polarization tensor in terms of our dimensionless variables is given by@xcite @xmath33 \int_{0}^{1}dx\frac{x(1-x)}{\left[{\tilde{\delta}}^2 + x(1-x)f(\zeta_l , y)\right]^{1/2 } } + \frac{8\alpha}{{\tilde{v}}_f^2}\int_{0}^{1}dx \label{eq6 } \\ & & ~\times \left\{\vphantom{\frac{{\tilde{\delta}}^2+\zeta_l^2x(1-x)}{\left[{\tilde{\delta}}^2 + x(1-x)f(\zeta_l , y)\right]^{1/2 } } } \frac{\tau}{2\pi}\ln\left[1 + 2\cos(2\pi lx)e^{-g(\tau,\zeta_l , y ) } + e^{-2g(\tau,\zeta_l , y)}\right ] \right . \nonumber \\ & & ~ -\frac{\zeta_l(1 - 2{\tilde{v}}_f^2)}{2}(1 - 2x ) \frac{\sin(2\pi lx)}{\cosh{g(\tau,\zeta_l , y)}+ \cos(2\pi lx ) } \nonumber \\ & & ~ \left . + \frac{{\tilde{\delta}}^2+x(1-x)[(1-{\tilde{v}}_f^2)^2\zeta_l^2- { \tilde{v}}_f^4y^2]}{\left[{\tilde{\delta}}^2 + x(1-x)f(\zeta_l , y)\right]^{1/2}}\ , \frac{\cos(2\pi lx)+e^{-g(\tau,\zeta_l , y)}}{\cosh{g(\tau,\zeta_l , y)}+ \cos(2\pi lx ) } \right\}. \nonumber\end{aligned}\ ] ] the properties of the reflection coefficients ( [ eq2 ] ) with the polarization tensor ( [ eq4 ] ) and ( [ eq6 ] ) were studied previously.@xcite in the framework of the hydrodynamic model discussed in sec . i the reflection coefficients on graphene take the more simple form@xcite @xmath34 here , the dimensionless characteristic wave number of the graphene is defined as @xmath35 , where the dimensional wave number is @xmath36 in eq . ( [ eq8 ] ) @xmath37 is the number of @xmath38-electrons per unit area , @xmath39 is the electron mass . note that the parameter @xmath40 of the hydrodynamic model does not depend on temperature . thus , the reflection coefficients ( [ eq7 ] ) and the free energy ( [ eq1 ] ) in the case of hydrodynamic model depend on the temperature only through the matsubara frequencies . a different situation arises in the dirac model . here , the reflection coefficients ( [ eq2 ] ) depend on @xmath3 not only through the matsubara frequencies , but also through the components of the polarization tensor . this links the dirac model of graphene to the drude - model approach to the thermal casimir force between real metals ( because the relaxation parameter of the drude model and , thus , the reflection coefficients are the explicit functions of temperature ) . in this respect the use of the hydrodynamic model of graphene in calculations of the casimir force is analogous to the plasma - model approach where the reflection coefficients depend on @xmath3 only through the matsubara frequencies . as to the reflection coefficient of electromagnetic oscillations on thick metallic plate ( semispace ) , they have the standard form@xcite @xmath41^{1/2}}{\varepsilon_l y+\left[y^2+\zeta_l^2(\varepsilon_l\mu_l- 1)\right]^{1/2 } } , \nonumber \\ & & r_{\rm te}^{(p)}(i\zeta_l , y)= \frac{\mu_l y-\left[y^2+\zeta_l^2(\varepsilon_l\mu_l- 1)\right]^{1/2}}{\mu_l y+\left[y^2+\zeta_l^2(\varepsilon_l\mu_l- 1)\right]^{1/2}}. \label{eq9}\end{aligned}\ ] ] here , both the dielectric permittivity @xmath42 and the magnetic permeability @xmath43 are calculated along the imaginary frequency axis . here , we investigate dependences of the free energy as functions of temperature and separation in the casimir interaction of a suspended graphene sheet with the dielectric plate made of different dielectrics . as the most typical dielectric materials we consider silicon ( si ) and sapphire ( al@xmath44o@xmath45 ) which possess relatively large values of the static dielectric permittivity [ @xmath46 and 10.102 , respectively ] but quite different behaviors along the axis of imaginary frequencies ( for si it is caused by electronic polarization , whereas for al@xmath44o@xmath45 by both electronic and ionic polarizations ) . we begin with the free energy of interaction of graphene described by the dirac model with si plate . the dielectric permittivity of si along the imaginary frequency axis was obtained@xcite by means of the kramers - kronig relations from the tabulated optical data for the complex index of refraction.@xcite it is assumed to be temperature - independent . computations were performed by eq . ( [ eq1 ] ) , where the reflection coefficients on graphene are given by eqs . ( [ eq2 ] ) , ( [ eq4 ] ) and ( [ eq6 ] ) , and on silicon by eq . ( [ eq9 ] ) with @xmath47 and @xmath48 specified above , over the temperature interval from 0 to 300k at two separation distances @xmath49 nm and @xmath50 m . for the mass gap parameter of the dirac model of graphene only the upper bound is known.@xcite for a suspended graphene we choose the realistic upper bound @xmath51ev . for a graphene deposited on substrate , @xmath1 can be several times larger.@xcite taking this upper bound into account , we perform all computations for @xmath52 , 0.05 , 0.01 , and 0ev . the computational results for the free energy per unit area are presented in fig . [ fg1 ] as functions of temperature , where the lines from top to bottom correspond to @xmath1 decreasing from 0.1ev to 0ev , ( a ) at the separation @xmath49 nm and ( b ) at @xmath50 m . as can be seen in fig . [ fg1](a , b ) , for each mass gap @xmath53 there exists the temperature interval where the free energy remains nearly constant with increasing temperature . the width of these intervals quickly decreases with decreasing @xmath1 . thus , at @xmath49 nm and @xmath54ev the free energy is nearly constant up to @xmath55k . at the same separation , but with @xmath56ev , the same property holds only up to @xmath57k . the widths of intervals , where the free energy remains nearly constant with increasing @xmath3 , are also narrowed with the increase of separation [ see fig . [ fg1](b ) ] . note that computations performed for @xmath58ev lead to nearly the same numerical results as for @xmath59 [ the lowest lines in fig . [ fg1](a , b ) ] . figure [ fg1 ] suggests that within the temperature interval , where the free energy is nearly flat , the thermal correction to the casimir energy at zero temperature should be relatively small . we confirm this conclusion by the direct computation of the thermal correction to the casimir energy defined as @xmath60 for the same values of parameters , as in fig . [ fg1 ] . the computational results for the thermal correction as a function of temperature are presented in fig . [ fg2 ] at separations ( a ) @xmath49 nm and ( b ) @xmath50 m . the lines labeled 1 , 2 , 3 , and 4 correspond to the values of mass gap parameter @xmath52 , 0.05 , 0.01 , and @xmath61ev , respectively . as expected , for each line with @xmath53 there is some interval where the thermal correction remains nearly zero . these intervals are just the same where the casimir free energy in fig . [ fg1 ] remains nearly flat . it is interesting to note that the thermal corrections are monotonously decreasing functions of temperature and have the same negative sign as the free energy . this contrasts with the drude model approach to the casimir force between real metals where the thermal correction over a wide temperature interval is positive making the free energy the nonmonotonous function of temperature.@xcite it is instructive also to compute the relative thermal correction to the casimir energy defined as @xmath62 the computational results are presented in fig . [ fg3 ] in percent as functions of temperature at separations distances ( a ) @xmath49 nm and ( b ) @xmath50 m . the lines 1 , 2 , 3 , and 4 are labeled in the same way as in fig . as can be seen in fig . [ fg3 ] , all the relative thermal corrections are monotonously increasing functions of temperature whose character depends crucially on the magnitude of a mass gap parameter @xmath1 . for example , at @xmath49 nm [ fig . [ fg3](a ) ] the relative thermal correction of line 1 ( @xmath54ev ) is nearly equal to zero below @xmath55k but achieves 120.44% at @xmath63k . to compare , at the same separation the relative thermal correction of line 4 ( @xmath58ev ) , which is much larger than the corrections of lines 13 at low temperatures , achieves only 59.40% at @xmath63k . it is seen that at room temperature the relative thermal correction in the casimir interaction of graphene with si is rather large even at relatively short separations . from fig . [ fg3](b ) it follows that at @xmath50 m the relative thermal correction at room temperature exceeds 4000% , i.e. , the absolute thermal correction exceeds the casimir energy at zero temperature by a factor of 40 . this makes graphene interesting for experimental investigation of thermal effects in the casimir force . next we consider the interaction of graphene described by the dirac model with sapphire plate . the dielectric permittivity of sapphire along the imaginary frequency axis is well described@xcite in the ninham - parsegian approximation and was already used@xcite in computations of the casimir force . computations of the free energy as a function of temperature were performed in the same way as for silicon . the computational results for @xmath49 nm are presented in fig . [ fg4](a ) by the four lines from top to bottom for @xmath52 , 0.05 , 0.01 , and @xmath61ev , respectively . similar to fig . [ fg1](a ) , there are temperature intervals where the free energy remains nearly constant . for sapphire , however , the respective magnitudes of the free energy are smaller than for a silicon . skipping the computational results at @xmath50 m [ which are similar to those presented in fig . [ fg1](b ) ] , we present in fig . [ fg4](b ) the more detailed computational results for the free energy in the temperature interval from 0k to 200k , where the lines from top to bottom correspond to @xmath52 , 0.09 , 0.08 , 0.07 , 0.06 , and 0.05ev , respectively . keeping in mind that the exact value of @xmath1 is not known , computations of this kind can be useful for the determination of @xmath1 from the comparison between experiment and theory . for sapphire , the computational results for the absolute and relative thermal corrections are similar to those for silicon ( see figs . [ fg2 ] and [ fg3 ] ) . because of this we do not present them here . here , we present the computational results for the interaction of graphene described by the dirac model with dielectric plates as a function of separation . the same equations and dielectric functions , as in sec . iiia , are used . taking into account that the casimir free energy strongly depends on separation , we normalize the results obtained on the casimir energy per unit area between two parallel plates made of ideal metal @xmath64 in fig . [ fg5](a ) the quantity @xmath65 at @xmath66k is plotted as a function of separation over the region from 50 nm to @xmath67 m . the lines from bottom to top correspond to the mass gap parameter equal to 0.1 , 0.05 , and @xmath68ev , respectively . in all cases the magnitude of the free energy decreases monotonously with the increase of separation ( the increase of @xmath65 is fig . [ fg5 ] is explained by the fact that @xmath69 decreases with separation faster than @xmath70 ) . in fig . [ fg5](b ) similar results are shown at @xmath63k . here , to avoid an overlap of the lines , we consider a more narrow separation region from 50 nm to @xmath71 m . the bottom and top lines correspond to @xmath52 and @xmath68ev , respectively . as can be seen from fig . [ fg5](b ) , at @xmath63k the dependence of the casimir free energy on the mass gap parameter of the dirac model is very weak . moreover , at separations above 50 nm the magnitude of the free energy @xmath6 decreases with separation as @xmath72 , so that the quantity @xmath65 is nearly linear function of separation . similar computations were performed for a graphene interacting with a sapphire plate . in fig . [ fg6](a ) the computational results for @xmath65 as a function of separation at @xmath66k are presented , where the lines from bottom to top correspond to @xmath73 , 0.05 , and @xmath68ev , respectively . these results are similar to those presented in fig . [ fg5](a ) for a silicon plate . again , at @xmath66k the casimir free energy is strongly affected by the value of @xmath1 . to illustrate this in more detail , in fig . [ fg6](b ) we plot @xmath65 over the separation region from 50 nm to @xmath74 m where the lines from bottom to top correspond to @xmath52 , 0.09 , 0.08 , 0.07 , 0.06 , and 0.05ev . at @xmath63k , similar to fig . [ fg5](b ) , the dependence of the free energy on @xmath1 becomes very weak and one obtains @xmath65 increasing nearly linear with the increase of @xmath2 . the asymptotic behavior of the casimir free energy at high temperature ( or , equivalently , at large separations ) can be obtained by considering the zero - frequency contribution to the lifshitz formula ( [ eq1 ] ) . this corresponds to large values of the dimensionless parameter @xmath30 introduced after eq . ( [ eq4 ] ) . by putting @xmath18 , @xmath75 in eq . ( [ eq4 ] ) one arrives at @xmath76 , \nonumber\end{aligned}\ ] ] where @xmath77^{1/2}. \label{eq14}\ ] ] equation ( [ eq13 ] ) can be identically rearranged to the form @xmath78 . \label{eq15}\end{aligned}\ ] ] the application region of eq . ( [ eq23 ] ) depends on the specific values of parameters . thus , for si at @xmath63k eq . ( [ eq23 ] ) leads to less than 1% errors in the values of the free energy , as compared with the results of numerical computations , at @xmath91 nm for graphene with @xmath54ev and at @xmath92 m for graphene with @xmath93ev . for si at @xmath66k and graphene with @xmath54ev eq . ( [ eq23 ] ) is not yet applicable at @xmath94 m and for graphene with @xmath93ev works well for @xmath95 m . it is interesting to compare the asymptotic expression ( [ eq23 ] ) with other results obtained in the literature . thus , using the nonlocal dielectric function in the random phase approximation , the free energy of graphene interacting with a dielectric substrate ( sio@xmath44 ) at large separations was found@xcite to decrease as @xmath96 . this is not in accordance with the main term of our result ( [ eq23 ] ) which demonstrates the classical limit , as is expected at large @xmath2 . note that another work@xcite models the dielectric properties of graphene by the drude - type function and arrives at the @xmath72 scaling for graphene - graphene interaction which satisfies the classical limit . the case of graphene interacting with metallic plate is of special interest . as was mentioned in sec . i , there are two different theoretical approaches to the description of casimir effect between real - metal plates . the drude model approach takes into account the relaxation properties of conduction electrons . in the framework of this approach , the imaginary part of the dielectric permittivity of the drude model is used to extrapolate @xmath97 , obtained from the measured optical data , to zero frequency . by contrast , the plasma model approach disregards relaxation processes and extrapolates @xmath98 to zero frequency by means of the simple plasma model . although the drude model approach may seem preferable , as it takes into account some really existing property of metals , the experimental situation more likely favors the plasma model approach . in a series of precise independent measurements performed by the two experimental groups@xcite the drude model approach was excluded at a high confidence level ( several experiments@xcite also excluded the influence of free charge carriers , that are present in dielectric materials at room temperature , on the casimir force ) . the two experiments that support the drude model approach@xcite are not independent measurements of the casimir force ; they are based on fitting procedures between measured data for the total force and theoretical predictions using hypothetical models for the electric contribution to it . here , we show that in the interaction of graphene described by the dirac model with metallic plate the results obtained are not sensitive to the approach used ( either drude or plasma ) . this is , however , not the case when graphene is described by the hydrodynamic model ( see sec . numerical computations of the casimir free energy per unit area between graphene described by the dirac model and au plate were performed by using eqs . ( [ eq1 ] ) , ( [ eq2 ] ) , ( [ eq4])([eq6 ] ) and ( [ eq9 ] ) with @xmath47 . the dielectric permittivity of au along the imaginary frequency axis was described either by the generalized drude - like model with temperature - dependent relaxation parameter@xcite or by the generalized plasma - like model.@xcite these models use the six - oscillator approximation for the optical data extrapolated to zero frequency by means of simple drude and plasma models , respectively , with the plasma frequency @xmath99ev and the relaxation parameter at room temperature @xmath100ev . at lower @xmath3 the lower values of @xmath101 according to the standard theory of electron - phonon interaction have been used.@xcite the computational results using the drude- and plasma - model approaches are found to be indistinguishable . thus , at @xmath66k the relative difference between the casimir free energies computed using both approaches achieves the maximum value of 0.02% at @xmath102 nm , does not depend on @xmath1 in the limits of our computational accuracy , and decreases with the increase of separation . at @xmath63k this difference achieves the maximum values of 0.06% at @xmath59 and 0.07% at @xmath54ev , and again decreases with increasing @xmath2 . figure [ fg7 ] presents the casimir free energy per unit area as a function of temperature ( a ) for @xmath49 nm and ( b ) for @xmath50 m . it can be seen that fig . [ fg7 ] demonstrates the same characteristic features , as figs . [ fg1 ] and [ fg4 ] plotted for dielectric plates , but the magnitudes of the free energy for the case of metallic plate are larger . the most important novel qualitative effect , which is found for both dielectrics and metals , is that for each nonzero @xmath1 the free energy is nearly unchanged with the increase of @xmath3 within some temperature interval . in fig . [ fg8 ] we present the computational results for the absolute thermal correction to the casimir energy at zero temperature , defined in eq . ( [ eq10 ] ) , as a function of temperature . the lines 1 , 2 , 3 , and 4 correspond to the values of the mass gap parameter @xmath52 , 0.05 , 0.01 , and @xmath61ev , respectively . this figure is analogous to fig . [ fg2 ] plotted for si . it demonstrates that for a metallic plate the thermal correction in the graphene - plate geometry behaves qualitatively in the same way as for a dielectric plate , but with slightly larger magnitudes of the thermal correction . the computational results for the relative thermal correction , defined in eq . ( [ eq11 ] ) are presented in fig . [ fg9 ] as a function of temperature . the lines 1 , 2 , 3 , and 4 again correspond to the same respective @xmath1 , as in fig . figure [ fg9 ] is analogous to fig . [ fg3 ] plotted for a dielectric plate ( silicon ) . for a metallic plate at @xmath49 nm the relative thermal correction at room temperature appears only slightly larger than for a dielectric plate . at @xmath50 m at room temperature the relative thermal correction for au is smaller than for si . this is explained by different values of the casimir energy at zero temperature . keeping in mind that in most experiments on the casimir force the temperature is preserved constant and measurements are performed at different separation distances , here we present the computational results for a free energy of graphene - metal interaction as a function of separation . in fig . [ fg10](a ) the free energy of graphene interacting with au plate normalized on the casimir energy between ideal metal planes ( [ eq12 ] ) is shown . the three lines from bottom to top correspond to the values of mass gap parameter @xmath52 , 0.05 , and @xmath68ev , respectively . the obtained values of the free energy are larger than for graphene interacting with si plate [ compare with fig . [ fg5](a ) ] . at @xmath63k the dependence of the computational results on the mass gap parameter becomes not so pronounced as in fig . [ fg10](a ) . this can be seen in fig . [ fg10](b ) where the bottom and top lines correspond to @xmath54ev and @xmath68ev , respectively . it is interesting to consider the casimir interaction of graphene with a ferromagnetic metal . it was shown@xcite that at room temperature the ferromagnetic properties of real metals may influence the casimir force only through the contribution of the zero - frequency term of the lifshitz formula . recently the gradient of the casimir force between a nonmagnetic au sphere and a magnetic metal ( ni ) plate has been measured.@xcite we have computed the casimir free energy per unit area between a graphene described by the dirac model and ni plate using the same formalism , as for an au plate . the dielectric permittivity of ni along the imaginary frequency axis was found from the tabulated optical data@xcite extrapolated to zero frequency either by the drude or by the plasma model with the plasma frequency @xmath103ev and the relaxation parameter at room temperature @xmath104ev.@xcite the value of @xmath105 for the static magnetic permeability of ni has been used . it was found that relative differences in the computational results for the free energy of graphene - ni interaction , when ni is described using the drude- and plasma - model approaches , are as small as computed above for the interaction of graphene with an au plate . the influence of magnetic properties on the free energy was also shown to be negligibly small . the relative difference between the free energies of graphene - ni and graphene - au interactions for graphene with @xmath54ev computed at @xmath63k is equal to 6% at @xmath49 nm and decreases to 1% at @xmath50 m . it is less for smaller values of the mass gap parameter . note that even these small differences are not due to magnetic properties of ni but due to different plasma frequency and optical properties of ni as compared to au . now we derive the analytic expression for the casimir free energy of graphene described by the dirac model interacting with metallic plate at @xmath106 . the contribution of the tm reflection coefficient for graphene interacting with dielectric plate was obtained in eq . ( [ eq22 ] ) . taking into account that for metallic materials @xmath107 defined in eq . ( [ eq20 ] ) is equal to unity , one obtains from eq . ( [ eq22 ] ) @xmath108 . \label{eq24}\ ] ] calculating the integral with respect to @xmath13 and using eq . ( [ eq17 ] ) , we arrive at @xmath109 \label{eq25 } \\ & & ~~ \approx -\frac{k_bt}{16\pi a^2}\zeta(3)\left[1- \frac{\pi\tilde{v}_f^2}{2\alpha\tau\ln\left(2\cosh \frac{\pi\tilde{\delta}}{\tau}\right)}\right ] , \nonumber\end{aligned}\ ] ] where @xmath110 is the riemann zeta function . this result in the special case @xmath111 was obtained in ref . note that for a dielectric plate considered in sec . iiic the contribution of the tm mode was in fact equal to the total free energy because @xmath83 for dielectrics . for metals this is in general not so ( see below ) . now we are in a position to consider metallic plates made of nonmagnetic and magnetic metals described within both the drude and the plasma model approaches and in all cases find the high - temperature behavior of the total casimir free energy . we begin with a nonmagnetic metal described by the drude model approach . in this case from eq . ( [ eq9 ] ) one obtains that @xmath83 and in accordance with eq . ( [ eq31 ] ) the te contribution to the free energy @xmath122 vanishes . thus , for the plate made of a nonmagnetic drude metal the total free energy of graphene - metal interaction at high temperature is given by eq . ( [ eq25 ] ) . next we consider a nonmagnetic metal described by the plasma - model approach . in this case from eq . ( [ eq9 ] ) we have @xmath123 where the parameter @xmath124 is defined as @xmath125 and @xmath126 is the effective penetration depth of electromagnetic oscillations into the metal . substituting eq . ( [ eq32 ] ) in eq . ( [ eq31 ] ) and integrating with respect to @xmath13 , one obtains @xmath127 the total asymptotic expression for the free energy at high temperature is given by the sum of ( [ eq25 ] ) and ( [ eq34 ] ) . note that the contribution of the te mode ( [ eq34 ] ) is a negligibly small correction because @xmath128 . we are coming now to the consideration of magnetic metals described by the drude model . from eq . ( [ eq9 ] ) it follows @xmath129 the substitution of this reflection coefficient in eq . ( [ eq31 ] ) results in @xmath130 this term is again negligibly small , as compared to @xmath131 , so that the total casimir free energy at high temperature is well described by eq . ( [ eq25 ] ) . finally we consider the asymptotic expression for the free energy of graphene interacting with a magnetic metal described by the plasma model . in this case from eq . ( [ eq9 ] ) we get @xmath132 . \nonumber\end{aligned}\ ] ] this is similar to eq . ( [ eq32 ] ) with the replacement of @xmath124 for @xmath133 . thus , instead of eq . ( [ eq34 ] ) , one obtains @xmath134 . \label{eq38}\end{aligned}\ ] ] the total asymptotic expression for the free energy at high temperature is given by the sum of eq . ( [ eq25 ] ) and negligibly small addition ( [ eq38 ] ) depending on the properties of magnetic metal . the obtained analytic expressions were found in good agreement with the results of numerical computations within appropriate temperature ( separation ) intervals . on this section we compare the computational results for the free energy of graphene - plate interaction obtained using two different models of graphene discussed in secs . i and ii . we find separation regions where the predictions of both models are distinct and similar and compare respective asymptotic expressions for the free energy at high temperature ( large separations ) . keeping in mind the possibility to compare theoretical predictions for the casimir force with the experimental data , we calculate the free energy of graphene - plate interaction as a function of separation for both dielectric and metallic plates . computations were performed using the lifshitz formula ( [ eq1 ] ) where the reflection coefficients ( [ eq17 ] ) for graphene in the framework of the hydrodynamic model were used . the computational results for @xmath65 as a function of separation are presented in fig . [ fg11 ] by the dashed lines ( a ) at @xmath66k and ( b ) at @xmath63k . in the same figure the respective results for @xmath65 computed using the dirac model of graphene are reproduced from fig . [ fg5](a , b ) by the solid lines . the solid lines in fig . [ fg11](a ) from bottom to top correspond to @xmath52 , 0.05 , and @xmath68ev , respectively . note that in the scale used in fig . [ fg11](b ) the two lines of fig . [ fg5](b ) overlap . they are shown as a single solid line in fig . [ fg11](b ) . as is seen in fig . [ fg11](a ) , at @xmath66k the hydrodynamic model of graphene predicts much larger magnitudes of the casimir free energy than the dirac model . thus , at @xmath66k the predictions of the hydrodynamic model for @xmath70 at @xmath135 m is by factors of 36.0 and 6.7 larger than predictions of the dirac model with @xmath54ev and @xmath136ev , respectively . at @xmath137 m the respective factors are 97.9 and 4.2 . at @xmath63k [ see fig . [ fg11](b ) ] the predictions of the hydrodynamic model are larger than the predictions of the dirac model by the factors of 2.5 and 1.3 at @xmath135 m and @xmath137 m , respectively . as is seen in fig . [ fg11](b ) , at @xmath63k , @xmath138 m the asymptotic regime of large @xmath30 is already achieved and the predictions of the hydrodynamic and dirac models almost coincide . at @xmath66k [ fig . [ fg11](a ) ] the asymptotic regime of large @xmath30 is achieved at much larger separations than those shown in the figure . now we compare the predictions of the hydrodynamic and dirac models of graphene interacting with a metallic plate . all computations were performed using the same formalism as above . we considered the plates made of a nonmagnetic metal au and a magnetic metal ni . each of these metals was described either using the drude- or the plasma - model approach . the computational results for the normalized free energy @xmath65 are presented in fig . [ fg12 ] at @xmath63k as a function of separation . in this figure , the solid line is reproduced from fig . [ fg10](b ) [ note that in the scale of fig . [ fg12 ] the two solid lines in fig . [ fg10](b ) overlap ] . this line is obtained using the dirac model of graphene interacting with a metallic plate , be it au or ni ( see sec . the dashed and dotted lines 1 , 2 , 3 , and 4 are obtained from computations using the hydrodynamic model of graphene . the computational data shown by the dashed lines 1 and 3 were found for an au plate described by the plasma- and drude - model approaches , respectively . the data shown by the dotted lines 2 and 4 were computed for a ni plate also described by the respective plasma- and drude - model approaches . as can be seen in fig . [ fg12 ] , the free energy of graphene - metal interaction with graphene described by the hydrodynamic model strongly depends on the metal used ( magnetic or nonmagnetic ) and on the chosen approach to the description of dielectric properties . only for the nonmagnetic metal ( au ) described by the drude model ( line 3 ) the behavior of the free energy is qualitatively similar to the case of dielectric plate and nears the prediction of the dirac model at large separations [ compare with fig . [ fg11](b ) ] . as an example , the casimir free energy of graphene - metal interaction computed using the hydrodynamic model of graphene at @xmath50 m , @xmath63k is larger than the same quantity computed using the dirac model by factors of 2.37 ( for an au plate described by the plasma model ) , 2.15 ( for a ni plate described by the plasma model ) , 1.93 ( for an au plate described by the drude model ) , and 1.55 ( for a ni plate described by the drude model ) . this allows comparison between different theoretical predictions and experimental data . it is interesting that for lines 13 the magnitude of @xmath6 predicted by the hydrodynamic model is always larger than for the predictions of the dirac model ( solid line ) . as to the line 4 ( ni described by the drude model ) , the prediction for @xmath70 from the hydrodynamic model becomes less than from the dirac model at @xmath139 m and remains so at larger separations . thus , at @xmath94 m the ratio between the predictions of hydrodynamic and dirac models is equal to 0.42 . as was mentioned in sec . iiic , at high temperature ( large separations ) the zero - frequency term of the lifshitz formula ( [ eq1 ] ) alone determines the casimir free energy . the reflection coefficients on the graphene , described by the hydrodynamic model , at zero frequency follow from eq . ( [ eq7 ] ) @xmath140 taking into account that for a hydrodynamic model at @xmath63k the high - temperature regime starts at @xmath141 m , we find from eq . ( [ eq8 ] ) that at these separations @xmath142 . substituting eq . ( [ eq39 ] ) for the reflection coefficient @xmath143 into the zero - frequency term of the lifshitz formula ( [ eq1 ] ) , for the tm contribution to the free energy at high temperature one obtains @xmath144 . \label{eq40}\ ] ] for a dielectric plate , using eq . ( [ eq20 ] ) and integrating in eq . ( [ eq40 ] ) , we get @xmath145 as noted in sec . iiic , for a dielectric plate the contribution of the te mode to the free energy vanishes . thus , eq . ( [ eq41 ] ) provides the complete expression for the free energy of graphene - dielectric interaction at high temperature . equation ( [ eq41 ] ) coincides with the first term in eq . ( [ eq23 ] ) obtained for graphene described by the dirac model [ remind that for the dirac model of graphene the asymptotic expression ( [ eq23 ] ) becomes applicable at much smaller separations at the same room temperature ; see fig . [ fg11](b ) ] . for a metallic plate , replacing @xmath107 with unity , we arrive at @xmath146 this result coincides with the first term of the asymptotic expression ( [ eq25 ] ) obtained for graphene described by the dirac model . equation ( [ eq42 ] ) provides the total asymptotic expression for the free energy only in the case when a nonmagnetic metal of the plate is described by the drude model ( see the dashed line 3 approaching the solid line in fig . [ fg12 ] when the separation distance increases ) . now we consider the contribution of the te mode to the free energy of graphene - metal casimir interaction . from eq . ( [ eq39 ] ) we have @xmath147 where @xmath148 takes the maximum value @xmath149 and decreases with further increase of separation . then , for the contribution of the te mode to the free energy at high temperature one obtains @xmath150 . \label{eq44}\end{aligned}\ ] ] for a magnetic metal described by the plasma model we use the reflection coefficient ( [ eq37 ] ) and substitute it in eq . ( [ eq44 ] ) . all calculations are similar to the case of nonmagnetic metal , but the quantity @xmath124 is replaced for @xmath133 . thus , instead of eq . ( [ eq45 ] ) , we obtain @xmath162 } 1 - 2\left[\beta+2\delta\sqrt{\mu(0)}\right]\right . \label{eq53 } \\ & & ~~\left . + \beta^2\left[1+\frac{\pi^2}{6\zeta(3)}\right]+ 2\delta\sqrt{\mu(0)}\left[\beta+\delta\sqrt{\mu(0)}\right ] \frac{\pi^2}{3\zeta(3)}\right\}. \nonumber\end{aligned}\ ] ] in this equation , however , it is impermissible to neglect by the quantity @xmath133 as compared to @xmath163 . substituting the numerical values of constants to eq . ( [ eq53 ] ) , we find @xmath164 \right . \nonumber \\ & & ~~\left . + 2.368\beta^2 + 5.474\delta\sqrt{\mu(0)}\left[\beta+\delta\sqrt{\mu(0)}\right ] \right\}. \label{eq54}\end{aligned}\ ] ] by combining eq . ( [ eq54 ] ) and eq . ( [ eq42 ] ) , we arrive at the total free energy of graphene interacting with a magnetic metal described by the plasma model @xmath165 \right\}. \label{eq55}\end{aligned}\ ] ] the main contribution to eq . ( [ eq55 ] ) is by a multiple two larger than the main contribution to eq . ( [ eq25 ] ) obtained for the dirac model of graphene ( compare the dotted line labeled 2 and the solid line in fig . [ fg12 ] ) . the obtained analytic asymptotic expressions for the free energy of graphene described using the hydrodynamic model in graphene - metallic plate geometry is in good agreement with the results of numerical computations . thus , for an au plate described by the drude- and plasma - model approaches at @xmath94 m , @xmath63k , the results of analytic and numerical calculations differ by 0.35% and 1.4% , respectively . for ni the same relative differences are equal to 9.3% and 3% . note that relatively large deviation obtained for ni plate described by the drude - model approach is explained by the fact that in this case at @xmath94 m the linear asymptotic regime is not yet achieved ( see the dotted line labeled 4 in fig . [ fg12 ] ) . to conclude , we emphasize that all results for the casimir free energy obtained in this and previous sections are simply convertable to the casimir force @xmath166 in the experimentally relevant configuration of a sphere above a plate used in most of experiments on measuring the casimir force . this can be done by means of the pfa which states that @xmath167 where @xmath168 is the radius of the sphere . in our case , where @xmath169 is the casimir free energy between a graphene sheet and a material plate , the force @xmath166 defined in eq . ( [ eq56 ] ) can be considered as the casimir force acting between a graphene sheet and a material sphere of radius @xmath168 . as was mentioned in sec . i , the error introduced by the use of the pfa was recently proved@xcite to be smaller than @xmath170 ( i.e. or order of 0.1% for the typical values of parameters ) . keeping in mind that many effects considered above far exceed 100% , the use of the pfa in the comparison between experiment and theory is fully justified . in the foregoing , we have investigated the casimir free energy and the thermal correction to the casimir energy at zero temperature for a suspended graphene sheet interacting with a material plate , either dielectric or metallic . in so doing graphene was described by the fully relativistic dirac model with temperature - dependent polarization tensor . the dielectric properties of the plate were described by the frequency - dependent dielectric permittivity taking into account the interband transitions of core electrons . for a metallic plate both the drude- and plasma - model approaches suggested in the literature have been used . the main novel result obtained for both dielectric and metallic plates is that for graphene with any nonzero mass gap parameter @xmath1 there exists temperature interval where the casimir free energy remains nearly constant . this happens under the condition @xmath171 , which should be satisfied with a large safety margin . if this condition is satisfied , the thermal correction to the casimir energy at zero temperature remains negligibly small . we have also demonstrated that under the condition @xmath172 the thermal correction becomes relatively large . this makes possible large thermal corrections for a graphene sheet interacting with material plate at rather low temperature ( short separations ) . with respect to the interaction with a metallic plate , it was shown that for graphene described by the dirac model the computational results for the free energy are nearly independent on whether the drude- or plasma - model approach to the dielectric permittivity of metal is used . to a large extent the free energy of graphene interacting with metallic plate is also independent on whether metal is nonmagnetic or magnetic if graphene is described by the dirac model . in all cases considered ( dielectric or metallic plate , nonmagnetic or magnetic , described by the drude- or plasma - model approach ) the analytic asymptotic expressions for the casimir free energy at high temperature ( large separations ) have been obtained and compared with the results of numerical computations . the casimir free energies obtained using the dirac model of graphene were compared with those calculated using the hydrodynamic model . it was shown that at moderate temperatures ( separations ) the magnitudes of the free energy computed using the hydrodynamic model of graphene differ significantly from that computed using the dirac model . this can be used for the experimental test of these models . at large separations ( high temeperature ) the theoretical predictions from both models of graphene nearly coincide for a dielectric plate and for a nonmagnetic metallic plate described by the drude model . for a nonmagnetic metallic plate described by the plasma model and for a magnetic plate described by any model the hydrodynamic and dirac descriptions of graphene lead to quite different results for the free energy at large separations ( high temperature ) . this fact can be also used for the experimental test of different models . in this respect the investigation of the interaction between graphene and metamaterials@xcite is also of large interest . we have also found analytic asymptotic expressions for the free energy at high temperature ( large separations ) when graphene described by the hydrodynamic model interacts with a dielectric plate or with a plate made of a nonmagnetic or magnetic metal . in the last two cases both the drude- and plasma - model approaches have been used for a description of the dielectric properties of metal . the calculation results obtained from the asymptotic expressions were found in a very good agreement with the results of numerical computations . this work was supported by the dfg grant bo 1112/211 . g.l.k . and v.m.m . are grateful to the institute for theoretical physics , leipzig university , where this work was performed , for kind hospitality . 99 m. s. dresselhaus , physica status solidi ( b ) * 248 * , 1566 ( 2011 ) . a. h. castro neto , f. guinea , n. m. r. peres , k. s. novoselov , and a. k. geim , rev . phys . * 81 * , 109 ( 2009 ) . b. alemn , w. regan , s. aloni , v. altoe , n. alem , c. girit , b. geng , l. maserati , m. crommie , f. wang , and a. zettl , acs nano * 4 * , 4762 ( 2010 ) . a. bogicevic , s. ovesson , p. hyldgaard , b. i. lundqvist , h. brune , and d. r. jennison , phys . * 85 * , 1910 ( 2000 ) . e. hult , p. hyldgaard , j. rossmeisl , and b. i. lundqvist , phys . b * 64 * , 195414 ( 2001 ) . j. jung , p. garca - gonzlez , j. f. dobson , and r. w. godby , phys . rev . b * 70 * , 205107 ( 2004 ) . j. f. dobson , a. white , and a. rubio , phys . lett . * 96 * , 073201 ( 2006 ) . a. n. rudenko , f. j. keil , m. i. katsnelson , and a. i. lichtenstein , phys . b * 84 * , 085438 ( 2011 ) . i. v. bondarev and ph . lambin , phys . b * 70 * , 035407 ( 2004 ) . e. v. blagov , g. l. klimchitskaya , and v. m. mostepanenko , phys . b * 71 * , 235401 ( 2005 ) . g. barton , j. phys . a * 37 * , 1011 ( 2004 ) . g. barton , j. phys . a * 38 * , 2997 ( 2005 ) . g. w. semenoff , phys . * 53 * , 2449 ( 1984 ) . d. p. divincenzo and e. j. mele , phys . b * 29 * , 1685 ( 1984 ) . m. bordag , j. phys . a : math . gen . * 39 * , 6173 ( 2006 ) . m. bordag , b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . b * 74 * , 205431 ( 2006 ) . e. v. blagov , g. l. klimchitskaya , and v. m. mostepanenko , phys . b * 75 * , 235413 ( 2007 ) . m. bordag , i. v. fialkovsky , d. m. gitman , and d. v. vassilevich , phys . b * 80 * , 245406 ( 2009 ) . i. v. fialkovsky , v. n. marachevsky , and d. v. vassilevich , phys . rev . b * 84 * , 035446 ( 2011 ) v. churkin , a. b. fedortsov , g. l. klimchitskaya , and v. a. yurova , phys . b * 82 * , 165433 ( 2010 ) . t. e. judd , r. g. scott , a. m. martin , b. kaczmarek , and t. m. fromhold , new . j. phys . * 13 * , 083020 ( 2011 ) . b. e. sernelius , europhys . 95 * , 57003 ( 2011 ) . j. sarabadani , a. naji , r. asgari , and r. podgornik , phys . b * 84 * , 155407 ( 2011 ) . v. b. bezerra , r. s. decca , e. fischbach , b. geyer , g. l. klimchitskaya , d. e. krause , d. lpez , v. m. mostepanenko , and c. romero , phys . rev . e * 73 * , 028101 ( 2006 ) . g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , rev . phys . * 81 * , 1827 ( 2009 ) . j. s. hye , i. brevik , j. b. aarseth , and k. a. milton , j. phys . gen . * 39 * , 6031 ( 2006 ) . g. gmez - santos , phys . b * 80 * , 245424 ( 2009 ) . m. chaichian , g. l. klimchitskaya , v. m. mostepanenko , and a. tureanu , phys . a * 86 * , 012515 ( 2012 ) . s. a. jafari , j. phys . : cond . mat . * 24 * , 205802 ( 2012 ) . k. pyatkovskiy , j. phys . : cond . mat . * 21 * , 025506 ( 2009 ) . v. p. gusynin , s. g. sharapov , and j. p. carbotte , new j. phys . * 11 * , 095013 ( 2009 ) . gusynin and s. g. sharapov , phys . b * 73 * , 245411 ( 2006 ) . m. bordag , g. l. klimchitskaya , u. mohideen , and v. m. mostepanenko , _ advances in the casimir effect _ ( oxford university press , oxford , 2009 ) . g. bimonte , t. emig , r. l. jaffe , and m. kardar , europhys . lett . * 97 * , 50001 ( 2012 ) . l. p. teo , m. bordag , and v. nikolaev , phys . d * 84 * , 125037 ( 2011 ) . g. bimonte , t. emig , and m. kardar , appl . . lett . * 100 * , 074110 ( 2012 ) . a. o. caride , g. l. klimchitskaya , v. m. mostepanenko , and s. i. zanette , phys . a * 71 * , 042901 ( 2005 ) . b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . a * 72 * , 022111 ( 2005 ) . , vol ii , ed . e. d. palik ( academic , new york , 1991 ) . l. bergstrm , adv . colloid interface sci . * 70 * , 125 ( 1997 ) . d. drosdoff and l. m. woods , phys . b * 82 * , 155459 ( 2010 ) . r. s. decca , e. fischbach , g. l. klimchitskaya , d. e. krause , d. lpez , and v. m. mostepanenko , phys . d * 68 * , 116003 ( 2003 ) . r. s. decca , d. lpez , e. fischbach , g. l. klimchitskaya , d. e. krause , and v. m. mostepanenko , ann . ( n.y . ) * 318 * , 37 ( 2005 ) . r. s. decca , d. lpez , e. fischbach , g. l. klimchitskaya , d. e. krause , and v. m. mostepanenko , phys . d * 75 * , 077101 ( 2007 ) . r. s. decca , d. lpez , e. fischbach , g. l. klimchitskaya , d. e. krause , and v. m. mostepanenko , eur . j. c * 51 * , 963 ( 2007 ) . chang , a. a. banishev , r. castillo - garza , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . b * 85 * , 165443 ( 2012 ) . f. chen , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , optics express * 15 * , 4823 ( 2007 ) . f. chen , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . b * 76 * , 035338 ( 2007 ) . j. m. obrecht , r. j. wild , m. antezza , l. p. pitaevskii , s. stringari , and e. a. cornell , phys . rev . lett . * 98 * , 063201 ( 2007 ) . g. l. klimchitskaya and v. m. mostepanenko , j. phys . a : math . theor . * 41 * , 312002(f ) ( 2008 ) . chang , a. a. banishev , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . lett . * 107 * , 090403 ( 2011 ) . a. a. banishev , c .- c . chang , r. castillo - garza , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . b * 85 * , 045436 ( 2012 ) . a. o. sushkov , w. j. kim , d. a. r. dalvit , and s. k. lamoreaux , nature phys . * 7 * , 230 ( 2011 ) . d. garcia - sanches , k. y. fong , h. bhaskaran , s. lamoreaux , and h. x. tang , phys . lett . * 109 * , 027202 ( 2012 ) . g. bimonte , phys . a * 81 * , 062501 ( 2010 ) . r. zandi , t. emig , and u. mohideen , phys . b * 81 * , 195423 ( 2010 ) . v. b. bezerra , g. l. klimchitskaya , v. m. mostepanenko , and c. romero , phys . rev . a * 69 * , 022119 ( 2004 ) . b. geyer , g. l. klimchitskaya , and v. m. mostepanenko , phys . b * 81 * , 104101 ( 2010 ) . g. l. klimchitskaya , b. geyer , and v. m. mostepanenko , int . j. mod a * 25 * , 2293 ( 2010 ) . a. a. banishev , c .- c . chang , g. l. klimchitskaya , v. m. mostepanenko , and u. mohideen , phys . b * 85 * , 195422 ( 2012 ) , vol i , ed . e. d. palik ( academic , new york , 1985 ) . m. a. ordal , r. j. bell , r. w. alexander jr . , l. l. long , and m. r. querry , appl * 24 * , 4493 ( 1985 ) . v. n. marachevsky , int . j. mod * 14 * , 435 ( 2012 ) . d. drosdoff and l. m. woods , phys . a * 84 * , 062501 ( 2011 ) .

we use galaxy groups selected from the sloan digital sky survey ( sdss ) together with mass models for individual groups to study the galaxy - galaxy lensing signals expected from galaxies of different luminosities and morphological types . we compare our model predictions with the observational results obtained from the sdss by mandelbaum et al . ( 2006 ) for the same samples of galaxies . the observational results are well reproduced in acdm model based on the wmap 3-year data , but acdm model with higher , such as the one based on the wmap 1-year data , significantly over - predicts the galaxy - galaxy lensing signal . we model , separately , the contributions to the galaxy - galaxy lensing signals from different galaxies : central versus satellite , early - type versus late - type , and galaxies in haloes of different masses . we also examine how the predicted galaxy - galaxy lensing signal depends on the shape , density profile , and the location of the central galaxy with respect to its host halo . [ firstpage ] dark matter - large - scale structure of the universe - galaxies : haloes - methods : statistical

0807.4934

according to the current paradigm of structure formation , galaxies form and reside inside extended cold dark haloes . while the formation and evolution of dark matter haloes in the cosmic density field is mainly determined by gravitational processes , the formation and evolution of galaxies involves much more complicated , and poorly understood processes , such as radiative cooling , star formation , and all kinds of feedback . one important step in understanding how galaxies form and evolve in the cosmic density field is therefore to understand how the galaxies of different physical properties occupy dark matter haloes of different masses . theoretically , the connection between galaxies and dark matter haloes can be studied using numerical simulations ( e.g. , katz , weinberg & hernquist 1996 ; pearce et al . 2000 ; springel 2005 ; springel et al . 2005 ) or semi - analytical models ( e.g. white & frenk 1991 ; kauffmann et al . 1993 , 2004 ; somerville & primack 1999 ; cole et al . 2000 ; van den bosch 2002 ; kang et al . 2005 ; croton et al . these approaches try to model the process of galaxy formation from first principles . however , since our understanding of the relevant processes is still poor , the predicted connection between the properties of galaxies and dark matter haloes needs to be tested against observations . more recently , the halo occupation model has opened another avenue to probe the galaxy - dark matter halo connection ( e.g. jing , mo & brner 1998 ; peacock & smith 2000 ; berlind & weinberg 2002 ; cooray & sheth 2002 ; scranton 2003 ; yang , mo & van den bosch 2003 ; van den bosch , yang & mo 2003 ; yan , madgwick & white 2003 ; tinker et al . 2005 ; zheng et al . 2005 ; cooray 2006 ; vale & ostriker 2006 ; van den bosch et al . this technique uses the observed galaxy luminosity function and clustering properties to constrain the average number of galaxies of given properties that occupy a dark matter halo of given mass . although the method has the advantage that it can yield much better fits to the data than the semi - analytical models or numerical simulations , one typically needs to assume a somewhat ad - hoc functional form to describe the halo occupation model . a more direct way of studying the galaxy - halo connection is to use galaxy groups , provided that they are defined as sets of galaxies that reside in the same dark matter halo . recently , yang et al . ( 2005 ; 2007 ) have developed a halo - based group finder that is optimized for grouping galaxies that reside in the same dark matter halo . using mock galaxy redshift surveys constructed from the conditional luminosity function model ( e.g. yang et al . 2003 ) and a semi - analytical model ( kang et al . 2005 ) , it is found that this group finder is very successful in associating galaxies with their common dark matter haloes ( see yang et al . 2007 ; hereafter y07 ) . the group finder also performs reliably for poor systems , including isolated galaxies in small mass haloes , making it ideally suited for the study of the relationship between galaxies and dark matter haloes over a wide range of halo masses . however , in order to interpret the properties of the galaxy systems in terms of dark matter haloes , one needs to know the halo mass associated with each of the groups . one approach commonly adopted is to use some halo mass indicator ( such as the total stellar mass or luminosity contained in member galaxies ) to rank the groups . with the assumption that the corresponding halo masses have the same ranking and that the mass function of the haloes associated with groups is the same as that given by a model of structure formation , one can assign a halo mass to each of the observed groups . this approach was adopted by y07 for the group catalogue used in this paper . there are three potential problems with this approach . first , the approach is model - dependent , in the sense that the assumption of a different model of structure formation will lead to a different halo mass function , and hence assign different masses to the groups . second , even if the assumed model of structure formation is correct , it is still not guaranteed that the mass assignment based on the ranking of group stellar mass ( or luminosity ) is valid . finally , even if all groups are assigned with accurate halo masses , the question how dark matter is distributed within the galaxy groups remains open . clearly , it is important to have independent mass measurements of the haloes associated with galaxy groups to test the validity of the mass estimates based on the stellar mass ( luminosity ) ranking . gravitational lensing observations , which measure the image distortions of background galaxies caused by the gravitational field of the matter distribution in the foreground , provide a promising tool to probe the dark matter distribution directly . in particular , galaxy - galaxy weak lensing , which focuses on the image distortions around lensing galaxies , can be used to probe the distribution of dark matter around galaxies , hence their dark matter haloes . the galaxy - galaxy lensing signal produced by individual galaxies is usually very weak , and so one has to stack the signal from many lens galaxies to have a statistical measurement . the first attempt to detect such galaxy - galaxy lensing signal was reported by tyson et al . more recently , with the advent of wide and deep surveys , galaxy - galaxy lensing can be studied for lens galaxies of different luminosities , stellar masses , colors and morphological types ( e.g. brainerd et al . 1996 ; hudson et al . 1998 ; mckay et al . 2001 ; hoekstra et al . 2003 ; hoekstra 2004 ; sheldon et al . 2004 ; mandelbaum et al . 2005 , 2006 ; sheldon et al . 2007a ; johnston et al . 2007 ; sheldon et al . 2007b ; mandelbaum et al . 2008 ) . given that galaxies reside in dark matter haloes , these results provide important constraints on the mass distribution associated with galaxies in a statistical way . in this paper , we use the galaxy groups of y07 selected from the sloan digital sky survey ( sdss ) , together with mass models for individual groups , to predict the galaxy - galaxy lensing signal expected from sdss galaxies . we compare our model predictions with the observational results obtained by mandelbaum et al . ( 2006 ) for the same galaxies . our goal is threefold . first , we want to test whether the method of halo - mass assignment to groups adopted by y07 is reliable . since the method provides a potentially powerful way to obtain the halo masses associated with the galaxy groups , the test results have general implications for the study of the relationship between galaxies and dark matter haloes . second , we want to examine in detail the contributions to the galaxy - galaxy lensing signal from different systems , such as central versus satellite galaxies , early - type versus late - type galaxies , and groups of different masses . such analysis can help us interpreting the observational results . finally , we would like to study how the predicted galaxy - galaxy lensing signal depends on model assumptions , such as the cosmological model and the density profiles of dark matter haloes . in a companion paper ( cacciato et al . 2008 , hereafter c08 ) , we use the relationship between galaxies and dark matter haloes obtained from the conditional luminosity function ( clf ) modeling ( yang et al . 2003 ; van den bosch et al . 2007 ) to predict the galaxy - galaxy cross correlation and to calculate the lensing signal , while here we directly use the observed galaxy groups and their galaxy memberships . this paper is organized as follows . in section [ sec_analysis ] we define the statistical measure that characterizes the galaxy - galaxy lensing effect expected from the mass distribution associated with the galaxy groups . we provide a brief description of the galaxy group catalogue and the models of the mass distribution associated with galaxy groups in section [ sec_sdss ] . we present our results in section [ sec_results ] and conclude in section [ sec_conclusions ] . unless specified otherwise , we adopt a @xmath0cdm cosmology with parameters given by the wmap 3-year data ( spergel et al . 2007 , hereafter wmap3 cosmology ) in our analysis : @xmath2 , @xmath3 , and @xmath4 , @xmath5 . galaxy - galaxy lensing provides a statistical measure of the profile of the tangential shear , @xmath6 , averaged over a thin annulus at the projected radius @xmath7 around the lens galaxies . this quantity is related to the excess surface density ( hereafter esd ) around the lens galaxy , @xmath8 , as @xmath9 where @xmath10 is the average surface mass density within @xmath7 , and @xmath11 is the azimuthally averaged surface density at @xmath7 . note that , according to this relation , @xmath12 is independent of a uniform background . in the above equation , @xmath13 is the critical surface density in comoving coordinates , with @xmath14 and @xmath15 the angular distances of the lens and source , @xmath16 the angular distance between the source and the lens , and @xmath17 the redshift of the lens . by definition , the surface density , @xmath11 , is related to the projection of the galaxy - matter cross - correlation function , @xmath18 , along the line - of - sight . in the distant observer approximation @xmath19\,d\chi\ , , \ ] ] where @xmath20 is the mean density of the universe and @xmath21 is the line - of - sight distance from the lens . the cross - correlation between galaxies and dark matter can , in general , be divided into a 1-halo term and a 2-halo term . the 1-halo term measures the cross - correlation between galaxies and dark matter particles in their own host haloes , while the 2-halo term measures the cross - correlation between galaxies and dark matter particles in other haloes . in the present work , we are interested in the lensing signals on scales @xmath22 where the observational measurements are the most accurate . as we will show in [ sec_results ] , on such scales the signal is mainly dominated by the 1-halo term . nevertheless , our model also takes the contribution of the 2-halo term into account . more importantly , since central galaxies ( those residing at the center of a dark matter halo ) and satellite galaxies ( those orbiting around a central galaxy ) contribute very different lensing signals ( e.g. natarajan , kneib & smail 2002 ; yang et al . 2006 ; limousin et al . 2007 ) , it is important to model the contributions from central and satellite galaxies separately . as an illustration , in fig . [ fig : sig ] we show the esds expected from a single galaxy in a host halo of mass @xmath23 . the solid line represents the lensing signal expected for the central galaxy of the halo . while the dotted and dashed lines show the lensing signal of a satellite galaxy residing in a sub - halo of @xmath24 with a projected halo - centric distance @xmath25 and @xmath26 from the center of the host halo , respectively . in the calculation , the dark matter mass distribution in the host halo is assumed to follow the navarro , frank & white ( 1997 ) profile and that in the sub - haloes is assumed to follow the hayashi et al . ( 2003 ) model . these models are described in detail in [ sec : distrib ] . in order to estimate the esd , we sample these profiles with mass particles and project the positions of all particles to a plane perpendicular to the line of sight . the @xmath11 is then estimated by counting the number of dark matter particles in a annulus with radius @xmath7 centred on the selected galaxies . [ fig : sig ] shows clearly that the lensing signals of the central and the satellite are quite different . the esd of the central galaxy follows the mass distribution of the host halo , decreasing monotonically with @xmath7 . the esd of a satellite , on the other hand , consists of two parts : one from the subhalo associated the satellite , which contributes to the inner part , and the other from the host halo , which dominates at larger @xmath7 . this simple model demonstrates clearly that , in order to model the galaxy - galaxy lensing signal produced by a population of lens galaxies , one needs to model carefully the distribution of matter around both host haloes and subhaloes . to do this , we need not only to identify the haloes in which each lens galaxy resides , but also to model the mass and density profile of each host halo and subhalo . in addition we also need to model the distribution of dark matter relative to galaxies . in the following section , we describe our modeling with the use of observed galaxy groups . . the solid line represents the lensing signal for the central galaxy in such a halo . the dotted line represents the lensing signal of a satellite galaxy residing in a sub - halo of mass @xmath27 which has a projected distance @xmath28 from the center of the host halo . the dashed line is the same as the dotted line , except that the subhalo s projected distance is @xmath29 from the center of the host halo.,scaledwidth=50.0% ] our analysis is based on the sdss galaxy group catalogue constructed by y07 . the groups are selected with the adaptive halo - based group finder developed by yang et al . ( 2005 ) , from the new york university value added galaxy catalog ( nyu - vagc ; blanton et al . 2005 ) which is based on the sdss data release 4 ( adelman - mccarthy et al . only galaxies with redshifts in the range @xmath30 , and with redshift completeness @xmath31 , are used in the group identification . the magnitudes and colors of all galaxies are based on the standard sdss petrosian technique ( petrosian 1976 ; strauss et al . 2002 ) , and have been corrected for galactic extinction ( schlegel , finkbeiner & davis 1998 ) . all magnitudes have been k - corrected and evolution - corrected to @xmath32 following the method described in blanton et al . ( 2003 ) . in y07 , three group samples were constructed using galaxy samples of different sources of galaxy redshifts . our analysis is based on sample ii , which includes 362,356 galaxies with redshifts from the sdss and 7091 galaxies with redshifts taken from alternative surveys : 2dfgrs ( colless et al . 2001 ) , psoz ( saunders et al.2000 ) or from the rc3 ( de vaucouleurs et al . there are in total 301,237 groups , including those with only one member galaxy . the group finder has been applied to mock catalogue to test the completeness and purity of the groups in y07 . about @xmath33 of the groups have a completeness @xmath34 and @xmath35 groups with @xmath36 , where @xmath37 is defined as the ratio between the number of true members that are selected as the members of the group and the number of the total true members of the group . an important aspect of the group catalog construction is the determination of the halo mass , @xmath38 , of each group . in y07 , two estimators are adopted . the first , @xmath39 , is estimated using the ranking of the characteristic luminosity of a group , which is the total luminosity of all member galaxies in the group with @xmath40 ( hereafter referred to as @xmath41 ) . the second , @xmath42 , is estimated using the ranking of the characteristic stellar mass , @xmath43 which is defined to be the total stellar mass of group members with @xmath44 . for each galaxy the stellar mass is estimated from its absolute magnitude and color using the fitting formula given by bell et al . ( 2003 ) . the basic assumption of the ranking method is that there is a one - to - one relation between @xmath45 ( or @xmath41 ) and the group mass . using the dark matter halo mass function predicted by a model of structure formation , one can assign a halo mass to each group according to its @xmath45 - ranking ( or @xmath41 - ranking ) . in this paper , we use the mass function obtained by warren et al . note that this one - to - one mapping is applicable only when the group sample is complete . in y07 , three complete samples are constructed in three redshift ranges . only groups in the complete samples are used in the ranking . the mass of other groups are estimated by a linear interpolation based on the @xmath46-@xmath38 relation ( or the @xmath41 - @xmath38 ) obtained from the complete sample . detailed tests using mock galaxy redshift samples have shown that the 1-@xmath47 error of the estimated halo mass is @xmath48 dex ( y07 ) . in addition , the two mass estimators , @xmath39 and @xmath42 , agree remarkably well with each other , with a scatter that decreases from about 0.1 dex at the low - mass end to about 0.05 dex at the high - mass end . since the correlation between @xmath43 and halo mass is somewhat tighter than that between @xmath41 and halo mass , we adopt @xmath42 as our fiducial halo mass throughout . as we demonstrate in [ sec : depend ] , using @xmath39 instead yields results that are fairly similar . [ fig : ml ] shows the relation between the host halo mass , @xmath42 , and the galaxy stellar mass @xmath49 ( the lower two panels ) or the galaxy luminosity @xmath50 ( the upper two panels ) . results are shown separately for central galaxies ( left panels ) and satellite galaxies ( right panels ) . as one can see , the stellar mass ( luminosity ) of central galaxies is quite tightly correlated with their host halo masses . however , for satellite galaxies of a given stellar mass ( or luminosity ) , their host halo mass covers a very large range , reflecting the fact that many low - mass galaxies are satellites in massive haloes . the distributions of host halo masses , @xmath42 , for central or satellite galaxies in different luminosity bins are shown in fig . [ fig : dis ] . on average , brighter central galaxies reside in more massive haloes . for faint galaxies , the halo - mass distribution is broader , again because many faint galaxies are satellites in massive systems . in the group catalogue , the mass assignment described above is used only for groups where the brightest galaxy is brighter than @xmath51 . this is because the mass ranking used in the group catalog is based on the total stellar mass ( or total luminosity ) of the member galaxies that are brighter than @xmath51 . the groups with no galaxies brighter than this magnitude thus have no assigned rank . as described in y07 , the reason for choosing this maginitude is a compromise between having a complete sample in a relatively large volume and having more groups that are represented by a number of member galaxies . for groups in which all member galaxies have @xmath52 , a different method has to be adopted . in modeling the luminosity function and stellar mass function of the central galaxies based on the same sdss group catalogue as used here , yang et al . ( 2008 ) obtain an average relation between the luminosity ( or stellar mass ) of the central galaxy and the halo mass down to @xmath53 . we adopt this relation to assign halo masses to all groups ( including those containing only one isolated galaxy ) represented by centrals with @xmath54 . for convenience , the halo masses obtained in this way are also referred to as @xmath42 ( based on the stellar mass of central galaxies ) and @xmath39 ( based on the @xmath55-band luminosity of the central galaxies ) , respectively . with the group catalogue described above , we can model the dark matter distribution by convolving the halo distribution with the density profiles of individual haloes . in our modeling of the density profiles , the host halo of a group is assumed to be centered on the central galaxy . there are two ways to define a central galaxy : one is to define the central in a group to be the galaxy with the highest stellar mass , and the other is to define the central to be the brightest member . for most groups these two definitions give the same results , but there are very few cases ( less than @xmath56 ) where different central galaxies are defined . in our fiducial model , we define the most massive galaxies ( in term of stellar mass ) to be the central galaxies . in a hierarchical model , a dark matter halo forms through a series of merger events . during the assembly of a halo , most of the mass in the merging progenitors is expected to be stripped . however , some of them may survive as subhaloes , although the total mass contained in subhaloes is small , typically @xmath57 ( van den bosch , tormen & giocoli 2005 ) . some of the subhaloes are associated with ` satellite galaxies ' in a halo . in our modeling of the galaxy - galaxy weak lensing , we only take into account subhaloes associated with satellite galaxies , treating other subhaloes as part of the host halo . giocoli , tormen & van den bosch ( 2008 ) provide a fitting function of the average mass function for subhaloes at the time of their accretion into the parent halo of a given mass . using this mass function , we first sample a set of masses for each group mass . we then set the mass originally associated with a satellite galaxy according to the stellar mass ranking of the satellites in the group . here we implicitly assume that the initial subhalo mass function is the same as the mass function of the subhalos that host satellites . this assumption is not proved by any observations , and we have to live with it since a more realistic model is not currently available . fortunately , subhalos only contribute a small fraction to the total lensing signal on small scales . the uncertainty here will not have a significant impact on any of our conclusions . to obtain the final mass in the subhalo at the present time , the evolution of the subhaloes needs to be taken into account . in other words , we need to know the fraction of the mass that is stripped and how the structure of a subhalo changes after the stripping . here it is convenient to introduce a parameter @xmath58 which is the retained mass fraction of the subhalo . gao et al . ( 2004 ) studied the radial dependence of the retained mass fraction @xmath58 from a large sample of subhaloes in a large cosmological simulation . in their work , @xmath58 is considered as a function of @xmath59 , where @xmath60 is the distance of the subhalo from the center of the host halo and @xmath61 is the virial radius of the host halo . the simulation of gao et al . gives @xmath62 we will adopt this in our modeling of the masses associated with subhaloes . however , in the group catalogue , only the projected distance , @xmath63 , from the group center is available . the 3d - distance , @xmath60 , is obtained by randomly sampling the nfw profile of the host halo with the given projected radius @xmath63 . thus , the mass assigned to a subhalo is determined by the following three factors : ( 1 ) the stellar mass of the satellite galaxy ; ( 2 ) the host halo mass ; ( 3 ) the distance between the satellite and the center of the host . here the host halo mass comes into our calculation in two ways . it not only determines the subhalo mass function , but also affects the parameter @xmath58 in eqs.[eq : gao ] . note that the accretion history of the host halo may also affect the value of @xmath58 . we have to neglect such effect because it is unclear how to model the accretion histories for individual groups . for host haloes , we use the following nfw profile ( navarro , frank & white 1997 ) to model the mass distribution : @xmath64 where @xmath20 is the mean density of the universe , @xmath65 is a scale radius , related to virial radius @xmath66 by the concentration , @xmath67 , and @xmath68 is a characteristic over - density related to the average over - density of a virialized halo , @xmath69 , by @xmath70 we adopt the value of @xmath69 given by the spherical collapse model ( see nakamura & suto 1997 ; henry 2000 ) . numerical simulations show that halo concentrations are correlated with halo mass , and we use the relations given by macci et al . ( 2007 ) , converted to our definition of halo mass . note that here we use @xmath65 , instead of the conventional notation @xmath60 , to denote the scale radius of the nfw profile , as @xmath60 has been used to denote the distance of a subhalo from the center of its host . for sub - haloes , we model their density profiles using the results obtained by hayashi et al . ( 2003 ) , who found that the density profiles of stripped sub - haloes can be approximated as @xmath71 where @xmath72 is a dimensionless factor describing the reduction in the central density , and @xmath73 is a cut - off radius imposed by the tidal force of the host halo . for @xmath74 and @xmath75 , @xmath76 reduces to the standard nfw profile @xmath77 . here @xmath77 is calculated using the mass of the subhalo at the time of its accretion into the host halo . both @xmath72 and @xmath78 depend on the mass fraction of the sub - halo that remains bound , @xmath58 . based on @xmath79-body simulations , hayashi et al . obtained the following fitting formulae relating @xmath72 and @xmath80 to @xmath58 : @xmath81 @xmath82 it should be pointed out , though , that there are substantial uncertainties in modeling the mass distribution around individual satellite galaxies . in particular , many of the results about subhaloes are obtained from @xmath79-body simulations , and it is unclear how significant the effect of including baryonic matter is . fortunately , the total mass associated with satellite galaxies is small ( see e.g. weinberg et al . furthermore , the contribution of the subhaloes associated with the satellite galaxies to the galaxy - galaxy lensing signal is confined to small scales . we therefore expect that these uncertainties will not change our results significantly . with the mass distributions described above , we use a monte - carlo method to sample each of the profiles with a random set of mass particles . note that the halo mass assigned to a group in the sdss group catalog is @xmath83 , which is the mass enclosed in the radius , @xmath84 , defined such that @xmath85 . we therefore sample the particle distribution within @xmath84 . after all the particles in each halo are sampled , we project the positions of all the particles to a plane and calculate @xmath12 by stacking galaxies in each of the luminosity bin . since the mass distribution is isotropic , an arbitrary direction can be chosen for the stacking . thus , the projection effect is naturally included in our calculation . each of the particles has a mass of @xmath86 . our test using particles of lower masses shows that the mass resolution adopted here is sufficient for our purpose . using 2 times more particles leads to a difference of about @xmath87 at @xmath88 , and almost no difference at @xmath89 . before presenting our model predictions , we first describe the observational results that we will use for comparison . the observational results to be used were obtained by mandelbaum et al . ( 2006 ) , who analyzed the galaxy - galaxy lensing effects using galaxies in a sample constructed from the sdss dr4 spectroscopic sample . their sample of lensing galaxies is similar to the galaxy sample used in y07 to construct the group catalogue used here . the only difference is that mandelbaum et al s sample includes all galaxies with redshifts in the range @xmath90 , while the galaxies in y07 s group catalogue are in @xmath30 . since , as to be described below , we are interested in the lensing signals around galaxies of given luminosity and morphological type , this difference in redshift range is not expected to have a significant impact on our results . for the faint luminosity bins , our galaxy samples should be almost identical to that of mandelbaum et al . ( 2006 ) , because all faint galaxies are at @xmath91 in the sdss catalog . for bright galaxies we also expect the statistic properties of the two samples to be similar . both mandelbaum et al . ( 2006 ) and y07 applied similar evolution correction and k correction , so that the evolution in the galaxy population has been taken into account , albeit in a simple way . as we will see , even for the the two brightest bins , the lensing signal is dominated by halos with masses @xmath92 , and the change in the halo mass function around this mass is small between @xmath93 and @xmath94 . following mandelbaum et al . ( 2006 ) , we split the galaxy sample into 7 subsamples according to galaxy luminosity . table [ table : t1 ] shows the properties of these subsamples : the luminosity range covered by each subsample , the mean redshift , the mean luminosity , and the fraction of late - type galaxies . as expected the mean redshifts of brightest bins are different from the corresponding redshifts in mandelbaum et al . .the properties of galaxy samples . in each case , the absolute - magnitude range , the mean redshift , the mean luminosity , and the fraction of late - types are listed . note that @xmath95 [ cols="^,^,^,^,^",options="header " , ] [ table : t1 ] also following mandelbaum et al . ( 2006 ) , we split each galaxy subsample into two according to galaxy morphology . the separation is made according to the parameter @xmath96 generated by the photo pipeline . the value of @xmath97 is obtained by fitting the galaxy profile , in a given band , to a model profile given by @xmath98 , where @xmath99 and @xmath100 are the de vaucouleurs and exponential profiles , respectively . as in mandelbaum , we use the average of @xmath97 in the @xmath101 , @xmath55 and @xmath102 bands . galaxies with @xmath103 are classified as early - type , while those with @xmath104 as late - type . it should be pointed out that we did not carry a ray - tracing simulation to predict the galaxy - galaxy lensing results . instead , we directly calculate the excess surface density around sdss galaxies . thus , our calculation does nt deal with source galaxies . on the other side , in mandelbaum et al . ( 2006 ) the source galaxies are carefully weighted , and lensing signals are calibrated , to reduce any bias in the observational measurements ( see mandelbaum et al . 2005 for details ) . thus , we assume the observational results are unbiased , and compare them directly with our model predictions . the observation data used here was kindly provided by r. mandelbaum . the data used in fig . [ fig:4bl ] has been published in mandelbaum et al . ( 2006 ) where lensing signal is calculated for early and late type galaxies separately . r. mandelbaum also provided the lensing data combining early and late type galaxies for us to make the comparisons presented in all other figures . note that here we only show the comparison of the lensing signal for galaxies divided according to luminosity . the errorbars on the observational points are @xmath105 statistical error . the systematic error of the galaxy - galaxy lensing has been discussed in detail in mandelbaum et al . the test has been carried out for three source samples : @xmath106 , @xmath107 , and high redshift lrgs . the overall systematic error is found to be comparable to or slightly larger than the statistical error shown here . in fig . [ fig:4bs ] we show the lensing signal around galaxies in different luminosity bins obtained from our fiducial model , which has model parameters as described in the last section and assumes the wmap3 cosmology . here the esd is plotted as a function of the projected distance @xmath7 from galaxies . the solid line shows the averaged esd of all galaxies in the corresponding luminosity bin . the amplitude of the predicted esd increases with galaxy luminosity , reflecting the fact that brighter galaxies on average reside in more massive haloes , as shown in figs . [ fig : ml ] and [ fig : dis ] . these results are to be compared with the data points which show the observational results obtained by mandelbaum et al . ( 2006 ) for the same luminosity bins . overall , our fiducial model reproduces the observational data reasonably well , especially for bright galaxy bins where the observational results are the most reliable . the reduced @xmath108 is @xmath109 combining all the luminosity bins . the best match is for l5f , with a reduced @xmath108 of @xmath110 . given that we do not adjust any model parameters , the @xmath108 indicates a good agreement . for the three low - luminosity bins , the predicted esd is lower than the corresponding observational result . for l1 and l2 , the observational data are very uncertain . for l3 , if we take the observational data points at face - value , the discrepancy with the model prediction is significant . as described in [ ssec_massassign ] , for groups which do not contain any member galaxies with @xmath111 , their halo masses are _ not _ obtained from the ranking of @xmath43 , but from the average stellar mass - halo mass relation of central galaxies that is required to match the observed stellar mass function of central galaxies . while all the galaxies in the bright luminosity bins have their host halo mass assigned by ranking method , the fraction of the galaxies in halos that have masses assigned according to the mass - halo mass relation is about @xmath112 in l3 and about @xmath113 in l2 and l1 . it is possible that this relation underestimates the halo mass . in order to see the effect caused by such uncertainties , we have used a set of parameters from yang et al . ( 2008 ) that are still allowed by the observed stellar mass function but give larger halo masses to the hosts of faint central galaxies . this increases the predicted esd for l3 by @xmath114 , not sufficient to explain the discrepancy . indeed , this discrepancy is not easy to fix . in the observational data , the amplitudes of the esd for l3 at @xmath115 - @xmath116 are actually slightly higher than for the brighter sample l4 , while in our model the esd for l3 is always lower than that for l4 . there is an effect that may help to reduce the discrepancy between our model prediction and the observational results for low - mass galaxies . since the group catalogue is only complete down to certain halo mass limit at different redshift ( see y07 ) , additional assumptions have to be made in order to model the distribution of the haloes below the mass limit . in the model described above , we have assumed that the haloes below the mass limit have a random distribution , so that they do not contribute to the esd . however , in reality these low - mass haloes are correlated with the more massive ones . as a result , our assumption will underestimate the 2-halo term of the esd . in order to understand how the predicted esd is produced , we also show separately the contributions from different sources . the dotted lines show the esd contributed by the 1-halo term of central galaxies . for all the luminosity bins , this term dominates the esd at small @xmath7 . for the brightest two bins , this term dominates the esd over the entire range of @xmath7 studied . this reflects the fact that almost all the galaxies in these two bins are central galaxies and the haloes in which they reside are more extended . the dashed lines show the esd contributed by the 1-halo term of satellite galaxies . this term first decreases with @xmath7 and then increases to a peak value before declining at large @xmath7 . this owes to the fact that , in the inner part , the lensing signal produced by satellite galaxies is dominated by the subhaloes associated with them , while at larger @xmath7 the lensing signal produced by satellite galaxies is dominated by their host halos . the value of @xmath7 at which the esd reaches the minimum corresponds roughly to the average halo - centric distance of the subhaloes in the luminosity bin . note that the contribution of satellites to the total esd is only important at large @xmath7 in the low - luminosity bins . this reflects the fact that a significant fraction of the low - luminosity satellites reside in massive haloes . note that although the 1-halo satellite term dominates the lensing signal at @xmath117@xmath118 for faint luminosity bins , the discrepancy between the observation and our prediction can not be simply solved by boosting up the satellite contribution . the reason is that the increase of the 1-halo satellite term requires the increase of the host halo mass of the satellites , which will make the 1-halo central term of the bright bins increase as well , causing significant discrepancy for the bright bins . finally , the dash - dotted lines represent the contribution of the 2-halo term . as expected , this term is relevant only on relatively large scales . in our model this term never dominates at @xmath119 . however , as discussed before this term may be underestimated here . in c08 , the 2-halo term is found to be comparable to the 1-halo satellite term even at @xmath120 . unfortunately the 2-halo term in c08 may be overestimated because halo - exclusion effect is not properly included . the large fluctuations seen in l6f are due to the small number of galaxies in this luminosity bin . since the halo mass of each lensing galaxy is known in our model , we can also examine the contributions to the total esd in terms of the halo mass . [ fig : mass ] shows the results where the host haloes are split into three bins of @xmath42 : @xmath121 ( dotted lines ) ; @xmath122 ( dashed lines ) ; and @xmath123 ( dashed - dotted lines ) . as one can see , the lensing signals in brighter bins are dominated by more massive haloes . very massive haloes with @xmath124 are not the dominant contributor , even for the brightest luminosity bin considered here , because the total number of galaxies hosted by such haloes are relatively small . for the low - luminosity bins , relatively massive haloes dominate the esd at large @xmath7 , because a significant fraction of the low - luminosity satellites are hosted by massive haloes ( see figs . [ fig : ml ] and [ fig : dis ] ) . in fig . [ fig:4bl ] we present the results separately for early - type and late - type galaxies . for a given luminosity bin , the predicted esd has a higher amplitude for early - type galaxies , clearly due to the fact that early - type galaxies are more likely to reside in massive haloes ( see e.g. van den bosch , yang & mo 2003 ) . for the faint samples , l1 and l2 , the behavior of the predicted esd for early - type galaxies resembles that of satellite galaxies in these luminosity bins , while the predicted esd for the late - type galaxies looks like of the central galaxies in the corresponding luminosity bins . this , again , reflects the fact that faint early - type galaxies are mostly satellites in massive haloes , while the faint late - type population is dominated by the central galaxies in low - mass haloes . the dashes lines in fig . [ fig:4bl ] show the results obtained using @xmath39 as halo mass , rather than @xmath42 ( see [ ssec_massassign ] ) . for early - type galaxies , the results based on this halo mass estimate are very similar to those based on @xmath42 . however , for the late - type samples , the esds obtained using @xmath39 are significantly higher than those obtained using @xmath42 , especially for the brighter samples . this is mainly due to the fact that late - type galaxies contain significant amounts of young stars , so that their stellar mass - to - light ratios are relatively low . consequently , they are assigned a larger halo mass based on their luminosity than based on their stellar mass . the model predictions are compared with the observational results of mandelbaum et al . ( 2006 ) . here again , the model prediction based on @xmath42 matches the observational data for the four bright samples . for the three faint bins , the model predictions are again lower than the observational results . as shown in fig . [ fig:4bl ] , it seems that our model prediction agrees better with the observation for the late type galaxies . for example , for l3 our model prediction matches the observation reasonably well for the late - type subsample ( with the reduced @xmath108 equal to @xmath125 ) , while the discrepancy is quite large for the early - type subsample ( with reduced @xmath108 equal to @xmath126 ) . since faint , early - type galaxies are preferentially satellite galaxies in relatively massive halos while faint , late - type galaxies are mostly central galaxies in relatively low mass halos , the above results seem to indicate that the discrepancy is due to the underestimate of the 1-halo satellite term in the model . unfortunately , there is no simple modification of the model that can fix the discrepancy . since the spatial distribution of galaxies is fixed by the observation , the only change that can be made is in halo properties . as we will see in the following subsection , increasing the halo concentration even by a factor of two can only increase the predicted esd by about 20% for the l3 sample , insufficient to explain the discrepancy . an increase in the halo masses assigned to galaxy groups can reduce the discrepancy for faint galaxies , but it would also significantly over - predict the esd for bright galaxies . the other possibility is that the halo masses of isolated , faint early - type galaxies are significantly underestimated . for example , isolated early - type galaxies may reside in much more massive halos than that given by their stellar masses . in order to explain the discrepancy , the halo masses for these galaxies need to be larger by a factor of at least 3 . this will not affect significantly the prediction for bright galaxies , but can boost the prediction for l3 by 50% at @xmath127 . unfortunately , it is still unclear if the required halo mass increase is feasible in current models of galaxy formation . the esd signal predicted by the model outlined in [ sec : fiducialmodel ] is based on several assumptions . beside the underlying cosmology and the halo mass assignment which are the most crucial model ingredients ( see [ sec : cosmodependence ] and [ sec : halomassassignment ] ) , the model requires a concentration - halo mass relation and it assumes that central galaxies reside at rest at the centre of a _ spherical _ dark matter halo . in this subsection , we test how our results are affected by these assumptions . the concentration parameter is a measure of the amount of dark matter in the central regions of the haloes . accordingly , different models for the concentration - halo mass relations are expected to result in different predictions for the esd signal ( at least on small scales ) . the fiducial model described in [ sec : fiducialmodel ] uses the model of macci et al . ( 2007 ) . however , other models are also available in the literature ( e.g. bullock et al . 2001 ; eke , navarro & steinmetz 2001 ) , which predict concentration - mass relations that are slightly different ( see c08 for an assessment of the impact on galaxy - galaxy lensing ) . furthermore , the presence of a ( central ) galaxy in a dark matter halo may have an impact on its concentration via , for example , adiabatic contraction ( e.g. blumenthal et al . 1986 ) , which is not accounted for in the concentration - mass relations obtained from pure @xmath79-body simulations . finally , attempts to measure halo concentrations observationally have thus far given conflicting results ( e.g. van den bosch & swaters 2001 ; comerford & natarajan 2007 , and referenecs therein ) . to examine how the lensing predictions depend on changes in halo concentation we consider two models : model ch , in which the concentrations are 2 times as high as in our fiducial model , and model cl , in which the concentrations are 2 times smaller . the left column of fig . [ fig : com ] shows the predictions of models ch ( dashed lines ) and cl ( dotted lines ) compared to our fiducial model ( solid lines ) . results are only shown for four luminosity bins , as indicated . note that the model with higher ( lower ) halo concentrations predicts esds that are higher ( lower ) . the effect is stronger on scales where the 1-halo central term dominates ( see fig [ fig:4bs ] ) . accordingly , in the case of the brightest sample ( l6f ) , models cl and ch differ from the fiducial model on scales up to @xmath128 , while in sample l4 the differences are only appreciable out to @xmath129 . we conclude that the predicted esd depends quite strongly on the assumed halo concentrations , indicating that galaxy - galaxy lensing has the potential to constrain the density profiles of dark matter haloes ( see mandelbaum et al . 2008 ) . in this paper , we have assumed that the concentration of a halo depends only on its mass , we have ignored the possible halo age - dependence of the concentration - mass relation ( e.g. wechsler et al . 2002 ; zhao et al . 2003 ; lu et al . if for a given mass older halos have higher concentration , and if the formation of a galaxy in a halo depends strongly on the formation history of the halo , then the age - dependence of the halo concentration must be taken into account . unfortunately , it is unclear how to connect the halo age ( hence halo concentration ) with the properties of galaxies . for a given halo mass , the dispersion in the concentration is about 0.12 dex ( e.g. jing 2000 ) , which corresponds to a change of about 20% in the predicted esd on @xmath130 . in our fiducial model , central galaxies are assumed to reside at the center of their dark matter haloes . however , as shown in van den bosch et al . ( 2005 ) , in haloes with masses @xmath131 there is evidence to suggest that central galaxies are offset from their halo centers by @xmath132 percent of the virial radius . a similar result was obtained by berlind et al . ( 2003 ) using sph simulations of galaxy formation . to examine how such an offset impacts on the esds , we consider two additional models . following van den bosch et al . ( 2005 ) and yang et al . ( 2006 ) , we assume that central galaxies in haloes with @xmath133 are offset from their halo centers by an amount that is drawn from a gaussian distribution with zero mean . we use two different values for the dispersion of the distribution : @xmath134 of the virial radius ( model ` dev1 ' ) and @xmath135 of the virial radius ( model ` dev2 ' ) . the corresponding lensing predictions are shown in the middle column of fig . [ fig : com ] as the dotted ( dev1 ) and dashed ( dev2 ) lines . note that the offsets only affect the lensing signal for the brightest sample ( l6f ) , where a larger offset results in a stronger suppression of the esd on small scales ( @xmath136 ) . for fainter samples , no ( significant ) differences with respect to the fiducial model are apparent , which owes to the fact that fainter centrals typically reside in haloes with @xmath137 which do not have an offset ( at least in our models ) . as a final test , we examine the impact of halo shapes on the galaxy - galaxy lensing signal . in our fiducial model , dark matter haloes are assumed to be spherically symmetric . however , @xmath79-body simulations show that , in general , they are triaxial rather than spherical . jing & suto ( 1998 ) proposed a fitting formula for triaxial dark matter haloes , which has been applied to both strong and weak lensing analyses ( e.g. oguri , lee & suto 2003 ; oguri & keeton 2004 ; tang & fan 2005 ) . in order to examine the impact of our assumption of halo sphericity , we consider an alternative model ( model ` tri ' ) , in which we assume that the dark matter density distribution is given by @xmath138 , where @xmath7 specifies an ellipsoidal surface : @xmath139 here @xmath140 are the three principal semi - axes of the ellipsoid . we set @xmath141 , where @xmath142 is the nfw profile , so that the total mass within a sphere of radius @xmath143 in @xmath142 is equal to the mass within the elliptical shell at @xmath7 . for the axis ratios we adopt the distribution function given by jing & suto ( 2002 ) : @xmath144^{0.7}}\nonumber \\ & \times & \exp\left\{-\frac{\left[(a / c)(m_{\rm vir}/m_{nl})^{0.07[\omega(z)]^{0.7 } } -0.54\right]^2 } { 2(0.113)^2}\right\ } \nonumber \\\end{aligned}\ ] ] and @xmath145\,,\end{aligned}\ ] ] where @xmath146 is the characteristic mass scale , on which the rms of the top - hat smoothed over - density is equal to @xmath147 . in practice , we proceed as follows : for each dark matter halo we first draw the axis ratio @xmath148 and @xmath149 using eqs . ( [ pac ] ) and ( [ pab ] ) , respectively . next we draw a random 3d orientation of the principal axes , and project the dark matter particles along the ( fiducial ) line - of - sight . next , for each halo , we determine the major axis of the projected distribution which we align with the major axis of the central galaxy . this assumption is motivated by observational claims that the major axis of a central galaxy tends to aligned with that of its host halo ( e.g. yang et al . 2006 ; faltenbacher et al . 2008 ; wang et al . 2008 ) . as shown in the right - hand panels of fig . [ fig : com ] , changing from spherical haloes ( solid lines ) to triaxial haloes ( dotted lines ) has almost no impact on the predicted esds . this should not come as an entire surprise , since the esds are azimuthally averaged over many haloes , which have random orientations on the sky . note that here we assume that the halo is perfectly aligned with the central galaxy . however , the observational results mentioned above actually suggest a misalignment . we have tried a model in which the orientation of the host halo is uncorrelated with that of the central galaxy . the change in the results is very small compared with the model assuming perfect alignment . another very important assumption in our model prediction is the cosmological model in the calculations of the halo mass function and the geometrical properties of spacetime . the redshifts of galaxies considered here are restricted to @xmath150 , and the impact of changing cosmological parameters on the spacetime geometry is quite small in our analysis . on the other hand , changing cosmological parameters can change the halo masses assigned to individual groups , which may have significant impact on the expected lensing signal . in our fiducial model , we adopt the cosmology parameters obtained from the wmap 3-year data , with @xmath151 , @xmath3 , @xmath152 , and @xmath5 ( spergel et al . 2007 ) . as comparison , we will show some results obtained assuming another set of cosmological parameters with @xmath153 , @xmath154 , @xmath155 , and @xmath156 , which is strongly supported by the first year data release of the wmap mission ( see spergel et al . 2003 ) and has been considered in many previous studies . in what follows we will refer to this second set of parameters as the wmap1 cosmology . note that the cosmological parameters given by the recent wmap 5-year data ( komatsu et al . 2008 ) are in between those of wmap1 and wmap3 . [ fig : wmp ] compares the esd predicted by the fiducial model using the wmap3 cosmology ( solid lines ) and that predicted by the wmap1 cosmology ( dotted lines ) . as one can see , the esd predicted by wmap1 is significantly higher than that predicted by wmap3 , especially for bright galaxies . most of this increase is due to changes in the halo mass function , which causes ( massive ) groups to be assigned a larger halo mass . the changes in the halo concentrations and the spacetime geometry play only a minor role . a comparison with the sdss data clearly favors the wmap3 cosmology over the wmap1 cosmology , especially for the brighter luminosity bins . the reduced @xmath108 for the wmap1 cosmology is @xmath157 , much larger than @xmath109 for the wmap3 cosmology . this result is in good agreement with that obtained in c08 using the clf model . thus , we conclude that the galaxy - galaxy lensing data either prefer the wmap3 cosmology , which has a relatively low @xmath1 , or our halo mass assignment is in serious error . in the following subsection we show that the uncertainties in our halo mass assignments are unlikely to change our results significantly . we therefore conclude that the galaxy - galaxy lensing data prefer a @xmath0cdm model with a relatively low @xmath1 . if we use wmap5 parameters , the model prediction is in between wmap1 and wmap3 , which is still too high to matched the observed esd of bright galaxies . in our model , the masses of groups are assigned according to the stellar - mass ranking and the halo mass function predicted by the adopted cosmology . the underlying assumption is that the mass function of the host haloes of groups is the same as that predicted by the cosmological model . however , even if the cosmological model adopted is a good approximation to the real universe , the observed halo density may be different from the model prediction because of cosmic variance introduced by the finite observational volume . the effect of such variance is expected to be most important for massive haloes , because the number density of such systems is small . if , for example , the number density of massive haloes in the observational sample is , due to cosmic variance , smaller than the model prediction , the mass assignment with the use of the theoretical halo mass function would assign a higher halo mass to groups . consequently , the esd of bright galaxies , which are biased toward massive haloes , would be overestimated . here we test the importance of such effects by considering the uncertainties due to poisson fluctuations . we use the halo mass function predicted with the wmap3 cosmology to generate a set of random halo samples , each of which contains the same number of groups as the observational sample . the halo masses in each of these samples is then ranked in descending order and the halo mass of a given rank is then assigned to the group with the same rank in stellar mass . we find that the scatter in the esd obtained in this way is negligibly small , even for the brightest sample . the reason is that , even for the brightest sample , the esd is dominated by haloes with intermediate masses , @xmath158 ( see fig . [ fig : mass ] ) , and the total number of such groups is quite large . another uncertainty in the mass assignment may arise from fiber collisions . in the sdss survey , no two fibers on the same sdss plate can be closer than 55 arcsec . although this fiber collision constraint is partially alleviated by the fact that neighboring plates have overlap regions , @xmath159 percent of all galaxies eligible for spectroscopy do not have a measured redshift . our analysis here is based on the group catalog constructed from galaxy sample ii ( see y07 ) , in which many of the galaxies missed due to fiber collisions are not included . consequently , the total stellar mass ( or the total luminosity ) of some of the groups may be underestimated , thus introducing a bias in the ranking . this bias is expected to be stronger in richer systems because they have higher projected galaxy number density and are more likely to suffer from fiber collisions . however , this effect is not likely to have a big impact on our results , because it only changes the relative ranking of the groups that have similar stellar mass ( luminosity ) and because the galaxy - galaxy lensing signals are averaged in relatively broad bins of galaxy luminosity . in order to quantify this effect we carry out a similar analysis using the group catalog constructed from sample iii , where a galaxy affected by fiber collisions is assigned the redshift of its nearest neighbor ( see y07 for details ) . in this case , the situation is the opposite to that in sample ii , because here some galaxies may be wrongly assigned to groups in the foreground or the background due to the wrong redshifts assigned to some of the fiber - collision galaxies . the stellar mass and the luminosity of some of the groups will therefore be overestimated . in our test , we use sample iii to construct the group and to set mass to groups according to their ranking in sample iii , but we calculate the esd only around galaxies that are in sample ii . the results obtained in this way are very similar to those based on sample ii alone , suggesting that the effect of missing fiber - collision galaxies in the mass assignment is not important . however , if the fiber - collision galaxies are themselves included in the calculation of the esd , i.e. if we use galaxies in sample iii to calculate the esd , the amplitude of the esd is significantly larger than that obtained from the galaxies in sample ii . the reason for this is that the galaxies affected by fiber collisions are preferentially located in high density regions , so that they are more likely associated with a massive halo . the observational results of mandelbaum et al . ( 2006 ) are based on galaxies that have spectroscopic redshifts , and so a fair comparison between the observational results and our model predictions can only be made with the use of galaxies in sample ii . our test above therefore demonstrates that our conclusions about the comparison between the observational data and the model predictions are robust against the uncertainties due to fiber collisions . the scatter in the relation between halo mass and stellar mass ( or luminosity ) may also produce some uncertainties in our model prediction . as shown in mandelbaum & seljak ( 2007 ) , the scatter is in partial degeneracy with cosmology model . in our investigation , we have fixed cosmological parameters and allowed no dispersion in the halo mass - total stellar mass relation . if , for example , we assume a log normal distribution with a dispersion of 0.3 index in halo mass for a given stellar mass , the predicted esd would be a few percent larger than that predicted by the fiducial model . in this paper we model the galaxy - galaxy lensing signal expected for sdss galaxies , using the galaxy groups selected from the sdss to represent the dark matter haloes within which the galaxies reside . we use the properties of the dark halo population , such as mass function , density profiles and shapes , expected from the current @xmath0cdm cosmogony to model the dark matter distribution in each of the groups identified in the sdss volume . the use of the real galaxy groups allows us to predict the galaxy - galaxy lensing signals separately for galaxies of different luminosity , morphological types , and in different environments ( e.g. central versus satellite galaxies ) . we check the robustness of our model predictions by changing the assumptions about the dark matter distribution in individual groups ( such as the shape , density profile , and center offset of dark matter haloes ) , as well as the cosmological model used to predict the properties of the halo population . we compare our model predictions with the observational data of mandelbaum et al . ( 2006 ) for similar samples of lens galaxies . although there is some discrepancy for lens galaxies in the low - luminosity bins , the overall observational results can be well understood in the current @xmath0cdm cosmogony . in particular , the observed results can be well reproduced in a @xmath0cdm model with parameters based on the wmap3 data , but a @xmath0cdm model with a significantly higher @xmath1 , such as the one based on the wmap1 data , significantly over - predicts the galaxy - galaxy lensing signal . our results also suggest that , once a correct model of structure formation is adopted , the halo masses assigned to galaxy groups based on ranking their stellar masses with the halo mass function , are statistically reliable . the results obtained imply that galaxy - galaxy lensing is a powerful tool to constrain both the mass distribution associated with galaxies and cosmological models . in the future , when deep imaging surveys provide more sources with high image quality in the sdss sky coverage , the galaxy - galaxy lensing signals produced by the sdss galaxies can be estimated to much higher accuracy . the same analysis as presented here is expected to provide stringent constraints on the properties of the dark matter haloes associated with galaxies and galaxy systems , as well as on cosmological parameters . we thank rachel mandelbaum for providing the sdss lensing data in electronic format . part of the computation was carried out on the sgi altix 330 system at the department of astronomy , peking university . li ran is supported by the national scholarship from china scholarship council . xy is supported by the _ one hundred talents _ project of the chinese academy of sciences and grants from nsfc ( nos . 10533030 , 10673023 ) . hjm would like to acknowledge the support of nsf ast-0607535 , nasa aisr-126270 and nsf iis-0611948 . zuhui fan would like to acknowledge the supports from nsfc under grants . 10373001 , 10533010 , and 10773001 , and 973 program ( no . 2007cb815401 ) . adelman - mccarthy j. k. , et al . , 2006 , apjs , 162 , 38 bell e. f. , mcintosh d. h. , katz n. , weinberg m. d. , 2003 , apjs , 149 , 289 berlind , a. a. , weinberg d. h. , 2002 , apj , 575 , 587 berlind a. a. , et al . , 2003 , apj , 593 , 1 blanton m. r. , et al . , 2003 , aj , 125 , 2348 blanton m. r. , et al . , 2005 , aj , 129 , 2562 blumenthal g. r. , faber s. m. , flores r. , primack j. r. , 1986 , apj , 301 , 27 brainerd t. g , blandford r. d. , smail i. , 1996 , apj , 466 , 623 bullock j. s. , kolatt t. s. , sigad y. , somerville r. s. , kravtsov a. v. , klypin a. a. , primack j. r. , dekel a. , 2001 mnras , 321 , 559 cacciato m. , van den bosch f. c. , more s. , li r. , mo h. j , yang x. , 2008 , submitted to mnras ( c08 ) cole s. , lacey c.g . , baugh c.m . , frenk c.s . , 2000 , mnras , 319 , 168 colless m. , et al . , 2001 , mnras , 328 , 1039 comerford j. m. , natarajan p. , 2007 , mnras , 379 , 190 cooray a. , sheth r. , 2002 , phr , 372 , 1 cooray a. , 2006 , mnras , 365 , 842 croton d. j. , et al . , 2006 , mnras , 367 , 864 de vaucouleurs g. , de vaucouleurs a. , corwin h.g . , buta r.j . , paturel g. , fouque p. , 1991 , third reference catalogue of bright galaxies , ( springel - verlag heidelber ) eke v.r . , navarro j.f . , steinmetz m. , 2001 , apj , 554 , 114 faltenbacher a. , li c. , mao s. , van den bosch f. c. , yang x. , jing y. p. , pasquali a. , mo h. j. , 2007 , apj , 662 , 71 giocoli c. , tormen g. , van den bosch f.c . , 2008 , mnras , 386 , 2135 gao l. , white s. d. m. , jenkins a. , stoehr f. , springel v. , 2004 , mnras , 355 , 819 hayashi e. , navarro j. f. , taylor j. e. , stadel j. , quinn t. , 2003 , apj , 584 , 541 henry , j. p. , 2000 , apj , 534 , 565 warren m. s. , abazajian k. , holz d. e. , teodoro l. , 2006 , apj , 646 , 881 wechsler r. h. , bullock j. s. , primack j. r. , kravtsov a. v. , dekel a. , 2002 , apj , 568 , 52 weinberg d. h. , colombi s. , dav r. , katz n. , 2008 , apj , 678 , 6

we discuss-ray emissions from the outer gap accelerators of middle - aged pulsars for part of the series of our studies . a two - dimensional electrodynamic model is used to solve the distribution of accelerating electric fields with electron and positron pair creation and radiation processes in the magnetic meridional plane . we compute the curvature radiation and the synchrotron radiation by solving the evolution of the lorentz factor and the pitch angle . the calculated spectra are compared with observed phase - averaged spectra . we also use a three - dimensional geometrical model to discuss the pulse profiles . we argue that the outer gap of middle - aged pulsars occupies the whole region between the last - open field lines and the critical magnetic field lines , which are perpendicular to the rotational axis at the light cylinder . we assume that there is no outer gap accelerator inside the light cylinder between the rotational axis and the critical magnetic field lines . for the geminga pulsar , we demonstrate that the outward curvature radiation dominates in the spectrum above 10 mev , while the inward synchrotron radiation dominates below 10 mev . we find that the computed spectrum is consistent with the observations in x - ray through-ray bands . with the pulse morphology of the-ray emissions , we argue that the inclination angle and the viewing angle for the geminga pulsar are and , respectively . we also apply our method to another four middle - aged radio pulsars , whose spin - down power and distance from the earth expect the possibility of detection of-ray emissions from those pulsars by fermi telescope . applying the inclination angle and the viewing angle inferred from radio polarization characteristics , the visibility of the-ray emissions from the outer gap is discussed . we show that-ray emissions from psrs b0355 + 54 , b1449 - 64 and b1929 + 10 will probably be detectable by fermi telescope . for psr b0740 - 28 , on the other hand , -ray emissions will not be detected because the-ray beam from the outer gap will be oriented in a different direction from the viewing angle . [ firstpage ] pulsars : middle - aged pulsars radiation mechanisms : non - thermal

0810.3767

the energetic gamma - ray experiment telescope ( egret ) aboard the _ compton gamma - ray observatory _ detected the gamma - ray emissions from six pulsars ( thompson 2004 ) , five of which show pulsed radio emissions . the sixth , geminga pulsar , has known as a radio quiet @xmath0-ray pulsar or with very weak radio emissions ( vats et al . 1999 ) . because of the observed high brightness temperature of radio emissions from pulsars , the radio radiation is a coherent mechanism . on the other hand , the emission mechanisms in optical through @xmath0-ray bands are non - coherent and non - thermal processes , which have been discussed with the polar cap model , the slot gap model and the outer gap model in the literature . the polar cap accelerator has been considered as the origin of radio emission from pulsars ( ruderman & sutherland 1975 ) . the configuration of the dipole magnetic field in inner magnetosphere region provides a natural explanation of position - angle curve of polarization of the radio pulse . for @xmath0-ray emissions , however , the polar cap model has to invoke both a small inclination angle and a small viewing angle measured from the rotational axis to reproduce the observed wide separation of the two peaks appeared in the pulse profiles ( daugherty & harding 1996 ) . the slot gap model ( arons 1983 ) is an extension of the polar cap model . the possibility of a high - altitude extension of the accelerator near the last - open field lines was proposed . the recent slot gap model explains the spectrum in optical to @xmath0-ray bands for the crab pulsar ( muslimov & harding 2003 , 2004 ; harding et al . 2008 ) and the pulse profiles in @xmath0-ray bands for the vela pulsar ( dyks & rudak 2003 ) . the outer gap accelerator model assumes a strong acceleration between the null charge surface of the goldreich - julian charge density and the light cylinder ( cheng , ho & ruderman 1986a , b ) . the goldreich - julian charge density is described by @xmath3 with @xmath4 being the rotation frequency , @xmath5 the magnetic field component projected to the rotation axis , and @xmath6 the speed of light . the recent studies of the outer gap model also explain the spectra and pulse profiles measured by egret ( hirotani 2007 ; tang et al . 2008 ; takata et al . 2008 ) . although numerous efforts have been done to discriminate emission models with @xmath0-ray data , the origin of the non - thermal emission remains inconclusive . in addition to the @xmath0-ray data of egret , emission properties in lower energy bands have also been provided . for example , harding et al . ( 2002 ) found that the pulse profile of the vela pulsar in rxte bands shows four or five peaks in a single period instead of the double peak structure measured by egret . the detail polarization features of optical pulsed emissions from the crab pulsar have been measured by kanbach et al ( 2005 ) . integral also measured the polarization of the @xmath0-ray emissions from the crab pulsar for the first time ( dean et al . 2008 ) . in the future , furthermore , the polarization of x - ray and soft @xmath0-ray emissions from the crab pulsar will probably be able to be measured by ongoing projects such as pogo ( kataoka et al . 2005 ) and the nuclear compton telescope ( chang et al . both the _ astro -rivelatore gamma a immagini leggero _ ( agile ) and recently lunched _ fermi gamma - ray space telescope _ ( fermi , hereafter ) will measure more detail features of the phase - resolved spectra and the pulse profiles in 1 mev through 100 gev bands . in particular , fermi telescope is expected to measure pulsed @xmath0-ray emissions from both many radio pulsars and from unidentified egret sources . a comprehensive theoretical study with observations in multi - wavelength bands is desired and will be important to discriminate emission models . takata et al . ( 2007 ) and takata & chang ( 2007 ) discussed non - thermal emissions from the young pulsars , psrs b0531 + 21 ( crab ) and b0540 - 69 , with the outer gap model . in takata et al . ( 2007 ) , they demonstrated that the outer gap model can explain observed polarization properties of the optical pulsed emissions from the crab pulsar . takata & chang ( 2007 ) extended their calculation of the polarization to higher energy band and predicted the results of future observations . also , they produced observed phase - resolved spectra in optical through @xmath0-ray bands for the crab pulsar . they predicted that @xmath0-ray flux on the earth from psr b0540 - 69 is about @xmath7 , which may be a target for fermi observation . takata et al . ( 2008 ) discussed the emission process in optical through @xmath0-ray bands for the vela - like young pulsars . solving the electrodynamics of the outer gap , they first pointed out that inward emission by ingoing particles is an important component below 10 mev bands , while only outward emission contributes to the spectrum above 10 mev as the traditional outer gap model has assumed . they produced the observed multi - peak structure of the pulse profile in x - ray bands of the vela pulsar with the inward and the outward emissions . they predicted that the pulse profile at 1 - 10 mev has four peaks in a single period , which can be checked in future observations . we had studied the non - thermal emissions from the crab - like and the vela - like pulsars . in this paper , therefore , we discuss the non - thermal emissions from the outer gap of the middle - aged pulsars for part of the series of our studies . using a two - dimensional electrodynamic model , we will solve the accelerating electric field with electron and positron pair creation and radiation processes ( section [ model ] ) . we will also discuss the x - ray and @xmath0-ray pulse profiles using a three - dimensional geometrical model ( section [ geomod ] ) . we apply the outer gap model to the geminga pulsar ( section [ gemingapul ] ) . the calculated spectrum will be compared with the observed phase - averaged spectrum in x - ray through @xmath0-ray bands . ( section [ spectr ] ) . we discuss the inclination angle , which is angle between the rotational axis and the magnetic axis , and the viewing angle with the observed properties of the @xmath0-ray pulse profile for the geminga pulsar . ( section [ lightr ] ) . also by assuming that the radio beam from the geminga pulsar is oriented in a different direction from the line of sight , we will further constrain the magnetic inclination angle ( section [ radioq ] ) . we will also apply the model to another middle - aged radio pulsars ( section [ other ] ) , whose timing properties are similar to the geminga pulsar . we will show candidates for detections by fermi telescope . with its high sensitivity , fermi telescope will also be able to identify @xmath0-ray quiet radio pulsars , although their @xmath0-ray emissions should be detectable in the sense of magnitude of the spin - down luminosity and its distance . with the present outer gap study , we will argue that psr b0740 - 28 is a candidate of the @xmath0-ray quiet pulsars due to misalignment of the directions of the @xmath0-ray beam and the line of sight . -ray photons are from the outer gap region . the critical field line ( thick solid line ) is defined by the magnetic field line on which the goldreich - julian density vanishes at the light cylinder.,width=264,height=264 ] figure [ pulsar ] shows a schematic picture of the magnetosphere model for middle - aged pulsars . in general , we expect that the outer gap can arise between the last - open field lines and the magnetic field lines ( so called critical field lines , hereafter ) on which the goldreich - julian charge density vanishes at the light cylinder . between the rotational axis and the critical field lines , the null charge surface is located beyond the light cylinder so that the goldreich - julian charge density does not change its sign all the way from the stellar surface to the light cylinder . therefore , we expect that there is no outer gap in the region between the rotational axis and the critical field lines . for a static dipole field , the polar angle of the critical field line on the stellar surface in the magnetic meridional plane is @xmath8 , \label{crit}\ ] ] where @xmath9 is the magnetic inclination angle , @xmath10 is the stellar radius , @xmath11 is the light radius , and @xmath12 is the angle of the null charge surface of the goldreich - julian density ; @xmath13 we define the trans - field thickness of the maximally extending gap in the magnetic meridional plane with @xmath14 , where @xmath15 is the polar cap angle . for the static dipole field , the polar cap angle is expressed by @xmath16 , \label{rim}\ ] ] where @xmath17 for the middle - aged pulsars , we can expect that the thickness of the outer gap becomes to be comparable with the maximum thickness @xmath18 as follows . the potential drop on the stellar surface between the critical magnetic field line and the polar cap rim is @xmath19 which becomes @xmath20 volt for the geminga pulsar . we can show that almost whole potential drop @xmath21 is required to explain the gev emissions from the geminga pulsar . for the geminga pulsar , a spectral cut off around 2 gev was measured ( fierro et al . we estimate whole potential drop in the gap , @xmath22 , from the cut off energy . we assume that @xmath0-ray photons are emitted via the curvature radiation , and that the particles that emit the cut off photons are accelerated by the whole potential drop along the magnetic field lines in the gap . the typical strength of the accelerating electric field is estimated from @xmath23 . we assume the force balance between the acceleration and the radiation back reaction force to calculate a typical lorentz factor of the accelerated particles ; @xmath24 , where @xmath25 is the curvature radius of the magnetic field line . using the typical energy of the curvature photons @xmath26 , the estimated potential drop in the outer gap is @xmath27 where we used @xmath28 . we find that the estimated magnitude of the potential drop is comparable with the available potential drop defined by equation of ( [ drop ] ) for the geminga pulsar . the observed @xmath0-ray emissions from the geminga pulsar imply that most of the region between the last - open field lines and the critical field lines will be occupied by the outer gap for the middle - aged pulsars . for younger pulsars , the outer gap accelerator occupies about 10 % of the whole region between the last - open field lines and the critical field lines . for example , the available potential drop defined by equation of ( [ drop ] ) for the vela pulsar becomes @xmath29 volt . the observed cut off energy around 3 gev indicates the potential drop in the gap is about @xmath30 volt . therefore , only about 10 % of the available potential drop is required to explain the @xmath0-ray emissions from the vela pulsar . from the dynamic point of view , we can also argue that the outer gap of the middle - aged pulsars occupies most of the region between the last - open field line and the critical field line . the @xmath0-rays emitted by the accelerated particles may convert into electron and positron pairs by the pair creation process with a soft x - ray field from the stellar surface . if typical energy of the curvature photons have an energy such that the pair creation condition @xmath31 is satisfied , the outer gap can not develop further due to screening effects of the pairs . in such a case , the position of the upper boundary of the gap is determined by the pair creation condition . for the geminga pulsar , however , the pair creation in the outer magnetosphere is not efficient . the temperature of the stellar surface is about @xmath32 mk ( kargaltsev et al . 2005 ) , so the threshold energy of the @xmath0-rays for the pair creation is @xmath33gev , where @xmath34 is the boltzmann constant . from equation of ( [ drop1 ] ) , the potential drop of @xmath35 volt in the gap is required to accelerate the particles to emit 6 gev photons . however it will be impossible to accelerate the particles because the available potential drop in the outer gap @xmath36 volt estimated from equation of ( [ drop ] ) is smaller than the required potential drop for the pair creation . the particles can not be accelerated in the outer gap to emit 6 gev photons with the curvature radiation process . therefore , we expect that the outer gap of the middle - aged pulsars can develop to occupy the region between the last - open field lines and the critical field lines . as we said in the first paragraph in this section , we do not expect that the outer gap accelerator arises beyond the critical field lines , because the goldreich - julian charge density does not change its sign all the way from the stellar surface to the light cylinder we assume that the radio emission originates from the polar cap accelerator , while the emissions in optical through @xmath0-ray bands originate from the outer gap accelerator . electrodynamically speaking , because of the different flow directions of the electric currents in the polar cap and in the outer gap , the current conservation forces that both acceleration regions do not exit on the same magnetic field lines . in this paper , we anticipate that the outer gap accelerator is extending from the last - open field line to the critical field lines , and the polar cap accelerator extends in a region between the critical field lines and the rotational axis . this assumption will be rejected if the radio emissions take place on the last - open field line . in fact , there are no observational evidences . for example , the phenomenological study with the polarization characteristics for the radio emissions ( dyks et al . 2004 ) indicates that the outer rim of the radio beam is emitted on the magnetic fields located well above the last - open field lines . in this section , we briefly summarize our method of the electrodynamic study of the outer gap model . a more detail description of our model can be found in the previous paper ( takata et al . we solve the accelerating electric field with the radiation and the pair creation processes in the magnetic meridional plane , which includes the rotation axis and the magnetic axis . in the electrodynamic study , we use a static dipole field to solve the structure of the outer gap . we consider a magnetized rotator in which an inclination angle @xmath9 between the rotational axis and magnetic axis is smaller than @xmath37 . the stationary electric potential , @xmath38 , for the accelerating field in the observer frame is obtained from ( shibata 1995 ; mestel 1999 ) @xmath39 , \label{poisson}\ ] ] where @xmath40 is the space charge density , and @xmath41 is the goldreich - julian charge density , for which we use @xmath42 . for the steady state , the continuity equations of the particles are written as @xmath43 where @xmath44 is the velocity along the field line , @xmath45 is the source term due to the pair creation process , and @xmath46 and @xmath47 denote the number density of the outgoing ( positrons ) and ingoing ( electrons ) particles , respectively . in the outer magnetosphere , @xmath48 pair creation process contributes to the source term @xmath45 . we simulate the pair creation process with a monte carlo method , which is explicitly described in takata et al . ( 2004 , 2006 ) . to solve the poisson equation ( [ poisson ] ) , we impose the boundary conditions on the four boundaries , which are called as inner , outer , upper and lower boundaries . the inner and outer boundaries are defined by the surfaces on which the accelerating electric field is vanishes , that is , @xmath49 . on the inner , upper and lower boundaries , the accelerating potential is equal to zero , that is , @xmath50 . on the inner boundary , because both conditions @xmath49 and @xmath50 are imposed , we can not know the position of the inner boundary in advance . we solve the position of the inner boundary with the dynamics . the lower boundary is defined by the last - open field line . for the upper boundary , we define with a magnetic field line . for the outer boundary , we define it with a curve line , which is perpendicular to the magnetic field lines in the magnetic meridian plane . for such a case , the upper part of the our gap extends beyond the light cylinder , if we put the outer boundary of the lower part of the gap close to ( but inside ) the light cylinder . the present static poisson equation of ( [ poisson ] ) will not give the correct solutions for the electric field outside the light cylinder . as we will see later ( figure [ geminga1 ] ) , however , most @xmath0-rays are emitted from the lower part of the gap . the spectral features will be less affected by the structure of the upper part of the gap . we expect that the results do not change much between the present approximated and correct treatments . to calculate the spectra , we take into account the emissions inside the light cylinder . to calculate the synchrotron and curvature radiations , we solve an evolution of the particle s momentum using the electric field distribution in the outer gap obtained by the method described in section [ gapst ] . the equation of motion for momenta of the parallel ( @xmath51 ) and perpendicular ( @xmath52 ) to the magnetic field lines are , respectively , described as ( harding et al . 2005 ; hirotani 2006 ) @xmath53 and @xmath54 where @xmath55 is the pitch angle , @xmath56 represents the radiation drag of the synchrotron and curvature radiation , and the second term on the right hand side on equation ( [ perpeq ] ) represents the adiabatic change along the dipole field line . the radiation drag , @xmath56 , of the synchrotron - curvature radiation is described by ( cheng & zhang 1996 ) , @xmath57 where @xmath58 @xmath59 @xmath60 the model parameters are the magnetic inclination angle @xmath9 and the current components @xmath61 , where @xmath62 ( or @xmath63 ) is the current carried by the positrons ( or electrons ) coming into the gap through the inner ( or outer ) boundary . we also use the outer boundary as the model parameter . as the lower boundary of the gap , we define the last - open field line . for the upper boundary of the middle age pulsars , which we discuss in the paper , we use the critical field line defined by equation of ( [ crit ] ) . setting the model parameters , we solve the position of the inner boundary , the accelerating electric field , and the current component carried by particles created in the gap . the created current , and the resultant calculated flux are determined by the value of the injected currents at the inner boundary @xmath62 and at the outer boundary @xmath63 . in fact , the magnitude of the created current is less affected by the current @xmath62 injected at the inner boundary , but is determined by @xmath63 injected at the outer boundary , because most of pairs are created between the ingoing @xmath0-rays and the outward propagating surface x - rays . in this paper , we choose the value of the component @xmath63 so that the created pairs produce the comparable @xmath0-ray flux with the observations . the created current and the @xmath0-ray flux do not monotonically increase with the current component @xmath63 . this is because an increase of the created current strengthens screening of the accelerating electric field . with a smaller electric field , the @xmath0-ray photons with a lower energy are produced , and therefore a smaller amount of the pairs are created . namely , the increment of the created current is determined under the competition between the increase of @xmath63 and the screening of the electric field . for example , if we assume the current @xmath64 in units of goldreich - julian value , we obtain the created current of @xmath65 for the geminga pulsar with @xmath66 . then if we double the value of @xmath63 , we find from the numerical results that the increments of the created current and the total luminosity are factors of @xmath67 and @xmath68 , respectively . we find that the computed @xmath0-ray flux is not sensitive to the detailed values of the current components injected at the boundaries . since the magnetic field must be modified by the rotational and the plasma effects the vicinity of the light cylinder ( muslimov & harding 2005 ) , the last - open field line will be also different from the conventional magnetic surface , which is tangent to the light cylinder for the vacuum case . for example , romani ( 1996 ) assumed the last - open lines with the field lines which have the polar angle @xmath69 at the stellar surface , where @xmath70 is the conventional polar cap angle ( [ rim ] ) for static dipole field . in the studies of the force - free magnetosphere , @xmath71 was obtained by contopoulos et al . ( 1999 ) and @xmath72 by gruzinov ( 2005 ) . in the paper , we use @xmath71 for the last - open field lines . the position of the last - open field lines affects the accelerating electric field , spectrum and the pulse profile . for example , if we adopt a vacuum solution , the typical electric field in the gap is described as @xmath73 with @xmath74 begin the stellar magnetic field ( cheng et al . 2000 ; hirotani 2006 ) . then as the last - open field line is located on the magnetic field line with a larger polar angle on the stellar surface , we obtain a stronger accelerating electric field because ( i ) the trans - field thickness of the gap , @xmath75 , becomes large and ( ii ) the typical curvature radius of the magnetic field line , on which most @xmath0-rays are emitted , becomes small . as a result , a spectrum of the radiation from the gap becomes hard . for the pulse profiles , the phase separation between two peaks becomes wide as the last - open field shift to inner magnetosphere from the canonical field line , which is tangent to the light cylinder ( takata & chang 2007 ) . in fact , the calculated @xmath0-ray spectrum and pulse profiles with the last - open field lines for @xmath71 are more consistent with the observations for the geminga pulsar than those for @xmath76 ( section [ result ] ) . the observed pulse profile provides an important tool for diagnose the geometry of the emission region in the pulsar magnetospheres . in particular , number of the peaks in a single rotational period , positions of the peaks in the pulse profile , and morphology of the pulse profiles at different energy bands will discriminate the emission models . in the framework of the outer gap model ( and also the slot gap model ) , the peaks appear in the pulse profiles because the photons emitted various point pileup due to the effect of the special relativity correction ; the aberration of the emitting direction and the flight time ( romani & yadigaroglu 1995 ; dyks & rudak 2003 ) . to compute the pulse profile , we anticipate that the emission direction coincides with the particle motion . in the observer frame , the particle motion along the field lines in the north hemisphere is described by @xmath77 , where the plus ( or minus ) sign represents the outgoing ( or ingoing ) particles , @xmath78 is the velocity along the magnetic field line and is determined by the condition that @xmath79 , @xmath80 is the unit vector of the magnetic field , for which we adopt a rotating dipole field in the observer frame , @xmath81 is the co - rotating velocity , and @xmath82 is the unit vector of the azimuthal direction . the emission direction , @xmath83 , is interpreted in terms of the viewing angle @xmath84 and the pulse phase @xmath85 , where @xmath86 and @xmath87 is the azimuthal emission direction and the radial distance to the emission point , respectively . we use a rotating magnetic dipole field in the observer frame to compute the pulse profile . the distribution of emissivity affects the pulse profiles . for the distribution of the emissivity in the meridional plane , we use the results of the two - dimensional electrodynamic model , which solves the trans - field distribution of the emissivity in the magnetic meridional plane ( see section [ lightr ] ) . for the structure of the azimuthal direction , it is difficult to model without solving a three - dimensional dynamics . for the first order approximation , however , we may apply the function form of the accelerating electric field for the vacuum solution . in the vacuum solution , the typical value of the electric field along a field line is determined by @xmath88 with @xmath89 being the curvature radius at the null charge surface and @xmath90 the magnetic azimuthal angle of the magnetic field on the stellar surface measured from the magnetic meridional plane ( cheng el al . 1986a ; cheng et al . in such a case , the number of the curvature photons emitted per second by a particle and the typical energy of the emitted photons are proportional to @xmath91 and @xmath92 , respectively , if we assume the force balance between the particle acceleration and the radiation drag . for example , we obtain @xmath93 for the magnetic meridional plane ( @xmath94 ) and @xmath95 for @xmath96 , if we use @xmath71 for the last - open field lines with the inclination angle @xmath66 . in this paper , therefore , when we compute the pulse profile , we assume that ratio of the photon number emitted on the field lines of @xmath97 and of @xmath96 is @xmath98 = ( 0.72/0.24)^{-3/4}\sim0.42 $ ] . we do not consider the emissions from the magnetic field lines on which the radial distance to the null charge point is larger than @xmath99 . in this section , we apply the model to the geminga pulsar and constrain the magnetic inclination and viewing angles . first , we will show structure of the outer gap that occupies most of the region between the last - open line and the critical field line ( section [ gapgeo ] ) . then we will show that the calculated spectra are consistent with the observations from x - ray through @xmath0-ray bands ( section [ spectr ] ) . we use the properties of the observed @xmath0-ray pulse profile to constrain the magnetic inclination and viewing angles of the observer ( section [ lightr ] ) . for the geminga pulsar , the observed pulse profile in @xmath0-ray bands shows double peak structure , and the phase separation of the two peaks is about 0.5 phase . finally , we further constrain the magnetic inclination angle by discussing visibility of the radio beam in section [ radioq ] , in which we will assume that the radio beam from the geminga pulsar is oriented in a different direction from the line of sight . for the geminga pulsar , we find that the calculated spectrum with the stellar dipole momentum inferred from the rotating dipole model is too soft compared with the observation . for the present calculations , we used double of the dipole momentum inferred from the rotating dipole model . -ray photons are radiated from the lower part outer gap inside the light cylinder . the results are for @xmath100.,width=529,height=264 ] and @xmath100 , respectively.,width=377,height=264 ] figure [ gapstr ] summarizes the outer gap structure of the geminga pulsar for the inclination angle that @xmath66 , and for @xmath101 , that is 0.3 % of the goldreich - julian currents are injected at both inner and outer boundaries . for the last - open field line , we use the magnetic field line that has the polar angle @xmath102 , as we discussed in section [ modelp ] in the left panel in figure [ gapstr ] , the thick solid line shows the boundary of the obtained gap structure . the dotted and dashed - dotted line represent the critical field line and the null charge surface of the goldreich - julian charge density , respectively . as the computed boundary of the gap shows , we can find the solution that the outer gap can occupy whole region between the last - open field line and the critical field line , as we discussed in section [ magneto ] . with the right panel of figure [ gapstr ] , we display the current density carried by the created pairs ( solid line ) and the total electric potential drop in the gap ( dashed line ) , that is potential drop between the inner and outer boundaries , as a function of the polar angle , @xmath103 , of the magnetic field lines on the stellar surface . in the abscissa , the polar angle satisfying the condition that @xmath104 is corresponding to the last - open field line , and the angle of @xmath105 is corresponding to the upper boundary of the gap . from the solid line , we can see that the maximum current density of about 30 % of the goldreich - julian value runs through along the magnetic field line that penetrates the gap at the height of about 40% of the gap thickness measured from the last - open field line . from the dashed line , we also find that the largest electronic potential drop of @xmath106 volt appears on the magnetic field line penetrating the gap at the height of about 20% of the thickness . this indicates that the strength of the accelerating electric field becomes maximum on the magnetic field line penetrating the gap at the height of about 20% of the thickness . because the emitting power of each magnetic flux tube is proportional to the product of the current density and the potential drop at each field line , we can read that most of emissions come from the height between 20 % and 40 % of the gap thickness ( see also figure [ geminga1 ] ) . figure [ ediftri ] shows the distribution of the electric field along the magnetic field line , on which the strength of the electric field attains the maximum value in the outer gap . we compare the results for @xmath100 ( solid line ) and @xmath107 ( dashed line ) . as the figure [ ediftri ] shows , a larger accelerating electric field is obtained if we assume a larger polar cap angle . the typical strength of the accelerating electric field is proportional to @xmath108 ( cheng et al 2000 ; hirotani 2006 ) . in the present case , the curvature radius on the field line on which the electric field attains the maximum value is typically @xmath109 for @xmath100 and @xmath110 for @xmath107 . this means that the ratio of strengths of the accelerating electric field for two cases is @xmath111 , which describes the difference of the strengths of the electric field seen in figure [ ediftri ] . the typical energy of the photons emitted by the curvature radiation is proportional to @xmath112 . the ratio of the typical energies becomes @xmath113 . as a result , the calculated spectrum for @xmath100 is harder than that for @xmath107 . -ray bands . the solid line represents the total emissions which includes the outward ( dashed line ) and the inward ( dotted lines ) emissions . the filled circles and the dashed - dotted line represent the observed phase - averaged spectrum ( after kargaltsev et al . 2005).,width=264,height=264 ] -ray emissions . each line represents the spectrum from the height below 20 % of the thickness measured from the last - open field line , ( solid line ) , between 20 % and 40 % ( dashed line ) , between 40 % and 60 % ( dotted line ) , and above 60 % ( dashed - dotted line).,width=264,height=264 ] figure [ geminga ] compares the calculated spectrum of the curvature and synchrotron processes with the observations in x - ray through @xmath0-ray bands . the result is for the last - open field lines with the polar angle @xmath100 , and the model parameters are the same with that of figure [ gapstr ] . in figure [ geminga ] , the filled circles and the dashed - dotted line show the observed phase - averaged spectrum . the dashed line and the dotted line represent the emissions of outgoing positrons and of the ingoing electrons , respectively . the solid line represents the total emissions . for each line , the components of the curvature and the synchrotron radiations are combined . in fact , the curvature radiation dominates in the spectrum above 100 mev , while the synchrotron radiation dominates below 10 mev . from figure [ geminga ] , we see that the total spectrum ( solid line ) is consistent with the observations in x - ray through @xmath0-ray bands . if we use the canonical last - open field line with the polar angle @xmath107 , the spectral cut off is lactated at the energy factor of 1/1.4 smaller than that seen in the solid line of figure [ geminga ] , as we discussed in the last paragraph of section [ gapgeo ] . therefore , the calculated spectrum for the last - open field line with @xmath100 is more consistent with that for the canonical last - open field line . from figure [ geminga ] , we can see that the curvature radiation of outgoing positrons ( dashed line ) dominate in the total emissions above 100 mev . this is because most of the pairs are created the vicinity of the inner boundary by the inward propagating @xmath0-rays and the outward propagating surface x - rays . the created electrons and positrons are accelerated inwardly and outwardly by the electric field , respectively . the positrons created around the inner boundary are accelerated by almost whole potential drop in the gap , while the created electrons are accelerated by the potential drop only between the inner boundary and the point where they were created . therefore , the total radiation power is larger for the outward emissions than the inward emissions . below 10 mev , on the other hand , we find that the synchrotron radiation of the ingoing electrons dominate in the calculated spectrum . this is because the synchrotron emissions with strong magnetic field near the stellar surface contribute to the spectrum in x - ray bands . -ray bands for the geminga pulsar . the upper left , the upper right , the lower left , and the lower right panels show the pulse profiles for the inclination angle of @xmath114 , @xmath115 , @xmath116 and @xmath117 . the solid and dashed lines in each pane represent the pulse profiles of the viewing angle @xmath118 ( or @xmath119 ) and @xmath120 ( or @xmath121 ) , respectively . the results are for @xmath100.,width=529,height=529 ] and the viewing angle @xmath118 ( or @xmath122 ) . the solid and dashed lines represent the pulse profiles for the outward and inward emissions , respectively . above 100 mev bands , only outward emissions contribute to the spectrum ( figure [ geminga ] ) so that the pulse profile ( solid line ) becomes double peak structure . for around 10 mev bands , both the outward and inward emissions contribute to the pulsar profile , and the pulse profile has the four peaks in a single period . for x - ray bands , the inward emissions dominate in the pulse profile , which has two peaks ( dashed line ) in a single period . , width=264,height=264 ] we discuss the @xmath0-ray pulse profiles , and constrain the inclination angle and the viewing angle using the three - dimensional geometrical model described in section [ geomod ] . as figure [ geminga ] shows , the outward emissions dominate in the total spectrum above 100 mev . in this section , therefore , we consider only the outward emissions to compute a pulse profile . for the trans - field distribution of the emissivity in the poloidal plane , we use the results of the two - dimensional electrodynamic study . in figure [ geminga1 ] , we can read the trans - field distribution of the emissions above 10 mev bands . for example , the dashed line represents the spectra of the emissions from the height between 20 % and 40 % of the thickness measured from the last - open field line . by comparing the magnitude of the fluxes between the solid and dashed lines in figure [ geminga1 ] , we may put an emissivity at the height below 20% of the gap about one order smaller than that at the height between 20% and 40% of the gap . figure [ light ] summarizes the pulse profiles for the magnetic inclination angle , @xmath114 ( upper left panel ) , @xmath123 ( upper right panel ) , @xmath66 ( lower left panel ) and @xmath124 ( lower right panel ) . we calculate the emissions from the whole region between the last - open field lines and the critical field lines . the results are for the last - open field lines with the polar angle @xmath125 . if we used the canonical last - open field lines with @xmath126 to computed pulse profile , we found that no geometries ( inclination angle and viewing angle ) can produce the observed phase separation between the two peaks with the @xmath0-ray spectrum ; we obtained narrower phase separation ( @xmath127 ) than the observation @xmath128 . in figure [ light ] , the solid and dashed lines are results for the viewing angle of @xmath118 ( or @xmath119 ) and @xmath129 ( or @xmath130 ) . as the same with the conventional outer gap model , the first peak in the pulse profile is made by the emissions from leading part of the gap according to the rotational direction , while the second peak is made by the emissions from trailing part of the gap . we can see from figure [ light ] that the morphology of the pulse profiles depends on the inclination and viewing angles . by comparing the observed pulse profile , therefore , we can constrain the inclination and viewing angles . first we can rule out nearly aligned rotator , as represented by the pulse profile for @xmath114 . with such a small inclination angle , a sharp second peak does not appear in the pulse profile for any viewing angles of the observer . secondary , we would rule out the viewing angle far away from @xmath131 as demonstrated by the pulse profile ( dashed line ) with @xmath129 in figure [ light ] . in such a viewing angle , the pulse profile has a single broad peak or the two peaks with a narrower separation of the phases compared with the observations ( @xmath132 phase ) . therefore , a larger inclination angle and the viewing angle that @xmath133 , say @xmath134 , are preferred to explain the observed pulse profile of @xmath0-ray bands . in this section , we discuss the expected pulse profile in soft @xmath0-ray and the x - ray bands . as the calculated spectrum in figure [ geminga ] shows , the inward emissions from the inward particles contribute to the total spectrum below 10 mev . specifically , both the inward and the outward emissions contribute to the total emissions around 10 mev , while the only inward emissions dominate into the total emission in x - ray bands . therefore , we expect that the pulse profile below 10 mev is not simply an extension of that of the @xmath0-ray bands above 100 mev . figure [ light2 ] shows the pulse profile for the outward ( solid ) and the inward ( dashed ) emissions for @xmath66 and @xmath118 ( or @xmath119 ) . from the dashed line in figure [ light2 ] , we find that the pulse profile of the inward emissions has also double peak structure . for the inward emissions , the first peak is made by the emissions from the trailing part of the gap and the second peak is made by the emissions of the leading part of the gap . for the calculated spectrum around 10 mev in figure [ geminga ] , both the outward and inward emissions contribute to the spectrum . the pulse profiles in 10 mev bands will be made by both the outward ( solid line in figure [ light2 ] ) and inward ( dashed line in figure [ light2 ] ) emissions . therefore , we expect that the pulse profile in soft @xmath0-ray bands will have four peaks in a single period . in x - ray bands , the inward emissions dominate in the total emissions ( figure [ geminga ] ) . therefore we expect the double peak structure of the pulse profile in the x - ray bands for the geminga pulsar . by comparing between the predicted pulse profiles in @xmath0-ray ( solid line ) above 10 mev bands and x - ray bands ( dashed line ) , the pulse peaks in x - ray band will appear in the leading phases of the pulses of the @xmath0-ray bands . with the present results , the observed differences of the pulse morphologies in the different wavelength are strong tool to diagnose the emission processes in the pulsar magnetosphere . for the vela pulsar , the outer gap model can explain the morphology differences between the pulse profiles in rxte and egret bands ( takata et al . for the geminga pulsar , the spectrum and the pulse profiles observed by @xmath135 and @xmath136 indicated that the non - thermal x - rays and @xmath0-rays are produced by a distinct mechanism ( jackson et al . 2002 ) . kargaltsev et al . ( 2005 ) analyzed x - ray spectrum and pulsations observed by @xmath137 . they found that the pulse profile of the non - thermal x - rays in 2 - 8 kev bands has two peaks in a single period , and that the spectrum and pulse profile are not simply an extension of that of the egret bands . the observed pulse morphologies for the geminga pulsars are different in the different wave length . as we discussed above , the present model also predicts the double peak structure in x - ray pulse profile and the energy dependent pulse profiles from x - ray through @xmath0-ray bands . the results of the present model are generally consistent with the observed pulse profiles in x - ray and @xmath0-ray bands . for the phase separation between two peaks in @xmath138-ray bands , however , the observations in 2 - 8 kev bands imply about 0.35 ( 0.65 ) phase ( kargaltsev et al . 2005 ) , while the present model predicts about 0.5 phase ( dashed line in figure [ light2 ] ) . so , it may be required an examination of the x - ray pulse profile with a more realistic magnetic field configuration . ) for different inclination angles @xmath114 ( upper left ) , @xmath115 ( upper right ) , @xmath116 ( lower left ) and @xmath117 ( lower right ) . the dotted line represent viewing angle of @xmath139 . , width=529,height=529 ] the geminga pulsar is the only known radio quiet ( or weak emissions ) @xmath0-ray pulsar . based on the following reasons , we assume that the radio beam of the geminga pulsar is oriented in a different direction from the line of sight . first , there are many radio pulsars around the geminga pulsar on @xmath140 diagram . second , lin & chang ( 2008 ) searched candidate periods in unidentified egret sources for the next geminga . they found that most of the candidate timing properties belong to that of vela - like pulsars . if their candidate periods are true , their results would suggest that the major cause for a radio quiet is likely geometrical rather than intrinsic . the radio emission model with the assumption that the radio `` quiet '' for the geminga pulsar is caused by geometry also provides a tool to study the magnetic inclination angle and the viewing angle with the @xmath0-ray emission model . for the emission height of the outer cone of the radio emissions , we apply the empirical model developed by kijak & gil ( 2003 ) . they found a relation among the emission height , the rotation period , the period time derivative and the frequency of the radio wave . assuming the rim of the radio beam are emitted vicinity of the critical magnetic field line ( defined by @xmath141 ) , their relation is described as @xmath142 where @xmath10 is the stellar radius and @xmath143 is the frequency of the radio wave . figure [ radio ] shows photon mapping of the rim of the radio beam emitted at the height described by equation of ( [ height ] ) for different inclination angles . we used the rotating dipole field as the magnetic field of the magnetosphere . in figure [ radio ] , the solid and dashed lines represent the emissions for the frequencies of @xmath144mhz and 400 mhz , respectively . for the references the dotted line represents line of sight for the viewing angle of @xmath118 . for a small inclination angle represented by @xmath114 ( upper left panel in figure [ radio ] ) , an observer with @xmath2 , which is preferred to explain observed @xmath0-ray pulse profile , does not measure the radio emission . this is because the magnetic pole points in a completely different direction from the line of sight for @xmath133 . for largely inclined pulsar such as @xmath124 ( lower right panel ) , on the other hand , it is highly possible to be measured the radio emissions by the observer with @xmath2 , because the magnetic pole points close to the line of sight . for the geminga pulsar , therefore , a smaller inclination angle is preferred to explain the radio `` quiet '' observations with the @xmath0-ray detections . using the observed pulse profiles in the @xmath0-ray observations discussed in section [ lightr ] , we safely eliminated the possibility of the nearly aligned rotator such as @xmath114 because predicted single peak structure ( upper left panel in figure [ light ] ) is not consistent with the observed double peak structure of the @xmath0-ray pulse . combing those results obtained by analysis in the @xmath0-ray and the radio `` quiet '' observations , we conclude that the geminga pulsar has a moderate magnetic inclination angle of @xmath1 , and we observe the geminga pulsar with a viewing angle of @xmath2 . .timing properties of the middle - aged pulsars . @xmath145 is the rotational period , @xmath146 is the time derivative of the rotational period , @xmath147 is the stellar magnetic field inferred from the dipole field model , @xmath148 is the pulse age , @xmath149 is the distance , @xmath150 is the spin - down luminosity measured on earth , @xmath9 is the magnetic inclination angle , and @xmath151 is the viewing angle . for the magnetic inclination angle @xmath9 and the viewing angle @xmath151 , we refer the present paper for the geminga pulsar , rankin ( 1993 ) for psrs b0355 + 54 , b0740 - 28 and b1449 - 64 , and everett and weisberg ( 2001 ) for psr b1929 + 10 . in the last column , we show the predicted detectability of pulsed @xmath0-ray emissions by fermi telescope with @xmath152 in 50 hours . [ cols="^,^,^,^,^,^,^,^,^,^ " , ] we apply the present outer gap model to another middle - aged pulsars . using atnf pulsar catalogue ( manchester et al . 2005 , see http://www.atnf.csiro.au/research/pulsar/psrcat ) , we chose four middle - aged pulsars , psrs b0355 + 54 , b0740 - 28 , b1449 - 64 , and b1929 + 10 , to discuss the detectability of the @xmath0-ray emissions from the outer gap by fermi telescope . these four pulsars were chosen with the following conditions . if the pulsars are at a large distance from the earth , fermi telescope could not measure the @xmath0-ray emissions from the pulsars . a rough idea of the @xmath0-ray flux on the earth is described by @xmath153 , where @xmath154 is efficiency , @xmath155 is the spin - down luminosity , and @xmath149 is the distance to the pulsar . our four pulsars have the spin - down luminosity so that @xmath156 with @xmath157 of the geminga pulsar is larger than @xmath158 , which is close to the lower limit of fermi telescope detection with @xmath152 in 50 hours . second , there is the possibility that we can not measure the @xmath0-ray emissions , because the @xmath0-ray beam is oriented in a different direction from the line of sight . because the geometry of the @xmath0-ray beams depends on the emission models , the detectability of the @xmath0-ray emissions depends on the models . if we know the viewing angle and the magnetic inclination angle of the pulsar , we can test the models with fermi observations . on these ground , we chose those four pulsars because the magnetic inclination and viewing angles are inferred from the radio polarization data ( rankin 1993 ; everett & weisberg 2001 ) . in table 1 , we summarize the timing properties of the four pulsars and the geminga pulsar . the represented results in this section were computed with ( i ) the double of the dipole inferred from the rotating dipole model and with ( ii ) the last - open field lines with the polar angle @xmath100 . -ray spectrum ( left ) and pulse profile ( right ) for psr b0355 + 54 . for @xmath159 , the observe with @xmath160 measures only the outward emissions . , title="fig:",width=264,height=264 ] -ray spectrum ( left ) and pulse profile ( right ) for psr b0355 + 54 . for @xmath159 , the observe with @xmath160 measures only the outward emissions . , title="fig:",width=264,height=264 ] psr b0355 + 54 has a rotational period of @xmath161 s ( manchester et al . 1972 ) , which is the fastest in the four pulsars we chose . the distance from the earth is @xmath162 kpc ( hobbs et al . psr b0355 + 54 is known to emit the x - ray ( mcgowan et al . 2006 ) and probably tev @xmath0-rays ( senecha et al . 1995 ) from its pulsar wind nebula . the polarization data on the radio wave can be fitted by the rotational vector model ; the magnetic inclination and the viewing angles are inferred as @xmath163 and @xmath160 , respectively ( rankin 1993 ) . figure [ b0355 ] shows the expected spectrum ( left panel ) and the pulse profile ( right panel ) in @xmath0-ray bands for psr b0355 + 54 . for the current components injected at the inner and outer boundaries , we used @xmath164 . for the inclination angle @xmath159 , the observer with the viewing angle @xmath165 can not measure the inward emissions from the outer gap the model predicts the emission flux around 1 gev are about @xmath166 erg / s , which will be able to be measured by fermi telescope . with the viewing angle @xmath160 , the expected pulse profile has a broad peak in a single rotational period , as right panel in figure [ b0355 ] shows . psr b0740 - 28 is younger and faster rotating pulsar than the geminga pulsar ( bonsignori - facondi et al 1973 ) . the distance to the pulsar is @xmath167 kpc ( hobbs et al . 2004 ) and the magnitude of @xmath168 is the largest in the four pulsars . the pulsar wind powered by psr b0740 - 28 was discovered in optical bands ( jones et al . 2002 ) . the magnetic inclination angle @xmath169 and the viewing angle @xmath170 are inferred with the polarization characteristics of the radio wave ( rankin 1993 ) . we will not expect the detection of the @xmath0-ray beam by fermi telescope from psr b0740 - 28 , because the @xmath0-ray beam will be oriented in a different direction from the line of sight . because the viewing angle with @xmath170 points between the rotational and magnetic axes in the magnetic meridional plane , the observer measures mainly emissions inside the null charge surface and on the magnetic field lines toward rotational axis . however , the outer gap does not predict a strong emissions from that region , but predicts a strong emission beyond the null charge surface and on the magnetic field lines curved away from the rotational axis . the present outer gap model expects that the strong emissions from the outer gap of psr b0740 - 28 is oriented in a different direction from the line of sight . for the slot gap model ( harding et al . 2008 ) , on the other hand , a detection of @xmath0-ray emissions from psr b0740 - 28 may be expected , because the model assumes the emissions from the polar cap region to the vicinity of the light cylinder above the whole last - open field lines ( dyks & rudak 2003 ) . therefore , psr b0740 - 28 will be a good candidate to discriminate the outer gap model and the slot - gap model with fermi telescope . -ray spectrum ( left ) and pulse profile ( right ) for psr b1449-64.in the left panel , the solid , dashed and dotted lines represent the spectra of the total , outward and inward emissions . in the right panel , the solid and dashed show the pulse profile of the outward and inward emissions , respectively.,title="fig:",width=264,height=264 ] -ray spectrum ( left ) and pulse profile ( right ) for psr b1449-64.in the left panel , the solid , dashed and dotted lines represent the spectra of the total , outward and inward emissions . in the right panel , the solid and dashed show the pulse profile of the outward and inward emissions , respectively.,title="fig:",width=264,height=264 ] psr b1449 - 64 has the rotational period @xmath171 and the period time derivative @xmath172 s / s ( large et al . the distance to the pulsar is @xmath173 kpc ( siegman et al . 1993 ) . the magnetic inclination of @xmath174 and the viewing angles of @xmath175 are used to fit the polarization data of the radio wave with the rotating vector model ( rankin 1993 ) . figure [ b1449 ] shows the expected spectrum ( left panel ) and the pulse profile ( right panel ) in @xmath0-ray bands for psr b1449 - 64 . the expected pulse profile shows the double peak structure . we adopt @xmath176 for the current components injected at the inner and the outer boundaries . -ray spectrum ( left ) and pulse profile ( right ) for psr b1929 + 10.the lines correspond to same case as figure [ b1449].,title="fig:",width=264,height=264 ] -ray spectrum ( left ) and pulse profile ( right ) for psr b1929 + 10.the lines correspond to same case as figure [ b1449].,title="fig:",width=264,height=264 ] psr b1929 + 10 is known as one of the closet pulsar with the distance @xmath177 kpc ( hobbs et al . psr b1929 + 10 is the oldest pulsar @xmath178 yrs in the pulsars we chose with its rotation period @xmath179 s and the period time derivative @xmath180 s / s . the x - ray emissions from both the pulsar and the pulsar wind nebula have been reported ( becker et al . 2006 ; hui and becker 2008 ) . the inferred magnetic inclination and the viewing angle of the observer from the radio polarization are @xmath181 and @xmath182 , respectively ( everett and weisberg 2001 ) . we use @xmath183 for the currents injected the inner and outer boundaries . because of the closeness to the earth of psr b1920 + 10 , the expect @xmath0-ray flux can significantly exceed the lower limit of fermi telescope detection with @xmath152 in 50 hours ( figure [ b1929 ] ) , although the spin - down power and therefore the @xmath0-ray luminosity are the smallest in the four pulsars . for the pulse profile , we expect single peak in a single rotation period . we summarize the results of our work as follows . 1 . we have studied the non - thermal emissions of the geminga pulsar with the outer gap accelerator model . we solved the accelerating electric field with the pair creation and the radiation processes in the magnetic meridional plane . we demonstrated that the calculated spectrum of the curvature and the synchrotron emissions is consistent with the observation in x - ray through @xmath0-ray bands . the outward curvature radiation dominates in the spectrum of @xmath0-ray bands , while the inward synchrotron emissions dominate in x - ray bands . we demonstrated that the @xmath0-ray pulse above 100 mev has two peaks , which are made by the outward emissions . around 10 mev bands , we expect four peaks in a single period with both the outward and inward emissions . in x - ray bands , the pulse profile has two peaks made by the inward emissions . the pulse phase in x - ray bands are not in phase with that in the @xmath0-ray bands . the model predicts the energy dependent pulse morphology in x - ray through @xmath0-ray bands , which is also indicated by the observations . ( jackson et al . 2002 ; kargaltsev et al . 2005 ) . 3 . we discussed the inclination angle and the viewing angle for the geminga pulsar with the pulse morphology . the observed double peak structure in egret bands rules out nearly aligned rotator . assuming that the radio beam from the geminga pulsar is oriented in a different direction from the line of sight , a greatly inclined magnetic axis is also ruled out . the observed phase separation of two peaks in @xmath0-ray pulse profiles is explained by a viewing angle @xmath184 . we conclude that the geminga pulsar has a inclination angle that @xmath1 and a viewing angle that @xmath2 . 4 . we applied our method to another four middle - aged radio pulsars , whose spin - down power and distant from the earth expect the potential of the detection of the @xmath0-ray emissions from those pulsars by fermi telescope . applying the inclination angle and the viewing angle inferred from the radio polarization characteristics , we discussed the visibility of the @xmath0-ray emissions from the outer gap . we predict that @xmath0-ray emissions from psrs b0355 + 54 , b1449 - 64 and b1929 + 10 are detectable by fermi telescope . for psr b0740 - 28 , the @xmath0-ray beam from the outer gap will be oriented in a different direction from the viewing angle . the observations of the psr b0740 - 28 by fermi telescope will probably be useful for studying the emission site of not only the @xmath0-rays but also the radio wave in the pulsar magnetospheres . the authors appreciate fruitful discussion with k.s . cheng , k.hirotani , k.s . cheng , s. shibata , and r.taam . authors also thank anonymous referee for his / her insightful comments on the manuscript . this work was supported by the theoretical institute for advanced research in astrophysics ( tiara ) operated under academia sinica and national science council excellence projects program in taiwan administered through grant number nsc 96 - 2752-m-007 - 002-pae . 99 arons j. , 1983 , apj , 266 , 215 becker w. , et al . , 2006 , apj , 645 , 1421 bonsignori - facondi s.r . , salter c.j . & sutton j.m . , 1973 , a&a , 27 , 67 chang h .- k . , boggs s. , chang y .- h . for the nct collaboration , 2007 , adspr , 40 , 1281 cheng k.s . , ho c. & ruderman m. 1986a , apj , 300 , 500 cheng k.s . , ho c. & ruderman m. 1986b , apj , 300 , 522 cheng k.s . , & zhang l. 1996 , apj , 463 , 271 cheng k.s . , ruderman m. & zhang l. 2000 , apj , 537 , 964 contopoulos i. , kazanas d. & fendt c. 1999 , apj , 511 , 351 daugherty j.k . & harding , a.k . , 1996 , apj , 458 , 278 dean et al . , 2008 , sci , 321 , 1183 dyks j. & rudak b. , 2003 , apj , 598 , 1201 dyks j. , rudak b. & harding a.k . , 2004 , apj , 607 , 939 everett j.e . & weisberg j.m . , 2001 , apj , 553 , 341 fierro j.m . , michelson p.f . , nolan p.l . & thompson d.j . , 1998 , apj , 494 , 734 grizov a. , 2005 , phys . 94 , 021101 harding a.k . , stern j.v . , dyks j. & frackowiak m. , 2008 , apj , 680 , 1378 harding a.k . , strickman m.s . , gwinn c , dodson r. , moffet d. & mcculloch p. , 2002 , apj , 576 , 376 harding a.k . , usov v.v . , muslimov a.g . , 2005 , apj , 622 , 531 hirotani k. , 2006 , apj , 652 , 1475 hirotani k. , 2007 , apj , 662 , 1173 hobbs g. , lyne a.g . , kramer m. , martin c.e . & jordan c. , 2004 , mnras , 353 , 1311 hui c.y . & becker w. , 2008 , a&a , 486 , 485 jackson m.s . , halpern j.p . , gotthelf e.v . & mattox j.r . , 2002 , apj , 578 , 935 jones d.h . , stappers b.w . & gaensler b.m . , 2002 , a&a , 389l , 1 kanbach g. , sowikoska a. , kellner s. & steinle h. , 2005,2005 , in bulik t. , rudak b. , madejski g. , eds , aip conf . 801 , astrophysical sources of high energy particles and radiation . american institute of physics , new york , p. 306 proceeding , 801 , 306 kargaltsev o.y . , pavlov g.g . , zavlin v.e . & romani r.w . , 2005 , apj , 625 , 307 kataoka j. , et al . , 2005 , proc . spie , 5898 , 133 kijak j. & gil j. 2003 , a&a , 397 , 969 large m.i . , vaughan a.e . & wielebinski r. , 1969 , natur , 223 , 1249 lin l.c .- c . & chang h .- k . , 2008 , mnras , 387 , 729 manchester r.n . , hobbs g.b . , teoh a. , hobbs m. , 2005 , aj , 129 , 1993 manchester r.n . , taylor j.h . & huguenin g.r . , 1972 , nat . , 240 , 74 mcgowan k.e . , vestrand w.t . , kennea j.a . , zane s. , cropper m. & crdova f.a . , 2006 , apj , 647 , 1300 mestel l. , 1999 , stellar magnetism , international series of monographs of muslimov a.g . & harding a.k . 2003 , apj , 588 , 430 muslimov a.g . & harding a.k . 2004 , apj , 606 , 1143 muslimov a.g . & harding a.k . 2005 , apj , 630 , 454 rankin j.m . , 1993 , apjs , 84 , 145 romani r.w . & yadigaroglu i .- a . 1995 , apj , 438 , 314 romani r.w . 1996 , apj , 470 , 469 ruderman m.a . & sutherland p.g . , 1975 , apj , 196 , 51 senecha v.k . , bhat c.l . , rawat h.s . , rannot r.c . , sapru m.l . , kaul r.k . & tickoo a.k . , 1995 , a&a , 302 , 133 shibata s. , 1995 , mnras , 276 , 537 siegman b.c . , manchester r.n . & durdin j.m . , 1993 , mnras , 262 , 449 tang p.s . anisia , takata j. , jia , j.j . , cheng k.s . , 2008 , apj , 676 , 562 takata j. , shibata s. & hirotani k. 2004 , mnras , 354 , 1120 takata j. , shibata s. , hirotani k. & chang h .- k . 2006 , mnras , 366 , 1310 takata j. & chang h .- k . , 2007 apj , 670 , 677 takata j. , chang h .- k . & cheng k.s . , 2007 apj , 656 , 1044 takata j. , chang h .- k . & shibata s. , 2008 , mnras , 386 , 748 thompson d.j . , 2004 , in cheng k.s . , romero g.e . , eds , cosmic gamma ray sources . dordrecht , kluwer , p. 149 vats h.o . , singal a. k. , deshpande m.r . , iyer k.n . , oza r. , shah c.r . & doshi s. , 1999 , mnras , 302 , l65

we present a comparison of the frequencies of carbon - enhanced metal - poor ( cemp ) giant and main - sequence turnoff stars , selected from the sloan digital sky survey ( sdss ) and the sloan extension for galactic understanding and exploration ( segue ) , with predictions from asymptotic giant - branch ( agb ) mass - transfer models . we consider two initial mass functions ( imfs)a salpeter imf , and a mass function with a characteristic mass of 10 . previous observations suggest that the carbon abundances of red giants are altered during red - giant branch evolution due to mixing of their convective outer layers , resulting in a reduction of the observed carbon - abundance ratios . thus , in order to derive more accurate estimates of cemp frequencies for stars in the milky way , it is preferable to make use of sdss / segue main - sequence turnoff stars , which are not expected to experience significant dilution . however , because of the difficulty of identifying moderately carbon - enhanced stars ( [ c / fe ] ) among warm , metal - poor turnoff stars , owing to their much weaker ch-bands , we derive a correction function to compensate for the resulting undercounts of cemp stars . these comparisons indicate good agreement between the observed cemp frequencies for stars with [ fe / h ] and a salpeter imf , but not with an imf having a higher characteristic mass . thus , while the adopted agb model works well for low - mass progenitor stars , it does not do so for high - mass progenitors . our results imply that the imf shifted from high- to low - mass dominated in the early history of the milky way , which appears to have occurred at a `` chemical time '' between [ fe / h ] and [ fe / h ] . the corrected cemp frequency for the turnoff stars with [ fe / h ] is much higher than the agb model prediction from the high - mass imf , supporting the previous assertion that one or more additional mechanisms , not associated with agb stars , are required for the production of carbon - rich material below [ fe / h ] .

1310.3277

numerous spectroscopic studies of metal - poor ( [ fe / h ] @xmath9 ) candidates identified by the hk survey ( beers et al . 1985 , 1992 ) and the hamburg / eso survey ( he s ; wisotzki et al . 1996 ; christlieb et al . 2001 , 2008 ; christlieb 2003 ) have revealed that the frequency of carbon - enhanced stars increases strongly with decreasing [ fe / h ] . these stars , now known as carbon - enhanced metal - poor ( cemp ) stars , were originally defined as stars with metallicity [ fe / h ] @xmath10 and carbon - to - iron ratios [ c / fe ] @xmath11 ( beers & christlieb 2005 ) . generally , the frequency of c - rich stars increases from a few percent at higher metallicity to on the order of 20% for [ fe / h ] @xmath12 , 30% for [ fe / h ] @xmath7 , 40% for [ fe / h ] @xmath13 , and 75% for [ fe / h ] @xmath14 ( beers et al . 1992 ; norris et al . 1997 ; rossi et al . 1999 ; beers & christlieb 2005 ; cohen et al . 2005 ; marsteller et al . 2005 ; rossi et al . 2005 ; frebel et al . 2006 ; lucatello et al . 2006 ; norris et al . 2007 ; carollo et al . 2012 ; norris et al . 2013 ; spite et al . 2013 ; yong et al . this increasing trend of cemp - star frequency with declining [ fe / h ] is again confirmed from the many thousands of cemp stars found among the several hundred thousand stars with available spectra from the sloan digital sky survey ( sdss ; fukugita et al . 1996 ; gunn et al . 1998 , 2006 ; york et al . 2000 ; stoughton et al . 2002 ; abazajian et al . 2003 , 2004 , 2005 , 2009 ; pier et al . 2003 ; adelman - mccarthy et al . 2006 , 2007 , 2008 ; aihara et al . 2011 ; ahn et al . 2012 ) and the sloan extension for galactic understanding and exploration ( segue-1 ; yanny et al . 2009 ) , and segue-2 ( c. rockosi et al . , in preparation ) as described by lee et al . ( 2013 ) . there exist a number of subclasses within the cemp classification , as originally defined by beers & christlieb ( 2005 ) , which may provide direct clues to the nature of their likely progenitors . stars in the cemp-@xmath15 subclass exhibit over - abundances of @xmath15(low)-process elements such as ba and sr , the cemp-@xmath16 subclass includes stars with enhanced @xmath16(apid)-process elements such as eu , and the cemp-@xmath17 stars exhibit elemental abundance patterns associated with both the @xmath16-process and the @xmath15-process . the cemp - no subclass exhibits no over - abundances of the neutron - capture elements . the cemp-@xmath15 ( and cemp-@xmath17 ) subclasses of cemp stars are the most commonly found to date ; high - resolution spectroscopic studies show that around 80% of the cemp stars are categorized as cemp-@xmath15 ( or cemp-@xmath17 ) ( aoki et al . 2007 , 2008 ) . the favored mechanism for the production of the high [ c / fe ] ratios found for cemp-@xmath15 ( cemp-@xmath17 ) stars is mass transfer of carbon - enhanced material from the envelope of a now - defunct asymptotic giant - branch ( agb ) star to its ( presently observed ) binary companion ( suda et al . 2004 ; herwig 2005 ; komiya et al . 2007 ; sneden et al . 2008 ; masseron et al . 2010 ; bisterzo et al . 2011 , 2012 ) . observational evidence now exists to suggest that the cemp-@xmath17 stars ( and other @xmath16-process - element rich stars ) were enhanced in @xmath16-process elements in their natal gas clouds by previous generations of supernovae ( sne ) , and did not require a contribution of @xmath16-process elements from a binary companion ( see hansen et al . 2013 ) . the limited amount of long - term radial - velocity monitoring available for cemp stars indicates variations for almost all of the cemp-@xmath15 stars , confirming their binary status ( lucatello et al . 2005a ) . in addition , the cemp-@xmath15 stars are mostly , though not exclusively ( e.g. , norris et al . 2013 and references therein ) , found among metal - poor stars with [ fe / h ] @xmath18 . on the other hand , cemp - no stars are found most commonly among the extremely metal - poor ( emp ) stars , with [ fe / h ] @xmath19 ( aoki et al . 2007 ; norris et al . 2013 ) . existing radial - velocity monitoring of these objects indicates that they are found in binary systems no more frequently than other metal - poor stars ( t. hansen et al . , in preparation ) . norris et al . ( 2013 ) found no cemp-@xmath15 stars among 18 cemp stars with [ c / fe ] @xmath20 and [ fe / h ] @xmath21 , as well as no discernible variations of their radial velocities . although there is general consensus on the origin of cemp-@xmath15 stars , the likely progenitor or progenitors of the cemp - no stars are still under discussion . suggested models include massive , rapidly rotating , mega metal - poor ( mmp ; [ fe / h ] @xmath22 ) stars , which produce large amounts of c , n , and o due to distinctive internal burning and mixing episodes ( meynet et al . 2006 , 2010 ; chiappini 2013 ) , and faint ( low - energy ) sne associated with the first generations of stars , which experience extensive mixing and fallback during their explosions , and eject large amounts of c and o , but not heavier metals ( umeda & nomoto 2003 , 2005 ; tominaga et al . 2007 , 2013 ; ito et al . 2009 , 2013 ; nomoto et al . nevertheless , the origin of the cemp - no star phenomenon is yet to be fully resolved ( see norris et al . 2013 , which summarizes other possible progenitors of the c - rich stars ) . previous authors have attempted to understand the large fractions of cemp stars at low metallicity , as well as the different subclasses of the cemp stars , by invoking agb models with different masses . furthermore , there have been several efforts to constrain the form of the early initial mass function ( imf ) by reproducing the observed frequencies of cemp stars , as well as the number ratios of the different cemp subclasses . abia et al . ( 2001 ) , for example , claimed that the large number of carbon - enhanced stars found among stars of very low metallicity could be accounted for if the imf in the early history of the galaxy was dominated by higher mass stars . lucatello et al . ( 2005b ) and komiya et al . ( 2007 ) utilized population - synthesis models with an imf biased towards massive stars to compare with the fractions of observed cemp stars , and concluded that an imf comprising a larger number of intermediate- to high - mass stars could reproduce the larger fraction of the cemp stars among metal - poor stars ( [ fe / h ] @xmath23 ) better than the present - day ( salpeter ) imf . recently , suda et al . ( 2013 ) made use of the number ratios of cemp / emp , cemp - no / cemp , and nemp / cemp giant stars ( where nemp stands for nitrogen - enhanced metal - poor ) from the stellar abundances for galactic archeology ( saga ; suda et al . 2008 ) database to constrain the parameters in their binary population - synthesis model . they considered several imfs , and proposed that the imf changed from high - mass dominated in the early galaxy to low - mass ( @xmath24 @xmath0 ) dominated at present , and that this transition occurred around a metallicity of [ fe / h ] @xmath25 . the above studies carried out comparisons of the cemp fractions derived from small samples of stars comprising mostly giants . however , observational evidence indicates that the c - rich material at the surface of a giant could be easily depleted by extra mixing of cno - processed material from its interior during the so - called first dredge - up episode ( spite et al . 2005 , 2006 ; lucatello et al . 2006 ; aoki et al . it is also known that more luminous red giant - branch ( rgb ) stars are more affected by such mixing ( spite et al . 2005 , 2006 ) . therefore , if such mixing does occur , the overall cemp frequencies as estimated from giants are expected to be a lower limit . in order to avoid this complication , the best way forward would appear to be comparing model predictions with the observed cemp frequencies based on unevolved stars , such as dwarfs or main - sequence turnoff stars . the current cemp stars that have been studied with high - resolution spectroscopy are mostly giants ( for which it is simpler to obtain high s / n spectra , due to their relative brightness and moderate temperatures , which allows for lines of interest to be measured with less uncertainty ) . the numbers of observed dwarf and turnoff stars with similar observations are in any case too small to derive statistically meaningful results for different subclasses of cemp stars . in this study , we make use of stars with available carbon - to - iron ratios ( [ c / fe ] ) and [ fe / h ] , based on medium - resolution ( @xmath26 ) spectroscopy obtained during the course of the sdss , segue-1 , and segue-2 , in order to derive accurate frequencies of cemp stars among giants and turnoff stars as a function of [ fe / h ] . the derived cemp frequencies are then compared with the predictions from agb binary - synthesis models that employ the two different imfs explored by suda et al . ( the results of these comparisons should provide more stringent constraints on the imf of the milky way , and clues to the existence of progenitors _ other than agb stars _ that are capable of producing large amounts of carbon - enhanced material in the early universe . this paper is outlined as follows . in section 2 , we describe the selection criteria used to assemble the sample for this study . section 3 presents and discusses results of the comparison of the cemp frequencies for giants and main - sequence turnoff stars with the binary - synthesis model predictions , and describes a procedure for correcting the anticipated undercounts of cemp stars among warm , metal - poor turnoff stars . our conclusions are presented in section 4 . the sdss , segue-1 , and segue-2 surveys have produced an unprecedented number of high - quality medium - resolution stellar spectra , covering stars in various evolutionary stages , and spanning a wide range of metallicity ( @xmath274.0 @xmath28 [ fe / h ] @xmath29 ) . a total of about 600,000 stars are potentially suitable for examination of the properties of the milky way s stellar populations . the resolving power of the spectra is @xmath26 , over the wavelength range 38209100 . below we simply refer to these stars ( spectra ) as sdss / segue stars ( spectra ) . accurate estimates of the atmospheric parameters for most of the sdss / segue stars are derived using the latest version of the segue stellar parameter pipeline ( sspp ; lee et al . 2008a , 2008b , 2011 ; allende prieto et al . 2008 ; smolinski et al . the typical external errors obtained by the sspp are 180 k for @xmath30 , 0.24 dex for @xmath31 , and 0.23 dex for [ fe / h ] , respectively ( smolinski et al . 2011 ) . in addition , estimates of the carbonicity , [ c / fe ] , is obtained following the prescription of lee et al . ( 2013 ) , for stars with @xmath32 @xmath30 @xmath33 6700 k , where accurate [ c / fe ] can be determined . as reported by lee et al . , uncertainties in the determination of [ c / fe ] are smaller than 0.35 dex for sdss / segue spectra with s / n @xmath34 @xmath35 . in order to derive reliable frequencies of the cemp stars among the field stellar populations , we follow the selection criteria of lee et al . briefly , we first exclude all stars located in the directions of known open and globular clusters . for stars that were observed more than once , we keep only the parameters derived from the highest s / n spectrum . we then restrict the sample to stars with spectra having s / n @xmath36 @xmath35 , effective temperatures in the range 4400 k @xmath33 @xmath30@xmath33 6700 k , and metallicities in the range @xmath274.0 @xmath33 [ fe / h ] @xmath37 , so that our estimates of [ c / fe ] are as reliable as possible . we then visually inspect each spectrum with [ fe / h ] @xmath38 , in order to reject spectra such as cool white dwarfs , or those with emission - line features in the cores of their caii lines , or other spectral defects that could lead to spurious determinations of metallicity by the sspp . additionally , we visually examine the spectra for all stars with [ c / fe]@xmath20 , and exclude stars with poor estimates of [ fe / h ] and/or [ c / fe ] . following lee et al . , for the purpose of deriving the cemp frequencies we do not include stars with [ c / fe ] @xmath39 and indication of upper limit estimate , and consider these stars to have unknown carbon status . by application of the above procedures , we are left with a sample of about 247,350 stars . for the purpose of our analysis , we consider stars with 4400 k @xmath33 @xmath30 @xmath33 5600 k and 1.0 @xmath33 @xmath31 @xmath28 3.2 as giants , and stars with 5600 k @xmath33 @xmath30 @xmath33 6600 k and 3.2 @xmath33 @xmath31 @xmath28 4.5 as main - sequence turnoff stars . the distance to each star is estimated following the prescriptions of beers et al . ( 2000 , 2012 ) , which obtains photometric distance estimates with errors on the order of 10%20% . note that we have added stars to our sample from table 1 of yong et al . ( 2013 ) , and one object from caffau et al . ( 2011 ) , with determinations based on high - resolution spectroscopic analyses , in order to increase the number of stars with [ fe / h ] @xmath40 . this results in better number statistics for the calculation of the cemp frequencies in the extremely and ultra metal - poor regime . figure [ fig : cfedist ] shows the distribution of [ c / fe ] for sdss / segue stars in various bins of metallicity , decreasing from the upper to lower panels . main - sequence turnoff stars are shown as solid histograms ; giants are shown as dashed histograms . inspection of this figure reveals that the overall distribution of [ c / fe ] gradually shifts ( for both turnoff stars and giants ) to higher [ c / fe ] with decreasing [ fe / h ] , with a tail extending toward higher [ c / fe ] appearing as the metallicity decreases . as the metallicity decreases below [ fe / h ] @xmath8 , this trend continues for the turnoff stars , but it is not as evident for the giants . another interesting feature seen in figure [ fig : cfedist ] is that , for [ fe / h ] @xmath7 , the turnoff stars are distributed over a wide range of [ c / fe ] , whereas the giants are mostly concentrated in the region of [ c / fe ] @xmath41 . the red vertical lines in figure [ fig : cfedist ] indicate the mean values of [ c / fe ] for the turnoff stars ( solid line ) and giants ( dashed line ) , respectively . on average , the giants appear to exhibit lower carbonicity ( by about 0.2 dex ) than the turnoff stars , down to [ fe / h ] = @xmath42 . the mean value of [ c / fe ] appears to increase with decreasing metallicity , as also found by carollo et al . ( 2012 , their figure 11 ) . they reported that the degree of carbon enhancement significantly increased from [ c / fe ] @xmath43 at [ fe / h ] = @xmath44 to [ c / fe ] @xmath45 at [ fe / h ] = @xmath46 , somewhat higher than our values ( it should be noted that a different sample of stars , as well as a different method for determination of [ c / fe ] , were employed by these authors ) . the difference in the distribution of [ c / fe ] between the turnoff stars and the giants may be explained by the different masses of the convective envelopes between the two evolutionary stages . because a giant has a much larger convective envelope , its surface material experiences more mixing , leading to reduction of the carbonicity . on the other hand , the turnoff stars have shallower convective envelopes , so that their surface abundances may not be expected to greatly change . as a result , the overall carbon abundance for the turnoff population is expected to be higher than that of the giant population , even if they were born with the same initial carbon abundance . bonifacio et al . ( 2009 ) also noted a difference in the mean [ c / fe ] between giants and turnoff stars of similar metallicities , finding a difference of about 0.2 dex ( giants being lower ) for stars with [ fe / h ] @xmath23 . they argued that this difference arises because the giants suffer from extra mixing due to first dredge - up , and have their surface carbon abundance reduced . they also suggested the stellar models employed in the analysis could contribute to this discrepancy ; they found a smaller difference between the giants and turnoff stars when deriving [ c / fe ] from a 3d , rather than a 1d model atmosphere . the black filled circles in figure [ fig : cempcom ] represent our derived differential frequencies for c - rich giants as a function of [ fe / h ] . from inspection of this figure , it appears that the cemp frequencies do not increase for [ fe / h ] @xmath47 , but rather , remain relatively constant , in contrast to the results of previous studies . as discussed by spite et al . ( 2005 , 2006 ) , lucatello et al . ( 2006 ) , and aoki et al . ( 2007 ) , this may be in part due to cn - processing , which converts carbon to nitrogen at the bottom of a star s convection zone , and in turn reduces the carbon abundance in the envelope at the time of first dredge - up . whether or not a giant experiences such mixing can be identified by measuring its @xmath48c/@xmath49c or [ c / n ] ratios , as both will be lower for an object that has gone through such an event . unfortunately , these ratios are difficult to assess from the sdss / segue spectra over the full range of metallicities we consider . in the metallicity regime of [ fe / h ] @xmath50 , our derived cemp frequencies of @xmath511% agree with the previously claimed fraction of classical ch or ba stars in the solar neighborhood ( luck & bond 1991 ) . regarding the effect of the first dredge - up , it is worth mentioning the following arguments from a theoretical point of view . according to suda et al . ( 2004 ) , first dredge - up might not play a significant role in decreasing the carbon abundance on the surface of a giant , based on an 0.8 @xmath0 model for he 0107 - 5240 ( a cemp - no star with [ fe / h ] = @xmath52 ; christlieb et al . they examined the effect of the first dredge - up following the accretion of c- and o - rich matter onto the star ( here assuming an agb mass - transfer scenario ) , and found that if its envelope is significantly c - rich ( [ c / fe ] @xmath53 ) , then after first dredge - up the surface carbon abundance changed little . this is because the surface carbon abundance is too large ( [ c / h ] @xmath54 ) prior to the dredge - up to be significantly depleted by the relatively small amount of matter in the hydrogen - burning shell , @xmath55 0.02 @xmath0 . even in cases of emp stars for which the initial carbon abundance is small , suda & fujimoto ( 2010 ) showed that the effect of the first dredge - up is also limited , due to the shallower convective envelopes in metal - poor ( as compared to metal - rich ) stars . according to their model calculation , the change of the cno abundance before and after the first dredge - up was on the order of one percent therefore , they did not notice a large impact on the surface carbon abundances after first dredge - up for stars with [ fe / h ] @xmath56 . the blue open circles in figure [ fig : cempcom ] are the predicted frequencies of cemp giants , as a function of metallicity , from agb binary - synthesis models with an imf peaked at 10 @xmath0 , while the open squares are the predicted frequencies from models using a salpeter imf , adopted from suda et al . ( 2013 ) . in their model , they included a mechanism referred to as `` pulsation - driven mass loss '' ( wood 2011 ) , which was argued to suppress the previously predicted over - production of nemp stars by izzard et al . ( 2009 ) and pols et al . it appears that the predicted cemp frequencies for the high - mass dominated imf are in relatively good agreement with the observed cemp giant frequencies for [ fe / h ] @xmath23 , but the model predicts too many c - rich stars above [ fe / h ] = @xmath42 . the predicted cemp frequencies from a salpeter imf are in good agreement with our derived frequencies for the metal - rich region ( [ fe / h ] @xmath50 ) , while the predicted cemp frequencies are far too low for stars with [ fe / h ] @xmath23 . these are similar results to those found by suda et al . ( 2013 ) , who used the giants in the saga database to compare the observed cemp frequencies with their model predictions . one of the reasons that suda et al . ( 2013 ) employed giants to derive the cemp frequency is that one can ignore effects such as atomic diffusion , which can alter the surface abundances of dwarfs and turnoff stars more significantly than in giants ( e.g. , richard et al . 2002a , b ; korn et al . 2007 ; lind et al . 2008 ) . however , our derived frequencies show a much better agreement for the salpeter imf in the metallicity region [ fe / h ] @xmath50 . in the study of suda et al . ( 2013 ) , the model - predicted frequency of the cemp stars above [ fe / h ] = @xmath57 was not well - constrained , most likely due to the selection biases associated with the assembly of their sample from previous high - resolution spectroscopic studies ( which tended to emphasize the more metal - poor and/or carbon - enhanced stars ) . in contrast , the good agreement of the frequencies calculated from our considerably less - biased sdss / segue sample with the model prediction for [ fe / h ] @xmath4 suggests that the agb binary - synthesis model with a salpeter mass function used by suda et al . ( 2013 ) works well , at least in this metallicity regime . based on the results from the comparisons of the observed cemp frequencies with model predictions from the two different imfs , we conjecture that , for very low - metallicity ( [ fe / h ] @xmath23 ) stars , the distribution of the stellar masses was dominated by rather massive stars ( @xmath5110 @xmath0 or higher ) , while for the relatively more metal - rich stars ( [ fe / h ] @xmath50 ) , it appears that the imf did not much differ from a salpeter imf , which is biased towards low - mass progenitor stars ( @xmath24 @xmath0 ) . as previously claimed by suda et al . ( 2013 ) , our results also support the idea that there must exist a shift in the imf from a high - mass dominated to low - mass dominated form in the early history of the milky way , corresponding to a `` chemical time '' between [ fe / h ] @xmath5 and [ fe / h ] @xmath6 . by way of comparison , the binary population - synthesis model of izzard et al . ( 2009 ) was able to reproduce the ratio of nemp to very metal - poor ( vmp ; [ fe / h ] @xmath12 ) stars ( that is , c and n normal stars ) _ without _ introducing an imf dominated by higher mass stars , but not the high frequency of the cemp stars . pols et al . ( 2012 ) also argued , by comparing the observed number ratio of nemp to cemp stars with their model predictions , that they could derive a similar number ratio from a salpeter imf , and ruled out an imf peaked at 10 @xmath0 claimed by komiya et al . ( 2007 ) . another interesting result emerges when one partitions the giant sample based on distance from the galactic plane . the green squares in figure [ fig : cempcom ] are the cemp frequencies for giants with distances from the galactic mid - plane ( @xmath58 ) larger than 5 kpc , whereas the red triangles represent frequencies based on those with @xmath59 kpc . the figure clearly indicates that the more distant halo giants exhibit higher frequencies of c - rich stars , while the stars closer to the galactic plane tend to have lower frequencies of c - rich stars . this same trend with vertical distance was hinted at ( due to small number statistics ) in frebel et al . ( 2006 ) , and strongly confirmed in the much larger sample of sdss / segue calibration stars by carollo et al . carollo et al . argued that this result was likely due to the fact that the outer - halo population has about twice the frequency of cemp stars , at a given low metallicity , as the inner - halo population . a few possible reasons for the observed differences in the cemp frequencies between the two spatial regions might be suggested within the context of the agb model predictions . first , the progenitors of the inner - halo population ( which dominates for @xmath60 5 kpc ) and the outer - halo population ( which , at @xmath61 5 kpc , includes more outer - halo stars ) might have formed their stars at different times , with different imfs . because the outer - halo population has more cemp stars than the model prediction for [ fe / h ] @xmath47 , it is possible that the outer - halo population might have had an imf with more intermediate - mass stars than considered by the model . on the other hand , since the cemp frequencies of the inner - halo stars are lower than the model estimate , the inner - halo population might have had an imf with less intermediate - mass stars than the proposed imf . related ideas are discussed by tumlinson et al . ( 2007 ) . another possibility is that suda et al . ( 2013 ) assumed that all cemp stars , including cemp - no objects , formed from the agb binary scenario . if there were to exist other channels of carbon production at [ fe / h ] @xmath47 , such as faint sne or rapidly rotating massive stars ( producing cemp - no stars in the subsequent generation ) , as suggested by several studies , we then might expect larger frequencies of cemp stars than the agb binary - synthesis model prediction ( as seen in figure [ fig : cempcom ] ) , even if the carbon dilution of the giants due to extra mixing is taken into account . a more detailed discussion of this is provided below . in any event , the frequency difference we find can be understood ( as argued by carollo et al . 2012 ) as the result of a change in the dominant population with distance above the plane , from the inner - halo population to the outer - halo population . carollo et al . further argued that the inner halo is dominated by stars with modest carbon enhancement ( [ c / fe ] @xmath62 ) , while the outer halo has a greater portion of stars with large carbon enhancements ( [ c / fe ] @xmath63 ) , although considerable overlap still exists . they interpreted these results , as well as the increase in the global frequency of cemp stars with distance from the galactic plane , as evidence for the possible presence of additional astrophysical sources of carbon , beyond agb production alone , associated with the progenitors of the outer - halo stars . it is difficult to separate cemp-@xmath15 and cemp - no stars from our medium - resolution sdss / segue spectra , but determination of the ratio of cemp-@xmath15/cemp - no for stars in the inner- and outer - halo populations , based on high - resolution spectroscopy , will provide not only very strong constraints on the binary - synthesis model , but clues to the origin of the different cemp frequencies between the inner- and outer - halo populations . above we have compared the derived cemp frequencies from our sample of giants with the predicted cemp frequencies from agb binary - synthesis models by suda et al . however , as previously mentioned , observations suggest that a giant can suffer from dilution of the carbon - rich material in its envelope by mixing during the first dredge - up event , resulting in a lower overall carbon abundance and ( as a population ) lower frequencies of cemp stars . absent such dilution , we would expect that the actual frequencies of cemp stars among giants would be higher than shown in figure [ fig : cempcom ] . stars near the main - sequence turnoff region do not experience dredge - up episodes ; rather , they preserve unpolluted material on their surfaces . it might be possible that their surface abundances could be affected by atomic diffusion , but because the impact on the carbon abundance is not well known , we do not take this concern into consideration in this study . thus , we expect that one could obtain a more valid estimate of the frequencies of cemp stars in a given population by making this evaluation using main - sequence turnoff stars . since turnoff stars evolve quickly into giants during their evolution , we might expect that the frequencies of cemp stars inferred from stars near the main - sequence turnoff should be the same as for giants that have not yet mixed carbon - depleted material into their envelopes . therefore , it is desirable to compare the predictions from the models to the frequencies of cemp stars derived from the turnoff stars . however , additional complications exist . the stars located near the main - sequence turnoff are relatively warmer than the red giants , and as a result , for a given carbon abundance , the molecular ch @xmath3-band feature becomes significantly weaker . to make matters more difficult , at low metallicity ( assuming carbon is not enriched ) a star s ch @xmath3-band will also become lower in strength . even with high - resolution spectroscopy aoki et al . ( 2013 ) noted that , although they were able to detect the ch @xmath3-band for a star with [ c / fe ] @xmath64 and [ fe / h ] @xmath65 at @xmath30 @xmath66 k , they failed to measure the ch @xmath3-band for [ c / fe ] @xmath2 in their sample of very metal - poor stars . these effects become even more prevalent for medium - resolution spectra , hence the calculated cemp frequencies obtained from the turnoff stars may also be lower than the actual values . in order to address this difficulty , and to provide a check on just how many c - rich halo stars may have been misclassified as c - normal objects ( [ c / fe ] @xmath67 ) for stars around the turnoff region , we have performed the following experiment . following the prescription by lee et al . ( 2013 ) , we inject artificial noise ( with characteristics similar to that for a typical sdss / segue spectrum ) into the grid of synthetic spectra that are used to estimate [ c / fe ] . the noise - added synthetic spectra have s / n = 40 , 45 , and 50 , which are typical of the quality of the sdss / segue spectra in our study with @xmath59 kpc ( justification for this choice is provided below ) . at each s / n there are 25 different realizations . these spectra are processed through the sspp to determine estimates of [ c / fe ] . with the measured [ c / fe ] in hand , we then derive the cemp frequencies of the spectra , as a function of [ fe / h ] , which have an _ estimated _ [ c / fe ] less than @xmath68 among the spectra with @xmath69 [ c / fe ] @xmath70 for giants , turnoff stars , and dwarf stars ( note that we employ discrete [ c / fe ] values , @xmath71 , and @xmath72 for the synthetic spectra ) . for the purpose of this exercise , we define giants as models with parameters in the ranges 4500 k @xmath33 @xmath30 @xmath33 5500 k and 1.0 @xmath33 @xmath31 @xmath73 , turnoff stars for 5750 k @xmath33 @xmath30 @xmath33 6500 k and 3.5 @xmath33 @xmath31 @xmath74 , and dwarfs for 4500 k @xmath33 @xmath30 @xmath33 5500 k and 4.5 @xmath33 @xmath31 @xmath75 . figure [ fig : corfun ] shows the results of this experiment . the plus signs represent the actual frequencies of cemp stars , which are set to 1.0 . the green squares indicate giants , the red circles the turnoff stars , and the blue triangles the dwarfs . the bottom plot exhibits the residuals in the derived fractions ( simply 1 the fractions ) in each category . inspection of this figure reveals that some of the c - rich dwarfs and giants start to be classified as c - normal ( [ c / fe ] @xmath76 ) from around [ fe / h ] @xmath12 ; the number of the misclassified stars slowly increases with decreasing metallicity . by way of comparison , the c - rich turnoff stars begin to be misclassified as c - normal stars as metallicity drops below [ fe / h ] = @xmath77 ; this misclassification rapidly increases with declining metallicity , as expected . we now derive a correction function for capturing the `` true '' cemp frequency , as a function of [ fe / h ] , for the sdss / segue turnoff stars , based on the results of the test carried out above . we then use this correction function to adjust the frequency calculation , among the stars with @xmath78 [ c / fe ] @xmath2 , taking the `` missing '' c - rich stars into account . since the giants are more luminous than the turnoff stars , they can probe to greater distances in the galaxy . this increases the likelihood of introducing a greater number of giants than turnoff stars into a magnitude - limited sample ( the frequency of giants can also be influenced by the luminosity function of the halo field stars , as well as by shifts in the mix of stellar populations between the nearby and more distant halo stars ) . this possible population transition has already been noted in figure [ fig : cempcom ] , and discussed in detail in the previous section . thus , in order to make sure we are sampling the giants and turnoff stars in similar regions of the galaxy , we restrict our considerations to the stars with @xmath59 kpc for the calculation of the cemp frequencies . figure [ fig : cempcor ] shows the differential frequencies of c - rich stars for main - sequence turnoff stars , as a function of [ fe / h ] , as red open triangles ; the corrected frequencies are indicated as green filled triangles . the blue open circles represent the predicted turnoff cemp frequencies from an agb binary - synthesis model with an imf peaked at 10 @xmath0 , while the magenta open squares indicates the prediction obtained using a salpeter imf . for comparison , the filled black circles are the frequencies from the giants with @xmath60 5 kpc given in figure [ fig : cempcom ] . the behavior seen in this figure is consistent with our expectations , in that the cemp frequencies for the giants are lower than that for the turnoff sample , at least at the lowest metallicities . note that the corrected cemp frequencies for the turnoff stars are , on average , higher than the uncorrected frequencies , by about 5% . the figure also shows that the models produce higher cemp frequencies for the turnoff stars than for the giants for both imfs ( compare with figure [ fig : cempcom ] ) , which at least qualitatively agrees with the observations . the small difference in the model - predicted cemp frequencies between the giants and turnoff stars ( figures [ fig : cempcom ] and [ fig : cempcor ] ) may arise from the difference in the mass of the convective envelope ; a star that evolves to the giant stage has a much deeper convective zone , and hence more effective dilution , resulting in lower cemp frequencies derived for the giants . our derived cemp frequencies from the distance - restricted turnoff sample appears to be in good agreement with the model prediction based on a salpeter imf ( magenta open squares ) , down to [ fe / h ] = @xmath42 . this reaffirms that the agb model for progenitor stars in the low - mass range works well . however , the model estimate of the cemp frequency does not reproduce the observed frequencies for [ fe / h ] @xmath23 . unlike the case for the giants ( figure [ fig : cempcom ] ) , the observed cemp frequencies from our turnoff sample do not agree with the model estimation from the top - heavy imf for [ fe / h ] @xmath23 at all , as these remain roughly flat instead of growing dramatically with decreasing metallicity . one reason for the large discrepancy between the model estimates and the observed frequencies of cemp stars may be uncertainties of the model parameters adopted for producing carbon in the agb star , and subsequent processes that enrich ( or deplete ) the envelope with carbon . below we discuss known sources of uncertainty associated with the agb binary - synthesis model , which may result in changes of the predicted carbon abundance of the secondary star . most agb stars in the mass range of @xmath5118 @xmath0 can produce carbon , but whether or not they develop carbon - enriched envelopes depends on the efficiency of the third dredge - up ( tdu ) and helium - flash driven deep mixing ( he - fddm ; fujimoto et al . 1990 , 2000 ) events for [ fe / h ] @xmath23 . at present , it is not fully understand how such episodes depend on the mass and metallicity of an agb star . for example , lau et al . ( 2009 ) claimed , in a study of the evolution of agb stars with metallicity between z = 10@xmath79 and 10@xmath80 , that the he - fddm did not take place for z @xmath81 independent of the mass of a star , and that this event did not occur for a star with @xmath82 @xmath0 , regardless of its metallicity . however , suda & fujimoto ( 2010 ) found , from an extensive set of stellar evolution models , that the he - fddm event occurred for a star with @xmath83 @xmath0 for zero metallicity , while it occurred for a star with @xmath84 @xmath0 for @xmath85 [ fe / h ] @xmath86 . they also found that the tdu episode was restricted to a mass range of @xmath87 1.55 @xmath0 for [ fe / h ] = @xmath88 , and that this mass range becomes smaller as the metallicity decreases . in the adopted models from suda et al . ( 2013 ) , the increasing fractions of cemp stars comes from the he - fddm , which can enhance the surface carbon abundance by a factor of 10 , from [ c / h ] @xmath54 to @xmath89 , regardless of the initial metallicity of the models , for stars with masses of 0.83 @xmath0 . in this view , the value of [ c / fe ] for the secondary component ( the presently observed cemp star ) of a given binary increases with decreasing [ fe / h ] , so that a larger fraction of cemp stars can be achieved at lower metallicity . it is also assumed that the efficiency of the binary mass transfer and the mass - loss rates do not depend on metallicity . in addition , among these agb stars , the intermediate - mass ( @xmath5138 @xmath0 ) objects can be enriched with nitrogen by operation of the hot - bottom burning ( hbb ) process , which converts carbon into nitrogen by cn processing , and predicts the production of nemp stars ( johnson et al . however , the dependency of the hbb on the mass and metallicity of an agb star is yet not well - established . even taking into account the uncertainties in the parameters of the agb models , a more plausible interpretation of our results may be the existence of additional ( non - agb ) carbon - production mechanisms , as discussed in the introduction , which result in large frequencies of cemp stars in the metallicity regime [ fe / h ] @xmath7 . the current observations certainly favor this interpretation , since most cemp - no stars in the galaxy appear at [ fe / h ] @xmath40 , and these stars do not commonly exhibit the radial velocity variations that would be expected if membership in a binary system were required ( as in the agb mass - transfer scenario ) . the adopted models , however , assume that all cemp - no stars form from the agb binary mass - transfer scenario , rather than including additional sources that have been argued are likely to be present in the early universe . according to the agb models , low - mass ( @xmath90 @xmath0 ) agb stars efficiently create @xmath15-process elements by generating extra neutrons via the @xmath49c(@xmath91,n)@xmath92o reaction , while a weak @xmath15-process ( for light @xmath15-process elements ) operates by @xmath93ne(@xmath91,n)@xmath94 mg for the intermediate - mass stars . cemp - no stars could form in the agb mass - transfer scenario by suppressing the formation of the @xmath49c pocket for intermediate mass stars ( @xmath95 @xmath0 ; suda et al . thus , in order to preferentially produce cemp - no stars at low metallicity and maintain the observed ratio of cemp - no/(cemp-@xmath15 @xmath96 cemp - no ) , which is close to 0.5 at [ fe / h ] @xmath65 ( from table 1 of suda et al . 2013 ) , the models have to assume that the @xmath49c pocket does not form in stars with metallicity significantly below [ fe / h ] = @xmath42 . it has also been suggested ( komiya et al . 2007 ) that a secondary star with an agb primary having 0.8 @xmath0 @xmath97 @xmath0 could become a cemp-@xmath15 star following mass transfer , while systems that include an agb primary with @xmath95 @xmath0 could produce c without the enhancement of neutron - capture elements , leading to a cemp - no star following mass transfer . however , current agb models do not satisfactorily explain the absence of the @xmath49c pocket at low metallicities , even though this assumption has been invoked to explain the observed decreasing trend of [ pb / ba ] for cemp stars with [ fe / h ] @xmath23 ( aoki et al . 2002 ; suda et al . 2004 ; barbuy et al . 2005 ; cohen et al . 2006 ; aoki et al . it is also significant that ito et al . ( 2013 ) obtained a rather low upper limit on the abundance of lead ( log @xmath98(pb ) @xmath99 ) for the [ fe / h ] @xmath100 cemp - no star bd@xmath9644@xmath101493 , while previous predictions called for log @xmath98(pb ) @xmath102 at these low metallicities if the lead were produced by the @xmath15-process ( cohen et al . 2006 ) . chemical abundances of cemp - no stars observed with high - resolution spectroscopy ( e.g. , ito et al . 2009 , 2013 ; norris et al . 2013 ) support other scenarios for significant carbon production . their abundance patterns are similar to the predictions from massive , rapidly rotating , mmp stars ( meynet et al . 2006 , 2010 ) or faint sne that experience mixing and fallback ( umeda & nomoto 2003 , 2005 ; tominaga et al.2007 , 2013 ; kobayashi et al . 2011 ; ito et al . 2013 ; nomoto et al . if such mechanisms are the dominant sources of the large amounts of carbon produced at low metallicity , these scenarios also favor an imf that preferentially produces massive stars . in fact , one might also expect a rather abrupt `` break '' in the cemp frequencies when the primary carbon sources change in nature such a sudden change can be seen for the turnoff stars in figure [ fig : cempcor ] below [ fe / h ] @xmath8 . although we are not able to distinguish cemp - no stars from cemp-@xmath15 stars in our sample , given that the majority of the cemp - no stars are found with [ fe / h ] @xmath40 , our derived frequencies imply that non - agb related phenomenon may be the dominant mechanisms for producing large carbon abundances at extremely low [ fe / h ] . finally , we see that a similar behavior in the cemp frequencies with [ fe / h ] applies to the turnoff stars with @xmath103 [ fe / h ] @xmath104 as for the giants , suggesting that the proposed shift in an imf occurred over this chemical interval . we have compared our derived cemp frequencies from the sdss / segue giant sample with that predicted by agb binary - synthesis models with two different imfs a salpeter imf , and an imf with a characteristic mass of 10 @xmath0 . good agreement of the cemp frequencies for [ fe / h ] @xmath50 with the salpeter imf indicates that the adopted agb model works well for low - mass progenitor stars . qualitatively , better agreement with an imf biased to higher - mass progenitors is found for [ fe / h ] @xmath47 , suggesting that the nature of the imf shifted from one that is high - mass dominated in the early history of the milky way galaxy , to one that is now low - mass dominated . this transition appears to have occurred , in `` chemical time '' , between [ fe / h ] @xmath5 and [ fe / h ] @xmath6 , as other recent studies have argued ( e.g. , suda et al . 2011 , 2013 ; yamada et al . 2013 ) . as noted by other recent work , the more distant halo giants ( those with @xmath58 @xmath105 kpc ) exhibit higher frequencies of cemp stars compared to those closer to the galactic plane . a plausible explanation for this difference is the expected change of the dominant stellar populations from the inner - halo to the outer - halo population , coupled with the assumption that the outer - halo stars are associated with progenitors capable of producing large amounts of carbon without the accompanying production of heavy metals . thus , one might expect that the inner - halo population harbors a higher ratio of cemp-@xmath15/cemp - no stars , while the opposite may apply to the outer - halo population . tests of this hypothesis are underway ( d. carollo et al . , in preparation ) . the weak ch @xmath3-bands for moderately carbon - enhanced stars ( @xmath1 [ c / fe ] @xmath2 ) among warm , metal - poor main - sequence turnoff stars results in their likely having been undercounted by previous assessments of cemp frequencies . we have derived a correction function to compensate for this , making use of noise - added synthetic spectra . the corrected cemp frequencies for turnoff stars are , on average , higher by @xmath106% than with the uncorrected frequencies . both the corrected and uncorrected cemp frequencies derived from the turnoff sample exceed those of the giants for [ fe / h ] @xmath7 . we have made use of main - sequence turnoff stars with @xmath59 kpc to compute more realistic cemp frequencies than obtained by using giants ( or the combination of giants with other classes ) , corrected as mentioned above . for [ fe / h ] @xmath107 , our corrected cemp frequencies agree with the model predictions based on a salpeter imf , indicating that the agb model used in this study is probably not far from reality , at least as applied to low - mass stellar progenitors . however , unlike the case for the giant sample , the top - heavy imf model does not reproduce the observed trend of the cemp frequencies for the turnoff stars at all . the combination of these results from the giant and turnoff samples suggests that the current agb binary - synthesis model may not be suitable for creating carbon - enhanced envelopes for intermediate- to high - mass stars ( 38 @xmath0 ) . as the agb binary - synthesis model ( using a salpeter imf or a top - heavy imf ) predict far too low frequencies of cemp stars for our turnoff sample , there likely exists one or more additional mechanisms capable of producing carbon - rich stars below [ fe / h ] = @xmath88 , the metallicity regime where the cemp - no stars dominate over the subclass of cemp-@xmath15 stars . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . y.s.l . is a tombaugh fellow . t.s . was supported by the jsps grants - in - aid for scientific research ( 23224004 ) . this work was supported in part by grant phy 08 - 22648 : physics frontiers center / joint institute for nuclear astrophysics ( jina ) , awarded by the u.s . national science foundation . abazajian , k. , adelman - mccarthy , j. k. , ageros , m. a. , et al . 2003 , , 126 , 2081 abazajian , k. , adelman - mccarthy , j. k. , ageros , m. a. , et al . 2004 , , 128 , 502 abazajian , k. , adelman - mccarthy , j. k. , ageros , m. a. , et al . 2005 , , 129 , 1755 abazajian , k. , adelman - mccarthy , j. k. , ageros , m. a. , et al . 2009 , , 182 , 543 abia , c. , busso , m. , gallino , r. , et al . 2001 , , 559 , 1117 adelman - mccarthy , j. k. , ageros , m. a. , allam , s. s. , et al . 2006 , , 162 , 38 adelman - mccarthy , j. k. , ageros , m. a. , allam , s. s. , et al . 2007 , , 172 , 634 adelman - mccarthy , j. k. , ageros , m. a. , allam , s. s. , et al . 2008 , , 175 , 297 ahn , c. p. , alexandroff , r. , allende prieto , c. , et al . 2012 , , 203 , 21 aihara , h. , allende prieto , c. , an , d. , et al . 2011 , , 193 , 29 allende prieto , c. , sivarani , t. , beers , t. c. , et al . 2008 , , 136 , 2070 aoki , w. , beers , t. c. , christlieb , n. , et al . 2007 , , 655 , 492 aoki , w. , beers , t. c. , lee , y. s. , et al . 2013 , , 145 , 13 aoki , w. , beers , t. c. , sivarani , t. , et al . 2008 , , 678 , 1351 aoki , w. , norris , j. e. , ryan , s. g. , beers , t. c. , & ando , h. 2002 , , 580 , 1149 barbuy , b. , spite , m , spite , f. , et al . 2005 , , 429 , 1031 beers , t. c. , carollo , d. , ivezi , . , 2012 , , 746 , 34 beers , t. c. , chiba , m. , yoshii , y. , et al . 2000 , , 119 , 2866 beers , t. c. , & christlieb , n. 2005 , , 43 , 531 beers , t. c. , preston , g. w. , & shectman , s. a. 1985 , , 90 , 2089 beers , t. c. , preston , g. w. , & shectman , s. a. 1992 , , 103 , 1987 bisterzo , s. , gallino , r. , straniero , o. , cristallo , s. , & kppeler , f. 2011 , , 418 , 284 bisterzo , s. , gallino , r. , straniero , o. , cristallo , s. , & kppeler , f. 2012 , , 422 , 849 bonifacio , p. , spite , m. , cayrel , r. , et al . 2009 , , 501 , 519 caffau , e. , bonifacio , p. , franois , p. , et al . 2011 , , 477 , 67 carollo , d. , beers , t. c. , bovy , j. , et al . 2012 , , 744 , 195 chiappini , c. 2013 , an , 334 , 5951 christlieb , n. 2003 , rvma , 16 , 191 christlieb , n. , bessell , m. s. , beers , t. c. , et al . 2002 , , 419 , 904 christlieb , n. , green , p. j. , wisotzki , l. , & reimers , d. 2001 , , 375 , 66 christlieb , n. , schrck , t. , frebel , a. , et al . 2008 , , 484 , 721 cohen , j. mcwilliam , a. , shectman , s. , et al . 2006 , , 132 , 137 frebel , a. , christlieb , n. , norris , j. e. , et al . 2006 , , 652 , 1585 fujimoto , m. y. , iben , i. jr . , & hollowell , d. 1990 , , 349 , 580 fujimoto , m. y. , ikeda , y. & iben , i. jr . 2000 , , 529 , l25 fukugita , m. , ichikawa , t. , gunn , j. e. , et al . 1996 , , 111 , 1748 gunn , j. e. , carr , m. , rockosi , c. , et al . 1998 , , 116 , 3040 gunn , j. e. , siegmund , w. a. , mannery , e. j. , et al . 2006 , , 131 , 2332 hansen , t. , andersen , j. , & nordtrom , b. 2013 , xii international symposium on nuclei in the cosmos , arxiv:1301.7208 herwig , f. 2005 , , 43 , 435 ito , h. , aoki , w. , beers , t. c. , et al . 2013 , , 773 , 33 ito , h. , aoki , w. , honda , s. , & beers , t. c. 2009 , , 698 , l37 izzard , r. g. , glebbeek , e. , stancliffe , r. j. , & pols , o. r. 2009 , , 508 , 1359 johnson , j. a. , herwig , f. , beers , t. c. , & christlieb , n. 2007 , , 658 , 1203 kobayashi , c. , tominaga , n. , & nomoto , k. 2011 , , 730 , l14 komiya , y. , suda , t. , minaguchi , h. , et al . 2007 , , 658 , 367 korn , a. , grundahl , f. , richard , o. , et al . 2007 , , 671 , 402 lau , h. h. b. , stancliffe , r. j. , & tout , c. a. 2009 , , 396 , 1046 lee , y. s. , beers , t. c. , allende prieto , c. , et al . 2011 , , 141 , 90 lee , y. s. , beers , t. c. , masseron , t. , et al . 2013 , , in press lee , y. s. , beers , t. c. , sivarani , t. , et al . 2008a , , 136 , 2022 lee , y. s. , beers , t. c. , sivarani , t. , et al . 2008b , , 136 , 2050 lind , k. , korn , a. j. , barklem , p. s. , & grundahl , f. 2008 , , 490 , 777 lucatello , s. , beers , t. c. , christlieb , n. c. , et al . 2006 , , 652 , l37 lucatello , s. , gratton , r. g. , beers , t. c. , & carretta , e. 2005b , , 625 , 833 lucatello , s. , tsangarides , s. , beers , t. c. , et al . 2005a , , 625 , 825 luck , r. e. , & bond , h. e. 1991 , , 77 , 515 marsteller , b. , beers , t. c. , rossi , s. , et al . 2005 , nucl . a. , 758 , 312 masseron , t. , johnson , j. a. , plez , b. , et al . 2010 , , 509 , 93 meynet , g. , ekstrm , s. , & maeder , a .. 2006 , , 447 , 623 meynet , g. , hirschi , r. , ekstrom , s. , et al . 2010 , , 521 , 30 nomoto , k. , kobayashi , c. , & tominaga , n. 2013 , , 51 , 457 norris , j.e . , ryan , s.g . , & beers , t.c . 1997 , , 488 , 350 norris , j. e. , christlieb , n. , korn , a. j. , et al . 2007 , , 670 , 774 norris , j. e. , yong , d. , bessell , m. s. , et al . 2013 , , 762 , 28 pier , j.r . , munn , j. a. , hindsley , r. b. , et al . 2003 , , 125 , 1559 pols , o. r. , izzard , r. g. , stancliffe , r. j. , & glebbeek , e. 2012 , , 547 , 76 richard , o. , michaud , g. , & richer , j. 2002a , , 580 , 1100 richard , o. , michaud , g. , richer , j. , et al . 2002b , , 568 , 979 roederer , i. u. 2009 , , 137 , 272 rossi , s. , beers , t.c . , & sneden , c. 1999 , in the third stromlo symposium , asp conference series , eds . b.k . gibson , t.s . axelrod , and m.e . putman , 165 , p. 264 rossi , s. , beers , t. c. , sneden , c. , et al . 2005 , , 130 , 2804 smolinski , j. p. , lee , y. s. , beers , t. c. , et al . 2011 , , 141 , 89 sneden , c. , cowan , j. j. , & gallino , r. 2008 , , 46 , 241 spite , m. , caffau , e. , bonifacio , p. , et al . 2013 , , 552 , 107 spite , m. , cayrel , r. , hill , v. , et al . 2006 , , 455 , 291 spite , m. , cayrel , r. , plez , b. , et al . 2005 , , 430 , 655 stoughton , c. , lupton , r. h. , bernardi , m. , et al . 2002 , , 123 , 485 suda , t. , aikawa , m. , machida , m. n. , & fujimoto , m. y. 2004 , , 611 , 476 suda , t. , & fujimoto , m. y. 2010 , , 405 , 177 suda , t. , katsuta , y. , yamada , s. , et al . 2008 , pasj , 60 , 1159 suda , t. , komiya , y. , yamada , s. , et al . 2013 , , 432 , l46 suda , t. , yamada , s. , katsuta , y. , et al . 2011 , , 412 , 843 tominaga , n. , iwamoto , n. , & nomoto , k. 2013 , , submitted , arxiv:1309.6734 tominaga , n. , umeda , h. , & nomoto , k. 2007 , , 660 , 516 tumlinson , j. 2007 , , 664 , l63 umeda , h. , & nomoto , k. 2003 , , 422 , 871 umeda , h. , & nomoto , k. 2005 , , 619 , 427 wisotzki , l. , koehler , t. , groote , d. , & reimers , d. 1996 , a@xmath108as , 115 , 227 wood p. r. , 2011 , in qain s. , leung k. , zhu l. , kwok s. , eds , asp conf . 451 , proc . 9th pacific rim conference on stellar astrophysics astron . , san francisco , p. 87 yamada , s. , suda , t. , komiya , y. , aoki , w. , & fujimoto , m. y. 2013 , , in press , arxiv:1309.3430 yanny , b. , newberg , h. j. , johnson , j. a. , et al . 2009 , , 137 , 4377 yong , d. , norris , j. e. , bessell , m. s. , et al . 2013 , , 762 , 27 york , d. g. , adelman , j. , anderson , j. e. , jr . , 2000 , , 120 , 1579

we present results from a very deep ( 650 ks ) _ chandra _ x - ray observation of the galaxy group ngc 5813 , the deepest _ chandra _ observation of a galaxy group to date . earlier observations showed two pairs of cavities distributed roughly collinearly , with each pair associated with an elliptical shock front . the new observations confirm a third pair of outer cavities , collinear with the other pairs , and reveal an associated outer outburst shock , with measured temperature jump , at kpc . this system is therefore unique in exhibiting three cavity pairs , each associated with an unambiguous shock front , arising from three distinct outbursts of the central agn . as such , it is particularly well - suited to the study of ongoing agn feedback . the implied mean kinetic power is roughly the same for each outburst , demonstrating that the average agn kinetic luminosity can remain stable over long timescales ( myr ) . the two older outbursts have larger , roughly equal total energies as compared with the youngest outburst , implying that the youngest outburst is ongoing . we find that the radiative cooling rate and the mean shock heating rate of the gas are well balanced at each shock front , suggesting that agn outburst shock heating alone is sufficient to offset cooling and establish agn / icm feedback within at least the central 30 kpc . this heating takes place roughly isotropically and most strongly at small radii , as is required for feedback to operate . we suggest that shock heating may play a significant role in agn feedback at smaller radii in other systems , where weak shocks are more difficult to detect . we find non - zero shock front widths that are too large to be explained by particle diffusion . instead , all measured widths are consistent with shock broadening due to propagation through a turbulent icm with a mean turbulent speed of . finally , we place lower limits on the temperature of any volume - filling thermal gas within the cavities that would balance the internal cavity pressure with the external icm . the most stringent limit we find is kev .

1503.08205

early _ chandra _ and _ xmm - newton _ x - ray observations revealed that the amount of gas cooling to very low temperatures at the centers of cool core clusters is much less than what is expected from simple radiative cooling models ( david et al . 2001 ; peterson et al . 2001 ; peterson & fabian 2006 ) . the implication is that the diffuse x - ray emitting gas must be heated , either by pre - heating during cluster formation or by ongoing energy injection . the most likely heating mechanism is feedback due to energy injection by the central active galactic nucleus ( agn ) of the cd galaxy ( see mcnamara & nulsen 2007 and fabian 2012 for recent reviews ) . during this process , matter is accreted onto the central supermassive black hole ( smbh ) , which drives powerful jets . these jets evacuate cavities in the intracluster medium ( icm ) , which can drive shocks as they are inflated and subsequently rise buoyantly ( churazov et al . the energy contained in cavities and shocks is then available to heat the icm , which lowers the cooling rate of the gas and subsequently the smbh accretion rate . the ensuing decrease in agn heating allows the gas to once again cool and accrete onto the smbh , establishing a feedback loop that regulates the temperature of the icm . several studies have shown that , generally , the total enthalpy in cavities in cool core systems is sufficient to offset radiative cooling in individual galaxies , galaxy groups , and clusters ( brzan et al . 2004 ; rafferty et al . 2006 ; nulsen et al . 2007 ; hlavacek - larrondo et al . 2012 ) . however , the details of how and where this energy gets transferred to the icm are unclear . weak agn outburst shocks are also expected to heat the icm , although they are difficult to detect and clear examples are very rare . galaxy groups provide an excellent opportunity to study agn feedback . their lower temperatures are more easily measured with modern high angular resolution x - ray satellites , and their central agn can more easily disturb the diffuse gas due to their shallower gravitational potentials as compared with clusters . here we report on results from a very deep _ chandra _ observation of the central galaxy in the galaxy group ngc 5813 ( n5813 ) . this group is a relatively isolated sub - group in the ngc 5846 group ( mahdavi et al . 2005 ; machacek et al . 2011 ) . in randall et al . ( 2011 , hereafter r11 ) we presented results based on an initial 150 ks _ chandra _ observation of n5813 . the icm in this group has a remarkably regular morphology , with three pairs of roughly collinear cavities and associated surface brightness edges , and shows no clear signs of a recent merger event . ( note that we refer to the diffuse gas in this group as the icm rather than the intragroup medium ( igm ) throughout to stress the connection with feedback in clusters and to avoid confusion with the intergalactic medium . ) with clear , cleanly separated signatures from three distinct outbursts of the central agn and no other significant dynamical processes at work , n5813 is uniquely well - suited to the study of agn feedback . in this work , we focus on the implications for agn feedback and the outburst history of the central smbh . we assume an angular diameter distance to n5813 of 32.2 mpc ( tonry et al . 2001 ) , which gives a scale of 0.15 kpc/. all uncertainty ranges are 68% confidence intervals ( i.e. , 1@xmath5 ) , unless otherwise stated . the _ chandra _ observations that were used in the analysis we present here are summarized in table [ tab : obs ] . the aimpoint was on the back - side illuminated acis - s3 ccd for each observation . all data were reprocessed from the level 1 event files using ciao and caldb 4.4.3 . cti and time - dependent gain corrections were applied . lc_clean was used to remove background flares . the mean event rate was calculated from a source free region using time bins within 3@xmath5 of the overall mean , and bins outside a factor of 1.2 of this mean were discarded . there were no periods of strong background flares . the cleaned exposure times are given in table [ tab : obs ] , for a total time of 635 ks . diffuse emission from n5813 fills the image fov for each observation . we therefore used the caldb blank sky background files appropriate for each observation , normalized to match the 10 - 12 kev count rate in our observations to account for variations in the particle background . to generate exposure maps , we used a mekal model with @xmath6 kev , galactic absorption , and abundance of 30% solar at a redshift @xmath7 . the exposure corrected , background subtracted , 0.33 kev _ chandra _ image of the central region is shown in figure [ fig : core_ximg ] . to enhance the visibility of the diffuse emission bright point sources were removed , and the regions containing point sources were `` filled in '' using a poisson distribution whose mean was equal to that of a local annular background region . to better show faint structure , particularly in the outer regions , we fitted the x - ray image with a 2d beta - model using the software package sherpa ( freeman et al . 2001 ) and divided the image by the model to produce a residual image . the smoothed , point source free image and the residual image are shown in figure [ fig : ximg ] . these images clearly show three pairs of roughly collinear cavities , along an axis from the sw to the ne . the inner cavity pair is surrounded by bright rims ( at about 1 kpc ) , and each of the intermediate and outer cavity pairs is associated with an elliptical surface brightness edge ( at @xmath210 kpc and @xmath230 kpc , respectively ) . in r11 , we showed that the 1 kpc and 10 kpc edges are shock fronts that were driven during the expansion phase of their associated cavities . the new , deeper observations clearly show a third ( outer ) cavity pair and edge , only hinted at in earlier observations , and allow us to unambiguously identify the outer edge as an associated shock front ( [ sec : shprofs ] ) . these outer features are most clearly seen in the residual image ( figure [ fig : ximg ] ) . thus , these deep x - ray observations show clear signatures from three distinct outbursts of the central agn . we note that , here and throughout , by `` outburst '' we refer to the creation of a cavity pair and its associated shocks , rather than a rapid increase in agn jet power , since these features are in principle consistent with either a constant or variable jet power ( see [ sec : history ] ) . to look for structure in the faint emission at large radii , beyond the fov of _ chandra _ , we examined the available archival _ xmm - newton _ observations of n5813 . n5813 was observed with _ xmm - newton _ on three occasions : july 23 2005 ( obsid 0302460101 ) , february 11 2009 ( obsid 0554680201 ) , and february 17 2009 ( obsid 0554680301 ) . the event files were calibrated using the xmm - newton science analysis system ( sas ) version 13.5.0 , and the most recent calibration files as of 2014 july . the calibrated , cleaned event files were produced after filtering for particle background flares . this resulted in cleaned exposure times of 120 ks , 127 ks , and 83 ks for the mos1 , mos2 , and pn detectors , respectively . details of the data reduction are described in bulbul et al . ( 2012a ) . the background subtracted , exposure corrected , merged _ xmm - newton _ image is shown in figure [ fig : xmm_compare ] beside the _ chandra _ image . the 10 kpc shock edges are visible , as are the intermediate cavities and the ne outer cavity . the inner @xmath8 kpc cavities and rims are not resolved . a smoothed version of this image , with the intensity scale chosen to better show faint emission at large radii , is shown in figure [ fig : xmm_smo ] . the extended emission is roughly azimuthally symmetric , and shows no clear evidence of a fourth pair of cavities or shock edge beyond the 30 kpc edge detected with _ chandra_. since , for the purposes of this study , we are most interested in the detailed structure in the icm , which is best resolved with _ chandra _ , we will not consider the _ xmm - newton _ data further here . unless otherwise specified , all spectra were fitted in the 0.63.0 kev band using xspec , with an absorbed apec model with the absorption fixed at the galactic value of @xmath9 @xmath10 ( kalberla et al . 2005 ) and the abundance allowed to vary . spectra were grouped with a minimum of 40 counts per energy bin . anders & grevesse ( 1989 ) abundance ratios and atomdb 2.0.2 were used throughout . we note that using this version of atomdb gives systematically larger temperatures ( by roughly 0.050.1 kev , significantly larger than our typical statistical error ) as compared with atomdb 1.3.1 , which was used in r11 . this difference arises from significant changes in the ionization balance , particularly at the low ( @xmath11 kev ) temperatures we consider here ( see fig . 3 in foster et al . thus , we expect a systematic offset between the spectroscopic quantities that we derive here as compared with r11 . to study the thermal structure of the icm , we generated a smoothed temperature map using the method described in randall et al . ( 2008 ) , and also used in r11 . for each pixel in the temperature map , spectra were extracted from a surrounding circular region that contained 1500 source counts . the temperature map pixel value was then set equal to the best fitting temperature from fitting to these spectra . since the extraction regions are generally larger than the pixels , nearby pixels are correlated with one another , such that the temperature map is effectively smoothed on the scale of the local extraction region size . some advantages of this method are that there is no a priori assumption about the thermal structure of the gas , and experience has shown that such maps can resolve real structure on scales that are smaller than the size of the extraction regions . for comparison , we also constructed temperature maps using the contour binning method developed by sanders ( 2006 ) . here , extraction regions are defined based on surface brightness contours . regions were defined to achieve a signal - to - noise ratio of 38 ( corresponding to @xmath21,400 counts in bright regions ) in each region . this method assumes that the temperature distribution follows the surface brightness distribution ( which is frequently , but not always , the case ) . some advantages of this method are that each extraction region is independent of the others ( as long as the region widths are large compared with the local psf ) , and it is relatively cheap computationally ( as compared with the smoothed temperature map method ) . a comparison of the smoothed and contour binned temperature maps gives an indication of the extraction region size as a function of position in the smoothed map . smoothed temperature maps over a wider area and a higher resolution map of the core are shown beside the contour binned maps in figures [ fig : tmap ] and [ fig : core_tmap ] , respectively . both the smoothed and contour binned temperature maps reveal relatively hot emission associated with the bright rims surrounding the inner cavities and just inside the prominent 10 kpc surface brightness edges . these features were identified as shocks in r11 . the maps suggest additional temperature rises associated with the outer 30 kpc surface brightness edges , particularly to the nw . the temperature profiles across these edges are presented in [ sec : shprofs ] . the temperature maps also show a plume of cooler gas , roughly extending along the line defined by the cavities , from sw to ne . in r11 , we showed that this plume is consistent with arising from cool gas that has been lifted by the buoyantly rising x - ray cavities . we derived pseudo - pressure and pseudo - entropy maps from the smoothed temperature maps . the fitted apec normalization for each spectral map pixel is proportional to the volume integral of the square of the electron density along the line of sight . the square root of the normalization therefore gives an average `` projected '' density along the line of sight . we define the pseudo - pressure and pseudo - entropy as @xmath12 and @xmath13 , respectively , where @xmath14 is the fitted projected temperature and @xmath15 is the normalization scaled by the area of the extraction region . the wide field and core maps are shown in figures [ fig : press ] and [ fig : core_press ] . the pseudo - pressure maps reveal prominent pressure jumps across the 10 kpc shock fronts and increased pressure in the bright central rims ( i.e. , the 1 kpc shock ) . there is no obvious equivalent structure in the pseudo - entropy maps . this is not unexpected for weak shocks . for the strongest shock we detect , with @xmath16 , the rankine - hugoniot shock jump conditions for a @xmath17 gas give a pressure jump at the shock front of a factor of 3.7 , but a jump in the entropy index of only a factor of 1.12 . thus , entropy jumps are expected to be relatively difficult to detect , particularly in projection . the core pseudo - entropy map ( figure [ fig : core_press ] ) shows some filamentary structure in the center , on the scale of the bright shocked rims . however , unlike the features in the pseudo - pressure map , these features do not directly trace the bright rims . furthermore , the entropy is lower in these filaments as compared with the surrounding regions , in contrast to the small increase in entropy that is expected at the shock fronts . these structures are likely formed as the central low - entropy gas is pushed and pulled away from the center as the inner cavities expand and rise buoyantly . the central regions of the core temperature and pseudo - entropy maps are shown with the h@xmath18 contours from r11 overlaid in figure [ fig : halpha ] . as shown in r11 , the h@xmath18 emission follows the plume of cool x - ray emitting gas up to the location of the intermediate cavities , indicating that these cavities have uplifted cold phase gas from the core as they buoyantly rise . the h@xmath18 emission also clearly traces the sw inner cavity , suggesting that the central cold - phase gas is dynamically coupled to the cavity and is being pushed out as the cavity expands . the se 1 kpc shock appears to be passing through the h@xmath18 filament ( at least in projection ) without destroying it . finally , there is a correlation between the h@xmath18 emission and the detailed , small - scale , filamentary structure of the central low - entropy gas . this is consistent with the above suggestion that these entropy features likely arise from gas that has been displaced from the core by the cavities , rather than from entropy changes driven by the central shocks . since each extraction region in the above maps contained only 1500 net counts , the abundance values determined by these fits were not tightly constrained , and the corresponding abundance maps show little to no structure . abundance measurements in @xmath20.7 kev gas are challenging due to the large number of closely spaced emission lines , and the resulting degeneracy between line and continuum emission at _ chandra s _ energy resolution . we found that , for n5813 , achieving 1-@xmath5 errors on the order of 510% for the abundance required roughly 30,000 net counts . fortunately , the new deeper observations provided well over one million photons appropriate for use in spectral fitting , allowing us to map the abundance distribution in the gas . a smoothed abundance map , with 30,000 net counts per extraction region , is shown in figure [ fig : abund ] . the abundance map shows an apparent plume of low abundance gas that corresponds to the plume of uplifted cool gas seen in figure [ fig : tmap ] ( roughly 40 kpc long , extending from sw to ne ) . it has long been recognized that fitting multi - temperature spectra with a single thermal model can give anomalously low abundance values ( the so - called `` fe - bias '' effect , buote 1999 ) . to test for this effect , we fit multi - temperature models to the apparent central minimum and apparent low - metallicity ne plume . we find that the plume is well described by a two - temperature model , with @xmath19 kev , @xmath20 kev , and an abundance ( set equal for each temperature component ) of @xmath21% solar , consistent with the surrounding icm . in contrast , we find that , while including multiple temperature components for the central region ( in a @xmath22 kpc ellipse oriented ne to sw , along the plume ) does raise the fitted abundance , it is still significantly lower than the abundance of the surrounding icm , at @xmath23% solar . thus , we conclude that the apparent low abundance in the extended plume is a fitting artifact arising from the fe - bias effect , and that the measured abundance in the plume is consistent with that of the surrounding icm , while the apparent abundance minimum within the central few kpc can not be easily explained by this effect . we note that central abundance dips that apparently can not be explained solely by projection effects have been seen in other systems ( e.g. , blanton et al . 2003 ; panagoulia et al . in addition to the apparent low abundance plume , there appear to be increases in the projected abundance at the location of the shock fronts , in contradiction with what is expected from the fe - bias effect . a more detailed study of these features and the icm abundance in general in n5813 will be presented in an upcoming paper . we produced projected radial profiles by fitting spectra extracted from concentric annuli , centered on the centroid of the diffuse emission at large radii . each annulus was chosen to be at least 1wide and contain at least 1000 net counts in the 0.63.0 kev band . the inner annuli were limited by the size constraint ( and therefore contained more than 1000 net counts ) , whereas the outer annuli were wider and contained roughly 1000 net counts . the resulting temperature profile is shown in the top panel of figure [ fig : azprof ] . the positions of the shock fronts ( determined in [ sec : shock_structure ] ) are marked by vertical lines . even in the projected temperature profile in circular annuli ( which effectively smooths out the elliptical shock fronts ) , temperature enhancements associated with each shock are clearly visible . the central 1.5 radius region , which contains the central agn , has been excluded from the fit to the central region . the central temperature rise is due to the presence of the inner shock rims surrounding the inner cavities . as in r11 , we determined the 3d structure of the icm by performing an `` onion peeling '' deprojection analysis ( fabian et al . 1981 ) using concentric annuli , each of which contained at least 8000 net counts ( in projection ) in the 0.63.0 kev band . we fit each annulus with an absorbed apec model , with the abundance fixed at 50% solar ( consistent with results from spectral fits to the total diffuse emission and with typical values in the projected abundance map shown in figure [ fig : abund ] ) . we assume spherical shells of uniform density . given the rich morphology of n5813 , with elliptical shock edges , cavities , and other features , it is clear that the assumption of spherical symmetry does not strictly apply . thus , the deprojected profiles include some level of systematic uncertainty associated with the assumed geometry . the deprojected temperature , density , pressure , and entropy profiles are shown in figure [ fig : azprof ] . we see that the deprojected temperature is generally consistent with the projected temperature . we interpret this an indicating that the scale of the azimuthally averaged temperature gradient is generally small compared with the sizes of our extracted annuli , so that projection effects in any given annulus ( which are dominated by emission from adjacent annuli ) are relatively small . the small discontinuous jumps , or `` kinks '' , in the density , pressure , and entropy profiles around 11 kpc are associated with the sharp surface brightness edges to the nw and se from the intermediate shock . to confirm the temperature rises across the surface brightness edges seen in figures [ fig : tmap ] and [ fig : core_tmap ] , we derived temperature profiles in annular bin sectors across each of the edges . the profiles were centered on the center of curvature for each feature ( such that there was a bin boundary that traced the surface brightness edge ) , then the distances were corrected to measure the distance from the center adopted for the azimuthal profiles presented in [ sec : azprofs ] . the temperature profiles across the 10 kpc and 30 kpc shock fronts are shown in figures [ fig : midsh_ktprofs ] and [ fig : outsh_ktprofs ] , respectively , with the fitted shock front positions overlaid ( see [ sec : shock_structure ] ) . clear temperature rises are seen at each front , confirming that these features are shocks . in the case of the 10 kpc shocks , the deeper observations allow us to measure the temperature profiles with greater angular resolution ( on the scale of a few hundred pc ) as compared with r11 . in addition to the overall shift in normalization due to the updated atomdb ( see [ sec : spec ] ) , the peak change in temperature across the fronts is larger as compared with the more coarsely binned profiles in r11 . in contrast , the shock mach numbers that we derive here ( [ sec : shock_structure ] ) are consistent with those derived in r11 , since the surface brightness profile requires fewer counts per radial bin , allowing for smaller bins . this is consistent with our suggestion in r11 that the temperature jumps we found there were smaller than what was expected based on the mach numbers derived from the surface brightness jumps due to projection effects . in some cases , there appears to be shock heated gas just _ outside _ the positions of the shock front ( e.g. , the two bins to the right of the edge in figure [ fig : midsh_ktprofs ] , left ) . there are at least two factors that can contribute to this effect . first , there is a small uncertainty in the fitted locations of the shock fronts . second , we show in [ sec : shock_structure ] that the shock fronts have finite widths ( on the order of 0.5 kpc for the 10 kpc shock ) . when fitted with a discontinuous model , the preferred position of the edge will naturally be at the center of the smoothed front . to quantitatively characterize the structure of the shock fronts , we applied the edge fitting method described in r11 . we extracted the 0.3 - 3.0 kev surface brightness profiles across each of the three shocks ( at roughly 1 kpc , 10 kpc , and 30 kpc ) to the se and the nw , where the edges are sharpest . we excluded the nw side of the 1 kpc shock as this region shows a complicated morphology , with the shocks driven by the ne and sw inner cavities just beginning to overlap ( figure [ fig : core_ximg ] ) , and is thus not well described by our simple spherically symmetric model . although the 1 kpc se shock edge appears relatively regular , we can not rule out the possibility of some systematic bias in our results due to overlapping cavity rims . the level of this potential bias is difficult to characterize without detailed numerical simulations , although we do not expect it to strongly affect our results . each surface brightness profile was centered on the center of curvature for the corresponding edge , which was not coincident with the location of the agn or with the adopted center for the azimuthal profiles derived in [ sec : azprofs ] . these surface brightness profiles were converted into integrated emissivity profiles ( iem ) using the best fitting projected temperature in each bin ( determined from fitting a single absorbed apec model ) with the abundance fixed at 50% solar . the iem profiles were then fit by projecting a spherically symmetric discontinuous power - law density model . the free parameters of this model are the normalization , the inner density slope , the outer density slope , and the amplitude and location of the density jump . finally , the density jumps were converted into shock mach numbers using the standard rankine - hugoniot shock jump conditions for a @xmath17 gas . the fit to the nw 10 kpc shock is shown in figure [ fig : nwedge ] . the results of all of the fits are summarized in table [ tab : shocks ] ( note that all radii are scaled to give the distance from the center adopted for the azimuthal profiles presented in [ sec : azprofs ] , rather than the distance from the center of curvature of each front ) . we find mach numbers of roughly 1.8 , 1.5 , and 1.2 for the 1 kpc , 10 kpc , and 30 kpc shocks , respectively . these results confirm the presence of the outer shock at @xmath0 kpc , only hinted at in the shallower observations analyzed in r11 . this shock is associated with the same agn outburst that inflated the outermost cavity pair @xmath24 myr ago . the iem profiles of both the nw and se 10 kpc shock fronts show edges that are less sharp than the discontinuous power - law model . we therefore fit each edge with the same model but smoothed with a gaussian , with the width of the gaussian ( @xmath5 ) as an additional free parameter of the fit . the fit for the nw 10 kpc shock front is shown along with the discontinuous model fit in figure [ fig : nwedge ] . the smoothed model is a significantly better fit to the data for both the se and nw 10 kpc shock fronts , with both probabilities @xmath25% based on the f test . for the se 1 kpc shock , the smoothed model is a marginally better fit , with a probability of @xmath26% , while for the 30 kpc shock both models describe the data equally as well . the 1 kpc se and both 10 kpc shock fronts are shown to have finite widths at greater than 90% confidence . only upper limits can be placed on the widths of the 30 kpc shock fronts , where the surface brightness is relatively low . the shock front edges are expected to be blurred by _ chandra s _ psf , which varies across the field . table [ tab : shocks ] gives the size of the 50% encircled energy fraction ( eef ) region at the location of each shock , assuming a 0.74 kev plasma with an abundance of 50% solar and galactic absorption , and the fitted edge widths . the 90% eef is about twice as wide as the 50% eef for this model . for the se 1 kpc shock front , the edge width is on the order of the size of the psf . thus , the measured width of this edge may not be intrinsic to the front . however , for the nw and se 10 kpc shock fronts , the 90% confidence lower limit on @xmath5 is larger than the local psf . the difference is particularly significant for the nw 10 kpc shock , where the psf is smaller ( the optical axis is located nw of the central core ) . furthermore , we note that the best fit shock width is somewhat larger ( consistent within the errors ) for the nw 10 kpc shock front , as compared with the se 10 kpc front , despite the fact that the psf is more than twice as large in this region . this shows that the measured widths are not driven by psf blurring . we conclude that , in the case of the 10 kpc shock , the edge widths given in table [ tab : shocks ] are likely intrinsic to the edges themselves and not instrumental artifacts . in the case of the 30 kpc shock , only upper limits on the shock widths can be placed . the intrinsic widths may simply be due to irregularities in the shape of the shock front , due to variations in the shock strength with azimuthal angle created as the shock propagates through a non - uniform and/or turbulent icm . to test this effect , we fit the surface brightness profiles , as above , in sectors of varying angular width for the se and nw 10 kpc shock fronts , excluding very wide sectors that include the x - ray cavities . we found consistent density jumps for every case that we tried , although for very narrow sectors the uncertainties were large . thus , while we can not rule out the possibility that the finite shock front widths that we find are due to deformations in the shock fronts ( possibly at smaller angular scales than we are able to measure ) , we do find that determining the shock widths in more narrow sectors does not result in significantly smaller widths . possible implications of finite shock widths and the impact of a turbulent icm are discussed further in [ sec : transport ] . shocks are expected to heat the icm as they propagate . while there is a relatively large temperature jump at the shock front , this increase is transient as the gas subsequently expands and cools . the relevant quantity for lasting heating of the gas is the change in entropy . the equivalent amount of heat energy @xmath27 imparted to the gas by a shock due to a change in entropy @xmath28 is given by @xmath29 where @xmath30 is the total thermal energy of the gas , @xmath31 is the adiabatic index ( taken to be 5/3 ) , and @xmath32 and @xmath33 are the pressure and density , respectively . thus , each shock contributes a fraction @xmath34 of the total thermal energy in the gas . in the case of weak shocks , this fraction is small ( between 0.412% for the shocks we detect in n5813 ) . to compare with radiative cooling , we are interested in the cumulative effect of shock heating per local cooling time in the gas ( i.e. , the number of shocks per local cooling time and the total amount of heat energy deposited in the gas ) . we approximate the local cooling time just outside each shock edge as the time it would take the gas to radiate away all of its thermal energy based on its deprojected emissivity . based on the shock ages given in table [ tab : shocks ] , and keeping in mind that these ages are expected to be underestimates for young shocks and overestimates for older shocks , we assume an outburst repetition rate of one every 20 myr . the results from the above comparison of the shock heating and radiative cooling rates are summarized in table [ tab : heating ] . given the various statistical and systematic uncertainties associated with , e.g. , the shock mach numbers , the shock ages and outburst repetition rate , the deprojected cooling luminosity , etc . , these numbers should be considered to be rough estimates . nevertheless , there is good agreement between the number of shocks expected per cooling time ( column 3 ) and the number required to completely offset radiative cooling ( column 4 ) . we conclude that shock heating alone is sufficient to balance radiative cooling in the gas out to at least @xmath230 kpc . this extends our earlier result in r11 , where we found a balance between heating and cooling out to @xmath210 kpc . at 30 kpc , the cooling time ( @xmath35 gyr ) is approaching the expected age of the system , indicating that there is no need for additional heating beyond this radius to completely balance cooling and explain the lack of a strong cooling flow in this system . we can estimate the average cavity power @xmath36 by dividing the total enthalpy of the cavities ( estimated as @xmath37 for each cavity , for a total of @xmath38 @xmath39 ) by the age of the oldest cavity pair ( @xmath40 yr ; see r11 , table 3 ) . this gives @xmath41 @xmath39 . spectral fitting gives a total x - ray luminosity of @xmath42 @xmath39 within a radius enclosing all of the cavities ( @xmath43 kpc ) . thus , in n5813 we find that a few times the @xmath44 work required to inflate the cavities is in principle large enough to offset radiative cooling , as has generally been found in other systems ( e.g. , brzan et al . 2004 ; rafferty et al . 2006 ; nulsen et al . 2007 ; hlavacek - larrondo et al . 2012 ) . since we have shown that , in n5813 , shock heating alone is sufficient to balance cooling , one might ask whether the total heating from cavities and shocks leads to an `` overheating '' of the icm , breaking the balance between heating and cooling that is required for agn feedback to operate over long time scales . although cavities are expected to deposit some of their enthalpy in the icm as they rise buoyantly ( mcnamara & nulsen 2007 ) , this heating is not directly observed , and the details on how and where cavities heat the icm are poorly understood . furthermore , cavities can only directly heat the icm locally , as they rise radially , whereas roughly isotropic heating must take place close to the central agn to regulate feedback . mixing may help to distribute locally heated gas ( e.g. , zhuravleva et al . 2014 ) , although the gas motions must be small enough to not completely destroy the observed x - ray cavities . shocks , in contrast , heat the gas roughly isotropically and most strongly at smaller radii , and the thermal effect of shocks on the icm is , in some cases , directly observed . we suggest that agn outburst shocks may generally play a significant role in agn feedback , particularly at small radii where the mach numbers are higher . the cavities are then free to rise buoyantly , and release their energy to heat the icm at larger radii . the fact that cavities are observed at larger radii demonstrates that they do not necessarily release their enthalpy to the icm at small radii , consistent with this picture . weak shocks are generally difficult to detect , since the shock fronts are thin and easily masked by projection effects . they are more easily detected in n5813 due to its proximity , low gas temperature , and regular morphology . the fact that one of the handful of examples of confirmed outburst shocks known in clusters , in abell 2052 ( blanton et al . 2011 ) , was confirmed by a measured temperature jump only after very deep _ chandra _ observations and careful deprojection is consistent with this scenario . similarly , no temperature jump is found associated with the presumed outburst shock in the perseus cluster using extremely deep _ chandra _ observations , even though it is the x - ray brightest cluster ( graham et al . 2008 ) . measurements of the cavities and shocks allow us to estimate the total mechanical outburst energy for each of the three outbursts in n5813 . since the mechanical luminosity dominates the total agn energy output in n5813 ( r11 ) , as is generally the case for kinetic mode agn feedback ( fabian 2012 ) , this gives us information on the total outburst energy history of the agn . we take the cavity internal energy to be @xmath45 , where the pressure @xmath32 is taken from the azimuthally averaged pressure profile . the assumed cavity dimensions and derived internal energies are given in table [ tab : cavities ] . we use the cavity internal energy rather than the enthalpy ( which is the sum of the internal energy and the work required to inflate the cavity , and roughly equal to @xmath37 for cavities filled with relativistic particles ) since the work done during cavity expansion goes into driving shocks , the energy of which we account for separately . as in r11 , we estimate the outburst energy in shocks as @xmath46 where @xmath47 and @xmath48 are the pre- and post - shock pressures , respectively , and @xmath49 is the total volume enclosed by each roughly ellipsoidal shock surface . although equation [ eq : eshock ] is expected to slightly underestimate the shock energy as the mach numbers were larger at earlier times ( giving larger pressure jumps ) , in r11 we showed that this estimate agrees well with shock energies derived from simple 1d hydrodynamic simulations of a central point explosion by matching the simulated surface brightness profiles with the observations . the total mechanical energy output of each outburst is simply the sum of the total cavity and shock energies . we note that the ratio of the energy in cavities to that in shocks is between 0.150.3 for each outburst , and that in the case of the central outburst , where the cavities occupy a significant fraction of the volume enclosed by the 1 kpc shock surface , this ratio is somewhat lower than the expected value of @xmath50 . this is likely a result of underestimating the central gas pressure , and hence the central cavity energies , due to the complicated morphology in this region , which leads to systematic uncertainties associated with projection effects and a lack of knowledge on the volume filling factor of the x - ray emitting gas . however , we note that such an underestimate would only moderately affect our total outburst energy , well within the factor of a few at which we expect our simple estimates to be accurate , and hence would not affect our main conclusions . using the above estimates , we find total outburst energies of @xmath51 erg , @xmath52 erg , and @xmath53 erg for the outbursts associated with the 1 kpc , 10 kpc , and 30 kpc shocks , respectively . as in r11 , we find the outburst energy of the 1 kpc outburst to be more than a factor of ten less than that of the 10 kpc outburst . based on this , we concluded that either the 1 kpc outburst is ongoing , with the agn actively inflating the cavities , or that the mean jet power varies significantly over timescales on the order of 10 myr . here , we find remarkable agreement between the total outburst energies of the 10 kpc and 30 kpc outbursts . this suggests that the mean outburst ( or jet ) power , averaged over the outburst interval of a few 10 myr , is fairly stable , and that the total energy of the 1 kpc outburst is lower because this outburst is ongoing . in r11 , we estimated the outburst power as the total energy divided by the age of the outburst . here , we instead estimate this power as the total energy divided by the duration of each outburst , which we take to be the shock age for the youngest outburst ( table [ tab : shocks ] ) and the outburst interval ( @xmath54 myr , [ sec : feedback ] ) for the remaining outbursts . this gives mean outburst powers of @xmath55 @xmath39 , @xmath56 @xmath39 , and @xmath57 @xmath39 for the 1 kpc , 10 kpc , and 30 kpc outbursts , respectively . thus , we find the mean outburst power is indeed roughly constant , certainly within a factor of a few , which is the level at which we expect our rough estimates to be accurate . we note that in r11 we found a lower power for the 1 kpc outburst . this difference arises mainly from using results from our numerical model in r11 , which predicts a lower shock energy and larger shock age as compared with the estimates based on observations alone ( see table 2 in r11 ) . the source of this difference is likely to be a systematic error that arises when applying the spherically symmetric model to the 1 kpc shock , which clearly has a more complicated morphology ( see figures [ fig : core_ximg ] & [ fig : core_tmap ] ) . we stress that our result on the consistency of the outburst power only applies to the mean power averaged over the outburst interval of a few 10 myr . agn luminosities are known to vary by up to several orders of magnitude over very short ( observable ) timescales ( e.g. , harris et al.2009 ) . however , in principle the outburst signatures in n5813 could be created with a constant jet power . in this case , the cavities expand rapidly and drive shocks just after they are formed . the expansion rate drops as the _ fractional _ energy input rate decreases , and the shocks separate from the cavities and propagate outwards . as the cavities rise buoyantly , they eventually disconnect from the central jets ( once the buoyant speed exceeds the expansion rate ) , which then begin inflating a new pair of cavities and the process repeats ( similar models are discussed , e.g. , in fabian et al.2003 ) . therefore , we use the term `` outburst '' to refer to the creation of a cavity pair and its associated shock , rather than a rapid increase in jet power , since these features are consistent with a kinetic jet power that either also cycles in outbursts or is roughly constant . we conclude that the most likely scenario is that n5813 is in a `` steady state '' kinetic feedback mode , with the agn outbursts roughly equally spaced in time , and each outburst depositing a similar amount of total energy into the icm , such that the mean agn outburst power is roughly constant . the lower total power of the 1 kpc outburst is an indication that this outburst is ongoing , and has yet to deposit the bulk of its energy in the icm . the detection of 1.4 ghz radio emission filling the inner cavities ( from young , high - energy , non - thermal particles injected by the agn ) , along with shock heated cavity rims that have not yet separated from the cavities ( as they have clearly done for the intermediate and outer cavities ) , are consistent with this interpretation ( r11 ) . this is also consistent with results from allen et al . ( 2006 ) , who find evidence for accretion flows around central agn that are stable over a few million years . the fact that n5813 shows little evidence of a recent merger or other complicated icm `` weather '' , probably contributes to it being able to maintain such steady state feedback over long timescales . fits to the surface brightness profiles across the shock fronts reveal that the se and nw 10 kpc fronts have non - zero widths that can not be explained by _ chandra s _ psf ( [ sec : shock_structure ] ) . measurements of the widths of surface brightness edges in cluster x - ray observations have been used to place constraints on transport processes in the icm by comparing these widths with the local collisional mean free paths of the particles . edges that are significantly narrower than the particle mean free path indicate that coulomb diffusion is suppressed across the edge . in the case of cold fronts in clusters , the front widths are found to be significantly smaller than the particle mean free paths ( e.g. , vikhlinin et al . 2001a ; russell et al . 2012 ) , implying that coulomb diffusion is suppressed across the fronts . however , in cold fronts the suppression of transport processes ( and kelvin helmholtz instabilities ) is likely due to magnetic draping , where magnetic field lines are stretched along the cold front edge as the relatively cool , dense gas moves through the icm ( vikhlinin et al . 2001b ; asai et al . 2004 , 2005 , 2007 ; lyutikov 2006 ; dursi & pfrommer 2008 , zuhone et al.2011 ) . in this case we consider shock fronts which propagate through the icm , and therefore magnetic draping is not expected to occur . in the case of merger shocks , markevitch & vikhlinin ( 2007 ) find a bow shock width of roughly 35 kpc , on the order of the local particle mean free path , for 1e 0657 - 56 ( the bullet cluster ) , although this width is only marginally preferred over a zero width front . russell et al . ( 2012 ) examine shock widths in the merging cluster abell 2146 , which contains a leading bow shock and a trailing reverse shock . they find finite width shocks in each case , with the reverse shock width being significantly smaller , and the bow shock width being marginally smaller , than the particle mean free path . they conclude that transport processes are suppressed across these merger shocks . in the case of agn outburst shocks , croston et al . ( 2009 ) consider the thickness of the northeastern shock in centaurus a , but their results are inconclusive . it is therefore of interest to compare our measured shock widths with the coulomb mean free path of particles in the icm . in the region of the shock fronts there are four relevant mean free paths : that in the pre - shock region , @xmath58 , that in the post - shock region , @xmath59 , that of particles crossing the front from the post - shock to the pre - shock region , @xmath60 , and that of particles crossing the front from the pre - shock region to the post - shock region , @xmath61 . in our case , the largest and most relevant mean free path is @xmath62 , which is given by @xmath63 where @xmath64 is the pre - shock electron density , @xmath65 , @xmath66 is the post - shock gas temperature , @xmath67 is the pre - shock gas temperature , @xmath68/2y^2 $ ] , and @xmath69 is the error function ( spitzer 1962 ) . when calculating @xmath60 , pre - shock temperatures were taken from the projected temperature profiles just outside the shock fronts . post - shock temperatures were calculated from the pre - shock temperatures by applying the rankine - hugoniot jump conditions , rather than taken from the observed projected post - shock temperatures . we note that this is a conservative assumption from the perspective of finding a mean free path that is smaller than the shock width , as it gives a higher post - shock temperature ( that is not diminished by projection effects ) , which gives a larger value for @xmath60 . the relevant particle mean free path is given for each shock front in table [ tab : shocks ] . the uncertainty in the mean free path is dominated by the uncertainty in the pre - shock temperature , which is on the order of 10% , leading to a mean free path uncertainty of roughly 20% . in all cases , the particle mean free path is significantly smaller than the shock width , on the order of ten times smaller than the 90% confidence lower limit for the 1 kpc and 10 kpc shocks ( where lower limits on the shock widths can be placed ) . for a weak shock , the shock width is expected to be roughly @xmath70 , where @xmath71 is the effective particle mean free path ( landau & lifshitz 1987 ; mcnamara & nulsen 2007 ) , or 1.42 times @xmath71 for the weak shocks we consider here . thus , the apparent shock widths are too large to be explained by particle diffusion alone . we note that this result likely precludes using these shock fronts to place accurate constraints on the electron - ion equilibration timescale , as has been done , e.g. , by russell et al . ( 2012 ) . as noted in [ sec : shock_structure ] , a measured finite shock width may arise due to deformations in and broadening of a shock front as it propagates through a `` clumpy '' and/or turbulent icm . nulsen et al . ( 2013 ) provide an estimate of the expected shock width due to turbulence as a function of radius as the shock propagates through a uniformly turbulent icm . the shock width ( @xmath72 , defined as the rms of the displacement of sections of the shock front due to turbulence ) is determined by the distance traveled ( @xmath73 ) , the shock speed @xmath74 , the coherence length of the turbulence ( @xmath75 ) , and the rms turbulent speed ( @xmath76 ) . here , we can invert this relation to find the turbulent speed implied by the observed shock width : @xmath77 the rms turbulent speed implied by measurements of each shock front we consider are given in table [ tab : shocks ] . following nulsen et al . ( 2013 ) , we take the coherence length to be @xmath78 ( rebusco et al . remarkably , the implied turbulent speed is consistent with roughly 70 @xmath3 in each case . we note that , in principle , the turbulent speeds calculated here are upper limits on the true icm turbulence , as this calculation does not include the contribution of density inhomogeneities to shock broadening , projection effects , and the effects of an irregular shock front geometry ( which is approximated as spherically symmetric in sectors ) , all of which will act to increase the apparent shock width . however , as the calculation is only approximate , our results should not be taken as hard upper limits . the implied turbulent velocity of roughly 70 @xmath3 is reasonable compared with results from simulations ( 100300 @xmath3 ; lau et al . 2009 ) and observations ( e.g. , de plaa et al . 2012 ; sanders & fabian 2013 ; zhuravleva et al . while this value is on the low end of the reported range , we note that n5813 is a relatively isolated galaxy group ( r11 ) , which has a regular morphology and shows no signs of a recent major merger . thus , the level of merger - driven turbulence is expected to be low . additionally , we note that observational measurements will naturally be biased towards detecting systems with larger turbulent velocities . in several cases only upper limits can be placed , with all reported upper limits significantly larger ( a few to several 100 @xmath3 ) than the @xmath79 @xmath3 detection we report here ( sanders et al . 2011 ; bulbul et al . 2012b ; sanders & fabian 2013 ) . finally , this value is consistent with the range of turbulent velocities of 43140 @xmath3 recently reported by zhuravleva et al . ( 2014 ) for the ( relatively low mass ) virgo cluster . by considering the effects of resonant scattering on line ratios , de plaa et al . ( 2012 ) constrain the turbulent velocity in n5813 to be @xmath80 , significantly larger than our value of 70 @xmath3 . however , we do not view this as a serious discrepancy for several reasons . first , as mentioned above , the limits provided by equation [ eq : vturb ] are expected to be rough estimates . in particular , the coherence length @xmath75 is unknown and the only hard constraint is that @xmath81 . second , as noted by the authors themselves , the atomic data used to calculate the line ratios in de plaa et al . ( 2012 ) suffer from significant systematic uncertainty . for example , by comparing results derived with the spex code versus atomdb , they find inferred turbulent velocities for n5813 that differ by almost a factor of two . finally , we note that there are plausible physical origins for the higher level of turbulence found by de plaa et al . for example , they place constraints for the central @xmath82 kpc only , where the turbulent velocity may be larger than the total icm average due to the influence of the agn , whereas the estimates given here are averages over the entire region interior to each shock . the trend of the inferred turbulent speed decreasing with radius ( table [ tab : shocks ] ) , although not statistically significant , is consistent with this picture . it is interesting to note that the turbulent velocity inferred from the width of the central @xmath83 kpc shock is consistent with the range given by de plaa et al.(2012 ) , although , as noted above , the errors are large and the effects of the larger relative psf are unclear . additionally , we note that the resonant line ratio may also be affected by bulk flows in the icm ( e.g. , the radial gas flows behind the shock fronts ) , possibly leading to an over - estimate of the turbulence . we conclude that , while the measured shock widths are too large to be explained by coulomb diffusion , they are consistent with arising from the shocks propagating through a turbulent icm . the implied ratio of turbulent to thermal pressure support , estimated as the ratio of the turbulent energy to the thermal energy ( de plaa et al . 2012 ) , is on the order of a few percent ( @xmath84 ) for a turbulent speed of 70 @xmath3 . for each shock we consider , the local thermal diffusion coefficient @xmath85 , where @xmath86 is the sound speed , is significantly smaller ( by up to an order of magnitude ) than the implied turbulent diffusion coefficient @xmath87 . thus , turbulent diffusion dominates the microscopic transport in the icm ( at least in this case , and likely in other similar - size groups , which are expected to have similar bulk icm properties ) . although the x - ray cavities must be close to pressure equilibrium with their surroundings , radio observations show that generally the pressure support from relativistic non - thermal electrons , under the assumption of equipartition , is insufficient to balance the thermal pressure in the icm ( brzan et al . . additional pressure support might be provided by heavy non - thermal particles , or by high temperature thermal gas , with deep x - ray observations providing constraints on the contribution from the latter ( e.g. , schmidt et al . 2002 ; blanton et al . 2003 ; sanders & fabian ( 2007 ) ; rafferty et al . ( 2013 ) ) . here we place similar limits for the low mass galaxy group n5813 . to place lower limits on the temperature of thermal gas in the cavities we use a method similar to that employed in previous studies of other systems ( e.g. , sanders & fabian 2007 ; rafferty et al . 2013 ) . due to the complex thermal structure of the gas , we do not use comparison regions outside of the cavities as in sanders & fabian ( 2007 ) . rather , we place limits based on fits to spectra extracted from regions at the cavity centers , where the contrast is highest and the relative contribution from high temperature gas inside the cavities is expected to be the largest . each spectrum is fit with two absorbed apec model components . the first component models the contribution from the total projected emission , while the second models hotter thermal emission from within the cavities . to derive conservative estimates , the temperature , abundance , and normalization of the first component are allowed to vary . we find that within the cavity regions this single thermal component is sufficient to adequately model the total projected emission ( additional thermal components do not significantly improve the fits and are not well constrained ) . for the second component , the temperature is fixed at some value , the abundance is tied to that of the first component , and the normalization is allowed to vary . for an assumed cavity volume , the upper limit on the normalization gives an upper limit on the density . if we further assume that the cavity is in pressure equilibrium with its surroundings , that the pressure is equal to the average azimuthal pressure at the cavity radius ( figure [ fig : azprof ] ) , and that the thermal gas dominates the pressure support inside the cavities , an upper limit on the volume filling fraction of the gas in the cavities ( for the given temperature ) is obtained . a lower limit on the gas temperature is obtained by increasing the fixed temperature of the hot gas until the upper limit on the volume filling fraction is equal to unity . we concentrate on the cavities with the largest contrast and the most regular morphology . since they are closely spaced ( and therefore in a similar environment ) we fit the inner cavities together , assuming a cylindrical geometry with a radius of @xmath88 ( 0.488 ) kpc and a depth of @xmath89 ( 0.815 ) kpc for the northeastern ( southwestern ) central cavity . for the intermediate cavities , only the southwestern cavity is considered ( with @xmath90 kpc , @xmath91 kpc ) as it has a larger contrast and more regular shape compared with the northeastern intermediate cavity , which may be split into two cavities and may be connected to the northeastern central cavity ( r11 ) . similarly , we choose the northeastern outer cavity over its southwestern counterpart due to its larger contrast and more regular shape , and assume an oblate spheroidal geometry with major axis @xmath92 kpc and minor axis @xmath93 kpc in the plane of the sky . we find limits on the temperature of any volume filling thermal gas in the cavities of @xmath94 kev , @xmath95 kev , and @xmath96 kev for the inner , intermediate , and outer cavities , respectively ( for comparison , the limit placed by rafferty et al.2013 from deep _ chandra _ observations of the galaxy group hcg 62 is 4.3 kev ) . our most stringent constraint of @xmath97 kev for the intermediate cavities implies a gas density in the cavities that is at least 20 times lower than the azimuthal average at that radius ( @xmath98 @xmath99 ) . we note that this method relies on several assumptions . for instance , the hot temperature component is assumed to arise solely from gas contained within the cavities ( it may be the case that hot thermal gas supplies only part of the pressure support , while , e.g. , heavy non - thermal particles or magnetic fields make up the difference ) . additionally , the azimuthally averaged pressure profile , which is used to estimate the mean external pressure at each cavity , is derived under the assumption of spherical symmetry ( [ sec : azprofs ] ) , which is clearly violated at some level for this system ( figure [ fig : ximg ] ) . furthermore , some assumptions about the cavity geometries ( or , more specifically , their volumes ) must be made to calculate densities and pressures . the largest source of error is likely from the latter assumption . we find that an assumed uncertainty in cavity volume of 20% leads to an uncertainty of @xmath100% on the temperature limit ( a similar result was found by rafferty et al . 2013 ) . the residual image shown in figure [ fig : ximg ] reveals a `` channel '' of decreased surface brightness to the north , apparently connected to the ne outer cavity . to test the significance of this feature , we extracted the surface brightness profile in azimuthal bins roughly 9 wide and 64 long across the channel . the profile , shown in figure [ fig : nchan ] , shows a clearly significant dip across the channel , indicating that this is a real feature . since the northern channel appears to connect to the ne outer cavity , one interesting possibility is that this channel is the result of energy ( presumably mostly in the form of non - thermal particles ) leaking through the bright cavity rim and heating the icm . indeed , the temperature maps shown in figure [ fig : tmap ] show relatively high temperatures in the region of the northern channel , although the correlation between the high temperature region in the temperature map and the channel in the residual map is not exact . although cavities are expected to heat the icm ( see [ sec : feedback ] ) , such a direct detection of this heating has not yet been observed . we therefore extracted temperatures both in ( wider ) azimuthal bins across the northern channel , and in identical elliptical regions within and to either side of the channel ( at roughly the same distance from the center ) . in neither case did we find a significant temperature enhancement associated with the channel , with temperatures of all regions agreeing within roughly 1-@xmath5 . we conclude that , although it is possible that the northern channel is a result of energy leaking from the ne cavity and heating the icm , causing the gas in this region to expand to pressure equilibrium ( and subsequently cool adiabatically ) and hence the density ( and surface brightness ) to drop , we find little evidence to support this interpretation . in particular , we can not rule out the possibility that this feature is simply due to large scale structure in the icm , e.g. , associated with previous mergers or bulk motion of the group , even in this relatively relaxed system . another possibility is that this deficit is a left over cavity from an even older , fourth outburst that lies off of the axis defined by the other three cavity pairs , although the fact that it is at least partially within the 30 kpc shock front is difficult to reconcile with this interpretation . deep , low - frequency radio observations would be useful to both detect the putative low - energy non - thermal particles in the ne outer cavity and to see if this emission extends out along the northern channel , consistent with it arising from a leaking cavity . we have presented results from our analysis of a very deep _ chandra _ observation of the galaxy group n5813 . this observation confirms an outer cavity pair and associated shock ( only hinted at in r11 ) , giving a total of three pairs of roughly collinear cavities , with each pair associated with an elliptical shock front with a measured temperature jump . the derived mach numbers are 1.8 , 1.5 , and 1.2 for the 1 kpc , 10 kpc , and 30 kpc shocks , respectively . these features are cleanly separated signatures from three distinct outbursts of the central agn . we compare the shock heating rate to the radiative cooling rate locally at each shock front , and find that the rates closely balance one another . this demonstrates that shock heating alone is sufficient to offset gas cooling and enable agn feedback to operate within at least the central 30 kpc ( compared with 10 kpc in r11 ) . we suggest that shock heating is likely important at small radii in other cool core clusters and groups , but that in most cases the shocks are more difficult to detect due to higher icm temperatures , larger distances , and more complicated icm structure , or `` weather '' , instigated by mergers . we find that the total outburst energies of the old and intermediate age outbursts are roughly equal , while the energy of the young outburst is more than a factor of 10 less . however , the mean powers of all of the outbursts are roughly the same , within a factor of two , indicating that the mean kinetic luminosity of the agn has remained stable for at least 50 myr . we suggest that the young outburst is ongoing , having deposited only a fraction of its total energy into the icm . the proximity of the central cavities to the agn , the relatively high - frequency radio emission that fills the cavities , and the shock heated rims surrounding the cavities ( that have not yet separated from the cavities , as is the case for the older outbursts ) are all consistent with this scenario . we find that the 10 kpc ( and possibly the 1 kpc ) elliptical shock front is broadened by @xmath101 kpc ( @xmath102 kpc ) , more than ten times the particle mean free path and thus too broad to be caused by particle diffusion . while we can not rule out broadening due to a clumpy icm and/or projection effects , using rough estimates we find that the measured shock widths ( which are upper limits for the 30 kpc shock ) are all consistent with broadening due to propagation through a turbulent icm with a turbulent speed of @xmath103 @xmath3 . this is within , but on the low end of , the range of turbulent speeds expected based on simulations and other observations , and thus provides a robust upper limit on the turbulence since other factors may contribute to the total shock broadening . this suggests that transport due to turbulence dominates that of particle diffusion throughout the icm in n5813 . using spectral fits in the cavity regions , we place lower limits on the temperature of any volume filling gas that could completely balance the pressure within the cavity with that in the external icm . our most stringent limit of @xmath104 kev comes from the intermediate cavity pair . finally , we find a channel of decreased surface brightness extending north from the outer ne cavity . we suggest that this feature may be due to energy leaking from the cavity and heating the icm , although we find no temperature enhancement in the channel to support this scenario . support for this work was partially provided by the chandra x - ray center through nasa contract nas8 - 03060 , the smithsonian institution , and by _ chandra _ x - ray observatory grant go1 - 12104x . basic research in radio astronomy at the naval research laboratory is supported by 6.1 base funding . we thank a. foster , r. smith , h. russell , and c. sarazin for useful discussions . fabian , a. c. , sanders , j. s. , allen , s. w. , canning , r. e. a. , churazov , e. , crawford , c. s. , forman , w. , gabany , j. , hlavacek - larrondo , j. , johnstone , r. m. , russell , h. r. , reynolds , c. s. , salom , p. , taylor , g. b. , & young , a. j. 2011 , mnras , 418 , 215 nulsen , p. j. e. , jones , c. , forman , w. r. , david , l. p. , mcnamara , b. r. , rafferty , d. a. , brzan , l. , & wise , m. w. 2007 , in heating versus cooling in galaxies and clusters of galaxies , ed . h. bhringer , g. w. pratt , a. finoguenov , & p. schuecker ( berlin : springer ) , 210 zhuravleva , i. , churazov , e. , schekochihin , a. a. , allen , s. w. , arvalo , p. , fabian , a. c. , forman , w. r. , sanders , j. s. , simionescu , a. , sunyaev , r. , vikhlinin , a. , & werner , n. 2014 , nature , 515 , 85 lllc 5907&2005 apr 2&s1 , s2 , s3 , i3&48.1 + 9517&2008 jun 5&s1 , s2 , s3 , i3&98.3 + 12951&2011 mar 28&s1 , s2 , s3 , i2 , i3&73.7 + 12952&2011 apr 5&s1 , s2 , s3 , i2 , i3&142.3 + 12953&2011 apr 7&s1 , s2 , s3 , i2 , i3&31.7 + 13246&2011 mar 30&s1 , s2 , s3 , i2 , i3&45.0 + 13247&2011 mar 31&s1 , s2 , s3 , i2 , i3&35.7 + 13253&2011 apr 8&s1 , s2 , s3 , i2 , i3&116.7 + 13255&2011 apr 10&s1 , s2 , s3 , i2 , i3&43.0 + lcccccccc 1 kpc , se&1.37&@xmath105&0.11&3.3&@xmath106&@xmath107&@xmath108&0.17 + 10 kpc , se&9.2&@xmath109&0.15&25&@xmath110&@xmath111&@xmath112&1.3 + 10 kpc , nw&12.5&@xmath113&0.07&36&@xmath114&@xmath115&@xmath116&1.8 + 30 kpc , se&27.7&@xmath117&0.31&88&@xmath118&@xmath119&@xmath120&5.2 + 30 kpc , nw&29.9&@xmath121&0.06&163&@xmath122&@xmath123&@xmath124&5.1 + lccccc inner , sw&0.95&0.95&1.3&0.5 + inner , ne&1.03&0.93&1.4&0.6 + middle , sw&3.9&3.9&7.7&13.1 + middle-1 , ne&2.9&2.2&4.9&5.5 + middle-2 , ne&2.8&2.4&9.3&3.8 + outer , sw&5.2&3.0&22.2&3.9 + outer , ne&8.0&4.4&18.0&15.6 +

we present new calculations of x - ray polarization from accreting black holes ( bhs ) , using a monte - carlo ray - tracing code in full general relativity . in our model , an optically thick disk in the bh equatorial plane produces thermal seed photons with polarization oriented parallel to the disk surface . these seed photons are then inverse - compton scattered through a hot ( but thermal ) corona , producing a hard x - ray power - law spectrum . we consider three different models for the corona geometry : a wedge `` sandwich '' with aspect ratio and vertically - integrated optical depth constant throughout the disk ; an inhomogeneous `` clumpy '' corona with a finite number of hot clouds distributed randomly above the disk within a wedge geometry ; and a spherical corona of uniform density , centered on the bh and surrounded by a truncated thermal disk with inner radius . in all cases we find a characteristic transition from horizontal polarization at low energies to vertical polarization above the thermal peak ; the vertical direction is defined as the projection of the bh spin axis on the plane of the sky . we show how the details of the spectropolarization signal can be used to distinguish between these models and infer various properties of the corona and bh . although the bulk of this paper focuses on stellar - mass bhs , we also consider the effects of coronal scattering on the x - ray polarization signal from supermassive bhs in active galactic nuclei .

0912.0907

a recent flurry of new mission proposals has renewed interest in measuring x - ray polarization from a variety of astrophysical sources . the gravity and extreme magnetism smex ( _ gems _ ) mission , which has recently been approved for funding in the latest round of nasa small explorer proposals , should be able to detect a degree of polarization @xmath3 for a flux of a few mcrab @xcite . a similar detector for the international x - ray observatory ( _ ixo _ ) could achieve sensitivity roughly @xmath4 greater ( @xmath5 degree of polarization ; jahoda et al . 2007 , costa et al . 2008 ) . projects like these could potentially detect a large number of galactic and extra - galactic sources at the @xmath6 level , including stellar - mass black holes , magnetars , pulsar wind nebulae , and active galactic nuclei . in this paper , we focus on accreting black holes ( bhs ) in the `` hard '' or `` steep power law '' state @xcite , which are characterized by a broad - band spectrum with a thermal peak around 1 kev and a strong high - energy power - law component extending above 100 kev . early estimates suggest that the typical level of polarization from these sources should be a few percent in the @xmath7 kev range , depending on the geometry of the accretion system and the inclination to the observer @xcite . we also extend these results to agn , whose hard x - ray properties are in many ways qualitatively similar . symmetry demands that in the newtonian limit the observed polarization from a flat disk must be either parallel or perpendicular to its rotation axis . however , the effects of relativistic beaming , gravitational lensing , and gravito - magnetic frame - dragging can break that symmetry and give a non - trivial net rotation to the integrated polarization vector . because the temperature in an accretion disk should increase closer to the bh , where these relativistic effects are strongest , it was predicted long ago that the observed angle and degree of polarization of thermal disk emission should depend on photon energy @xcite . in more recent years , discussion of polarization in accreting black holes has expanded to include a number of other aspects of thermal disk emission , such as uv and x - ray emission from agn disks @xcite and `` lamp post '' models for irradiating the accretion disk with a non - thermal source above the plane @xcite . quite recently , @xcite investigated the effect of atmospheric optical depth on the disk s polarization signal , and @xcite applied the original calculations of thermal x - ray polarization to the problem of measuring the inclination of the inner accretion disk . @xcite , using a monte carlo ray - tracing code on data from shearing - box simulations , estimated the effects of magnetic fields on the polarization of thermal emission . @xcite ( hereafter sk09 ) showed the importance of including returning radiation @xcite when calculating the polarization of bhs in the thermal state , an effect previously noted by @xcite . in addition to the relatively well - understood thermal state , it has also long been known that most agn and stellar - mass bhs can produce significant levels of harder x - rays with energies well above the thermal peak . in both cases , the hard flux is thought to be produced via inverse compton scattering of the disk photons in a corona of hot ( yet thermal ) electrons ( e.g. @xcite ) . from the shape of this hard spectrum , the basic geometry and optical depth of the corona may be constrained @xcite . despite these constraints , there remain sizable uncertainties about the nature and geometry of the hard x - ray emitting region . popular models for the hard state of stellar - mass bhs include a cool disk truncated at large radius ( @xmath8 ) surrounding a hot , radiatively inefficient flow @xcite , or alternatively a more extended disk surrounded by an optically thick hot corona , possibly in the form of a hot wind @xcite . even less is known about the steep power - law ( spl ) state . as for the hard state , most popular models are based on the inverse compton scattering of thermal seed photons from a thin disk surrounded by a hot corona @xcite , but bulk comptonization of a converging accretion flow has also been suggested @xcite . although the geometry of the corona could be as simple as a uniform slab in the spl state @xcite , the hard state of galactic bh binaries , as well as the x - ray emission from agn , are more likely caused by clumpy , inhomogeneous coronae , possibly caused by magnetic flares @xcite . here we explore the x - ray polarization signatures of three simple models for the corona geometry : a smooth sandwich with uniform optical depth in the vertical direction ; an inhomogeneous model made of a finite number of spherical clouds , randomly distributed above the disk ; and a truncated thin disk surrounding a spherical corona . in all cases , seed photons are emitted from the thermal , optically thick disk and are up - scattered in the hot corona ( in the truncated disk case , we also include low - energy seed photons embedded in the corona ) . we find that in all cases , the spectrum comprises a thermal peak ( @xmath9 kev for stellar - mass bhs ; @xmath10 ev for agn ) and a power - law component dominating above that , which typically makes up @xmath11 of the total flux . the gross features of the polarization spectra are quite robust : horizontal polarization with a few percent amplitude at low energies , with a transition to vertical orientation above the thermal peak , where the polarization amplitude can be as large as @xmath12 . the specific details of the polarization spectrum , i.e. , the amplitude of polarization at low and high energies , and the shape and location of the transition , provide constraints on the global geometry , temperature , and optical depth of the corona , as well as the bh mass , spin , accretion rate , and the observer inclination angle . in this paper , we begin in section [ methods ] with a brief overview of the computational methods used in the calculations . in sections [ wedge][sphere ] we present our results for three different corona geometries : sandwich , hot spots / clumps , and a sphere embedded in a truncated disk , all in the context of galactic x - ray binaries . in section [ agn ] , we apply the results to agn . in section [ discussion ] we discuss possible applications to observations . although a detailed description of the ray - tracing code will be given in a companion paper @xcite , we give a brief summary of the relevant physics here . the basic geometry is described by a thin disk with inner radius @xmath2 , typically placed at the inner - most stable circular orbit ( isco ) , but possibly at smaller radii so long as @xmath2 is outside the horizon . in the latter case , the gas inside the isco follows plunging trajectories along geodesics with energy and angular momentum determined by the isco values . the orbital angular momentum of the disk is prograde and aligned with the bh spin axis . seed photons are emitted from the thin disk with a diluted black - body spectrum characterized by a hardening factor @xmath13 @xcite . the local flux is given by the novikov - thorne emissivity profile @xcite , but can be modified to include emission inside the isco , as described in sk09 . the seed photons have initial polarization parallel to the disk surface in the local fluid frame , as determined by the classical result for scattering - dominated atmospheres @xcite . the degree of polarization varies from zero for photons emitted normal to the disk surface up to @xmath14% for an inclination angle of @xmath15 . in addition to the polarization effects , the scattering of the outgoing flux causes limb - darkening , effectively focusing the emitted radiation in the direction normal to the disk surface . we define our seed photons in accordance with the polarization and limb - darkening factors tabulated as a function of emission angle in table xxiv in @xcite . after leaving the disk , the photon packets follow geodesic trajectories around the black hole , eventually reaching a distant observer , returning to the disk via gravitational deflection , or getting captured by the horizon . en route , many photons scatter in the corona . we describe these events with the classical thomson electron cross section . the differential probability for scattering along a path length @xmath16 is simply @xmath17 where @xmath18 is the opacity to electron scattering , @xmath19 is the local mass density of the corona , and @xmath20 is the optical depth . unlike some recent monte carlo ray - tracing codes ( e.g.@xcite ) that use the more physically accurate , energy - dependent klein - nishina cross section , the thomson cross section allows for the use of photon packets that include the entire broad - band spectrum , as described in @xcite . this simplification gives improved computational efficiency , albeit at the loss of physical accuracy for above @xmath21 kev , where the classical and relativistic electron cross sections begin to diverge . for most stellar - mass bh sources , and for most x - ray polarimetry missions in the foreseeable future , it should be quite safe to focus on photons with energies less than @xmath22 kev . at each step along the photon geodesic , the probability for scattering is calculated according to equation ( [ diff_prob ] ) , and then a uniform random number in @xmath23 determines whether that photon scatters at that location . ] . when the photon _ does _ scatter , we first transform the photon 4-momentum @xmath24 and polarization vector @xmath25 into the local inertial frame of the corona , then do a special - relativistic boost into the electron s rest frame , where the electron velocity is taken from an isotropic thermal distribution . in the electron s rest frame , the scattering event is treated completely classically , with a differential cross section given by @xcite @xmath26\ , , \ ] ] where @xmath27 cm is the classical electron radius , @xmath28 is the degree of polarization , @xmath29 is the angle between incident and outgoing photon directions , and @xmath30 is the polarization angle , measured with respect to the scattering plane . the angles @xmath29 and @xmath30 are selected appropriately from the probability distribution corresponding to ( [ dsigma_domega ] ) . the outgoing stokes parameters are given by @xmath31 where @xmath32 and @xmath33 are the components of the intensity with polarization parallel to and perpendicular to the scattering plane , respectively . these stokes parameters are then used to reconstruct the new polarization degree and angle through [ x_y ] @xmath34 and [ delta_psi ] @xmath35 the scattered photon packet is boosted back to the coronal fluid frame , then transformed back into the coordinate basis and continues propagating along its new geodesic . this `` transform boost scatter boost transform '' algorithm automatically carries out the inverse compton change in photon energy , giving the outgoing photon an average energy increase of @xmath36 , where @xmath37 is the electron lorentz factor ( @xmath38 for thermal electrons ) . the polarization 4-vector is parallel - transported simply by satisfying the constraints @xmath39 and @xmath40 and using the complex - valued walker - penrose integral of motion @xmath41 @xcite . analogous to the way that carter s constant @xcite can be used to constrain 4-velocity components , @xmath41 can be used to reconstruct the polarization vector at any point along the geodesic . the two orthonormality conditions stated above , along with the real and imaginary parts of @xmath41 , give a total of four equations for the four components of the polarization vector @xmath25 . some photons encounter the disk before reaching their final destination . when we treat models relevant to agn , photons striking the disk are absorbed because our primary interest is in photons @xmath42 kev , and agn disks are generally thought to have enough photo - ionization opacity to have little albedo at these energies . in the stellar - mass black hole case , disks are expected to be highly reflective . for these models , we use the results given in section 70.3 of @xcite for diffuse reflection , i.e. multiple scattering events in a semi - infinite plane . when a ray reaches an observer at infinity , the polarization vector is projected onto the detector plane . in our convention , the y - axis of the detector is parallel to the projected symmetry axis of the black hole and accretion disk , so @xmath43 corresponds to `` horizontal '' polarization and @xmath44 is `` vertical . '' finally , we integrate over all photon bundles to obtain the energy - dependent stokes parameters @xmath45 , @xmath46 , and @xmath47 , which in turn give @xmath48 and @xmath49 , our preferred observables for most of the results presented below . we begin by considering the simplest corona geometry , an isothermal layer of uniform vertical optical depth forming a wedge with constant opening angle @xmath50 . at each point in the disk , the atmosphere has an exponential profile in the vertical direction with @xmath51 $ ] . the vertically - integrated optical depth @xmath1 is constant , giving @xmath52 we set the local rest frame of the corona to be corotating with the underlying disk : @xmath53 , @xmath54 , and @xmath55 . @xmath56 is the orbital frequency ( measured at infinity ) of a planar circular orbit at radius @xmath57 , which in boyer - lindquist coordinates is @xmath58 figure [ schem_sandwich ] gives a schematic view of the accretion geometry , showing how the seed photons originate in the midplane , then scatter through the hot corona before reaching a distant observer . the coronal scattering has two major effects on the observed polarization signature . first , it changes the underlying spectrum by inverse compton scattering a portion of the photons , boosting them to higher energies , and thus producing a harder spectrum . second , it changes the amplitude and orientation of the net polarization at high energies , rotating it from horizontal to vertical , similar to the way returning radiation gets scattered into a vertical orientation by a thermal disk ( sk09 ) . the coronal polarization rotation effect is essentially the same as that described in @xcite , who considered the up - scattering of low - energy photons embedded in a hot electron corona with planar geometry , finding strong vertical polarization for high - energy photons and observers at high inclinations . the cause of this rotation in systems with moderate optical depth ( @xmath59 ) can be understood easily from the scattering geometry in figure [ schem_sandwich ] . photons initially emitted in directions near the disk plane are almost certain to scatter because they face a large optical depth . to reach an observer who views the disk nearly edge - on , they must stay in the disk plane even after scattering , but this requires a vertical polarization direction . additionally , those photons that scatter multiple times in the corona and are thus boosted to higher energies are geometrically more likely to move in the plane of the corona , parallel to the disk . this further increases the amplitude of their vertical polarization and preferentially scatters them to infinity with large emission angle , leading to a limb-_brightening _ effect at high energies with respect to the classical chandrasekhar result @xcite . for hot coronae with high optical depth , the multiple scatters cause the polarization and limb - darkening to tend towards the chandrasekhar limit , but with the photon energy dependence given by a wien spectrum . in the limit of a cold corona , where there is no transfer of energy via inverse compton scattering , there is no way to distinguish between photons coming directly from the disk atmosphere and those that are scattered in the corona , and the classical result is reproduced regardless of the coronal optical depth . to improve our qualitative understanding of the polarization effects of the global corona geometry , we plot in figure [ image_wedge ] a series of images of the bh accretion disk plus corona , sorted by the scattering history of each photon . figure [ image_wedge]a shows only photons that travel directly from the disk to the observer ; figure [ image_wedge]b shows flux from return radiation photons that are bent by gravitational lensing and scatter off the disk to the observer ; figure [ image_wedge]c is the flux from photons that scatter exactly once in the corona ; and figure [ image_wedge]d is made of photons that scatter multiple times in the corona . the observed intensity is color - coded on a logarithmic scale ( normalized to the peak intensity from all photons ) and the energy - integrated polarization vectors are projected onto the image plane with lengths proportional to the local degree of polarization . the black hole has a spin of @xmath60 , mass @xmath61 , and accretion rate such that the thermal flux alone totals @xmath62 of the eddington luminosity . the corona has a temperature of @xmath22 kev , scale height @xmath63 , and vertical optical depth @xmath64 . these parameters lead to a coronal luminosity @xmath65 times that of the thermal luminosity . the disk is rotating in the counter - clockwise direction , and the observer is located at an inclination of @xmath66 to the rotation axis . the intensity maximum at the left of each image is caused by relativistic beaming and doppler boosting . photons from the far side of the disk are bent by gravitational lensing , giving the appearance of a warped disk with the far side having a smaller effective inclination angle . not surprisingly , the direct image looks quite similar to that for the direct radiation from a simple thermal disk with no corona ( sk09 ) . the polarization is modest and predominantly horizontal , with relativistic effects most noticeable in the inner disk , where beaming and lensing combine to rotate the observed angle of polarization at higher energies . the degree of polarization is reduced above and to the left of center , where the effective inclination of the disk is smaller , while the opposite occurs in the receding section of the disk on the right side of the image . however , when carefully compared to the non - corona system ( fig . 1 of sk09 ) , we see a clockwise shift in the peak flux distribution because the level of direct flux from a given patch of the disk is dependent on the optical depth to scattering along the geodesic connecting that patch with the observer . for the wedge geometry , this means that where the effective inclination is smallest , the observed flux is greatest , leading to additional limb - darkening in the direct component . the return radiation image is also similar to that for the thermal disk alone , although now the presence of a scattering corona strongly suppresses any return flux from large radii . since most returning photons pass very close to the light orbit around @xmath67 , they would necessarily intersect the outer disk at very high angles of incidence , and thus face very large optical depth through the wedge corona . as shown in sk09 , the flux from returning radiation , while small compared to the total flux , is very highly polarized , and in a vertical direction , perpendicular to that of the direct flux . since the majority of the returning radiation originates in the hot inner regions of the disk , there is a dependence of the polarization direction on photon energy . as can be seen from figure [ image_wedge]b , the angle of polarization for the return radiation is _ negative _ , as measured in the image plane , with @xmath68 defined along the @xmath69-axis . this can also be understood from the global geometry of the system , as shown in figure [ schem_sandwich ] . photons emitted from the right side of the disk , deflected by the bh and incident on the left side at some finite angle , will scatter towards a high - inclination observer with large polarization in the @xmath70 orientation . of course , the opposite occurs for photons emitted from the left side and scattered on the right side , but they get preferentially scattered away from the observer by relativistic beaming . in figure [ image_wedge]c we show the flux from photons experiencing a single coronal scatter . since _ every _ photon in this image scatters exactly once , there is no preference for oblique emission , and photons emitted at all inclination angles are equally weighted . if anything , for @xmath64 , the seed photons emitted normal to the disk surface are more likely to scatter exactly once and then escape to an observer at high inclination , thus explaining the prevailing horizontal polarization seen here . lastly , figure [ image_wedge]d shows the multiply - scattered photons . these have strong net vertical polarization due to the geometric effects described above and shown schematically in figure [ schem_sandwich ] . unlike the returning radiation , the coronal flux has a net _ positive _ polarization angle @xmath71 when viewed edge - on . this can again be understood from the schematic image of the wedge corona in figure [ schem_sandwich ] . it is clear from this geometry that for two photons , both emitted in the plane of the sky at the same angle @xmath72 to the disk normal , the photon emitted in the direction _ away _ from the black hole will have a greater path length within the wedge corona . however , due to the increasing scale height and constant optical depth in the vertical direction , it will pass through lower - density gas . combining these two effects path length and coronal density as a function of emission angle we find that the optical depths for ingoing and outgoing photons are [ tau_in_out ] @xmath73\ ] ] and @xmath74 \ , .\ ] ] here @xmath75 is the opening angle of the corona and we restrict @xmath76 so that the outgoing optical depth is finite . since @xmath77 for each @xmath72 , the `` tilt '' of the polarization vector for corona scattering is opposite that of return radiation . for the region of the disk beamed towards the observer ( left side of images in fig . [ image_wedge ] ) , this means that photons emitted in the direction away from the bh are more likely to be scattered in the wedge corona , giving a net polarization oriented with @xmath71 , as can be seen in the lower panels of figure [ image_wedge ] . in practice , we ca nt ever really know whether a given photon came directly from the disk or was scattered in the corona or off the disk , but the images do provide important qualitative understanding of the geometric effects involved . for a more quantitative picture , we plot in figure [ wedge_grid ] the observed flux , polarization degree , and polarization angle as a function of energy for observer inclinations of @xmath78 , @xmath79 , and @xmath66 . we sort the photons into direct ( dotted curves ) , scattered ( dot - dashed curves ) , and total flux ( solid curves ) . for a bh mass of @xmath80 accreting at a rate @xmath81 , the thermal peak is near 1 kev , with essentially no flux from the disk above @xmath82 kev . with @xmath64 and @xmath83 kev , the compton @xmath84 parameter is of order unity , so that the up - scattered spectrum has @xmath85 @xcite . as the inclination of the observer increases , the optical depth to the disk increases , thus lowering the relative contribution from the direct radiation . in all cases , the up - scattered radiation dominates the flux above a few kev . the direct radiation behaves quite similarly to that described in @xcite , reproducing the chandrasekhar result at low energy , and then decreasing in fractional polarization at higher energies as the relativistic effects of the inner disk begin to dominate . as discussed in sk09 , the angular rotation in @xmath30 is greatest for small observer inclinations simply because there is a smaller degree of `` classical '' polarization to overcome . when considering only the scattered photons , the polarization is still given by the chandrasekhar limit at low energies . the scattered photons below the thermal peak have typically scattered only once in the corona , thus preserving their initial horizontal orientation ( see fig . [ image_wedge]c ) . at higher energies , we begin to sample the multiply - scattered photons that make up figure [ image_wedge]d , giving strong vertical polarization . thus there is a transition from horizontal to vertical orientation even when considering the scattered photons alone . when including all the flux ( solid curves ) , this transition point is shifted to somewhat higher energies than for the scattered photons alone due to the horizontal contribution of the direct thermal radiation . as mentioned above , the effects of return radiation and coronal scattering on the _ sign _ of the polarization angle are opposite : return radiation leads to a negative angle and the coronal scattering in a wedge geometry leads to a positive angle in the transition from horizontal to vertical polarization . however , some photons that scatter in the corona are also deflected by gravitational lensing and then scattered in the corona on the far side of the disk , leading to some ambiguity as to whether they should be classified as return photons or coronal photons . from figure [ wedge_grid ] we see that the relative contributions from these two effects are also dependent on the observer inclination , with low - inclination systems giving a negative tilt and high - inclination leading to a positive tilt . this distinction may be observable in practice by comparing the polarization in the hard state to that of the soft state , where the transition direction is independent of inclination and therefore can be used to define the negative orientation ( sk09 ) . using spectropolarimetry to infer physical properties of the source is certainly more complicated in the hard state than in the thermal state , for the simple reason that there are more free parameters in the underlying model , creating degeneracies in fitting the data . instead of attempting to quantify our ability to measure any particular property of the bh system , we present here the effects of separately changing individual parameters with respect to a fiducial model with @xmath61 , @xmath60 , @xmath86 , @xmath87 , @xmath88 , @xmath83 kev , and @xmath63 . in figure [ wedge_ledd ] we show the polarization degree and angle for the fiducial model , varying the luminosity in the thermal flux component from @xmath89 to @xmath90 . since in all cases the thermal peak is well below the corona temperature , the transition energy scales linearly with the disk temperature , which is in turn proportional to the quarter power of the luminosity . with multiple observations of a single source accreting at various rates as in figure [ wedge_ledd ] , one could confirm that the thermal part of the spectrum is indeed coming from a thin disk with constant @xmath91 @xcite . holding the net thermal flux fixed , but varying the spin , changes the emissivity profile of the disk , concentrating more of the emission close to the bh for larger spins ( and thus smaller @xmath92 ) . as in the thermal state ( sk09 ) , this leads to a greater fraction of return radiation , and thus the transition from horizontal to vertical polarization occurs at lower energy , as shown in figure [ wedge_spin ] . in the hard / spl states , the presence of a corona provides another , complementary mechanism for effecting the polarization transition above the thermal peak , regardless of the spin parameter . thus the dependence of the polarization signal on bh spin is weaker than in the thermal state . in this way , the sandwich corona acts as a veil around the inner disk , obstructing our view of the plunging region where the orbital dynamics are most sensitive to the spin . in figure [ wedge_taut ] , we simultaneously vary the coronal optical depth and temperature , maintaining a roughly constant compton @xmath84-parameter . in the wedge corona model , the transition from horizontal to vertical polarization is largely due to local geometric effects ( as opposed to more global effects like returning radiation ) . in this case , the optical depth and opening angle of the corona largely determine the scattering history of each photon , and thus the polarization degree and angle of the outgoing radiation . holding the size and shape of the corona constant , two photon packets experiencing the same number of scattering events will emerge on average with the same polarization signature . thus the energy of the transition is primarily a function of the scattering electron temperature , and hotter coronae lead to a higher - energy transition . similarly , to reach a given energy above the thermal peak , photons have to scatter more times in a cooler corona , resulting in a more constrained scattering history , and thus higher polarization . at the same time , when changing the optical depth of the corona , the overall scattering geometry _ does _ change slightly , so the shape of the transition also varies in figure [ wedge_taut ] . furthermore , for small optical depths , a larger fraction of the photons can return to the disk , giving the negative polarization angle characteristic of returning radiation . lastly , we show in figure [ wedge_hr ] the effect of varying the scale height of the corona . here , the fact that return radiation and coronal scattering rotate the polarization angle in opposite senses away from horizontal leads to a complicated dependence . the geometrically thinnest coronae behave similarly to the razor - thin thermal disks described in sk09 , giving a more gradual transition from horizontal to vertical with @xmath70 ) we see that when @xmath93 , there is no difference between photons emitted towards or away from the bh , leaving only the global effects of returning radiation . ] . as the opening angle of the corona increases , the transition flips to @xmath71 as the coronal scattering effects of equation ( [ tau_in_out ] ) becomes more important . for very large @xmath0 , the lower density of the corona allows longer path lengths and photons can sample a larger volume of the accretion flow . at low energies this loosening of the geometrical constraints leads to a deviation from the classical planar scattering atmosphere . at higher energies , it leads to a higher effective temperature as the bulk velocity of the corona ( which can be a substantial fraction of @xmath94 ) is added to the thermal velocities of the scattering electrons . the combination of these various effects eliminates the existence of any simple trends in figure [ wedge_hr ] . motivated by spectral models of agn @xcite , whose reasoning also applies to galactic bhs in the hard state @xcite , we also consider an inhomogeneous geometry where the hot coronal plasma is clumped into a large number of small , dense regions . these clouds are distributed randomly above the disk , roughly following a wedge geometry . for precisely the same reasons that originally motivated the clumpy corona model , it is not generally possible to reproduce the same spectrum as in the sandwich geometry . yet we find that by fixing the electron temperature and conserving the total mass in the corona , the net flux in hard x - rays @xmath95 is roughly the same . we describe the inhomogeneous corona with five parameters : aspect ratio @xmath0 , temperature @xmath96 , mean vertical optical depth @xmath1 , number of clouds per unit radius @xmath97 ( @xmath98 above the plane and @xmath98 below the plane ; @xmath99 is constant throughout the disk , giving a larger number of clouds per unit area in the inner disk ) , and an overdensity factor @xmath100 , where @xmath101 is the mean density of a wedge corona with the same scale height and optical depth . the distribution of clouds is chosen so as to reproduce the wedge corona in the limit of @xmath102 and @xmath103 . to do this , we require that the average mass in the inhomogeneous corona is equal to that of a wedge corona at each annulus @xmath104 . for spherical clouds of radius @xmath105 , the coronal mass above ( or below ) the disk at each annulus is given by : @xmath106 which gives a clump radius @xmath107^{1/3 } \,.\ ] ] the covering fraction can be estimated by calculating the probability that a given region of the disk is _ not _ covered by a coronal cloud . consider an annulus of area @xmath108 . the expectation value for the number of clouds over it is @xmath109 . if each cloud has cross sectional area @xmath110 , this probability is given by @xmath111.\ ] ] the covering fraction is simply @xmath112 . in figure [ image_hotspot ] we show a series of images of the bh accretion disk plus a clumpy corona , again sorted by photon history as in figure [ image_wedge ] . the corona is made up of 100 clouds distributed randomly with scale height ratio @xmath63 with constant probability per unit radius ( i.e. , @xmath113 ) inside radius @xmath114 . the mean local optical depth @xmath64 , and the density contrast @xmath115 . for the clumping algorithm described above , these parameters correspond to a covering fraction in the inner disk of @xmath116 , comparable to that inferred for the hard state of cyg x-1 by @xcite . as in figure [ image_wedge ] , the bh has spin @xmath60 , mass @xmath117 , thermal flux @xmath118 , corona temperature @xmath83 kev , and observer inclination angle @xmath66 . the direct flux from the disk , plotted in figure [ image_hotspot]a , clearly shows a number of distinct shadows where optically thick clouds block the disk . the reduced covering fraction of the corona leads to a greater flux from return radiation ( fig . [ image_hotspot]b ) , which otherwise behaves much the same as in the sandwich geometry , contributing a strong vertical component to the polarization signal . as in the direct image , the return radiation that eventually reaches the observer is also blocked in places by intervening coronal patches . in the lower panels of figure [ image_hotspot ] , we show the scattered photons ( fig.[image_hotspot]c : one scatter ; fig . [ image_hotspot]d : many scatters ) . now the polarization of the hard flux is significantly less coherent than in the sandwich geometry . many photons scatter multiple times within a single cloud , eventually emerging with no particular polarization direction . at the same time , the global geometry of the system is still essentially the same as the sandwich model , and photons are more likely to experience multiple scattering events by propagating from one coronal clump to another , constrained to move roughly parallel to the disk surface . as in the homogeneous model , this leads to a net vertical polarization in the observed flux at high energies . in figure [ hotspot_grid ] we plot the flux , polarization degree , and polarization angle as a function of energy for the inhomogeneous corona and observer angles of @xmath119 , @xmath79 , and @xmath66 . as in the sandwich geometry , here too we see that with increasing inclination , a greater fraction of the observed flux comes from scattered photons . due to the smaller covering factor , a greater fraction of the thermal flux can escape at all inclinations , but the spectrum is still dominated by comptonized photons above @xmath120 kev . from equation ( [ r_c ] ) , we see that the individual coronal clumps have typical optical depths of @xmath121 in the inner disk . this naturally leads to a harder power - law spectrum relative to that of the wedge geometry . the polarization signature for the inhomogeneous corona is similar to the sandwich geometry at low energies , but noticeably different above a few kev . below the thermal peak , a sandwich corona merely extends the classic chandrasekhar atmosphere , leaving the polarization unchanged . the scattering contribution ( dot - dashed curves ) is more weakly polarized when the gas is clumped because it is easier for a photon emitted at a large angle relative to the disk normal to scatter once off a coronal cloud and then reach an observer , contributing a vertical component to the polarization , slightly diluting the dominant horizontal signal at low energy . on the other hand , the degree of polarization at high energies is significantly lower for a clumpy corona than for the sandwich corona . this is because many high - energy photons scatter multiple times within a single cloud , ultimately emerging with little or no net polarization due to the spherical symmetry of the cloud . lastly , we note that when the corona is clumped , the _ shape _ of the transition from horizontal to vertical polarization more closely resembles that of the thermal state than that of a homogeneous sandwich corona : relatively smooth and in the @xmath70 direction . because of the small covering factor of the corona , thermal seeds emitted in the inner disk have a greater chance of returning to the disk or even scattering off a coronal clump on the far side of the bh , thus leading to a @xmath122 distribution quite similar to that of the thermal disk with no corona ( sk09 ) . not surprisingly , the polarization signature is sensitive to the compactness of the coronal hot spots , as shown in figure [ hotspot_rho ] . in the limit of @xmath123 , the signal more closely resembles the homogeneous sandwich result : a sharp transition at @xmath65 kev , with @xmath71 and the polarization degree rising above @xmath124 at high energies ( this case does not identically reproduce the homogeneous sandwich when @xmath125 is finite because , although the centers of the clouds must lie within @xmath126 of the plane , their outer portions may extend farther , resulting in an inhomogeneous geometry even when @xmath123 ) . as the compactness increases , the covering fraction and the overall symmetry decrease , giving a smoother transition and smaller amplitude above a few kev . in figure [ hotspot_nc ] we show the dependence of the polarization signal on the number of clumps in the corona , holding the overdensity fixed at @xmath115 . in the limit of @xmath127 , the homogeneous result is reproduced , and in the limit of @xmath128 we get the pure thermal result . the last model we consider is a spherical corona immediately surrounded by a truncated disk , a geometry motivated by a variety of spectral and timing observations ( e.g.@xcite ) . the disk extends in to radius @xmath2 with a novikov - thorne emissivity profile , and the corona is defined by its temperature @xmath96 and optical depth @xmath1 , as measured from the horizon out to @xmath2 . the electron density in the corona is taken to be constant , and in keeping with its spherical shape , the fluid is at rest in the frame of a zero - angular momentum observer @xcite . in order to most closely match the spectral properties of the sandwich corona , we also embed thermal seed photons with a uniform distribution inside the corona . the net flux and spectrum as a function of radius is determined by matching the shell - integrated flux to that of a standard novikov - thorne disk down to the isco . these coronal seed photons are emitted isotropically with zero polarization . in figure [ image_sphere ] we show the same photon - sorted images as in figure [ image_wedge ] , now for the spherical coronal model with @xmath129 . the direct disk flux behaves as expected : because it is emitted from regions that are almost entirely in the non - relativistic regime , its polarization is very close to the chandrasekhar value . there is now also a significant amount of unpolarized thermal flux coming directly from the coronal seeds . these seed photons can also scatter off the surrounding disk , in which case they are classified here as return radiation , again strongly polarized in the vertical direction . the corona - scattered flux shown in figures [ image_sphere]c , d closely resembles that of a scattering - dominated atmosphere around a star or planet : zero polarization from surfaces normal to the observer , and strongly - polarized ( but limb - darkened ) flux from the edges . however , the disk blocks a portion of the bottom half of the sphere , with the very bottom the most likely to be blocked . consequently , the net polarization from a spherical corona is non - zero and oriented in the vertical direction . we have not shown here the contribution from photons that scatter in the corona and _ then _ off the disk , but these also are strongly polarized in the vertical direction , much like the image in the upper - right panel . just as in the analogous figures for the sandwich and clumped models , figure [ sphere_grid ] shows that the total flux spectrum is dominated at low energies by the thermal contribution ( here the sum of the disk proper and the thermal seeds embedded in the corona ) , while it is dominated at high energies by inverse compton scattered photons . because the coronal density and temperature were specifically chosen to give the same compton @xmath84-parameter as in the sandwich geometry ( see fig.[wedge_grid ] ) , it is not surprising that the spectra agree so closely . the polarization signature is relatively straightforward to understand , especially in the context of figure [ image_sphere ] : the polarization from the direct radiation is almost exactly horizontal , decreasing steadily in amplitude with energy as a greater proportion of the observed flux comes from the unpolarized and higher - energy seeds in the corona . the scattered flux is consistently vertical , since even the lower - energy photons that have scattered only once have a net vertical orientation ( see the lower - left panel of fig.[image_sphere ] ) . because the unobscured coronal hemisphere looks much the same from each of the three viewing angles , the polarization amplitude at high energies is only weakly dependent on observer inclination . combining these effects gives the now - familiar behavior of horizontal polarization at low energies governed by the chandrasekhar limit transitioning to vertical orientation above the thermal peak . the spherical geometry leads to a particularly sharp transition , at least for @xmath130 , but it is still clear that a negative @xmath30 is preferred in the transition region . again this can be understood by the geometry of the return radiation scattering off the region of the disk most strongly beamed towards the observer . in this case , the coronal photons originating from the central region above the disk plane replace the return radiation that similarly had passed near the photon orbit , just above the bh . in figure [ sphere_redge ] we investigate the effects of varying @xmath2 , while keeping the total optical depth of the corona constant . since the thermal disk moves in with decreasing @xmath2 , we find the transition point moves to higher energies , as a greater fraction of the flux is coming directly from the disk and is therefore horizontally polarized . the shape of the transition grows broader with decreasing @xmath2 as relativistic effects become more important , rotating the polarization angle as in the thermal state . in the limit of large @xmath2 , where newtonian physics dominates the problem , the system becomes scale - invariant and symmetry constrains the polarization at any given energy to be exactly horizontal or exactly vertical . in the limit of very small @xmath2 , we reproduce the thermal result , plus a small power - law contribution to the spectrum at high energy due to the ultra - compact corona . holding the corona radius constant at @xmath131 , in figure [ sphere_taut ] we show the effects of varying the optical depth and temperature of the corona . as in figure [ wedge_taut ] for the sandwich geometry , here too we keep the compton @xmath84-parameter fixed so that the resulting spectra are nearly identical . however , unlike the sandwich model , for the spherical corona we find essentially no dependence on the coronal temperature or density . the reason is the essential simplicity of the scattering geometry : the polarization signature from the direct radiation alone is nearly the same , regardless of the coronal optical depth , and from figure [ sphere_grid ] we see that all scattered radiation is polarized exactly the same over the entire spectrum . thus the behavior of the net signal is simply a function of the relative flux in each component , and this is held constant by fixing the compton @xmath84-parameter . it has long been known that a significant fraction of the flux from agn is emitted in the x - ray band @xcite . as in stellar - mass bhs , this high - energy flux likely comes from lower - energy seed photons inverse compton scattered in a corona of hot electrons with @xmath132 kev . similar to the stellar - mass case , this leads to a relatively hard power - law spectrum with index @xmath133 @xcite . however , unlike the stellar - mass case , the temperature of the inner disk for an agn will be well below a kev , leading to a thermal peak in the uv band . furthermore , even when the disk is dominated by radiation pressure and electron scattering opacity , there should still be a substantial fraction of metals that are not fully ionized , producing a large opacity for absorption above @xmath134 1 kev . both of these agn features lower energy seed photons and an x - ray absorbing disk lead to important differences in the polarization signature as compared to the stellar - mass models described above in sections [ wedge]-[sphere ] . because the seed photons start off with lower energies , they must scatter more times in the corona in order to reach the @xmath135 kev band . for a thin sandwich corona , this means that the scattering geometry is even more constrained than in the stellar - mass case , forcing the photons to move in a plane parallel to the disk surface , leading to a stronger vertical polarization . the agn disk absorbs much of the incident x - ray flux from the corona , so the compton @xmath84-parameter is effectively smaller than that of a stellar - mass system with the same coronal properties because scattering sequences are halted once a photon strikes the disk ( we do not re - radiate the x - ray flux , but rather treat it as completely absorbed ) , thereby reducing the average path length the photons that escapes to infinity . an absorbing disk boundary condition with a sandwich corona therefore leads to an even higher degree of x - ray polarization because the photons are forced to scatter in a more constrained geometry before escaping the corona . in figure [ agn_flux ] we show broad - band spectra from an agn with central mass @xmath136 , spin parameter @xmath60 , thermal luminosity @xmath137 , and observer inclination angle @xmath78 . four different corona models are considered : a homogeneous wedge with @xmath63 , @xmath64 , and @xmath83 kev , and three clumpy models with the same scale height , mean optical depth , and temperature , but with @xmath138 and overdensities of @xmath123 , @xmath139 , and @xmath140 . the inhomogeneous models can be characterized by their covering fractions as measured in the inner disk at @xmath141 ; these overdensities correspond to @xmath142 , @xmath143 , and @xmath144 , respectively . as can be seen in figure [ agn_flux ] , a greater covering fraction leads to a softer spectrum , while an inhomogeneous corona with smaller yet denser clumps gives a larger optical depth for those photons that do not escape directly from the disk , in turn giving a larger compton @xmath84-parameter and harder spectrum . as mentioned above , the disk absorption in the agn models results in a softer spectrum than stellar - mass bhs with the same coronal parameters this can be seen by comparing figure [ agn_flux ] with the upper - left panels in figures [ wedge_grid ] ( wedge geometry ; @xmath145 ) and [ hotspot_grid ] ( clumpy geometry ; @xmath146 ) . in both cases , the spectral index @xmath147 is increased ( i.e. , softer ) in the agn case by about 0.5 . the effect of this absorption on the x - ray polarization can be seen in figure [ agn_c ] , which plots the degree and angle of polarization as a function of energy for the same model parameters used in figure [ agn_flux ] . as expected , for the wedge geometry with @xmath145 , the absorbing disk leads to a larger degree of polarization than the reflecting boundary condition used in stellar - mass bhs ( @xmath148 for agn compared with the @xmath149 shown in fig . [ wedge_grid ] for @xmath61 and @xmath78 ) . for the clumpy coronae , however , we find the opposite effect : including absorption actually _ reduces _ the degree of polarization . recall from section [ hotspot ] that inhomogeneous coronae generally produce a weaker polarization signal than the uniform - density sandwich geometry . this is because photons scattering multiple times in a single spherical cloud will ultimately escape with no net polarization . it is only through the _ global _ scattering geometry that a net vertical polarization is acquired , due to the small number of photons that scatter from one cloud to another , often reflecting off the disk surface at grazing incidence along the way . since many of these photons are lost to absorption in the agn case , global geometric effects play a smaller role , and thus the degree of polarization is diminished ( @xmath150 for agn versus @xmath151 for the stellar - mass case in fig . [ hotspot_grid ] ) . thus we find that for agn , x - ray polarization is even more sensitive to the inhomogeneity of the corona than for stellar - mass bhs . from figure [ agn_flux ] , we see that it should be possible to determine the agn covering fraction from the x - ray spectrum alone , without polarization information . if so , then the degree of polarization could rather be used to constrain the inclination of the agn disk , a parameter that can be quite difficult to constrain in many cases . in figure [ agn_i ] we compare the degree of polarization for a range of inclination angles and two different corona models : wedge geometry with @xmath145 ( solid curves ) and a clumpy geometry with @xmath152 ( dashed curves ) . clearly the inclination will be easier to measure for a smooth corona , but even for a clumpy corona with @xmath152 , a first - generation x - ray polarimeter should be able to distinguish between @xmath153 and @xmath119 . in the event that we can measure the covering fraction _ and _ inclination with other observations , the polarization can constrain other corona parameters like the number of clumps @xmath98 and their overdensity @xmath100 . from equation ( [ f_c ] ) , we see that for a constant covering fraction , @xmath154 , so increasing the number of clouds both decreases their characteristic size and increases their density . yet when averaged over a photon s entire path , a corona with larger @xmath98 will appear more homogeneous , despite the increased density perturbations on very small scales . this in turn leads to a greater degree of polarization , as shown in figure [ agn_nc ] . holding the inclination and covering fraction fixed at @xmath78 and @xmath152 , respectively , we vary the number density of coronal clumps . while all four cases plotted in figure [ agn_nc ] have nearly identical spectra , we see that polarization can distinguish them and thus give improved constraints on the corona geometry . the different models in figure [ agn_nc ] would also likely have observably different timing properties : if individual clumps evolve coherently , but are independent of one another , systems with smaller @xmath98 should produce greater amplitude flux variations in the x - ray band . in practice , agn polarization measurements will have some additional observational challenges not present in stellar - mass bh systems . first , the typical x - ray flux from nearby seyfert galaxies and quasars is at least an magnitude smaller than the brightest galactic bhs . this means that much longer observation times will be required to reach a comparable level of polarization sensitivity . in addition , type 1 agn are expected to have only modest inclinations ( @xmath155 ) because at higher inclinations our view of the inner disk is likely blocked by the surrounding dusty torus ( assuming the torus and disk are oriented in the same direction ) . there are also two potential sources of dilution . radio - loud agn generically have somewhat larger ratios of x - ray flux to optical flux than radio - quiet @xcite , a fact plausibly interpreted as due to a jet contribution ; in blazars , the jet contribution is almost certainly substantial . in addition , fe k@xmath147 emission often accounts for a few percent of the 210 kev flux . lastly , the much longer variability timescales for supermassive bhs make it unlikely that we could use the strategy discussed in section [ wedge ] , in which observations of multiple spectral states can jointly determine the orientation of the spin on the sky . this advantage would be absent in agn systems , which for the most part do not undergo significant state transitions on an observable timescale . using a monte carlo ray - tracing code in the kerr metric , we have explored a variety of models for x - ray - emitting coronae attached to accreting stellar - mass black holes in the hard / spl states or to radio - quiet agn . over a wide range of model parameters , the polarization swings from parallel to the disk plane ( horizontal " ) at low energy to perpendicular ( vertical " ) at high energy . the location of this transition is generically at an energy a few times that of the highest temperature found in the disk ; in the case of stellar - mass bhs , that means @xmath156 kev ; in agn , @xmath157 ev . the detailed properties of this transition , and the polarization amplitude at the low- and high - energy limits , provide the observer with information about the bh s mass , spin , inclination , and accretion rate , as well as the coronal properties : degree of ( in)homogeneity , vertical scale height , temperature , optical depth , and covering fraction . we do not claim that a single or even multiple polarization observations of a given source will unambiguously measure all these model parameters . what does seem possible , even with relatively low precision data , is to rule out large regions of parameter space , and in so doing challenge a number of the traditional paradigms and toy models for bh accretion geometry . for example , if we were to observe an upper limit of @xmath158 at @xmath82 kev from a stellar - mass system with known binary inclination @xmath159 , we would have only two choices : to abandon any form of smooth corona geometry in favor of a highly asymmetric , inhomogeneous accretion flow , or to posit a misalignment between the binary orbit and the inner accretion disk ( thus giving lower inclination and polarization than expected ) . for agn sources , where the expected disk inclination is lower ( @xmath155 ) , a null polarization measurement would rule out a smooth sandwich corona for anything but a nearly face - on system . as with any theoretical model based on a large number of parameters , we expect that certain physical properties of the bh system will be more easily constrained than others . for example , even if we assume prior knowledge of the observer inclination , bh mass and accretion rate , as well as the temperature , density , and geometry of the corona , figure [ wedge_spin ] suggests that , with a polarization sensitivity of @xmath3 , one could only just distinguish between a schwarzschild and extreme kerr bh in the hard state . however , this also means that even if we have relatively poor constraints on the spin parameter ( which is generally the case for most galactic bhs ) , we should still be able to measure other parameters robustly . by measuring the power - law slope of the spectrum around 10 kev , one could estimate the compton @xmath84-parameter with reasonable accuracy , but still face a degeneracy between the corona temperature and optical depth . again assuming a known inclination and corona geometry , we see from figure [ wedge_taut ] that this degeneracy could be broken with a polarization observation . as another example , if the coronal properties are well - known ( perhaps through more detailed spectral observations over a greater range of energies , which could allow an independent determination of the electron temperature ) , the polarization could be used to determine the disk inclination . while the inclination could in principle be measured directly from the low - energy polarization @xcite , first - generation polarimeters are likely not going to be very sensitive below @xmath160 kev , where in any case , magnetic turbulence in the disk may reduce the net polarization via faraday rotation @xcite . yet as we see from figure [ wedge_grid ] , a single high - energy polarization measurement at @xmath82 kev could give the inclination angle , assuming the other model parameters are known reasonably well . in practice , it is more likely that we could use some independent method ( e.g. optical light curves , fe line fitting , etc . ) to determine the disk inclination , and then use the amplitude of polarization in the @xmath161 kev band to determine the inhomogeneity of the corona . coronae with large - amplitude density fluctuations produce weaker polarization at high energies and a more gradual transition from horizontal to vertical orientation , as shown in figures [ hotspot_rho ] and [ hotspot_nc ] . @xcite showed that , using basic physical arguments about the covering fraction of the corona and the relative flux in the hard and soft components , the smooth sandwich corona could be distinguished from a patchy corona on the basis of the spectral properties alone . yet in section [ sphere ] , we have shown that the spectrum from a truncated disk around a spherical corona can be nearly indistinguishable from that of a wedge corona above an extended disk . however , the polarization signatures from these two models are quite distinguishable : the spherical geometry leads to a horizontal - vertical transition at lower energy and gives significantly weaker polarization at all energies . as seen in sections [ wedge ] and [ hotspot ] , the sense of rotation in the polarization transition ( i.e. , @xmath71 or @xmath162 ) contains additional information that constrains the corona scale height and level of inhomogeneity . the angle of polarization at low energies defines the projection of the disk plane on the sky , but not the direction of the bh spin ( as in many binary systems , the line of nodes is uniquely determined , but it is impossible to know which is the ascending node ) . from observations of the thermal state , we can use the sense of the rotation toward vertical polarization to determine the sense of orbital motion of the disk , thus defining the @xmath71 direction . given that result , the sign of @xmath30 in the coronal state can constrain the scale height of a wedge corona ( fig . [ wedge_hr ] ) or the covering fraction of a clumpy corona ( figs . [ hotspot_rho ] and [ hotspot_nc ] ) . additional constraints on clumping properties @xmath100 and @xmath98 can come from the amplitude of polarization above the thermal peak as follows : in stellar - mass systems , we might determine the inclination of the disk from radial velocity observations of the companion . in agn , the inclination may be inferred from the blue edge of a broad iron line @xcite . in either stellar - mass systems or agn , the slope of the hard power - law tail of the x - ray spectrum can give the covering fraction @xmath163 , which is related to @xmath100 and @xmath98 through equation ( [ f_c ] ) . for a given @xmath164 and @xmath163 , the degree of polarization at high energies can then be used to determine the coronal properties @xmath100 and @xmath98 . furthermore , with timing observations we expect an inverse correlation between the amplitude of hard x - ray luminosity fluctuations and the number of coronal clumps , due simply to poisson statistics . in a related way , polarization could provide valuable insight into the nature of poorly - understood timing phenomena such as quasi - periodic oscillations ( qpos ) . the qpos observed in stellar - mass bhs are most prominent in the hard x - rays ( @xmath165 kev ) , even in the spl state when most of the flux is at lower energies @xcite , suggesting that the qpos come from the hot corona . by comparing the polarization signal with the timing properties at different epochs , we should be able to constrain the location , size , and coherence of the regions from which qpos originate . for example , if we find that periods of large amplitude timing fluctuations correlate with low - amplitude polarization , then it is quite likely that both features are caused by a relatively small number of hot , dense clouds orbiting above the inner disk . lastly , in section [ agn ] , we showed how the special properties of agn could be exploited through x - ray spectropolarimetry . because the seed photons in such systems are at much lower energies , more scattering events are required to inverse compton scatter the seed photons to x - ray energies . additionally , because the relatively cool disk is largely opaque to absorption below @xmath82 kev , the range of available scattering angles is more restricted . together , these features particular to agn systems lead to a stronger ability to discriminate between different coronal covering fractions , as well as the characteristic size and density of the scattering clouds . combining spectral and polarization information , it should be possible to determine the coronal covering fraction and the disk inclination independently with moderate precision . in contrast to the galactic bhs , in the agn case it is unlikely that we will be able to observe a single system in multiple accretion states , yet the much longer time scales should allow for more detailed comparisons between polarization and variability . we would like to thank tim kallman , jean swank , and shane davis for helpful discussions and comments . this work was supported by the chandra postdoctoral fellowship program grant pf7 - 80051 ( jds ) and nsf grant ast-0908336 ( jhk ) . 99 agol , e. , & krolik , j. h. 2000 , apj , 528 , 161 bardeen , j. m. , press , w. h. , & teukolsky , s. a. 1972 , apj 178 , 347 bellazzini , r. , et al . 2006 , nucl . meth . , a560 , 425 black , j. k. , et al . 2003 , nucl . , 513 , 639 blandford , r. d. , & begelman , m. c. 2004 , mnras , 349 , 68 carter , b. 1968 , phys . rev . , 174 , 1559 chandrasekhar , s. 1960 . _ radiative transfer _ , dover , new york connors , p. a. , & stark , r. f. 1980 , nature , 269 , 128 connors , p. a. , piran , t. , & stark , r. f. 1980 , apj , 235 , 224 costa , e. , et al . 2008 , proc . spie , vol . 7011 - 15 , [ arxiv:0810.2700 ] cunningham , c. t. 1976 , apj , 208 , 534 davis , s. w. , blaes , o. m. , hirose , s. , krolik , j. h. 2009 , apj 703 , 569 dolence , j. c. , gammie , c. f. , moscibrodzka , m. , leung , p. k. 2009 , [ arxiv:09090708 ] done , c. , & zycki , p. t. 1999 , mnras , 305 , 457 done , c. , & trigo , m. d. 2009 , mnras submitted , [ arxiv:0911:3243 ] dovciak , m. , karas , v. , & matt , g. 2004 , mnras , 355 , 1005 dovciak , m. , muleri , f. , goodmann , r. w. , karas , v. , & matt , g. 2008 , mnras submitted , [ arxiv:0809.0418 ] elvis , m. , et al . 1978 , mnras , 183 , 129 esin , a. a. , et al . 2001 , apj , 555 , 483 gierliski , m. , et al . 1997 , mnras , 288 , 958 gierliski , m. , & done , c. 2005 , mnras , 347 , 885 haardt , f. , & maraschi , l. 1993 , apj , 413 , 507 haardt , f. , maraschi , l. , & ghisellini , g. 1994 , apj , 432 , l95 ingram , a. , done , c. , & fragile , p. c. 2009 , mnras , 397 , l101 jahoda , k. , black , k. , deines - jones , p. , hill , j. e. , kallman , t. , strohmayer , t. , & swank , j. 2007 , [ arxiv:0701090 ] laor , a. , netzer , h. , & piran , t. 1990 , mnras , 242 , 560 li , l .- x . , narayan , r. , & mcclintock , j. e. 2008 , apj submitted , [ arxiv:0809.0866 ] makishima , k. , et al . 2008 , pasj , 60 , 585 matt , g. , fabian , a. c. , & ross , r. r. 1993 , mnras , 264 , 839 mcclintock , j. e. , et al . 2001 , apj , 555 , 477 miller , j. m. , et al . 2006 , apj , 653 , 525 mushotzky , r. f. , done , c. , & pounds , k. a. 1993 , ara & a , 31 , 717 nandra , k. , et al . 1991 , mnras , 248 , 760 noble , s. c. , krolik , j. h. , & hawley , j. f. 2008 , apj submitted , [ arxiv:0808.3140 ] novikov , i. d. , & thorne , k. s. 1973 , in _ black holes _ , ed . c. dewitt & b. s. dewitt ( new york : gordon and breach ) pietrini , p. & krolik , j.h . 1995 , apj , 447 , 526 poutanen , j. , krolik , j. h. , & ryde , f. 1997 , mnras , 292 , l21 poutanen , j. , & fabian , a. c. 1999 , mnras , 306 , l31 reis , r. c. , fabian , a. c. , & miller , j. m. 2009 , mnras accepted , [ arxiv:0911.1151 ] remillard , r. a. , & mcclintock , j. e. 2006 , ara & a , 44 , 49 reynolds , c. s. , & nowak , m. a. 2003 , phys . reports , 377 , 389 rybicki , g. b. , & lightman , a. p. 1979 , _ radiative processes in astrophysics _ ( new york : wiley - interscience ) schnittman , j. d. , & krolik , j. h. 2009 , apj 701 , 1175 schnittman , j. d. , & krolik , j. h. 2010 , in preparation shakura , n. i. , & sunyaev , r. a. 1973 , a&a , 24 , 337 shen , shiyin , white , s.d.m . , mo , h.j . , voges , w. , kauffmann , g. , tremonti , c. & anderson , s.f . 2006 , mnras 369 , 1639 shimura , t. , & takahara , f. 1995 , apj , 445 , 780 stark , r. f. , & connors , p. a. 1977 , nature , 266 , 429 stern , b. e. , poutanen , j. , svensson , r. , sikora , m. , & begelman , m. c. 1995 , apj , 449 , l13 sunyaev , r. a. , & titarchuk , l. g. 1985 , a&a , 143 , 374 swank , j. , et al . , in _ x - ray polarimetry : a new window in astrophysics _ , rome 2009 , edited by r. bellazzini , e. costa , g. matt and g. tagliaferri titarchuk , l. , & shrader , c. 2002 , apj , 567 , 1057 turolla , r. , zane , s. , & titarchuk , l. 2002 , apj , 576 , 349 walker , m. , & penrose , r. 1970 , commun . phys . , 18 , 265 zdziarski , a. a. , & gierliski , m. 2004 , prog . theor . phys . suppl . , 144 , 99 zdziarski , a. a. , gierliski , m. , rao , a. r. , vadawale , s. v. , & mikalojewska , j. 2005 , mnras , 360 , 825

we study the collective dephasing process of a system of non - interacting atomic qubits , immersed in a spatially uniform magnetic field of fluctuating intensity . the correlation properties of bipartite states are analysed based on a geometric representation of the state space . particular emphasis is put on the dephasing - assisted generation of states with a high correlation rank , which can be related to discord - type correlations and allow for direct applications in quantum information theory . finally we study the conditions that ensure the robustness of initial entanglement and discuss the phenomenon of time - invariant entanglement .

1602.08973

ensembles of trapped , laser - cooled atomic particles provide some of the best - controlled experimental platforms to study quantum dynamics , to engineer effective interactions , or to generate specific quantum states @xcite . an essential requirement for such levels of control is the efficient isolation of the system from its environment , whose detrimental influence leads to the decay of coherent superpositions @xcite . the loss of coherence often also implies the loss of quantum correlations , such as entanglement , which are required , e.g. , to process quantum information . one persistently dominant source of error , common to most experiments on trapped atomic particles , is caused by intensity fluctuations of external electromagnetic fields , which are needed to lift degeneracies , to compensate background fields , or to manipulate the quantum state of the system @xcite . since such fields are typically generated by large coils outside the vacuum chamber , the resulting field is spatially homogeneous along the positions of the trapped particles . the unavoidable fluctuations of the field strength therefore lead to a _ collective _ dephasing process , which is formally described by an ensemble average over the fluctuating parameter @xcite . the collective nature of this noise provides new possibilities to protect coherent superpositions @xcite , and , as demonstrated in a recent experiment @xcite , to generate robust , strongly correlated , albeit separable quantum states . in this article we study the impact of the collective dephasing process on different types of correlations inscribed into quantum states , using the analytical description of the collective dephasing dynamics outlined in @xcite . our analysis focuses on the correlation rank , which in turn entails direct consequences for the discord - type correlations @xcite . specifically , in sec . [ sec : generate ] , we study how strongly correlated two - qubit states can be generated via collective dephasing , as a function of tunable external control parameters , e.g. , the magnetic field direction . we discuss specific applications of the produced quantum states in the context of quantum information protocols . in sec . [ sec : generalk ] we follow a complementary approach and analyse the protection of existing correlations during the collective dephasing process . we further discuss robust conditions that lead to the striking phenomenon of _ time - invariant _ entanglement : the perfect conservation of the initial entanglement , even for states whose purity is reduced due to the dephasing process . we begin by reviewing different notions of correlations in quantum states that will become relevant in the course of this article . _ quantum entanglement _ captures the non - classical correlations of a quantum state in a composite hilbert space @xmath0 , i.e. , it expresses the inability to characterize the full quantum state @xmath1 via a classical probability distribution @xmath2 and local density operators @xmath3 and @xmath4 on the local hilbert spaces @xmath5 and @xmath6 , respectively . in particular , a quantum state is _ separable _ ( i.e. , not entangled ) if it can be expressed as a convex linear combination of product states , @xmath7 determining whether a given mixed quantum state is entangled , or evaluating an appropriate measure to quantify entanglement , is generally a very hard task . only for the special case of two - qubit systems ( @xmath8 ) , an algebraic expression that exactly quantifies the entanglement of arbitrary mixed states is available @xcite . this measure , the _ concurrence _ , is determined on the basis of the eigenvalues @xmath9 of @xmath10 , labelled in decreasing order , where the spin - flipped state @xmath11 is obtained by collective application of the pauli matrix @xmath12 to @xmath13 , and complex - conjugation is performed in the computational basis . the concurrence of the state @xmath1 is then given by @xmath14 . for higher dimensional problems , only algebraic lower bounds are available @xcite . the _ quantum discord _ describes the disturbance of local measurements on quantum states of composite systems @xcite . as we will see in the following , these features related to the quantum - mechanical measurement process can be expressed via non - vanishing commutators , and can only occur in correlated quantum states . they can , however , also manifest in separable states that exhibit only classical correlations , and therefore describe a more general type of quantum properties than entanglement . formally , a state @xmath1 has zero discord if it can be written as @xcite @xmath15 where @xmath16 is an orthonormal basis of @xmath17 . this definition is equivalent to the following : a state @xmath1 has zero discord if and only if there exists a non - selective local projective measurement on @xmath17 that leaves the state invariant , i.e. , @xmath18 , where @xmath19 . the definition presented here considers measurements on @xmath5 and is straight - forwardly extended to measurements on @xmath6 . due to the asymmetry of the definition , one should always specify in which subsystem measurements are performed , when discussing quantum discord . notice that every zero - discord state is separable but the converse is not true . the two concepts coincide only in the case of pure states . in contrast to entanglement , local operations on one of the subsystems can generate discord @xcite , which confirms that discord is not a proper measure for correlations . to quantify the correlations of a bipartite quantum state , we employ the rank of an appropriately constructed correlation matrix ( the _ correlation rank _ ) , which is the minimal number of bipartite operator products needed to describe the density operator @xcite . for the formal definition , we write the density operator @xmath1 in an arbitrary basis of local hermitian operators @xmath20 and @xmath21 @xcite : @xmath22 where @xmath23 . here @xmath24 is the correlation matrix , a real - valued @xmath25-dimensional matrix whose rank @xmath26 is then called the correlation rank @xcite . employing a singular value decomposition , we find non - zero singular values @xmath27 and orthogonal matrices @xmath28 and @xmath29 such that @xmath30 . we obtain @xmath31 where @xmath32 and @xmath33 . the above decomposition can be regarded as a schmidt decomposition of a density matrix in terms of local operator bases @xcite . a more familiar application of the schmidt decomposition is known for pure bipartite quantum states , which are decomposed in terms of local vectors . the associated singular value decomposition yields the schmidt rank , which quantifies how entangled a pure state is . consequently , the correlation rank stands in close analogy to the schmidt rank . in general , the correlation rank does not quantify entanglement but rather total correlations , i.e. , any incompatibility with an uncorrelated product state , without an explicit distinction between classical and quantum nature of the correlations . in the special case of a pure state one obtains @xmath34 , where @xmath35 denotes that state s schmidt rank @xcite . conclusions about the local quantum nature of the state can be drawn from the correlation rank by realizing that its maximal value is bounded from above for all zero - discord states . generally , the correlation rank can not exceed @xmath36 , where @xmath37 . however , as can be seen from the definition ( [ eq.zerodiscord ] ) , states of zero discord with respect to measurements on @xmath38 are represented in terms of local projectors @xmath39 onto orthogonal subspaces . since there exist no more than @xmath40 orthogonal subspaces in @xmath38 , the correlation rank of zero - discord states is bounded by @xmath41 . this is directly related to the observation that the operators @xmath42 , which appear in eq . ( [ eq.svd ] ) , can be used to assess the discord of @xmath1 : the state @xmath1 has zero discord ( with respect to measurements on @xmath5 ) if and only if all of the @xmath42 commute @xcite . while local operations can not increase the correlation rank @xmath26 @xcite , they can change the commutativity of the local operators @xmath42 , and thereby generate discord @xcite without actually generating correlations @xcite . the correlation rank allows to distinguish separable states with high correlation rank ( @xmath43 ) from those that can be generated from states of zero discord via local operations , as characterized by a low correlation rank ( @xmath41 ) . separable operations of the form @xmath44 , with local operations @xmath45 and @xmath46 , can generate classical correlations among the particles and , consequently , are able to increase the correlation rank @xcite . the collective dephasing operation to be discussed in this article represents such a separable operation . the density operator of any bipartite system can be represented in terms of the operator bases @xmath47 , where @xmath48 denote vectors whose entries are the generators of su(@xmath40 ) , and @xmath49 denote the respective identity operators . we obtain the state s _ fano form _ as @xcite : @xmath50 where @xmath51 and @xmath52 are the ( generalized ) bloch vectors of the reduced subsystems , and @xmath53 is a real @xmath54 matrix that describes the correlations between the subsystems . because the state is completely characterized by @xmath51 , @xmath52 and the @xmath53 matrix , throughout this article we will use the compact notation @xmath55 @xcite . from the representation ( [ eq : fano - form ] ) it immediately follows that the correlation matrix of @xmath1 is given by @xmath56 whose rank rk is @xcite @xmath57 while for a rigorous proof of the above identity we refer to ref . @xcite , we remark that the result can be obtained via block - diagonalization of the matrix @xmath58 , where @xmath59 and @xmath60 are rank-4 matrices and , in this case , @xmath61 is a @xmath62 matrix called the schur complement ( of the submatrix @xmath63 ) . in the remainder of this article , we mostly employ the reduced bloch vectors and the @xmath53-matrix to investigate the impact of the collective dephasing process , which allows for an intuitive geometric description . a tool that we often employ to simplify our analysis are local _ unitary _ transformations , since these affect neither the state s entanglement properties ( by definition ) nor its correlation rank ( as we show explicitly later ) . following @xcite , we consider transformations @xmath64 such that @xmath65 . for every unitary transformation @xmath66 , there exists a rotation @xmath67 on the respective ( generalized ) bloch sphere such that @xmath68 . we then obtain the following transformation rules : @xmath69 if two matrices @xmath53 and @xmath70 can be transformed into each other by such an operation , we write @xmath71 . in particular , we can always diagonalize the @xmath53 matrix by applying unitary transformations to the underlying quantum state . we now show that such transformations do not change the rank of the correlation matrix . if @xmath72 , we have @xmath73 the rank of @xmath74 is defined by the image of the linear map @xmath75 : @xmath76 since @xmath67 and @xmath77 are orthogonal matrices , we have @xmath78 since the maps @xmath79 and @xmath80 are bijective , we finally obtain @xmath81 in the following , we analyse the impact of the collective dephasing process , to be introduced in the next section , on the different concepts that were introduced in this section , all of which are intimately related to correlations in quantum states . collective dephasing describes the dynamics of @xmath82 particles that share the same environment , but experience no direct particle - particle interactions among each other . the environment does not induce exchange of energy with the particles , and hence does not lead to dissipation , but rather to pure dephasing , i.e. , loss of phase relations without loss of populations . a physically relevant example of such a scenario was already mentioned in the introduction : when an ensemble of atomic dipoles is spatially confined in a region where the electromagnetic field is homogeneous , all dipoles share the same transition frequency . the fluctuations of the field then lead to a collective dephasing process @xcite . the quantum state that predicts the measurement results after many experimental repetitions is described by the ensemble average over the actual realisations of these fluctuations @xcite . let us consider magnetic dipoles , described by two - level systems , in a constant magnetic field @xmath83 . the hamiltonian @xmath84 of the @xmath82-particle system is given by @xmath85 , with @xmath86 , where here @xmath87 is a vector of pauli matrices on the hilbert space of the @xmath88th particle , and further constants are absorbed into @xmath89 . identifying @xmath90 , with @xmath91 , we write @xmath92 for a fixed magnetic field strength @xmath93 , the fully coherent dynamics of the total system is given by @xmath94 where @xmath95 is the @xmath82-particle density operator at time @xmath96 . the dephasing process is now described by an ensemble average over the fluctuations of @xmath83 . the physically intuitive reason for such a description is the necessity to repeat experiments many times to produce significant statistics for the efficient estimation of the populations . we make the following assumptions on the fluctuations of @xmath83 : * the _ direction _ @xmath97 of the magnetic field is constant , and the fluctuations only affect the field strength @xmath93 ; * the magnetic field may change from experimental run to the next , but within each run we assume the magnetic field to be constant . both of these assumptions can be motivated at the hand of state - of - the - art experiments on cold atoms or trapped ions : the external field influences the energy splitting of the atomic two - level systems through a zeeman effect , as described above , and , thus , the field is chosen strong enough to dominate over the effect of possible stray fields . the field therefore has a fixed direction ( satisfying the first of the two above assumptions ) , but small fluctuations of the supplying currents will produce weak intensity fluctuations of @xmath93 , on top of a relatively large mean value . the mean value determines the time scale of the atomic evolution , which is therefore much faster than the time scale on which fluctuations occur ( satisfying the second assumption ) . characterizing the intensity fluctuations with the probability distribution @xmath98 , the collective dephasing dynamics is described by @xmath99 in @xcite , the above integral was solved analytically without further assumptions , and in the following we will recall the resulting solution and some of its properties . the transient time evolution of a quantum state under collective dephasing ( [ eq.integral ] ) is determined by the characteristic function @xmath100 of the probability distribution @xmath98 . introducing the matrix elements @xmath101\end{aligned}\ ] ] and the hermitian operators @xmath102v_s^\dagger , \end{aligned}\ ] ] where @xmath103 , and @xmath104 represents the permutation @xmath105 in the hilbert space of @xmath82 qubits , we can express the collective dephasing dynamics with the following map @xcite : @xmath106 the matrix of elements @xmath107 is positive semi - definite , and can be diagonalized to obtain the canonical kraus form of the above map @xcite . one can further show @xcite that the map ( [ eq.colldeph ] ) is always completely positive and trace preserving @xcite . the obtained dynamics therefore exhibits the properties of dynamical maps associated with the dynamics of open quantum systems , indicating the equivalence of ensemble average approaches with open - system treatments based on a microscopic model for the environment and its coupling to the system @xcite . in the context of the present article , we are only interested in the asymptotic limit , which is described , _ independently _ of @xmath98 ( assuming that @xmath98 is absolutely integrable ) , by @xcite @xmath108 in performing this limit , we assume that the time evolution of the atomic ensemble is recorded for an interval long enough that the atomic evolution has reached its stationary state , but not too long to compromise with the assumption that the field strength can be considered constant during the evolution . in @xcite the map ( [ eq.colldeph ] ) was shown to conserve the trace of the @xmath53 matrix for bipartite systems , as defined in ( [ eq : fano - form ] ) . this integral of motion can , in fact , be understood as the manifestation of the more general conservation of angular momentum in the special case of @xmath109 . to see this , recall that the total spin @xmath110 commutes with the hamiltonian for every choice of the magnetic field @xmath83 , hence the expectation value of @xmath111 is conserved , even in the presence of the ensemble average over the fluctuations of the magnetic field . we express the squared total spin as @xmath112 with @xmath113 , and the index @xmath114 labels the spatial directions . we generalize the definition of the @xmath53 matrix to @xmath115 note that this definition reduces to the bipartite @xmath53 matrix , as introduced in ( [ eq : fano - form ] ) , in the special case of @xmath109 . the total angular momentum is expressed via the quantum mechanical expectation value , using eqs . ( [ eq : total - spin ] ) and ( [ eq.generalbeta ] ) , @xmath116 and from the time - independence of @xmath117 , we obtain the conservation of the trace of the generalized @xmath53 matrix : @xmath118 let us discuss the description of the collective dephasing of an initial two - qubit state @xmath1 into the stationary state @xmath119 , using the map ( [ eq : solution - stationary ] ) for @xmath109 : @xmath120 = \sum_{i=0}^2 \theta_i \rho \theta_i.\end{aligned}\ ] ] based on ( [ eq.krausops ] ) , the kraus operators @xmath121 can be explicitly given as @xcite @xmath122 to efficiently describe the impact of the collective dephasing on an arbitrary initial state @xmath55 , we now derive a description of its map on the level of the reduced bloch vectors @xmath51 and @xmath52 , together with the @xmath53 matrix . we first express the @xmath53-matrix of the initial state in terms of a diagonal , singular value decomposition [ recall ( [ eq.transform ] ) ] , as @xmath123 where @xmath124 and @xmath125 are normalized vectors , @xmath126 are non - negative real numbers , and the tensor product is defined element - wise as @xmath127 . by direct application of the operators @xmath121 and of the properties of the scalar product , ( [ eq : stat - bipartite ] ) leads to @xcite @xmath128\cdot \boldsymbol{\sigma } \otimes [ \bi w - ( \bi w \cdot \bi n ) \bi n]\cdot \boldsymbol{\sigma } \rbrace .\end{aligned}\ ] ] we can thus formulate _ transformation rules _ for vectors , and tensor products thereof , to express how they are altered by the collective dephasing , as a function of the direction @xmath97 of the fluctuating , external field : @xmath129 \otimes [ \bi w - ( \bi w \cdot \bi n ) \bi n ] \rbrace . \label{eq : corr - deph - beta}\end{aligned}\ ] ] the local vectors that determine the decomposition on the right - hand - side of ( [ eq : corr - deph - beta ] ) form an orthogonal basis consisting ( in the first subsystem ) of the direction @xmath97 of the magnetic field , the vector orthogonal to the plane spanned by @xmath130 and @xmath131 , and the vector orthogonal to these ; the same holds for the second subsystem when @xmath132 is replaced by @xmath133 . the decisive parameters are the angles @xmath134 and @xmath135 . notice that if either one of @xmath132 and @xmath133 is parallel or orthogonal to @xmath97 , some of the terms in ( [ eq : corr - deph - beta ] ) disappear . for now , we assume that @xmath136 and @xmath137 . in this case , we can introduce two orthonormal bases of @xmath138 as @xmath139 allowing us to re - express ( [ eq : corr - deph - beta ] ) as @xmath140 with the coefficients @xmath141 in this section we investigate to what extent the collective dephasing map ( [ eq : solution - stationary ] ) can generate or increase the correlations between the subsystems . the present section extends the analysis of few special cases provided in the theoretical treatment of the experiment reported in @xcite to a complete picture . we know the map is separable , hence it can not create entanglement ; however , it contains stochasticity , and therefore can create classical correlations between subsystems , thereby increasing the correlation rank @xmath26 of the initial state . for this reason , we focus on the analysis of the correlations in the asymptotic state , based on the correlation rank . as discussed in section [ sec.correlationsintro ] , a correlation rank of @xmath142 can be interpreted as a witness for non - zero discord . furthermore , we can exclude that the thereby detected discordant states can be generated by applying a local operation to a zero - discord state @xcite . for two - level systems we have @xmath143 , and , thus , the maximal correlation rank for zero - discord states is @xmath144 . states with @xmath145 or @xmath146 are considered strongly correlated , since their correlations are beyond the reach of any zero - discord states , and neither can be attained by states whose discord was generated by a local operation . we begin by considering an initially completely uncorrelated state , i.e. , a product state @xmath147 . in the fano form ( [ eq : fano - form ] ) this reads @xmath148 this means that the initial @xmath53 matrix is @xmath149 and the initial rank ( [ eq : corr - rank ] ) is @xmath150 , which is consistent with the state having no correlations @xcite . application of the collective dephasing map , assuming that @xmath151 does not coincide with the direction of the bloch vectors @xmath51 and @xmath52 of the respective reduced systems , yields @xmath152 with the set of orthonormal vectors ( [ eq : basis ] ) @xmath153 and coefficients ( [ eq : coeff ] ) @xmath154 the set of accessible final states is described by four real parameters , namely the norms @xmath155 of the reduced bloch vectors and their angles @xmath156 with the magnetic field direction . within the set of density matrices , which is a fifteen - dimensional real space , this represents a measure - zero set , therefore we can not synthesize arbitrary states by adjusting the parameters of the collective dephasing map or of the initial state . in the following , however , we specify conditions that lead to a given value of the correlation rank @xmath26 after collective dephasing . as discussed before , the vectors @xmath157 , defined in ( [ eq : basisr ] ) , span an orthonormal basis of @xmath138 . following sec . [ sec : unitary - equivalence ] , we can apply a local unitary operator @xmath158 ( associated with a rotation matrix @xmath77 , such that @xmath159 ) which does not change the properties of interest . we apply this rotation only to the second subsystem , as described by the unitary operator @xmath160 . this transforms the correlation matrix @xmath161 of the asymptotic state ( [ eq.asymptoticstate ] ) to the matrix @xmath162 , where @xmath163 , expressed in the basis @xmath157 , reads @xmath164 when expressed in the same basis , the bloch vectors after dephasing are given by @xmath165 the final correlation rank reads , using ( [ eq : corr - rank ] ) , @xmath166 the correlation rank of the final state can only have two values , determined by the relative orientations of the reduced bloch vectors and the direction of the magnetic field @xcite : * @xmath167 , if the magnetic field is parallel to one of the reduced bloch vectors : @xmath168 ; * @xmath169 , if the magnetic field has a different direction than both reduced bloch vectors : @xmath170 . the physical interpretation is immediate if we realise that , when the bloch vector of a subsystem coincides with the magnetic field direction , that subsystem is in an eigenstate of the local hamiltonian @xmath171 , and is consequently invariant under the action of the map . in this case , this atom can be treated separately , and the collective dephasing acts only on the remaining atoms , whose bloch vector differs from @xmath97 . let us suppose that @xmath172 : the atom described by @xmath173 is no longer affected by the collective dephasing process and the final state can be written as @xmath174 which is an uncorrelated product state with correlation rank @xmath175 . as is best seen from the integral representation ( [ eq.integral ] ) , this observation is easily generalized for systems of @xmath176 atoms . in general , for product states involving arbitrary local states @xmath177 of qubits @xmath88 that satisfy @xmath178=0,\end{aligned}\ ] ] the dephasing operation factorizes : @xmath179 where @xmath1 is an arbitrary quantum state of the remaining qubits . the states that satisfy the factorisation condition ( [ eq.factorizationcondition ] ) encompass all incoherent mixtures of eigenstates of @xmath180 and the identity operator . in short , application of ( [ eq : solution - stationary ] ) to an uncorrelated state produces a state with high correlation rank , as long as @xmath170 , but can not reach the maximal value of @xmath146 . yet , is it possible to transform the resulting @xmath145 state into an @xmath146 state by a consecutive , second application of the collective dephasing map ? since we consider collective dephasing to the asymptotic state , the second application would not have any effect unless we _ change _ the direction of the external field . this can be shown as follows : the application of the collective dephasing map along the direction @xmath181 yields a state with a @xmath53 matrix @xmath182 .\ ] ] if we define @xmath183 , the correlation rank of the state is given by @xmath184 , where @xmath185 + \epsilon^{\bi m}(\bi e_1 \otimes \bi e_1 ) + \epsilon^{\bi m } ( \bi e_3 \otimes \bi e_3 ) \end{aligned}\ ] ] where @xmath186 is non - zero if @xmath170 . we can now observe that , when @xmath187 , the term proportional to @xmath188 vanishes and @xmath189 , yielding @xmath190 ; alternatively , we can solve the equation @xmath191 with the constraint @xmath192 to obtain @xmath193 $ ] , which yields again @xmath190 . finally , one can show that the matrix @xmath74 is not of full rank , e.g. by solving @xmath194 as a function of @xmath195 given the parameter @xmath188 . in any case , using linear algebra or by direct computation , one can verify that there exists at most a measure - zero set of directions , other than the one of the first dephasing process , along which the correlation rank will not increase . therefore , an @xmath175 state can be converted into an @xmath146 state by a twofold application of the collective dephasing map , if the direction of the magnetic field is different for the second dephasing . after studying the influence of the collective dephasing on an initially uncorrelated state @xmath197 , we now turn to the discussion of an initial state with low correlation rank @xmath198 . this state is considered weakly correlated since it contains correlations that are compatible with either a state of zero discord , or a non - zero discord state that can be created from a state of zero discord with a local operation recall the discussion at the end of section [ sec.correlationsintro ] . we consider states with maximally mixed reduced subsystems , i.e. , systems with vanishing reduced bloch vectors , which can be written as @xmath199 . since these states are diagonal in the basis of bell states , they are also called _ bell - diagonal _ states . in the fano form ( [ eq : fano - form ] ) they are written as @xcite @xmath200 where @xmath201 and the initial @xmath53 matrix was expressed in terms of its singular value decomposition ( [ eq : svd ] ) , as @xmath202 . positivity of the state @xmath203 requires that @xmath204 . states of this form have always zero discord @xcite , as can be easily seen based on the cummutativity of the local operators @xcite , as discussed in sec . [ sec.correlationsintro ] . application of the collective dephasing map ( [ eq : corr - deph - beta ] ) produces another bell - diagonal state , @xmath205 , where the correlation matrix @xmath161 can be written ( analogously to the case discussed before ( [ eq.diagonalbeta ] ) , and possibly after suitable , local orthogonal transformations that do not alter the correlation properties ) as @xmath206 in order to find the correlation rank of the final state , by virtue of ( [ eq : corr - rank ] ) , we need to determine @xmath207 . except for the trivial case @xmath208 , we notice that also in this case the rank depends on the geometric features of the state , namely on the angle between the magnetic field direction and the left- and right - singular vectors of @xmath209 : the state has correlation rank @xcite * @xmath167 if either ( @xmath210 and @xmath211 ) , or ( @xmath212 and @xmath213 ) ; * @xmath214 if @xmath210 or @xmath212 ( but neither @xmath213 nor @xmath211 ) ; * @xmath169 if @xmath215 or @xmath216 ( but neither @xmath210 nor @xmath212 ) ; * @xmath217 otherwise . a weakly correlated ( @xmath218 ) bell - diagonal state can therefore be transformed into a state with maximal correlation rank ( @xmath146 ) , provided that the magnetic field direction does not coincide with some very specific choices , determined by the geometric characterisation of the initial state . * generation of werner states.*an important class of states for many applications of quantum information theory is given by the family of werner states , @xmath219 with the singlet state @xmath220 , and @xmath221 @xcite . written in the fano form ( [ eq : fano - form ] ) , this state has @xmath222 . collective dephasing can generate such a state from an @xmath196 bell - diagonal state of the form ( [ eq : class - corr - state ] ) when we choose the values for @xmath223 , @xmath224 and @xmath225 that solve the following system of equations : @xmath226 where @xmath227 and @xmath228 . summation of the two equations yields @xmath229 which proves that the system of equations ( [ eq : wern - state - sys ] ) admits solutions in the variables @xmath230 only if @xmath231 . in particular , when @xmath232 , the solutions lie on the lines @xmath233 ( see figure [ fig : werner - minus ] , left ) . the solutions are then found as : @xmath234 or @xmath235 we remark here that we have to simultaneously respect the conditions @xmath231 and @xmath204 ( for positivity of the state ) , which leads to the observation that , based on the present approach , it is impossible to generate werner states with @xmath236 . in fact , the werner states are separable precisely when @xmath237 and entangled when @xmath238 . this is consistent with the fact that the initial state was separable and that the map can not create entanglement . 0.48 ) for the werner state ( [ eq : werner - state ] ) ( left ) , and the system ( [ eq : general - werner - sys ] ) for the werner - like state ( [ eq : werner - like ] ) ( right ) , in the case @xmath239 . the blue ( darker ) circles are the contour lines of @xmath240 [ first equation in ( [ eq : wern - state - sys ] ) and ( [ eq : general - werner - sys ] ) ] , the orange ( lighter ) ones , those of @xmath241 ( second equation ) . the solutions of the two systems ( [ eq : wern - state - sys ] ) and ( [ eq : general - werner - sys ] ) are then the intersections of the curves.,title="fig : " ] 0.48 ) for the werner state ( [ eq : werner - state ] ) ( left ) , and the system ( [ eq : general - werner - sys ] ) for the werner - like state ( [ eq : werner - like ] ) ( right ) , in the case @xmath239 . the blue ( darker ) circles are the contour lines of @xmath240 [ first equation in ( [ eq : wern - state - sys ] ) and ( [ eq : general - werner - sys ] ) ] , the orange ( lighter ) ones , those of @xmath241 ( second equation ) . the solutions of the two systems ( [ eq : wern - state - sys ] ) and ( [ eq : general - werner - sys ] ) are then the intersections of the curves.,title="fig : " ] the same procedure can be followed to produce werner - like states @xmath242 where @xmath243 is one of the other bell states : @xmath244 these states have @xmath53 matrices with the eigenvalues @xcite : @xmath245 the values of @xmath224 and @xmath225 that yield those states are found by solving the system @xmath246 in fact , we will show in the next section that the @xmath53 matrix of the asymptotic state always has one non - degenerate eigenvalue , with the associated eigenspace spanned by @xmath97 , and one doubly degenerate eigenvalue , with the plane orthogonal to @xmath97 as eigenspace . thus , the three werner - like states are distinguished by the choice of the direction @xmath97 of the magnetic field . in all previous expressions we had arbitrarily chosen to identify @xmath97 with the @xmath247-direction , @xmath248 , thus the non - degenerate eigenvalue always appeared as second diagonal element , corresponding to @xmath249 . in general , we identify @xmath250 in order to generate these werner - like states we then solve the system of equations ( [ eq : general - werner - sys ] ) to find the relative angles between @xmath97 , @xmath132 and @xmath133 , and rotate the basis accordingly , to obtain the target state . in order to solve ( [ eq : general - werner - sys ] ) , the same conditions on @xmath114 hold , but the solutions for @xmath251 now lie on the lines @xmath252 ( see figure [ fig : werner - minus ] , right ) , with @xmath224 as before : @xmath253 or @xmath254 . for the cases with @xmath255 there will be twice as many solutions , which can be easily found numerically . for the genuine ( singlet - based ) werner state it does not matter how we choose to order the eigenvalues of the @xmath53 matrix ( by appropriate choice of the coordinate system relative to the magnetic field orientation ) , since they are all negative . this expresses these states invariance under arbitrary collective unitary rotations of the form @xmath256 , which effectively rotate the coordinate system . in general , the set of @xmath257-invariant states can be identified as a completely time - invariant set of states under the collective , multipartite dephasing operation , for arbitrary directions of the magnetic field @xcite . * generating resource states for entanglement distribution.*separable states can find applications in quantum information protocols such as _ entanglement distribution _ @xcite ; for experimental realisations based on separable states see @xcite . this protocol allows to increase the entanglement between two parties by exchanging a carrier particle , which is not necessarily entangled with the two parties . it is , however , necessary that the combined state contains non - zero discord between the two parties and the carrier . in fact it was recently further shown that discord alone is not sufficient , and that discordant mixtures of two pure product states are not able to distribute entanglement @xcite . by producing states with @xmath258 , we can ensure that these can not be written as mixtures of two product states . let us consider the following example of an initial two - qubit state @xcite @xmath259 with @xmath260 . when @xmath261 , these are the werner - like states , ( [ eq : werner - like ] ) , generated by @xmath249 ; in particular , the protocol can be realised with an initially separable state , which further imposes @xmath262 . as explained in the previous paragraph , these states can be generated from a weakly correlated ( @xmath196 ) state by action of the collective dephasing map . states with other values of @xmath105 are not accessible as the map only produces states with a @xmath53 matrix that contains a doubly - degenerate eigenvalue . in the three - partite scenario considered here , the werner - like state above can be generated on systems @xmath263 if the initial state contains no correlations with system @xmath264 , which is prepared in a maximally mixed state @xmath265 . by virtue of ( [ eq.factorization ] ) the qubit @xmath264 is invariant under the collective dephasing dynamics . the state @xmath266 can thus be generated by collective dephasing of a weakly correlated @xmath196 state of zero discord for the choice of parameters @xmath267 and , e.g. , @xmath268 . for these parameters the same state was shown to be useful for entanglement distribution @xcite . * entanglement activation from uncorrelated states through collective dephasing.*a three - partite scenario is also considered for the _ entanglement activation _ protocol @xcite , where initial discord between systems @xmath269 and @xmath93 is converted into entanglement across the partition @xmath270 , where @xmath74 is an initially independent ancilla system that interacts via a local unitary operation with system @xmath93 . formally , this can be interpreted as a local measurement process of the system @xmath93 , where @xmath74 is a measurement apparatus . thereby , the entanglement of the combined quantum state describing the apparatus and the system before readout ( state reduction ) of the apparatus is then directly linked to the discord of the initial bipartite state of @xmath269 and @xmath93 @xcite . an all - optical experimental realisation of this protocol confirmed this general theoretical result by modelling all three subsystems @xmath269 , @xmath93 and @xmath74 , as polarization qubits @xcite . based on the collective dephasing process , we can now extend this protocol such that also initial zero - discord states or even completely uncorrelated states can be used to generate entanglement . to this end , we subject an initial three - qubit state of the form @xmath271 to a three - partite collective dephasing process in direction @xmath97 , where we choose @xmath272 as an eigenstate of the operator @xmath273 , and @xmath274 may be an arbitrary state of qubits @xmath269 and @xmath93 . while in the previous section we have made use of the local invariance under the dephasing process , see ( [ eq.factorization ] ) , for the trivial case of the identity operator , here we prepare the third qubit in an eigenstate of the local hamiltonian to achieve the same effect . application of the collective dephasing map to the state ( [ eq.intentactstate ] ) yields @xmath275 by following the conditions provided in sections [ sec : prod - state ] and [ sec : class - corr ] , initial product states of bell - diagonal zero discord states @xmath274 can be easily converted into strongly correlated , non - zero discord states @xmath276 . based on the entanglement activation protocol , any local unitary operation on @xmath277 necessarily must generate distillable entanglement across the @xmath270 partition , and the amount of generated entanglement is lower bounded by the discord of the state @xmath276 @xcite . in the present section we only considered the asymptotic mapping of the collective dephasing map ( [ eq : solution - stationary ] ) after long times , but the results hold also for intermediate times @xmath96 , as described by the map ( [ eq.colldeph ] ) . in a recent experiment @xcite , _ local _ noise processes were used to generate discordant states ( see also @xcite ) , whose discord was subsequently activated into entanglement . * local quantum interferometry.*to end this section , we briefly comment on the use of correlated separable states in the context of local precision measurements @xcite . the field of quantum metrology is dedicated to developing methods that allow to estimate unknown parameters with the highest possible precision , often by exploiting the usage of entangled states @xcite . consider a setup , in which two incoming particles are sent into two different arms of an interferometer before being jointly measured . the parameter to be estimated is a local phase shift @xmath278 , imprinted on one of the particles by an unknown hamiltonian @xmath279 ( with fixed , non - degenerate spectrum ) through the unitary operation @xmath280 . in a worst - case scenario , the local hamiltonian may commute with the quantum state , which therefore renders any estimation of the phase shift impossible . this , however , is only possible if the quantum state has zero discord . in general , the worst - case estimation precision of @xmath278 is quantified by a particular measure of discord @xcite . consequently , the successful estimation of the phase in the above scenario requires the presence of non - zero discord between the two particles . using the results provided in sections [ sec : prod - state ] and [ sec : class - corr ] , the required discord can be easily generated by submitting the two parties to a collective dephasing process _ before _ sending them into the interferometer . in the present section , we have discussed the behaviour of the correlation rank under the action of the collective dephasing dynamics . we have seen in section [ sec : prod - state ] that completely uncorrelated initial states ( @xmath175 ) can be transformed into states with @xmath145 by a single application of collective dephasing , or into states of @xmath146 by a double application , provided that the magnetic field direction is changed before the second dephasing . weakly correlated states with @xmath196 are transformed into states with the maximum correlation rank @xmath146 for most choices of the magnetic field direction , as discussed in section [ sec : class - corr ] . since any value of @xmath281 implies the presence of nonzero discord , the strongly correlated states that can be conveniently generated by the collective dephasing process allow for direct applications in a series of tasks from quantum information theory , as shown in section [ sec.applications ] . in the previous section we saw how the collective nature of the ensemble averaged dephasing process , induced by a spatially homogeneous , fluctuating external field , can be used to generate strongly correlated quantum states , which have direct applications for specific tasks in quantum information processing . we now extend the discussion from the initially weakly correlated states to initial states with an arbitrary correlation rank @xmath282 . we thus shift our focus from the generation of strongly correlated quantum states , to the robustness of the correlations under the action of collective dephasing . the objective , then , is to control the magnetic field orientation , such that the preservation of these correlations under the collective dephasing is ensured . as in the previous section , we assume bell - diagonal states ( thus vanishing reduced bloch vectors @xmath283 ) . the matrices @xmath53 and @xmath284 have the same rank unless @xmath285 is a singular value of @xmath53 with left- and right - singular vectors @xmath286 and @xmath287 , respectively , and only in these cases can the correlation rank be reduced by non - zero reduced bloch vectors [ recall ( [ eq : corr - rank ] ) ] . the effect of the collective dephasing map on arbitrary initial states is easily investigated based on ( [ eq : stat - bipartite ] ) . furthermore , before application of the map ( [ eq : corr - deph - beta ] ) , we employ unitary transformations to bring the @xmath53 matrix of the initial state into diagonal form @xmath288 . its three eigenvalues then parametrize @xcite the fano form ( [ eq : fano - form ] ) of the initial state : @xmath289 where @xmath290 is the standard basis in @xmath138 . this transformation neither affects the correlation rank of the initial state , as discussed earlier , nor that of the final state , unless we start from a rank-2 state of the form ( [ eq : class - corr - state ] ) and we apply a magnetic field in the direction @xmath132 or @xmath133 , as dealt with in sec . [ sec : class - corr ] . in all other cases , there is at least one rank-1 matrix of the form @xmath291 in ( [ eq : svd ] ) that is transformed , according to ( [ eq : corr - deph - beta ] ) , into a rank-3 matrix by the collective dephasing map . application of the collective dephasing map ( [ eq : corr - deph - beta ] ) yields @xmath292 + 2 n_i^2 \bi n \otimes \bi n \rbrace,\end{aligned}\ ] ] with the normalized vectors @xmath293 notice that all the @xmath294 s and @xmath295 s belong to the plane orthogonal to @xmath97 and are mutually orthogonal : @xmath296 . this means that we may write all orthonormal bases @xmath297 , with @xmath298 , as a rotation about @xmath97 of , e.g. , @xmath299 by an angle @xmath300 ( in this case @xmath301 ) : @xmath302 by direct substitution we obtain @xmath303 ( \bi a_1 \otimes\bi a_1 + \bi b_1 \otimes\bi b_1 ) , \end{aligned}\ ] ] which shows that @xmath304 . the @xmath53 matrix of the final state , given by ( [ eq : beta - renorm ] ) , can thus be rewritten as @xmath305 + 2 n_i^2 \bi n \otimes \bi n \rbrace .\end{aligned}\ ] ] because @xmath306 is an orthonormal basis of @xmath138 , the above expression is a spectral decomposition of @xmath161 where two eigenvalues appear , one of which is two - fold degenerate : @xmath307 from these expressions the invariance of the eigenvalues under cyclic permutation of the indices is evident . the @xmath53 matrix of the state after application of the map is given by @xmath308 , \end{aligned}\ ] ] where we could have equivalently chosen @xmath309 and @xmath310 , instead of @xmath299 ( we have used above their equivalence under rotation of the basis about @xmath97 , which is orthogonal to all of them ) . to summarize , we find @xmath311 from the above explicit expressions for the eigenvalues we can immediately verify that the trace of the @xmath53 matrix is preserved under the application of the map [ recall the general result , ( [ eq.betaconserved ] ) ] : @xmath312 bell - diagonal states of two qubits allow for a simple geometric description of their correlations properties @xcite , which we will employ in the following . recall that bell - diagonal states are unambiguously characterized by their @xmath53 matrix ( [ eq : bd - state ] ) . since there exists a unique unitary operator that diagonalizes this matrix without changing the state s correlation properties , we can parametrize the correlation properties of bell - diagonal states by the three real eigenvalues of @xmath53 . it follows that this unitary operator defines an isomorphism that maps each bell - diagonal state to a point @xmath313 . positivity implies that the space of density matrices is isomorphic to a tetrahedron @xmath314 whose vertices represent the four bell states @xcite : @xmath315 in this tetrahedron we distinguish two regions : a central octahedron @xmath316 with vertices @xmath317 , which contains the separable states , and the four remaining corners , containing the entangled states . because each corner has one of the bell states as its vertex , we call the other entangled states in that corner `` bell - like '' . we notice that all @xmath318-like entangled states have negative coordinates , while the @xmath319-like states , with @xmath320 , have only the @xmath88-th coordinate negative , like @xmath319 itself . ) ( points with one vanishing and two equal coordinates ) . ] in figure [ fig : tetra - poi ] we show how the classes of states we are interested in are represented in the tetrahedron @xmath314 . the origin is the maximally mixed state @xmath321 , and the rotation axes of the octahedron correspond to the rank-2 states . the rank-3 states lie on the squares obtained by intersecting the octahedron with the planes orthogonal to its rotation axes . both these sets of states ( rank-2 and 3 ) have measure zero inside the tetrahedron . moreover , as discussed in sec . [ sec : correlations ] , all entangled states have rank 4 , but the converse is not true . we now turn to describing the action of the map ( [ eq : corr - deph - beta ] ) in this geometrical framework . in particular , the subset of states that is accessible by the map is defined by the conservation of the trace of the @xmath53 matrix ( sec . [ ssec : conserved - trace ] ) and the double degeneracy in the coordinates of the asymptotic state ( sec . [ ssec : eigenvalues ] ) . in the following we adopt greek letters to indicate cyclic indices , i.e. , @xmath322 denotes an even permutation of @xmath323 . whenever we apply the map ( [ eq : corr - deph - beta ] ) we transform to the reference frame where the direction of the magnetic field is the unit vector @xmath324 of the standard basis . this allows for the most general description , which is independent of the choice of reference frame . consequently , the set of points with two equal coordinates is a plane defined as @xmath325 while the conservation of @xmath326 defines another plane @xmath327 where , inside the tetrahedron , @xmath328 . because the trace of the @xmath53 matrix is conserved at all times , the trajectory of each point lies on the @xmath329 plane defined by the initial coordinates . however , since the final state must belong to @xmath330 , every state will asymptotically move to the intersection line defined by @xmath331 , whose existence is guaranteed by the fact that @xmath332 ( figure [ fig : par_planes ] ) . the position of the final state on the line defined by @xmath331 is determined by the magnetic field direction . let us use again @xmath333 and @xmath334 as the coordinates of the final states . we obtain @xmath335 since @xmath114 is fixed by the initial state , the coordinates depend only on @xmath97 . we remark here that , when the initial state is @xmath318-like , the intersection lines never cross the octahedron of separable states , because the planes @xmath329 are parallel to the octahedron face opposite to @xmath318 ( figure [ fig : par_planes ] ) . we deduce that the states in the @xmath318 corner move entirely in that corner . on the other hand , the entangled states in the other corners move inside their respective corner , but may also enter the octahedron . this entails significant implications for the entanglement preservation of the initial states from the different corners , and ultimately enables the effect of _ time - invariant _ entanglement for @xmath318-like states , as we will discuss in further detail later in this manuscript . 0.45 ) on the bell - diagonal states ( [ eq : bd - state ] ) contained in the tetrahedron . the octahedron , inscribed into the tetrahedron , represents the separable states.,title="fig : " ] [ fig : par_planes ] 0.45 ) on the bell - diagonal states ( [ eq : bd - state ] ) contained in the tetrahedron . the octahedron , inscribed into the tetrahedron , represents the separable states.,title="fig : " ] [ fig : isoconc ] in the geometric representation of bell - diagonal states , planes of equal purity are characterized by constant values of @xmath336 which correspond to concentric spheres around the tetrahedron s center , the maximally mixed state . from the kraus representation ( [ eq.colldeph ] ) of the collective dephasing map @xmath337 , it follows immediately that the purity of the final state is upper bounded , for all @xmath338 , by the initial state s purity : @xmath339 in most cases , however , the collective dephasing tends to push states towards the center of the tetrahedron . the preservation of the initial purity can be achieved for time - invariant states . whether there exist states outside of invariant subspaces for which the purity is preserved is unclear , but seems unlikely since the entire dynamics must lie on the circumference generated by the intersection of the sphere defined by @xmath340 and the plane @xmath329 . we now focus on the correlation rank of the final states , which directly translates into the number of non - vanishing coordinates of a point in the tetrahedron . the states in the @xmath318 corner never exhibit a vanishing coordinate , as displayed in figure [ fig : tetra - poi ] . moreover , as discussed in the previous section , the states in this corner never enter the octahedron , hence they never intersect the subsets of rank-2 and rank-3 states . we conclude that the correlation rank in the states in the @xmath318 corner is robust and always maximal , which also implies that these states are always discordant , and can not be created using local operations on classical states . for the remaining states , i.e. , those in the octahedron and in the other three corners , there is at most one intersection between the reachable rank-2 or rank-3 states , and the accessible final states defined by the line @xmath341 , where @xmath342 for those states . the magnetic field direction that yields such a state is found by solving either @xmath343 or @xmath344 under the normalisation constraint @xmath345 . using the kronecker - capelli theorem @xcite it is possible to show that there exists at most a 1-parameter family of directions that solve such equations . we conclude that , after the action of the collective dephasing map ( [ eq : corr - deph - beta ] ) , the state has almost always a correlation rank equal to 4 . this includes both scenarios where a state was already initially discordant and this property is preserved throughout the dephasing dynamics , as well as when the discord is generated by the action of the collective dephasing map , recall also the experiment reported in @xcite . let @xmath1 be a bell - diagonal state of two qubits and @xmath346 its eigenvalues in the basis of bell states . if @xmath347 , then the concurrence of @xmath1 can be written as @xcite @xmath348 inside the tetrahedron @xmath314 , where states are described by vectors @xmath349 , the concurrence has an isotropic form , @xmath350 which can be interpreted as the distance to the faces of the octahedron of separable states , parametrized by the equation @xmath351 . following the state space from these faces to either one of the corners of the tetrahedron , the concurrence increases from zero to one . let us remind here that the collective dephasing map , being separable , can not create entanglement . in our geometrical framework , this means that points initially in the octahedron @xmath316 are mapped onto @xmath352 . suppose that an initial state is inside the octahedron , i.e. , it has vanishing concurrence ( [ eq : iso - conc ] ) : @xmath353 . the concurrence ( [ eq : iso - conc ] ) in the final state , where the coordinates are @xmath354 from ( [ eq : lambda1 ] ) and ( [ eq : finallambda2 ] ) , is then @xmath355 and we have the following chain of inequalities : @xmath356 where we have used the normalisation condition @xmath357 , which implies @xmath358 . since the states in the @xmath359-corner are constrained to move on planes parallel to the octahedron face opposite to @xmath318 , these states remain on iso - concurrence planes at all times ( compare figs . [ fig : par_planes ] and [ fig : isoconc ] ) . more specifically , as shown in @xcite , the points in this corner are characterized by negative coordinates : @xmath360 . hence , the concurrence ( [ eq : iso - conc ] ) for these states can be rewritten as @xmath361 where we have used @xmath362 and , for entangled states in this corner , @xmath363 . in the other corners of the tetrahedron only one of the coordinates is negative , and we denote it with @xmath364 : @xmath365 , where @xmath366 represents the initial state . we then have @xmath367 let @xmath368 represent the final , asymptotic state of the system after collective dephasing , and let us once again denote the negative components of @xmath366 and @xmath368 with a subscript @xmath364 , i.e. , we have @xmath369 and @xmath370 , respectively . since the eigenvalue @xmath371 has a double degeneracy , but the points in the @xmath372 corners can only have a single negative component , the negative eigenvalue after collective dephasing must necessarily be @xmath373 the concurrence of the final state then reads @xmath374 it follows that @xmath375 , i.e. , the initial concurrence of states in the three triplet corners can only be preserved if the magnetic field is chosen along the special direction that characterizes the respective corner . moreover , ( [ eq : conc - final ] ) allows for an estimation of the loss of concurrence due to small deviations from @xmath324 . let us conclude this section by highlighting the relationship between the different types of states and the respective magnetic field directions : the concurrence of a @xmath376-like state is _ conserved _ under collective dephasing if and only if the magnetic field points in the @xmath324 direction , while in @xmath318-like states the concurrence is _ always conserved _ , _ independently _ of the direction of the magnetic field . this means that , independently of the field direction , one quarter of the entangled bell - diagonal states shows preserved concurrence , while an additional quarter of states that depend on the magnetic field direction shows the same effect . let us briefly remark on the applicability of the results of the present section to states that are not bell - diagonal , i.e. , which have non - vanishing reduced bloch vectors . to this end , let us consider a state @xmath377 and the local operation @xmath378 such that @xmath379 \doteq ( 0,0,\beta^\mathrm{bd})$ ] @xcite . the operation that transforms a general two - qubit into a bell - diagonal state , i.e. , sets the reduced bloch vectors to zero , is a local operation that can not increase entanglement @xcite , therefore @xmath380 , where the behaviour of the lower bound @xmath381 under collective dephasing can be controlled , e.g. , by manipulating the magnetic field direction as described above . despite the possibility to achieve conservation of entanglement in a bell - diagonal state , the concurrence of the more general state can still decrease due to the time evolution of the reduced bloch vectors . perfect conservation of concurrence in a general bipartite state is consequently only possible when concurrence is preserved in the corresponding bell - diagonal state , _ and _ when additionally _ both _ initial reduced bloch vectors are parallel to the magnetic field direction , which is the only possible case where the reduced bloch vectors of the initial and asymptotic state coincide , as per ( [ eq : corr - deph - bloch ] ) . let us review the insights gained in the previous section on the entanglement dynamics under collective dephasing employing the geometric representation of the tetrahedron picture . the state @xmath382 , represented as one of the corners of the tetrahedron , is completely unaffected by the collective dephasing process because it is an eigenstate of the hamiltonian of the system for _ every _ magnetic field direction . the state is therefore time - invariant . this is true also for all the werner states @xmath383 . in the geometric framework , these states constitute the rotation axis of the tetrahedron that passes through @xmath318 and the origin . for the other bell states , as well as the respective mixtures with the identity ( werner - like states ) , the previous remarks only hold provided the magnetic field is chosen according to the association rule ( [ eq : state - direc - rel ] ) , since in that case the respective bell state becomes an eigenstate of the hamiltonian . this explains and generalizes the observations reported in @xcite , where a dynamical evolution equivalent to the collective dephasing along the @xmath384 axis is applied to the werner and werner - like states . those obtained from the @xmath385 states are then categorized as ` robust ' , while those from @xmath386 are ` fragile ' . in fact , as we have shown above , only the werner state ( [ eq : werner - state ] ) is robust under collective dephasing in any field direction , while the other werner - like ( [ eq : werner - like ] ) states are robust only when a specific direction is chosen @xcite . the preservation of entanglement within decoherence - free subspaces @xcite which in the present case , due to the absence of a hamiltonian evolution , coincide with the above - mentioned time - invariant subspaces is a rather obvious phenomenon : if the state does not evolve in time , then its properties are naturally also conserved . our analysis , however , points out the non - trivial possibility of time - invariant entanglement of states beyond time - invariant subspaces @xcite , see also @xcite . in fact , as pointed out above , all states in the @xmath318 corner remain at a level of constant concurrence , even if they do not belong to the axis of time - invariant ( werner ) states . in the present section we have studied the behavior of initially correlated quantum states under the influence of the collective dephasing evolution . to this end , we have extended the analysis of the correlation rank @xmath26 under collective dephasing to states of arbitrary initial correlation rank . furthermore , we have presented a convenient geometrical setting to describe bell - diagonal states and their evolution under collective dephasing . this geometrical interpretation allowed us to demonstrate that highly correlated ( @xmath146 ) initial states retain their strong correlations , as quantified by @xmath26 , under the collective dephasing dynamics for almost all choices of the magnetic field direction . combining the a geometric description of the collective dephasing dynamics with a geometric interpretation of the concurrence an entanglement measure we have characterized a finite set of states showing time - invariant entanglement despite their incoherent evolution . some features of bipartite states can be directly generalized to a multipartite setting . for example , the generalized werner states , i.e. , those states that are invariant under @xmath387-operations @xcite , are the fixed points of the collective dephasing map for any number of qubits @xcite . in @xcite , the decay and the time - invariant preservation of multipartite entanglement properties was also provided . a compelling explanation for the mechanism that enables this phenomenon in a multipartite case is , however , presently unavailable . in the following we allude to two possible approaches towards achieving this goal . the conservation of @xmath326 in bipartite states of qubits can be related to the overlap between the state of the system @xmath95 and the singlet state @xmath388 @xcite . this is true for any state @xmath389 that is an eigenstate of the collective dephasing hamiltonian for an arbitrary choice of the magnetic field direction . these eigenstates are the multi - qubit singlet states @xcite , which exist only for an even number @xmath82 of qubits , and are a family of @xmath390 $ ] linearly independent states . the overlap with each of these states is an integral of motion , which in the bipartite case reduces to the conservation of entanglement for all magnetic field directions . it is however not clear whether the overlap with the _ multi_-qubit singlet states is related to entanglement , especially considering that we observe conserved entanglement properties for any and not just for an even number of qubits . another approach originates from the observation that the collective dephasing map describes a simultaneous rotation of all qubits , hence it has a set of fixed points , the rotation axis , and the ( hyper-)planes orthogonal to it are mapped onto ( a subset of ) themselves . the fixed points of the map are the werner states @xcite , and , in the tetrahedron of bell - diagonal states , the planes orthogonal to the rotation axis contain the states with the same amount of entanglement . in an analogous higher - dimensional picture , the set of states orthogonal to the family of multipartite werner states may lead to an interesting set of integrals of motion . to summarize , we provided a detailed analysis of the impact of a collective dephasing process on the correlation properties of bipartite states . based on the kraus representation of the dephasing process @xcite , we provided conditions that enable the generation of states with high correlation rank and non - vanishing quantum discord from uncorrelated or only weakly correlated states . using an intuitive geometric representation of the state space and the collective dynamics , we investigated the entanglement dynamics under collective dephasing . for initially entangled states , we provided conditions that ensure the complete preservation of the entanglement for all times under the dephasing dynamics . surprisingly , this is possible even for large families of states that do not belong to time - invariant subspaces , i.e. , states that change in time due to the incoherent dynamics . while some of the results could be generalized straight - forwardly to multipartite scenarios , a compelling picture describing time - invariant multipartite entanglement remains to be conceived . since collective dephasing represents one of the dominant sources of error for many experiments with trapped atomic particles , we expect that the results derived in the present article can be readily harnessed in state - of - the - art setups with trapped ions @xcite or ultracold atoms @xcite . _ note added._an experimental observation of time - invariant entanglement was reported in @xcite after completion of this manuscript . m. g. thanks the german national academic foundation for their support . this project has received funding from the european union s horizon 2020 research and innovation programme under grant agreement no 641277 . 10 girolami d , souza a m , giovannetti v , tufarelli t , filgueiras j g , sarthour r s , soares - pinto d o , oliveira i s and adesso g 2014 http://journals.aps.org/prl/abstract/10.1103/physrevlett.112.210401[_phys . lett . _ * 112 * 210401 ] .

given a digraph , let be the minimum semi - degree of . is-ordered hamiltonian if for every sequence of distinct vertices of there is a directed hamilton cycle which encounters in this order . our main result is that every digraph of sufficiently large order with is-ordered hamiltonian . the bound on the minimum semi - degree is best possible . an undirected version of this result was proved earlier by kierstead , srkzy and selkow . # 1 # 1 # 1&#2 = by -4 1= 1 < to # 1 [ firstthm]theorem [ firstthm]proposition [ firstthm]lemma [ firstthm]corollary [ firstthm]problem [ firstthm]definition [ firstthm]conjecture [ firstthm]theorem

0707.1577

the famous theorem of dirac determines the smallest minimum degree of a graph which guarantees the existence of a hamilton cycle . there are many subsequent results which investigate degree conditions that guarantee the existence of a hamilton cycle with some additional properties . in particular , chartrand ( see @xcite ) introduced the notion of a hamilton cycle which has to visit a given set of vertices in a prescribed order . more formally , we say that a graph @xmath7 is _ @xmath2-ordered _ if for every sequence @xmath3 of distinct vertices of @xmath7 there is a cycle which encounters @xmath3 in this order . @xmath7 is _ @xmath2-ordered hamiltonian _ if it contains a hamilton cycle with this property . kierstead , srkzy and selkow @xcite showed that for all @xmath8 every graph on @xmath9 vertices of minimum degree at least @xmath10 is @xmath2-ordered hamiltonian . this bound on the minimum degree is best possible and proved a conjecture of ng and schultz @xcite . several related problems have subsequently been considered : for instance , the case when @xmath2 is large compared to @xmath4 was investigated in @xcite ( but has not been completely settled yet ) . ore - type conditions were investigated in @xcite . for more results in this direction , see the survey by gould @xcite . it seems that digraphs provide an equally natural setting for such problems . our main result is a version of the result in @xcite for digraphs . the digraphs we consider do not have loops and we allow at most one edge in each direction between any pair of vertices . given a digraph @xmath0 , the _ minimum semi - degree @xmath11 _ of @xmath0 is the minimum of the minimum outdegree @xmath12 of @xmath0 and its minimum indegree @xmath13 . [ thm : ordham ] for every @xmath14 there is an integer @xmath15 such that every digraph @xmath0 on @xmath16 vertices with @xmath5 is @xmath2-ordered hamiltonian . our proof shows that one can take @xmath17 where @xmath18 is a sufficiently large constant . note that if @xmath4 is even and @xmath2 is odd the bound on the minimum semi - degree is slightly larger than in the undirected case . however , it is best possible in all cases . in fact , if the minimum semi - degree is smaller , it turns out that @xmath0 need not even be @xmath2-ordered . this is easy to see if @xmath2 is even : let @xmath0 be the digraph which consists of a complete digraph @xmath19 of order @xmath20 and a complete digraph @xmath21 of order @xmath22 which has precisely @xmath23 vertices in common with @xmath19 . pick vertices @xmath24 and @xmath25 . then @xmath0 has no cycle which encounters @xmath3 in this order . a similar construction also works if both @xmath2 and @xmath4 are odd . the construction in the remaining case is a little more involved , see @xcite for details . note that every hamiltonian digraph is @xmath26-ordered hamiltonian , so the case when @xmath27 in theorem [ thm : ordham ] is covered by the result of ghouila - houri @xcite ( theorem [ thm : gh ] below ) which implies that every digraph with minimum semi - degree at least @xmath28 contains a hamilton cycle . theorem [ thm : ordham ] can be used to deduce a version for edges which have to be traversed in a prescribed order by the hamilton cycle : we say that a digraph @xmath0 is _ @xmath2-arc ordered hamiltonian _ if , for every sequence @xmath29 of independent edges , @xmath0 contains a hamilton cycle which encounters @xmath29 in this order . @xmath0 is _ @xmath2-arc hamiltonian _ if it contains a hamilton cycle which encounters these edges in any order . @xmath0 is called _ hamiltonian @xmath2-linked _ if @xmath30 and if for every sequence @xmath31 of distinct vertices there are disjoint paths @xmath32 in @xmath0 such that @xmath33 joins @xmath34 to @xmath35 and such that together all the @xmath33 cover all the vertices of @xmath0 . thus every digraph @xmath0 which is hamiltonian @xmath2-linked is also @xmath2-arc ordered hamiltonian . indeed , if @xmath36 are the ( directed ) edges our hamilton cycle has to encounter then disjoint paths linking @xmath37 to @xmath34 for all @xmath38 yield the required hamilton cycle . [ thm : arcordham ] for all @xmath14 there is an integer @xmath15 such that every digraph @xmath0 on @xmath16 vertices with @xmath39 is hamiltonian @xmath2-linked and thus in particular @xmath2-arc ordered hamiltonian . the examples in @xcite show that in both parts of corollary [ thm : arcordham ] the bound on the minimum semi - degree is best possible . in fact , if the minimum semi - degree is smaller then one can not even guarantee the digraph to be @xmath2-arc ordered . a result of bermond @xcite ( see also @xcite ) implies that if @xmath40 then @xmath0 is @xmath2-arc hamiltonian . it easily follows that if @xmath41 , then @xmath0 is hamiltonian @xmath42-linked , i.e. hamiltonian connected ( see @xcite ) . this covers the case @xmath43 of corollary [ thm : arcordham ] . as observed in ( * thm 9.2.10 ) , if @xmath44 , then @xmath0 is hamiltonian @xmath26-linked , which covers the case @xmath45 of corollary [ thm : arcordham ] . corollary [ thm : arcordham ] can easily be deduced from theorem [ thm : ordham ] as follows : let @xmath46 and @xmath47 be distinct vertices where we aim to link @xmath34 to @xmath35 for all @xmath48 . let @xmath49 be the digraph obtained from @xmath0 by contracting @xmath34 and @xmath37 into a new vertex @xmath50 whose outneighbourhood is that of @xmath34 and whose inneighbourhood is that of @xmath37 . more precisely , let @xmath51 . then @xmath49 is the digraph obtained from @xmath52 by adding new vertices @xmath3 and defining the edges incident to these new vertices as follows . the outneighbours of @xmath50 are the outneighbours of @xmath34 in @xmath53 as well as all the @xmath54 for all those @xmath55 for which @xmath56 is an outneighbour of @xmath34 in @xmath0 ( where @xmath57 ) . similarly , inneighbours of @xmath50 are the inneighbours of @xmath37 in @xmath53 as well as all the @xmath54 for all those @xmath58 for which @xmath59 is an inneighbour of @xmath37 in @xmath0 . it is easy to check that @xmath60 and that a hamilton cycle in @xmath49 which encounters @xmath3 in this order corresponds to a spanning set of disjoint paths from @xmath34 to @xmath35 . a result of chen et al . * theorem 10 ) implies that the smallest minimum degree which guarantees an undirected graph to be @xmath2-arc ordered hamiltonian is @xmath61 . ( a graph is @xmath2-arc ordered hamiltonian if for any sequence of @xmath2 independent oriented edges there exists a hamilton cycle which encounters these edges in the given order and orientation . ) the smallest minimum degree which forces a graph to be @xmath2-linked was determined by kawarabayashi , kostochka and yu @xcite . it is not clear whether the minimum degree for hamiltonian @xmath2-linkedness is the same . the main tool in our proof of theorem [ thm : ordham ] is a recent result by the first authors ( theorem [ thm : ord ] below ) , which shows that the degree condition in theorem [ thm : ordham ] at least guarantees a @xmath2-ordered cycle ( but not necessarily a hamiltonian one ) . the strategy of the proof of theorem [ thm : ordham ] is to consider such a cycle of maximal length and to show that it must be hamiltonian . the same strategy was already applied in the proof of the undirected case in @xcite . however , both parts of the strategy are more difficult in the digraph case : the existence of a @xmath2-ordered directed cycle ( i.e. theorem [ thm : ord ] ) already confirms a conjecture of manoussakis @xcite for large @xmath4 . the hamiltonicity of a @xmath2-ordered cycle of maximal length is easier to show in the undirected case as one can consider ` local transformations ' of a given @xmath2-ordered cycle which reverse the orientation of certain segments of the cycle . this means that apart from some basic observations like lemma [ longloop ] below our proof is quite different from that in @xcite . @xcite [ thm : ord ] let @xmath2 and @xmath4 be integers such that @xmath62 and @xmath63 . then every digraph @xmath0 on @xmath4 vertices with @xmath5 is @xmath2-ordered . given a digraph @xmath0 , we write @xmath64 for its vertex set , @xmath65 for its edge set and @xmath66 for its order . we write @xmath67 for the edge directed from @xmath68 to @xmath69 . more generally , if @xmath19 and @xmath21 are disjoint sets of vertices of @xmath0 then an _ @xmath19-@xmath21 edge _ is an edge of the form @xmath70 where @xmath71 and @xmath72 . a digraph is _ complete _ if every pair of distinct vertices is joined by edges in both directions . given disjoint subdigraphs @xmath73 and @xmath74 of a digraph @xmath0 such that @xmath75 is spanning and a set @xmath76 , we write @xmath77 for the set of all those vertices @xmath78 which in the digraph @xmath0 receive an edge from some vertex in @xmath19 . @xmath79 is defined similarly . if @xmath19 consists of a single vertex @xmath68 , we just write @xmath80 etc . and put @xmath81 and @xmath82 . so in particular , @xmath83 is the outneighbourhood of @xmath68 in @xmath0 and @xmath84 is its outdegree . also , note that @xmath85 is the outneighbourhood of @xmath68 in the subdigraph @xmath86 $ ] of @xmath0 induced by @xmath87 and not its outneighbourhood in @xmath73 ( where @xmath88 ) . we let @xmath89 . if we refer to paths and cycles in digraphs then we always mean that they are directed without mentioning this explicitly . the _ length _ of a path is the number of its edges . given two vertices @xmath90 , an _ @xmath68-@xmath69 path _ is a path which is directed from @xmath68 to @xmath69 . given two vertices @xmath68 and @xmath69 on a directed cycle @xmath18 , we write @xmath91 for the subpath of @xmath18 from @xmath68 to @xmath69 . similarly , given two vertices @xmath68 and @xmath69 on a directed path @xmath92 such that @xmath68 precedes @xmath69 , we write @xmath93 for the subpath of @xmath92 from @xmath68 to @xmath69 . a digraph @xmath0 is _ strongly connected _ if for every ordered pair @xmath94 of vertices of @xmath0 there exists an @xmath68-@xmath69 path . @xmath0 is _ hamiltonian connected _ if for every ordered pair @xmath94 of vertices of @xmath0 there exists a hamilton path from @xmath68 to @xmath69 . ( so hamiltonian connectedness is the same as hamiltonian 1-linkedness . ) we will often use the following result of ghouila - houri @xcite which gives a sufficient condition for the existence of a hamilton cycle in a digraph . in particular , it implies a version of theorem [ thm : ordham ] for @xmath95 as any hamiltonian digraph is 2-ordered hamiltonian . [ thm : gh ] suppose that @xmath0 is a strongly connected digraph such that @xmath96 for every vertex @xmath97 . then @xmath0 is hamiltonian . the next result of overbeck - larisch @xcite provides a sufficient condition for a digraph to be hamiltonian connected . [ thm : hamconn ] suppose that @xmath0 is a digraph such that @xmath98 whenever @xmath67 is not an edge . then @xmath0 is hamiltonian connected . let @xmath0 be a digraph satisfying the conditions of theorem [ thm : ordham ] . let @xmath99 by any sequence of @xmath100 vertices of @xmath0 . we will often view @xmath101 as a set . @xmath101-cycle _ in @xmath0 is a cycle which encounters @xmath3 in this order . so we have to show that @xmath0 has a hamiltonian @xmath101-cycle . theorem [ thm : ord ] implies the existence of an @xmath101-cycle in @xmath0 . let @xmath18 be a longest such cycle and suppose that @xmath18 is not hamiltonian . let @xmath102 be the subdigraph of @xmath0 induced by all the vertices outside @xmath18 . our aim is to find a longer @xmath101-cycle by modifying @xmath18 ( yielding a contradiction ) . the purpose of this section is to collect the properties of @xmath18 and @xmath102 that we need in our proof of theorem [ thm : ordham ] . we let @xmath103 be the set of all those vertices on @xmath18 which receive an edge _ from _ some vertex in @xmath102 and we let @xmath104 be the set of all those vertices on @xmath18 which send an edge _ to _ some vertex in @xmath102 . given @xmath105 , we write @xmath106 for the set of all those vertices on @xmath18 which receive an edge from at least @xmath48 vertices in @xmath102 . thus @xmath107 . @xmath108 is defined similarly . given a vertex @xmath68 on @xmath18 , we will denote its successor on @xmath18 by @xmath109 and its predecessor by @xmath110 . [ hlemma1 ] @xmath102 is hamiltonian connected and @xmath111 for all vertices @xmath112 . moreover any digraph obtained from @xmath102 by deleting at most @xmath26 vertices is strongly connected and @xmath113 . we first show that any two ( not necessarily distinct ) vertices @xmath112 for which @xmath102 contains an @xmath68-@xmath69 path , @xmath92 say , satisfy the degree condition in the lemma . to see this , note that no vertex in @xmath114 is a predecessor of some vertex in @xmath115 . indeed , if @xmath116 and @xmath117 then by replacing the edge @xmath118 with the path @xmath119 we obtain a longer @xmath101-cycle , a contradiction . but this means that @xmath120 and thus @xmath121 as required . however , as @xmath100 this degree condition means that @xmath122 and so @xmath102 contains an @xmath69-@xmath68 path of length 2 . thus whenever @xmath102 contains an @xmath68-@xmath69 path it also contains a @xmath69-@xmath68 path . now let @xmath68 and @xmath123 be any two vertices of @xmath102 . what we have shown above applied with @xmath124 implies that @xmath125 and thus @xmath126 . note that by the above @xmath68 is joined to every vertex in @xmath127 with paths in both directions . similarly , @xmath128 and @xmath123 is joined to every vertex in @xmath129 with paths in both directions . as @xmath130 this means that @xmath68 is joined to @xmath123 with paths in both directions , i.e. @xmath102 is strongly connected . together with ( [ eq : h2 ] ) this in turn implies that @xmath131 for all vertices @xmath132 . in particular , @xmath102 is hamiltonian connected by theorem [ thm : hamconn ] . to show that any digraph @xmath133 obtained from @xmath102 by deleting at most 2 vertices is strongly connected note that @xmath134 for every @xmath135 . thus if @xmath136 then either @xmath137 is an edge or @xmath133 contains an @xmath69-@xmath68 path of length 2 . it now remains to prove the bounds on @xmath138 . consider any vertex @xmath139 . then @xmath140 and so @xmath141 . for the upper bound , note that no vertex in @xmath104 has a successor in @xmath103 . indeed , if @xmath142 is such a vertex in @xmath104 and @xmath143 is its successor then we could replace @xmath118 with a path through @xmath102 to obtain a longer @xmath101-cycle , a contradiction . but this means that some vertex of @xmath18 must have all its inneighbours on @xmath18 or all its outneighbours on @xmath18 . thus @xmath144 and so @xmath145 . recall that the proof of lemma [ hlemma1 ] implies the following . [ ftcor ] no vertex on @xmath18 which lies in @xmath104 has a successor in @xmath103 . the next result deals with the case when the vertices @xmath146 and @xmath147 are further apart . [ longloop ] suppose that @xmath148 are distinct and the interior of @xmath149 does not contain a vertex from @xmath101 . then there are no distinct vertices @xmath150 such that @xmath151 . suppose that such @xmath152 do exist . furthermore , we may assume that @xmath153 and @xmath154 are chosen such that they satisfy all these properties and subject to this @xmath155 is minimum . let @xmath156 denote the set of all vertices in the interior of @xmath149 . then our choice of @xmath153 and @xmath154 implies that @xmath157 and @xmath158 . moreover , by corollary [ ftcor ] no vertex in @xmath159 is a predecessor of some vertex in @xmath160 . thus @xmath161 and so @xmath162 this implies that @xmath163 and thus replacing the interior of @xmath149 with a hamilton path from @xmath164 to @xmath165 through @xmath102 ( which exists by lemma [ hlemma1 ] ) yields a longer @xmath101-cycle , a contradiction . the next two results will be used in the proof of lemma [ h2paths ] . [ strongconn ] let @xmath7 be a digraph such that @xmath166 for every vertex @xmath167 and @xmath168 for every pair of vertices @xmath169 . let @xmath170 and @xmath171 be distinct vertices of @xmath7 such that @xmath172 . then there exists a vertex @xmath173 such that @xmath174 is strongly connected . first note that @xmath175 since @xmath172 . pick @xmath176 . we will show that one of these @xmath177 can play the role of @xmath178 . let @xmath179 . note that @xmath180 for every vertex @xmath181 and @xmath182 for every pair of vertices @xmath183 . in particular , the latter condition implies that @xmath184 is strongly connected . thus @xmath184 has a hamilton cycle @xmath18 by theorem [ thm : gh ] . let @xmath185 denote the successor of @xmath186 on @xmath18 and let @xmath187 be its predecessor . put @xmath188 and @xmath189 . note that @xmath190 since @xmath191 and @xmath192 . similarly @xmath193 . clearly , if @xmath194 is an edge or @xmath195 , then @xmath196 is strongly connected and so we can take @xmath178 to be @xmath186 . so we may assume that neither of these is the case . but then @xmath197 . let @xmath198 be such that @xmath199 is maximal . similarly , let @xmath200 be such that @xmath201 is maximal . note that if @xmath202 then @xmath196 is strongly connected . so we may assume that this is not the case . but then @xmath142 must be the successor of @xmath203 on @xmath18 , @xmath204 must consist of precisely the vertices in @xmath205 and @xmath206 must consist of precisely the vertices in @xmath207 . let @xmath208 and @xmath209 . we may assume that @xmath7 does not contain an @xmath210-@xmath211 edge as otherwise @xmath196 is strongly connected . we will now show @xmath212 $ ] is complete and that @xmath186 receives an edge from every vertex in @xmath210 . so consider any vertex @xmath213 . then @xmath214 . together with the fact that there is no @xmath210-@xmath211 edge this shows that @xmath215 . thus @xmath212 $ ] is complete and @xmath186 receives an edge from every vertex in @xmath210 . similarly one can show that @xmath216 $ ] is complete and that @xmath186 sends an edge to every vertex in @xmath211 . now consider @xmath217 and @xmath218 . if for example @xmath219 then @xmath220 is strongly connected and so we can take @xmath178 to be @xmath217 . as one can argue similarly for @xmath218 , we may assume that @xmath221 and @xmath222 . if @xmath223 is an edge or @xmath224 then @xmath220 is strongly connected . ( here we used that @xmath225 since @xmath226 . ) if this is not the case , then @xmath227 implies the existence of some vertex @xmath228 with @xmath229 . if @xmath213 then @xmath230 is an @xmath211-@xmath210 edge avoiding @xmath222 and so @xmath231 is strongly connected . ( here we used that @xmath232 since @xmath233 . ) similarly , if @xmath234 then @xmath220 is strongly connected . altogether , this shows that we can take @xmath178 to be @xmath186 , @xmath217 or @xmath218 . [ lowdegh ] suppose that @xmath102 contains a vertex @xmath142 with @xmath235 . suppose that @xmath236 and @xmath237 are distinct vertices on @xmath18 . then @xmath238 or @xmath239 ( or both ) . let @xmath240 denote the set of all those vertices on @xmath18 which receive an edge from @xmath142 . let @xmath241 denote the set of all those vertices on @xmath18 whose predecessor sends an edge to @xmath142 . corollary [ ftcor ] implies that @xmath242 . since @xmath243 this shows that at most one vertex on @xmath18 lies outside @xmath244 . let @xmath123 be the vertex in @xmath245 ( if it exists ) . suppose first that @xmath246 ( this also covers the case when @xmath123 does not exist ) . . also either @xmath248 or @xmath249 . so let us assume that @xmath248 ( the case when @xmath249 is similar ) . we will show that @xmath238 . so suppose first that @xmath250 . then @xmath251 and thus @xmath252 , a contradiction to corollary [ ftcor ] . similarly , if @xmath253 then @xmath254 and thus @xmath255 , i.e. the predecessor of @xmath164 lies in @xmath104 , contradicting corollary [ ftcor ] . so suppose next that @xmath256 and thus , by corollary [ ftcor ] , the predecessor of @xmath123 does not lie in @xmath104 . this in turn implies that @xmath257 . moreover either @xmath258 or @xmath259 . so let us assume that @xmath258 . similarly as before one can show that @xmath260 . in our proof of theorem [ thm : ordham ] we will frequently need two disjoint paths through @xmath102 joining two given disjoint pairs of vertices on @xmath18 in order to modify @xmath18 into a longer @xmath101-cycle . the following lemma implies the existence of such paths provided that the pairs consist of vertices having sufficiently many neighbours in @xmath102 ( see also corollary [ corh2paths ] ) . [ h2paths ] suppose that @xmath261 and @xmath262 are disjoint subsets of @xmath263 such that @xmath264 and @xmath265 . then there are disjoint @xmath266-@xmath267 paths @xmath33 of length at least @xmath26 and such that all inner vertices of @xmath268 and @xmath269 lie in @xmath102 . moreover , if @xmath270 and if we even have that @xmath271 and @xmath272 then we can find such paths which additionally satisfy @xmath273 . by disregarding some neighbours if necessary we may assume that @xmath274 . our first aim is to show that for some @xmath275 there is an @xmath266-@xmath267 path @xmath33 which satisfies the following properties : 1 . the graph @xmath276 has a hamilton cycle @xmath277 . 2 . all @xmath135 satisfy @xmath278 . @xmath279 , i.e. @xmath33 contains at least 1 and at most 3 vertices from @xmath102 . 4 . if @xmath280 then @xmath281 and @xmath282 . if @xmath283 then @xmath284 and @xmath285 . if we have found such an @xmath48 , say @xmath280 , then our aim is to use the hamilton cycle @xmath277 in order to find @xmath269 . to prove the existence of such an @xmath48 , recall that lemma [ hlemma1 ] implies @xmath286 for every pair of vertices @xmath112 . thus condition ( ii ) will hold automatically if ( iii ) holds . now suppose first that there exists a vertex @xmath287 . take @xmath280 and take @xmath268 to be any @xmath288-@xmath289 path whose interior consists precisely of @xmath170 . then @xmath290 for every @xmath291 . as @xmath133 is strongly connected by lemma [ hlemma1 ] we can apply theorem [ thm : gh ] to find a hamilton cycle @xmath277 of @xmath133 . ( if @xmath292 then @xmath277 will consist of just a double edge . ) in the case when @xmath293 we proceed similarly . now suppose that @xmath294 and @xmath295 . then lemma [ lowdegh ] implies that @xmath296 for every @xmath139 . if there is an @xmath297-@xmath298 edge @xmath299 take @xmath300 and take @xmath268 to be any @xmath288-@xmath289 path whose interior consists of this edge . then @xmath301 for every @xmath291 and so again , as @xmath133 is strongly connected by lemma [ hlemma1 ] , we can apply theorem [ thm : gh ] to find a hamilton cycle @xmath277 of @xmath133 . in the case when there is an @xmath302-@xmath303 edge we proceed similarly . thus we may assume that @xmath304 and that there is no @xmath305-@xmath306 edge ( for @xmath307 ) . pick any vertex @xmath308 and let @xmath309 be a vertex such that @xmath310 and @xmath311 . ( the fact that we can choose such a @xmath171 follows from @xmath304 and our assumption that the sizes of the @xmath305 and the @xmath306 are equal . ) apply lemma [ strongconn ] with @xmath312 to find a vertex @xmath313 such that @xmath314 is strongly connected . take @xmath300 and @xmath268 to be any @xmath288-@xmath289 path whose interior consists of @xmath315 . then @xmath316 for every @xmath291 and so again @xmath133 contains a hamilton cycle @xmath277 by theorem [ thm : gh ] . our choice of @xmath170 and @xmath171 implies that ( iv ) holds . altogether , this shows that in each case for some @xmath48 there exists a path @xmath33 satisfying ( i)(iv ) . we may assume that @xmath280 . as mentioned before , our aim now is to use the hamilton cycle @xmath277 of @xmath133 in order to find an @xmath317-@xmath318 path @xmath269 through @xmath133 . in the case when @xmath319 this is trivial since by ( iv ) both @xmath302 and @xmath303 meet @xmath133 in at least one vertex . so suppose now that @xmath270 and @xmath320 and thus we wish to find a long @xmath317-@xmath318 path . to do this , let @xmath321 and @xmath322 . thus @xmath323 by ( iv ) . choose @xmath324 and @xmath325 to be distinct such that @xmath326 is maximum . if @xmath327 then we can take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath328 . so we may assume that @xmath329 . note that the choice of @xmath186 and @xmath330 implies that @xmath331 . moreover , all the vertices in @xmath204 must precede the vertices in @xmath206 on @xmath328 . ( indeed , if e.g. @xmath332 and @xmath333 are distinct vertices such that @xmath334 precedes @xmath178 , i.e. @xmath178 lies on @xmath335 then @xmath336 , contradicting the choice of @xmath186 and @xmath330 . ) thus @xmath337 and there are vertices @xmath338 and @xmath339 such that @xmath340 are distinct and appear on @xmath277 in this order . we now distinguish several cases . * case 1 . * _ there are @xmath341 such that @xmath342 is an edge . _ note that @xmath343 or @xmath344 by ( ii ) . suppose that the former holds ( the other case is similar ) . as @xmath329 this means that @xmath345 has at least @xmath346 outneighbours in the interior of @xmath347 and so we can find such an outneighbour @xmath142 with @xmath348 . but then we can take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath349 ( figure [ fig : h2edge ] ) . [ ] [ ] @xmath186 [ ] [ ] @xmath217 [ ] [ ] @xmath345 [ ] [ ] @xmath350 [ ] [ ] @xmath351 [ ] [ ] @xmath330 [ ] [ ] @xmath269 [ ] [ ] @xmath142 in case 1 . the left figure is for the subcase when @xmath343 and the right figure is for the subcase when @xmath344.,title="fig : " ] * case 2 . * _ for all @xmath341 @xmath342 is not an edge . _ * case 2.1 . * _ there exists some vertex @xmath352 . _ note that @xmath353 since by our assumption neither @xmath354 nor @xmath355 is an edge . as before , either @xmath356 or @xmath357 . suppose that the former holds ( the other case is similar ) . if @xmath358 lies in the interior of @xmath359 , let @xmath142 be an outneighbour of @xmath217 in the interior of @xmath360 with @xmath348 . then we can take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath361 . so we may assume that @xmath358 lies in the interior of @xmath360 . but then either the interior of @xmath362 contains at least @xmath363 outneighbours of @xmath217 or the interior of @xmath364 contains at least @xmath363 outneighbours of @xmath217 . if the former holds let @xmath142 be any outneighbour of @xmath217 in the interior of @xmath362 such that @xmath365 and take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath366 ( see figure [ fig : h2case22 ] ) . if the latter holds let @xmath142 be any outneighbour of @xmath217 in the interior of @xmath364 such that @xmath367 and take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath361 . [ ] [ ] @xmath186 [ ] [ ] @xmath217 [ ] [ ] @xmath368 [ ] [ ] @xmath142 [ ] [ ] @xmath358 [ ] [ ] @xmath269 [ ] [ ] @xmath330 [ ] [ ] @xmath350 in case 2.1 if @xmath358 lies in the interior of @xmath360 . the left figure is for the subcase when the interior of @xmath362 contains at least @xmath363 outneighbours of @xmath217 . the right figure is for the subcase when the interior of @xmath364 contains at least @xmath363 outneighbours of @xmath217.,title="fig : " ] * case 2.2 . * _ there exists some vertex @xmath369 . _ this case is similar to case 2.1 and we omit the details . * case 2.3 . * _ both @xmath370 and @xmath371 are empty . _ together with ( ii ) and our assumption that @xmath372 is not an edge this implies that @xmath373 . since @xmath374 is not an edge this means that @xmath375 is an edge . similarly it follows that @xmath376 is an edge . but as before either @xmath356 or @xmath377 . suppose that the former holds ( the other case is similar ) . then we can find an outneighbour @xmath142 of @xmath217 in the interior of @xmath347 with @xmath348 . but then we can take @xmath269 to be any @xmath317-@xmath318 path whose interior consists of @xmath378 . lemma [ h2paths ] immediately implies the following corollary , which is sometimes more convenient to apply . [ corh2paths ] suppose that @xmath379 and @xmath380 are distinct vertices on @xmath18 . then @xmath0 contains disjoint @xmath34-@xmath35 paths @xmath33 of length at least @xmath26 such that all inner vertices of @xmath268 and @xmath269 lie in @xmath102 . moreover , if @xmath270 and if we even have that @xmath381 and @xmath382 then we can find such paths which additionally satisfy @xmath273 . the last of our preliminary results gives a lower bound on the sizes of @xmath383 and @xmath384 . [ sizet5f5 ] we have that @xmath385 . moreover , @xmath386 and @xmath387 . to see the bound on @xmath388 , note that @xmath389 for every vertex @xmath139 and so @xmath390 . the proof for @xmath391 is similar . to prove the bound on @xmath392 , we double - count the number @xmath393 of edges in @xmath0 from @xmath104 to @xmath394 . since @xmath395 for any vertex @xmath139 we have that @xmath396 . on the other hand @xmath397 . before we can use this to estimate @xmath398 , we need an upper bound on @xmath388 . for this , recall that @xmath399 . together with corollary [ ftcor ] this shows that @xmath400 . altogether this gives @xmath401 the proof for @xmath402 is similar . the bound on @xmath403 follows since @xmath404 . indeed , the latter holds since lemma [ longloop ] implies that whenever @xmath405 are distinct and no vertex from @xmath101 lies in the interior of @xmath406 then @xmath407 meets @xmath406 in at most one vertex . throughout this section , we assume that the order @xmath4 of our given digraph @xmath0 is sufficiently large compared to @xmath2 for our estimates to hold . we will also omit floors and ceilings whenever this does not affect the argument . let @xmath101 , @xmath18 and @xmath102 be as defined at the beginning of section [ sec : prelim ] . recall that we assume that @xmath18 is not hamiltonian and will show that we can extend @xmath18 into a longer @xmath101-ordered cycle ( which would yield a contradiction and thus would prove theorem [ thm : ordham ] ) . given consecutive vertices @xmath405 , we call the path obtained from @xmath406 by deleting @xmath408 the _ interval from @xmath409 to @xmath408_. thus no vertex from @xmath101 lies in the interior of @xmath406 and @xmath18 consists of precisely @xmath410 disjoint intervals . in our proof of theorem [ thm : ordham ] we distinguish the following 4 cases according to the order of @xmath102 . recall that @xmath141 by lemma [ hlemma1 ] . * case 1 . * @xmath411 . recall that @xmath412 by lemma [ sizet5f5 ] and so at least one of the @xmath2 intervals of @xmath18 must contain at least @xmath413 vertices from @xmath383 . suppose that this is the case for the interval @xmath414 from @xmath409 to @xmath408 . recall that by lemma [ sizet5f5 ] at most @xmath415 vertices of @xmath18 do not lie in @xmath416 and by corollary [ ftcor ] no vertex in @xmath384 is the successor of a vertex in @xmath383 . since every maximal subpath of @xmath414 consisting of vertices from @xmath383 is succeeded by at least one vertex outside @xmath416 , it follows that @xmath414 contains a subpath @xmath19 which consists entirely of vertices from @xmath383 and satisfies @xmath417 . let @xmath418 be the subpath of @xmath19 consisting of its initial @xmath419 inner vertices and let @xmath420 be the subpath of @xmath19 consisting of its last @xmath419 inner vertices . let @xmath421 be the first vertex of @xmath19 . ( so @xmath422 is the first vertex of @xmath418 . ) consider any vertex @xmath178 on @xmath423 . lemma [ longloop ] implies that @xmath424 . thus @xmath425 and hence @xmath426 ( to see the third inequality recall that @xmath399 by lemma [ sizet5f5 ] . ) * case 1.1 . * _ there are vertices @xmath427 and @xmath428 such that @xmath429 is an edge . _ inequality ( [ eq : dega ] ) applied with @xmath430 implies that there exists a vertex @xmath431 such that the successor @xmath432 of @xmath203 lies in @xmath103 . recall that @xmath103 avoids @xmath423 and so @xmath432 must lie in @xmath433 . hence @xmath203 must lie in @xmath434 ( and thus in the interior of @xmath435 ) . as @xmath436 and as @xmath102 is hamiltonian connected by lemma [ hlemma1 ] , there is an @xmath437-@xmath432 path @xmath92 whose interior consists of precisely all the vertices in @xmath102 . but then the @xmath101-ordered cycle @xmath438 is hamiltonian , contradicting the choice of @xmath18 ( see figure [ fig : proofcase1 ] ) . [ ] [ ] @xmath186 [ ] [ ] @xmath185 [ ] [ ] @xmath437 [ ] [ ] @xmath217 [ ] [ ] @xmath203 [ ] [ ] @xmath432 [ ] [ ] @xmath92 [ ] [ ] @xmath18 [ ] [ ] @xmath187 [ ] [ ] @xmath186 [ ] [ ] @xmath437 [ ] [ ] @xmath217 [ ] [ ] @xmath439 [ ] [ ] @xmath440 [ ] [ ] @xmath441 [ ] [ ] @xmath442 [ ] [ ] @xmath268 [ ] [ ] @xmath269 into a longer @xmath101-ordered cycle in case 1.1 ( left ) and case 1.2 ( right).,title="fig : " ] * case 1.2 . * _ there are no such vertices @xmath427 and @xmath428 . _ let @xmath443 denote the set of all predecessors of vertices in @xmath384 . recall that @xmath103 avoids @xmath423 . thus @xmath443 avoids @xmath444 . now consider any vertex @xmath178 on @xmath420 . then @xmath425 since @xmath424 and thus @xmath445 by our assumption . but then using that @xmath446 by lemma [ sizet5f5 ] and arguing similarly as in ( [ eq : dega ] ) one can show that @xmath447 . together with the fact that @xmath448 this gives @xmath449 be the subpath of the interval @xmath414 preceding the first vertex in @xmath418 . so @xmath450 . let @xmath451 denote all the other intervals . for each @xmath38 let @xmath452 be the auxiliary bipartite graph whose vertex classes are @xmath453 and @xmath454 and in which @xmath455 is joined to @xmath456 if @xmath457 . recall that @xmath443 avoids @xmath444 . thus @xmath458 and so the edges of @xmath459 correspond to the edges of @xmath0 from @xmath443 to @xmath420 . together with ( [ eq : daw ] ) this implies that there is some @xmath48 such that @xmath460 thus @xmath452 is not planar and so there are vertices @xmath461 and @xmath462 such that the edges @xmath463 , @xmath464 ` cross ' in @xmath452 , i.e. such that @xmath439 lies in the interior of @xmath465 and @xmath186 lies in the interior of @xmath466 . recall that @xmath467 by the definition of @xmath443 and @xmath468 as @xmath420 consisted of inner vertices of @xmath19 . thus we can apply corollary [ corh2paths ] to obtain disjoint @xmath469-@xmath470 paths @xmath471 having all their inner vertices in @xmath102 and such that each @xmath471 contains at least one inner vertex ( where @xmath472 ) . thus @xmath473 is an @xmath101-ordered cycle with at least @xmath474 vertices ( note that it contains all the vertices of @xmath18 ) , contradicting the choice of @xmath18 ( see figure [ fig : proofcase1 ] ) . * * @xmath475 . the argument for this case is similar to that in case 1 . recall that @xmath476 by lemma [ sizet5f5 ] and so one of the @xmath2 intervals of @xmath18 must contain at least @xmath477 vertices from @xmath383 . suppose that this is the case for the interval @xmath414 from @xmath409 to @xmath408 . let @xmath421 be the first vertex on @xmath414 that lies in @xmath383 . let @xmath19 be the set consisting of the last @xmath478 vertices from @xmath383 lying in the interior of @xmath414 . for each @xmath71 let @xmath479 be the set of @xmath480 vertices of @xmath18 preceding @xmath178 . note that the definition of @xmath19 implies that @xmath479 lies in the interior of @xmath414 and that @xmath421 precedes the first vertex of @xmath479 . together with lemma [ longloop ] this shows that @xmath103 avoids @xmath423 and thus all of @xmath481 . in particular , @xmath424 . thus @xmath425 and so @xmath178 satisfies ( [ eq : dega ] ) . * case 2.1 . * _ there is a vertex @xmath71 for which @xmath482 receives an edge from some vertex @xmath483 . _ inequality ( [ eq : dega ] ) implies that there exists a vertex @xmath484 such that the successor @xmath432 of @xmath203 lies in @xmath103 . note that @xmath203 lies in the interior of @xmath485 since @xmath103 avoids @xmath481 . as @xmath486 and as @xmath102 is hamiltonian connected by lemma [ hlemma1 ] , there is an @xmath178-@xmath432 path @xmath92 whose interior consists precisely of all the vertices in @xmath102 . but then the cycle @xmath487 is @xmath101-ordered and contains @xmath488 more vertices than @xmath18 , a contradiction . * case 2.2 . * _ there is no such vertex @xmath71 . _ this case is similar to case 1.2 . let @xmath443 denote the set of all predecessors of vertices in @xmath384 again . let @xmath210 denote the set of all successors of vertices in @xmath19 . recall that @xmath103 avoids @xmath423 . thus @xmath443 avoids @xmath444 and thus in particular all the sets @xmath479 . consider any @xmath71 . then @xmath489 since @xmath490 by corollary [ ftcor ] . thus @xmath491 by our assumption . hence similarly as in case 1.2 one can show that @xmath492 . together with the fact that @xmath493 this gives @xmath494 be the subpath of the interval @xmath414 preceding the first vertex in @xmath210 . let @xmath451 denote all the other intervals . for each @xmath38 let @xmath452 be the auxiliary bipartite graph whose vertex classes are @xmath210 and @xmath454 and in which @xmath495 is joined to @xmath456 if @xmath496 is an edge of @xmath0 . note that @xmath458 since @xmath443 avoids @xmath444 . thus the edges of @xmath459 correspond to the edges from @xmath443 to @xmath210 . together with ( [ eq : daw2 ] ) this implies that there is some @xmath48 such that @xmath497 thus @xmath452 is not planar and so there are vertices @xmath498 and @xmath462 such that the edges @xmath499 , @xmath500 cross . as in case 1.2 we can apply corollary [ corh2paths ] to obtain disjoint @xmath501-@xmath502 paths having all their inner vertices in @xmath102 such that each @xmath471 contains at least one inner vertex ( where @xmath472 and @xmath501 is the predecessor of @xmath503 ) . thus @xmath504 is an @xmath101-ordered cycle with at least @xmath474 vertices ( note that it contains all the vertices of @xmath18 ) , contradicting the choice of @xmath18 . * case 3 . * @xmath505 . our first aim is to find vertices @xmath506 on @xmath18 with the following properties : 1 . @xmath506 occur on @xmath18 in this order and either all of these vertices are distinct or else @xmath507 and @xmath508 . 2 . @xmath101 avoids the interior of @xmath149 , the interior of @xmath509 as well as @xmath154 and @xmath164 . there are distinct vertices @xmath510 such that @xmath511 are edges . 4 . if @xmath512 ( and so @xmath506 are distinct ) then there are disjoint @xmath34-@xmath35 paths @xmath33 of length at least 2 such that all inner vertices of @xmath268 and @xmath269 lie in @xmath102 and @xmath273 . to prove the existence of such vertices , suppose first that @xmath513 and @xmath514 . then we can find two vertices @xmath515 and two vertices @xmath516 satisfying ( ii ) . then these vertices automatically satisfy ( iii ) . lemma [ longloop ] implies that they also satisfy ( i ) . finally , if they are all distinct then corollary [ corh2paths ] shows that they also satisfy ( iv ) . so suppose next that for example @xmath517 but @xmath514 . pick @xmath516 as before . to find @xmath153 and @xmath154 , first note that each vertex @xmath518 satisfies @xmath519 and so @xmath520 receives at least one edge from some vertex in @xmath521 . as each vertex in @xmath521 sends an edge to at most 7 vertices in @xmath102 , this means that there are at least @xmath522 independent edges from @xmath18 to @xmath102 . thus the interior of some interval of @xmath18 contains the endvertices of 16 of these independent edges which avoid @xmath164 and @xmath165 . let @xmath288 be the set of the first 8 endvertices of these edges on this interval and let @xmath317 be the set of the next 8 endvertices . then lemma [ h2paths ] implies that there are vertices @xmath523 and @xmath524 which together with @xmath164 and @xmath165 satisfy ( iv ) . by construction , @xmath506 are all distinct and satisfy ( ii ) and ( iii ) . again , lemma [ longloop ] implies that they also satisfy ( i ) . the cases when @xmath513 but @xmath525 and when @xmath526 are similar . so we have shown that there are vertices @xmath506 satisfying ( i)(iv ) . in what follows , we will frequently use the fact that any vertex @xmath527 receives an edge from all but at most @xmath528 vertices of @xmath18 . similarly , any vertex @xmath529 sends an edge to all but at most @xmath530 vertices of @xmath18 . * case 3.1 . * @xmath531 let @xmath420 be the set of @xmath532 vertices which immediately precede @xmath154 and let @xmath418 be the set of @xmath532 vertices which immediately precede @xmath420 . corollary [ ftcor ] implies that the successor @xmath533 of @xmath154 on @xmath18 does not lie in @xmath103 . thus @xmath533 receives an edge from some vertex @xmath427 since it receives an edge from all but at most @xmath530 vertices of @xmath18 . similarly , the predecessor @xmath534 of @xmath165 does not lie in @xmath104 and thus sends an edge to some vertex @xmath428 . lemma [ hlemma1 ] now implies that @xmath102 contains a hamilton path @xmath92 from @xmath535 to @xmath536 . but then the cycle @xmath537 is @xmath101-ordered and contains all vertices of @xmath18 except those in the interior of @xmath538 ( see figure [ fig : proofcase3 ] ) . but as @xmath539 this means that this new cycle is longer than @xmath18 , a contradiction . [ ] [ ] @xmath153 [ ] [ ] @xmath186 [ ] [ ] @xmath418 [ ] [ ] @xmath217 [ ] [ ] @xmath420 [ ] [ ] @xmath154 [ ] [ ] @xmath533 [ ] [ ] @xmath164 [ ] [ ] @xmath534 [ ] [ ] @xmath165 [ ] [ ] @xmath92 [ ] [ ] @xmath18 [ ] [ ] @xmath540 [ ] [ ] @xmath541 [ ] [ ] @xmath535 [ ] [ ] @xmath536 [ ] [ ] @xmath542 [ ] [ ] @xmath543 into a longer @xmath101-ordered cycle in case 3.1 ( left ) and case 3.2 ( right).,title="fig : " ] * case 3.2 . * @xmath544 the proof of this case is similar to that of case 3.1 . let @xmath418 be the set of @xmath532 vertices which immediately succeed @xmath164 and let @xmath420 be the set of @xmath532 vertices which immediately succeed @xmath418 . then the predecessor @xmath541 of @xmath164 sends an edge to some vertex @xmath428 and the successor @xmath540 of @xmath153 receives an edge from some vertex @xmath427 . then the @xmath101-ordered cycle @xmath545 is longer than @xmath18 , where @xmath92 is a hamilton path in @xmath102 from @xmath543 to @xmath542 ( see figure [ fig : proofcase3 ] ) . * case 3.3 . * @xmath546 let @xmath547 be a segment of the interior of @xmath548 such that @xmath549 and such that @xmath547 avoids @xmath101 . let @xmath550 be the set consisting of the first @xmath532 vertices on @xmath547 . let @xmath551 be the set consisting of the next @xmath532 vertices and define @xmath552 similarly . as by corollary [ ftcor ] the predecessor @xmath541 of @xmath164 does not lie in @xmath104 it must send an edge to some vertex @xmath553 . similarly the predecessor @xmath534 of @xmath165 sends an edge to some vertex @xmath554 , the successor @xmath540 of @xmath153 receives an edge from some vertex @xmath555 and the successor @xmath533 of @xmath154 receives an edge from some vertex @xmath556 . now lemma [ longloop ] implies that either @xmath557 or @xmath558 or both . if @xmath557 then every vertex in @xmath550 sends an edge to @xmath559 ( since every vertex outside @xmath560 sends an edge to all but at most @xmath530 vertices on @xmath18 ) . similarly , if @xmath558 then every vertex in @xmath559 receives an edge from some vertex in @xmath550 . so in both cases we can find a @xmath550-@xmath559 edge @xmath561 . but then the cycle @xmath562 is @xmath101-ordered and contains at least @xmath563 vertices more than @xmath18 , a contradiction ( see figure [ fig : proofcase33 ] ) . [ ] [ ] @xmath170 [ ] [ ] @xmath171 [ ] [ ] @xmath564 [ ] [ ] @xmath565 [ ] [ ] @xmath566 [ ] [ ] @xmath567 [ ] [ ] @xmath153 [ ] [ ] @xmath540 [ ] [ ] @xmath154 [ ] [ ] @xmath533 [ ] [ ] @xmath541 [ ] [ ] @xmath164 [ ] [ ] @xmath534 [ ] [ ] @xmath165 [ ] [ ] @xmath268 [ ] [ ] @xmath269 [ ] [ ] @xmath550 [ ] [ ] @xmath551 [ ] [ ] @xmath568 [ ] [ ] @xmath569 [ ] [ ] @xmath570 [ ] [ ] @xmath559 into a longer @xmath101-ordered cycle in case 3.3.,title="fig : " ] * case 3.4 . * _ none of cases 3.13.3 holds . _ in this case we must have that @xmath571 and can argue similarly as in case 3.3 ( see figure [ fig : proofcase34 ] ) . we omit the details . [ ] [ ] @xmath170 [ ] [ ] @xmath171 [ ] [ ] @xmath564 [ ] [ ] @xmath565 [ ] [ ] @xmath566 [ ] [ ] @xmath567 [ ] [ ] @xmath541 [ ] [ ] @xmath164 [ ] [ ] @xmath534 [ ] [ ] @xmath165 [ ] [ ] @xmath153 [ ] [ ] @xmath540 [ ] [ ] @xmath154 [ ] [ ] @xmath533 [ ] [ ] @xmath268 [ ] [ ] @xmath269 [ ] [ ] @xmath550 [ ] [ ] @xmath551 [ ] [ ] @xmath568 [ ] [ ] @xmath569 [ ] [ ] @xmath570 [ ] [ ] @xmath559 into a longer @xmath101-ordered cycle in case 3.4.,title="fig : " ] * case 4 . * _ none of cases 13 holds . _ together with lemma [ hlemma1 ] this implies that @xmath572 is even and @xmath573 . so @xmath574 . first note that any vertex @xmath518 satisfies @xmath575 moreover , if @xmath576 are distinct and if @xmath577 then by lemma [ longloop ] the special vertex @xmath408 succeeding @xmath409 on @xmath18 ( i.e. the unique vertex @xmath578 for which @xmath101 avoids the interior of @xmath406 ) can not lie in @xmath579 . thus @xmath580 and so @xmath581 * case 4.1 . * _ there exists some vertex @xmath582 . _ first note that by corollary [ ftcor ] the successor @xmath109 of @xmath68 on @xmath18 does not lie in @xmath103 . thus @xmath583 and so @xmath109 receives an edge from the predecessor @xmath110 of @xmath68 . pick any vertex @xmath584 . ( such a vertex exists since @xmath585 by ( [ eq : degh ] ) . ) note that @xmath586 since @xmath587 . by corollary [ ftcor ] the predecessor @xmath588 of @xmath69 does not send an edge to @xmath102 and so @xmath589 must be an edge ( since @xmath590 ) . now apply lemma [ hlemma1 ] to find an @xmath68-@xmath69 path @xmath92 of length at least 2 all whose inner vertices lie in @xmath102 . then @xmath591 is an @xmath101-ordered cycle which is longer than @xmath18 , a contradiction . * case 4.2 . * _ there is no vertex as in case 4.1 . _ together with ( [ eq : deghs ] ) this implies that we can find a vertex @xmath592 . we then argue similarly as in case 4.1 . this completes the proof of theorem [ thm : ordham ] . we are grateful to oliver cooley for a careful reading of the manuscript .

we investigate a communication setup where a source output is sent through a free noisy channel first and an additional codeword is sent through a noiseless but expensive channel later . with the help of the second message the decoder should be able to decide with zero - error whether its decoding of the first message was error - free . this scenario leads to the definition of a digraph parameter that generalizes witsenhausen s zero - error rate for directed graphs . we investigate this new parameter for some specific directed graphs and explore its relations to other digraph parameters like sperner capacity and dichromatic number . when the original problem is modified to require zero - error decoding of the whole message then we arrive back to the witsenhausen rate of an appropriately defined undirected graph . keywords : zero - error , graph products , sperner capacity , dichromatic number , witsenhausen rate

1406.0767

consider the following situation . alice writes a message to bob consisting of the numbers of several bank accounts to which bob has to send some money . she writes in a hurry ( she just got to know that the transfers are urgent if they do not want to pay delay punishment , but currently she has little time ) . therefore her characters are not very well legible , so bob may misread some numbers . however , there are some rules for the possible mistakes , e.g. , a @xmath0 may be thought to be a @xmath1 but never a @xmath2 . this relation between the possible digits need not be symmetric : it is possible that a @xmath3 is sometimes read as a @xmath2 but a @xmath2 may not be decoded as a @xmath3 . these rules of possible confusions are known both by alice and by bob . as alice is aware of the possibility that bob misread her message , later in the day she sends another message to bob , the goal of which is to make bob certain whether he read ( decoded ) the first message correctly or not . if he did he can transfer the money with complete confidence that he sends it to the right accounts . if he did not he will know that he does not know the account numbers correctly and so he better wait and pay the punishment than transfer the money to the wrong place . the second message will be received by bob correctly for certain , but it uses an expensive device , e.g. , alice sends it as an sms from another country after she has arrived there . ( now we understand why she was in a hurry : she had to arrive to the airport in time . ) for some reason , every character sent from this foreign country costs a significant amount of money for her . so she wants to send the shortest possible message that makes it sure ( here we insist on zero - error ) that bob will know whether his decoding of the original handwritten message was error - free or not . the problem is to determine the best rate of communication over the second channel as the length of the original message received tends to infinity . in section [ sec : dil ] we describe the abstract communication model for this scenario and show that the best achievable rate is a parameter of an appropriate directed graph . we will see that this parameter of a directed graph is a generalization of the parameter ( of an undirected graph ) called witsenhausen rate . ( this means that we also obtain a new interpretation of witsenhausen rate . ) in section [ sec : bounds ] we investigate the relationship with other graph parameters . these include sperner capacity and the dichromatic number . the former is a generalization of shannon s graph capacity @xcite to directed graphs . though originally defined to give a general framework for some problems in extremal set theory ( see @xcite ) , sperner capacity also has its own information theoretic relevance , see @xcite . the dichromatic number is a generalization of the chromatic number to directed graphs introduced in @xcite . using the above mentioned relations we determine our new parameter for some specific directed graphs . in section [ sec : comp ] we consider a compound channel type version of the problem parallel to @xcite . in section [ sec : extr ] some connections to extremal set theory are pointed out . in section [ sec : ambit ] we will consider the setup where the requirement is more ambitious and we want that alice s second ( the error - free but expensive ) message make bob able to decode the original message with zero - error . ( that is , he will know the message itself not only the correctness or incorrectness of his original decoding of alice s handwriting . ) we will see that this setting leads to the witsenhausen rate of an undirected graph related to the problem . this gives a further new interpretation of witsenhausen rate . all logarithms are meant to be of base @xmath4 . the abstract setting for our communication scenario is the following . we have a source whose output is sent through a noisy channel . ( this belongs to alice s handwriting . ) the input and output alphabets of this channel are identical and they coincide with the output alphabet of the source . it is known how the noisy channel can deform the input , in particular we know what ( input ) letters can become a certain , possibly different ( output ) letter on the other side . ( we always assume though that every letter can result in itself , that is get through the noisy channel without alteration . ) later another message is sent ( by the same sender ) to the same receiver . this second message is sent via a noiseless channel and its goal is to make zero - undetected - error decoding possible , i.e. , after having received this second message the receiver should be able to decide whether it decoded the first message correctly . the use of the noiseless channel is expensive , so the second message should be kept as short as possible . let the shortest possible message that satisfies the criteria have length @xmath5 when @xmath6 characters of the source output are encoded together . let @xmath7 denote the noisy channel . the efficiency of the communication is measured by the quantity @xmath8 that we call the dilworth rate of the noisy channel @xmath7 . ( for an explanation of the name see remark [ rem : name ] in subsection [ subsect : dilwdef ] . ) [ rem : cvsr ] note the special feature of the problem that we characterize a channel by a rate , that is with a parameter that , unlike channel capacity , we want to be as small as possible . the reason is that we measure the reliability of a channel not by the amount of information it can safely transfer but with the amount of information needed _ to be added _ for making the communication reliable . @xmath9 the relevant properties of @xmath7 are described by a directed graph @xmath10 having the ( common input and output ) alphabet as its vertex set and the following edge set . an ordered pair @xmath11 of two letters forms a directed edge of @xmath10 if and only if @xmath12 and the output of @xmath7 can be @xmath13 when it is fed by @xmath14 at the input . [ rem : ve ] as usual we will use @xmath15 to denote the vertex set and @xmath16 to denote the edge set of a directed graph @xmath17 . we will use similar notation for undirected graphs that we always consider to be the same as a _ symmetrically directed graph_. in such a graph an ordered pair @xmath18 of two vertices forms a directed edge if and only if the reversely ordered pair @xmath19 is also present in the digraph as a directed edge . we will use the term _ oriented graph _ for directed graphs that do not contain any edge together with its reversed version . that is @xmath17 is an oriented graph if @xmath20 implies @xmath21 . as it is also customary , the term _ digraph _ will be used as a synonym for `` directed graph '' . @xmath9 to express @xmath22 as a graph parameter we need the following notion . the and product @xmath23 of two directed graphs @xmath17 and @xmath24 is defined as follows . the vertex set of @xmath25 is the direct product @xmath26 and vertex @xmath27 sends a directed edge to @xmath28 iff either @xmath29 and @xmath30 or @xmath29 and @xmath31 or @xmath32 and @xmath30 . the @xmath6-th and power of a digraph @xmath24 , denoted by @xmath33 is the @xmath6-wise and product of digraph @xmath24 with itself . observe that this graph exponentiation extends to sequences of letters the relation between individual letters @xmath34 and @xmath35 expressing that feeding @xmath34 to the noisy channel @xmath7 may result in observing letter @xmath35 at the output . a sequence of letters at the input of @xmath7 can result in another such sequence at the output of @xmath7 if at each coordinate the character in the first sequence can result in the corresponding character of the second sequence . ( this includes the possibility that the character does not change when sent through @xmath7 . ) [ rem : ao ] the terminology of graph products is not completely standardized . the and product we just defined is also called normal product @xcite , strong direct product @xcite , or strong product @xcite . we follow the paper of alon and orlitsky @xcite when use the name and product , because we find this name informative . a similar remark applies to the or product that we will introduce later in definition [ defi : or ] . @xmath9 recall that the chromatic number @xmath36 of a graph @xmath37 is the minimal number of colors that suffice to color the vertices of @xmath37 so that adjacent vertices get different colors . if @xmath17 is a digraph , its chromatic number @xmath38 is understood to be the chromatic number of the underlying undirected graph . [ prop : chi ] @xmath39 it is easy to see that the above limit always exists . ( the reason is the submultiplicative behaviour of the chromatic number under the and product ) . @xmath9 * proof . * alice and bob can agree in advance in a proper coloring of @xmath40 with @xmath41 colors . alice can send bob the color of the vertex belonging to the original source output using @xmath42 bits . bob compares this to the color of the vertex representing the sequence he obtained as a result of his decoding . if the latter color is identical to the one alice has sent him , then he can be sure that his decoding was error - free . this is because any other sequence that could result in his decoded sequence is adjacent ( in @xmath43 ) to this decoded sequence , so its color is different . on the other hand , if alice sent a shorter message through the noiseless channel , then she could not have @xmath41 distinct messages and thus there must exist two adjacent vertices in @xmath43 that are encoded to the same codeword by alice ( for the noiseless channel ) . then one of the two sequences represented by these two adjacent vertices could result in the other one , while this other one could also result in itself . thus bob can not make the difference between these two sequences , one of which is the correct source output sequence while the other one differs from it . so receiveing this message bob could not be sure whether his decoding was error - free or not . @xmath44 the right hand side expression in proposition [ prop : chi ] can be considered as a digraph parameter that we will call the dilworth rate of the digraph @xmath10 . [ defi : dilrate ] for a directed graph @xmath24 we define its ( logarithmic ) _ dilworth rate _ to be @xmath45 the non - logarithmic dilworth rate is @xmath46{\chi(\vec{g}^{\wedge t}}).\ ] ] obviously , @xmath47 [ rem : name ] let @xmath48 be the directed graph on @xmath4 vertices with a single directed edge . if we consider the vertices of @xmath49 as characteristic vectors of subsets of a @xmath6-element set then @xmath50 can be interpreted as the asymptotic exponent of the minimum number of antichains ( sets of pairwise incomparable elements ) in the boolean lattice of these subsets that can cover all the subsets . the exact value of this minimum number is given ( easily ) by a special case of what is called the `` dual of dilworth s theorem '' @xcite ( also called mirsky s theorem , see @xcite ) . this connection to dilworth s celebrated result is the reason for calling our new parameter dilworth rate . note that the name sperner capacity was picked by the authors of @xcite for analoguous reasons : the sperner capacity of the digraph @xmath48 has a similar relationship with sperner s theorem @xcite . @xmath9 the and product is also defined for undirected graphs . considering undirected graphs as symmetrically directed graphs the definition is straightforward . witsenhausen considered the `` zero - error side - information problem '' that led him to introduce the quantity @xmath51 that is called the witsenhausen rate of ( the undirected ) graph @xmath52 . it is straightforward from the definitions that if @xmath24 is a symmetrically directed graph and @xmath52 is the underlying undirected graph ( that we consider equivalent ) , then @xmath53 . thus dilworth rate is indeed a generalization of witsenhausen rate to directed graphs . [ rem : nrfam ] we note that nayak and rose @xcite defines what they call `` the witsenhausen rate of a set of directed graphs '' . though formally this gives the dilworth rate of a directed graph , the focus of @xcite is elsewhere . when its motivating setup results in a family consisting of a single digraph , then this digraph is symmetrically directed . ( see also theorem [ thm : wm ] in section [ sec : comp ] . ) sperner capacity was introduced by gargano , krner and vaccaro @xcite . traditionally this parameter is defined by using the or product . [ defi : or ] the or product @xmath54 of directed graphs @xmath17 and @xmath24 has vertex set @xmath26 and @xmath27 sends a directed edge to @xmath28 iff either @xmath55 or @xmath56 . the @xmath6-th or power @xmath57 is the @xmath6-wise or product of digraph @xmath24 with itself . let @xmath58 denote the complete directed graph on @xmath59 vertices , that is the one we obtain from a(n undirected ) complete graph @xmath60 when substituting each of its edges @xmath61 by the two oriented edges @xmath11 and @xmath62 . the ( directed ) complement of a digraph @xmath24 is the directed graph @xmath63 on vertex set @xmath64 having edge set @xmath65 now we note the straightforward relation of the and and or powers that @xmath66 . the ( logarithmic ) sperner capacity of digraph @xmath24 is defined ( see @xcite ) as @xmath67 where @xmath68 denotes the _ symmetric clique number _ , that is the cardinality of the largest symmetric clique in digraph @xmath17 : the size of the largest set @xmath69 where for each @xmath70 both @xmath71 and @xmath72 are edges of @xmath17 . using the above relation of the and and or products , sperner capacity ( of the complementary graph @xmath73 ) can also be defined as @xmath74 where @xmath75 stands for the independence number ( size of the largest edgeless subset of the vertex set ) of graph @xmath17 . this is the definition given in @xcite . ( the authors of @xcite call this value the sperner capacity of @xmath24 . ) when @xmath52 is an undirected ( or symmetrically directed ) graph , then @xmath76 , the _ shannon capacity _ of graph @xmath52 ( see @xcite ) . we will need a sort of probabilistic refinement of our capacity - like parameters called their `` within - a - type '' versions , see @xcite . first we need the concept of @xmath77-typical sequences , cf . @xcite . [ defi : peps ] let @xmath78 be a finite set , @xmath79 a probability distribution on @xmath78 , and @xmath80 . a sequence @xmath81 in @xmath82 is said to be @xmath77-typical if for every @xmath83 we have @xmath84 , where @xmath85 . we denote the set of @xmath77-typical sequences in @xmath82 by @xmath86 . when @xmath87 we also write @xmath88 for @xmath89 . if @xmath90 we say that the _ type _ of @xmath81 is @xmath79 . let @xmath91 stand for either @xmath33 or @xmath57 . for a directed ( or undirected ) graph @xmath17 and @xmath69 we denote by @xmath92 $ ] the digraph induced by @xmath17 on the subset @xmath93 of the vertex set . we also use the shorthand notation @xmath94 $ ] . let @xmath95 be either of the following graph parameters of the directed graph @xmath24 : independence number , clique number , chromatic number , clique cover number ( which is the chromatic number of the complementary graph ) , symmetric clique number , or transitive clique number . ( the latter is the size of the largest subset @xmath93 of @xmath64 the elements of which can be linearly ordered so that if @xmath96 precedes @xmath97 then the oriented edge @xmath18 is present in @xmath98 . ) let the asymptotic parameter @xmath99 be defined as @xmath100 while @xmath101 stands for @xmath102 . [ defi : captyp ] the parameter @xmath103 of a digraph @xmath24 within a given type @xmath79 is the value @xmath104 we note that for several of the allowed choices of @xmath95 and @xmath91 we obtain a graph parameter that already exists in the literature . for example , when @xmath105 and the power we look at is the or power , we get sperner capacity within a given type , that has an important role in the main results of the papers @xcite . if we choose @xmath106 and the or power , we obtain the functional called graph entropy , which is defined in @xcite and has several nice properties , see @xcite , as well as important applications , see e.g. @xcite . when @xmath106 but the exponentiation is the and power , then we arrive to the within a type version of dilworth rate @xmath107 . the special case of this for an undirected graph @xmath52 was already known under name `` complementary graph entropy '' that could justifiably be called `` witsenhausen rate within a given type '' . this parameter was introduced by krner and longo @xcite and further investigated by marton @xcite . although this within - a - type version of witsenhausen s invariant was introduced earlier than the non - probabilistic version ( cf . @xcite ) , for the sake of consistancy we denote it by @xmath108 . note that marton @xcite proved the important identity @xmath109 where @xmath110 is the entropy of the probability distribution @xmath111 . this holds for any probability distribution @xmath79 on @xmath112 . along the same lines one can also prove the following theorem . we give its proof for the sake of completeness . we will use the notion of fractional chromatic number @xmath113 in the proof . let @xmath114 denote the set of independent sets in @xmath52 . a function @xmath115 is a fractional coloring of @xmath52 if for every vertex @xmath116 we have @xmath117 , that is the sum of the weights @xmath118 puts on independent sets containing @xmath97 is at least @xmath1 . ( a proper colorig is also a fractional coloring : the color classes get weight @xmath1 , the other independent sets get weight @xmath3 . ) the fractional chromatic number is @xmath119 , that is the minimum ( taken over all fractional colorings ) of the total weight put on independent sets by a fractional coloring @xmath118 . ( formally we should write infimum but it is known that the minimum is always attained . see the book @xcite for a detailed account on fractional graph parameters . ) we will need the following properties of the fractional chromatic number . [ defi : vtrans ] a directed graph @xmath24 is vertex - transitive if for any two vertices @xmath120 it admits an automorphism that maps @xmath96 to @xmath97 . if @xmath37 is a vertex - transitive graph , then @xmath121 ( for a proof see @xcite , proposition 3.1.1 on page 41 . ) for every graph @xmath37 we have @xmath122{\chi(f^{\odot t}})=\lim_{t\to\infty}\sqrt[t]{\chi_f(f^{\odot t}}).\ ] ] the latter follows from lovsz s result @xcite stating that @xmath123 and the obvious inequality @xmath124 that holds for all ( finite simple ) graphs . [ thm : sumhp ] let @xmath24 be a directed graph and @xmath79 an arbitrary fixed probability distribution on @xmath64 . then @xmath125 * proof . * note that by the well - known ( and more or less trivial ) inequality @xmath126 for every graph @xmath37 , we have @xmath127 . clearly , this relation also holds if we have a directed graph @xmath17 in place of the undirected graph @xmath37 . this is straightforward since @xmath38 and @xmath75 are defined to be identical to the corresponding parameter of the underlying undirected graph @xmath37 . it is also well - known ( cf . e.g. @xcite ) that @xmath128 . the last two relations immediately give @xmath129 for the reverse inequality let us fix a sequence of probability distributions @xmath130 on the vertex set of our graph so that @xmath131 and @xmath132 notice that @xmath133 is a vertex - transitive graph , since every sequence forming an element of @xmath134 can be transformed into any other such sequence by simply permuting the coordinates . thus @xmath135 @xmath44 using standard techniques of the method of types , cf . @xcite we can already state our lower bound on @xmath136 . we need the fact that fixing the length @xmath6 the number of distinct types of a sequence over some fixed alphabet is only a polynomial funcion of @xmath6 ( cf . the type counting lemma 2.2 in @xcite ) , while the parameters we investigate are asymptotic exponents of some graph parameters that grow exponentially as @xmath6 tends to infinity . with this in mind we can write @xmath137 [ thm : lowerb ] @xmath138 * proof . * using the above equalities , we obtain @xmath139 this gives the lower bound in the statement . @xmath44 note that sperner capacity is unkown for many graphs , so the lower bound above usually does not give a known numerical value . still , there are some examples of graphs where sperner capacity is known and is non - trivial . a basic example is the cyclically oriented triangle , or more generally , any cyclically oriented cycle . first we formulate a consequence of the above formula . [ cor : vtrans ] if @xmath24 is a vertex - transitive digraph then @xmath140 * proof . * let @xmath141 denote the uniform distribution on the vertex set of @xmath24 . if @xmath24 is vertex - transitive then by symmetry @xmath142 and @xmath143 . combining these equalities with theorem [ thm : lowerb ] we obtain @xmath144 and thus the statement . @xmath44 now we will use this corollary to determine the dilworth rate of the cyclically oriented @xmath145-length cycle @xmath146 for every @xmath145 . note that the complement of a cyclically oriented cycle is a cyclically oriented cycle of the same length together with all diagonals as bidirected ( or equivalently , undirected ) edges . for @xmath147 there are no diagonals , so the cyclic triangle is isomorphic to its complement . the sperner capacity of the cyclic triangle was determined in @xcite , cf . also @xcite , and its value is @xmath148 . this result was generalized by alon @xcite , who proved that the sperner capacity of a digraph @xmath24 is bounded from above by @xmath149 where @xmath150 and @xmath151 stand for the maximum outdegree and maximum indegree of @xmath17 , respectively . the indegree and outdegree of a vertex @xmath97 is the number of edges at @xmath97 that are oriented towards or outwards @xmath97 , respectively . @xcite for a further generalization of alon s result . ) on the other hand , sperner capacity is bounded from below by ( the logarithm of ) the transitive clique number , the number of vertices in a largest transitively directed complete subgraph , denoted by @xmath152 . ( this is an easy observation which implies that substituting @xmath153 by @xmath154 in the definition of sperner capacity gives the same value , i.e. it gives an alternative definition of sperner capacity , see @xcite and also proposition 4 and the remark following it in @xcite . ) note that a transitively directed complete subgraph meant here is not necessarily induced . it is allowed that some reverse edges are also present on the same subset of vertices . [ cor : ck ] the dilworth rate of the cyclically oriented @xmath145-cycle is @xmath155 * proof . * let the directed complement of @xmath146 be denoted by @xmath156 . since @xmath157 , alon s above mentioned result implies that the sperner capacity of @xmath156 is at most @xmath158 . it is easy to see that @xmath159 , so the lower bound mentioned above is also @xmath158 . since the above two bounds coincide , the sperner capacity of @xmath160 is equal to @xmath161 . using that @xmath162 is vertex transitive corollary [ cor : vtrans ] implies the statement . @xmath44 note that corollary [ cor : ck ] shows that the dilworth rate is a true generalization of witsenhausen rate since @xmath163 if @xmath164 . [ defi : acyc ] call a subset of the vertex set of a directed graph @xmath24 acyclic if it induces an acyclic subgraph . the latter means that there is no oriented cycle on these vertices . the acyclicity number @xmath165 of a directed graph @xmath24 is the number of vertices in a largest acyclic subset of @xmath64 . note that unlike for a transitive clique we do not allow reverse edges in an acyclic subgraph . let @xmath166 be an odd number . the following tournaments ( oriented complete graphs ) are also generalizations of the cyclic triangle . let @xmath167 and @xmath168 is an edge iff @xmath169 for some @xmath170 . ( figure 1 shows the tournament @xmath171 . ) note that it holds for every directed graph that reversing all of its edges does not change the value of either its sperner capacity or of its dilworth rate . this implies that if @xmath172 is a tournament then we have @xmath173 and @xmath174 . by @xmath175 all the four values are equal . . ] [ cor : tourn ] for all odd integers @xmath176 we have @xmath177 * proof . * we know that @xmath178 holds for all directed graphs ( cf . @xcite and also the discussion before corollary [ cor : ck ] for an equivalent statement concerning the complementary digraph ) . this gives @xmath179 by @xmath180 ( see the note before stating the corollary ) alon s result can be applied implying that our lower bound is sharp . since @xmath181 is vertex - transitive we can apply corollary [ cor : vtrans ] to complete the proof . @xmath44 observe that corollary [ cor : tourn ] shows not only that the value of the dilworth rate of an oriented graph may differ from the witsenhausen rate of the underlying undirected graph , but also that the difference can be arbitrarily large . indeed , denoting the complete graph on @xmath182 vertices by @xmath183 we have @xmath184 for every @xmath185 . the left hand side of the inequality is bounded above by @xmath186 , while the right hand side goes to infinity with @xmath182 . now we show that the ( logarithm of the ) _ dichromatic number _ defined in @xcite is an upper bound on the dilworth rate . [ def : dic ] the _ dichromatic number _ @xmath187 of a directed graph @xmath24 is the minimum number of acyclic subsets that cover @xmath64 . a partition of @xmath64 into acyclic subsets will be called a _ directed coloring _ or _ dicoloring_. we note that an undirected edge ( meaning a bidirected edge ) is considered to be a @xmath4-length cycle , therefore its two endpoints can not be both contained in an acyclic set . this shows that for undirected ( equivalently , symmetrically directed ) graphs the dichromatic number is equal to the chromatic number . we do not use the term `` acyclic coloring '' , because it is already used for a completely different concept , see @xcite . @xmath9 [ thm : upb ] for any directed graph @xmath24 @xmath188 * proof . * let us fix a directed coloring of digraph @xmath24 consisting of @xmath189 acyclic subsets ( `` color classes '' ) . for each @xmath190 let @xmath191 denote the color class that contains @xmath97 . now consider @xmath33 . it has @xmath192^t$ ] vertices . for each sequence @xmath193 we attach the sequence of colors @xmath194 . there are @xmath195 such color sequences , so this gives a partition of @xmath196 into @xmath195 partition classes . we also give another partition of @xmath196 according to types . two vertices are in the same partition class if their type is the same . by the type counting lemma 2.2 in @xcite , we know that this latter partition has at most @xmath197 , that is a polynomial number ( in @xmath6 ) of classes . now let @xmath198 be the common refinement of these two partitions . we have @xmath199 by the foregoing . now we show that each partition class @xmath200 induces an independent set in @xmath33 . let two sequences @xmath201 and @xmath202 belong to the same @xmath200 , that is , they have the same type and @xmath203 . let @xmath204 be an index for which @xmath205 . since @xmath206 and @xmath207 are in the same color class of a valid dicoloring , we can not have both @xmath208 and @xmath209 present in the graph as a directed edge . if neither is present then there is no edge between @xmath210 and @xmath211 . if @xmath212 , then we know that @xmath213 , so the oriented edge @xmath214 can not be an edge of @xmath33 . we need to show that neither the opposite oriented edge @xmath215 can be present in @xmath33 . this is because if @xmath212 , then there should be another position @xmath216 for which @xmath217 . if this is true , then @xmath218 and so @xmath219 . the existence of @xmath216 with the above property holds for the following reason . consider all coordinates @xmath220 , for which @xmath221 . denote the set of these @xmath220 s @xmath222 . since vertices @xmath223 with the same `` color '' @xmath191 induce an acyclic subdigraph , we can put these vertices into a linear order so , that @xmath224 implies that @xmath97 precedes @xmath225 in this linear order . so if @xmath212 , then for our edge @xmath208 @xmath206 precedes @xmath207 . however , since @xmath210 and @xmath211 has the same type it can not happen that for each @xmath226 @xmath227 precedes @xmath228 in this linear order . so there must be a coordinate @xmath216 where @xmath229 precedes @xmath230 and this implies the claimed properties . thus each partition class @xmath200 is independent indeed , so the undirected graph underlying @xmath33 can be properly colored with @xmath231 colors . this implies that @xmath232{\chi([\vec{g}_h]^{\wedge t})}\le \liminf_{t\to\infty}\sqrt[t]{(t+1)^{|v(\vec{g})|}k^t}=k=\chi_{\rm dir}(\vec{g}).$ ] @xmath44 as a strengthening of the previous theorem we will show that we can also write a natural fractional relaxation of the dichromatic number on the right hand side above . ( we could not prove this right away , as the weaker statement will be used in the proof . ) to prove this stronger statement we need some preparation , in particular we will use the following observations . first note , that @xmath233 holds for any digraph @xmath17 . this is simply because independent sets in @xmath17 are special acyclic sets , so any proper coloring of @xmath17 is also a directed coloring of @xmath17 . [ prop : submult ] the dichromatic number is submultiplicative with respect to the and product , i.e. @xmath234 in particular , @xmath235^t.\ ] ] a straightforward consequence of proposition [ prop : submult ] is that the limit @xmath236{\chi_{\rm dir}(f^{\wedge t})}$ ] exists . * proof of proposition [ prop : submult ] * let @xmath237 and @xmath238 be optimal directed colorings of the digraphs @xmath17 and @xmath24 , respectively . using these colorings we define the function @xmath239 as follows . for @xmath240 let @xmath241 . observe that the and product of two acyclic ( sub)graphs results in an acyclic ( sub)graph . assume for contradiction that @xmath242 and @xmath243 are acyclic subsets of @xmath15 and @xmath64 , respectively , and @xmath244 $ ] contains a directed cycle . ( recall that @xmath245 $ ] denotes the digraph @xmath246 induces on @xmath247 . ) let its vertices be @xmath248 in the ( cyclic ) order the cycle defines , i.e. @xmath249 is an edge of @xmath250 for all @xmath251 where addition is intended modulo @xmath145 . we may assume without loss of generality that not all @xmath252 s are equal . then in the sequence @xmath253 we have for all @xmath254 either @xmath255 or @xmath256 ( addition is again modulo @xmath145 ) and for some @xmath257 the second case occurs . but then there must be a directed cycle in @xmath258 $ ] contradicting the assumption that @xmath242 is acyclic . the above implies that @xmath259 is a directed coloring of @xmath25 . as it uses @xmath260 colors the statement is proved . [ lem : nemno ] for any digraph @xmath17 and positive integer @xmath145 we have @xmath261^k.\ ] ] * proof . * fix an arbitrary positive integer @xmath145 . we can write @xmath262{\chi(\vec{f}^{\wedge mk})}\\ = & \lim_{m\to\infty}\sqrt[k]{\sqrt[m]{\chi([\vec{f}^{\wedge k}]^{\wedge m})}}\\ = & \sqrt[k]{\lim_{m\to\infty}\sqrt[m]{\chi([\vec{f}^{\wedge k}]^{\wedge m})}}\\ = & \sqrt[k]{r_{\rm d}(\vec{f}^{\wedge k } ) } , \end{array}\ ] ] that implies the statement . @xmath44 [ prop : asympt ] for any digraph @xmath17 we have @xmath122{\chi_{\rm dir}(\vec{f}^{\wedge t})}=r_{\rm d}(\vec{f}).\ ] ] * proof . * by @xmath263 we have @xmath122{\chi_{\rm dir}(\vec{f}^{\wedge t})}\le \lim_{t\to\infty}\sqrt[t]{\chi(\vec{f}^{\wedge t})}=r_{\rm d}(\vec{f}).\ ] ] for the reverse inequality we can write @xmath264{r_{\rm d}(\vec{f}^{\wedge t})}\le\lim_{t\to\infty}\sqrt[t]{\chi_{\rm dir}(\vec{f}^{\wedge t})},\ ] ] where the equality follows by noticing that lemma [ lem : nemno ] above is valid for all positive integers @xmath6 and the inequality is a consequence of @xmath265 applied for @xmath266 . @xmath44 [ def : fracdic ] let the set of subsets of the vertex set inducing an acyclic subgraph in a digraph @xmath24 be @xmath267 . a function @xmath268 is called a _ fractional directed coloring _ ( or fractional dicoloring ) if for @xmath269 we have @xmath270 the _ fractional dichromatic number _ of @xmath24 is @xmath271 where the minimum is taken over all fractional directed colorings @xmath118 of @xmath24 . note the obvious inequality @xmath272 for any digraph @xmath24 . we will need the following lemma . [ lem : dicfsub ] for any digraphs @xmath17 and @xmath24 we have @xmath273 * proof . * let @xmath34 and @xmath118 be optimal fractional directed colorings of @xmath17 and @xmath24 , respectively . we use the observation , already verified in the proof of proposition [ prop : submult ] , stating that if @xmath274 and @xmath275 then the direct product @xmath276 is in @xmath277 , i.e. @xmath276 induces an acyclic subdigraph in @xmath25 . now give the following weights @xmath278 to the acyclic sets of @xmath25 . if @xmath279 has a product structure , i.e. @xmath280 for some @xmath281 and @xmath282 , then let @xmath283 . if @xmath7 is not of this form , then let @xmath284 . for any @xmath285 we have @xmath286 thus @xmath278 is a fractional dicoloring of @xmath25 . now we have @xmath287 this completes the proof . @xmath44 [ cor : powfsub ] for any digraph @xmath24 and any positive integer @xmath6 we have @xmath288^t.\ ] ] @xmath44 we also need the following result . [ prop : asyfrac ] for any digraph @xmath17 we have @xmath122{\chi_{{\rm dir},f } ( \vec{f}^{\wedge t})}=r_{\rm d}(\vec{f}).\ ] ] for the proof we need some preparation . a hypergraph @xmath289 consists of a vertex set @xmath290 and an edge set @xmath291 , where the elements of @xmath291 are subsets of @xmath78 . a covering of hypergraph @xmath292 is a set of edges the union of which contains all elements of @xmath293 . let @xmath294 denote the minimum number of edges in a covering of @xmath295 . a fractional covering of a hypergraph @xmath289 is a function @xmath296 satisfying for every @xmath297 that @xmath298 . the fractional covering number is @xmath299 where the minimization is over all fractional covers @xmath118 . clearly , @xmath300 . lovsz proved in @xcite ( cf . also @xcite ) that @xmath301 where @xmath302 , that is the cardinality of a largest edge in @xmath292 . for a directed graph @xmath24 let @xmath303 where @xmath304 , i.e. it consists of the acyclic subsets of vertices in @xmath24 . it is straightforward that @xmath305 and @xmath306 while @xmath307 . thus the above result implies that @xmath308 * proof of proposition [ prop : asyfrac ] * we have @xmath236{\chi_{{\rm dir},f}({\vec{f}}^{\wedge t})}\le \lim_{t\to\infty}\sqrt[t]{\chi_{\rm dir}({\vec{f}}^{\wedge t})}=r_{\rm d}({\vec{f}})$ ] by proposition [ prop : asympt ] and the obvious inequality @xmath309 applied to @xmath310 . for the reverse inequality we write @xmath311{\chi_{\rm dir}(\vec{f}^{\wedge t})}\\ \le&\lim_{t\to\infty}\sqrt[t]{\chi_{{\rm dir},f}(\vec{f}^{\wedge t})(1+\log a(\vec{f}^{\wedge t}))}\\ = & \left(\lim_{t\to\infty}\sqrt[t]{\chi_{{\rm dir},f}(\vec{f}^{\wedge t})}\right)\times\\&\times\left(\lim_{t\to\infty}\sqrt[t ] { ( 1+\log a(\vec{f}^{\wedge t}))}\right)\\ = & \lim_{t\to\infty}\sqrt[t]{\chi_{{\rm dir},f}(\vec{f}^{\wedge t})}. \end{array}\ ] ] @xmath44 [ thm : frupb ] for any directed graph @xmath24 @xmath312 * proof . * from the above we have @xmath313{\chi_{{\rm dir},f}(\vec{g}^{\wedge t})}\\&\le \lim_{t\to\infty}\sqrt[t]{[\chi_{{\rm dir},f}(\vec{g})]^t } = \chi_{{\rm dir},f}(\vec{g } ) . \end{array}\ ] ] @xmath44 there are several directed graphs @xmath24 for which the above upper bound is sharp . in particular , corollaries [ cor : ck ] and [ cor : tourn ] can be proved using theorem [ thm : frupb ] instead of vertex - transitivity . ( an optimal fractional dicoloring of @xmath171 is shown on figure 2 . ) . ] note however , that theorem [ thm : frupb ] is not always tight : for the graph ( symmetrically directed digraph ) @xmath314 , we have @xmath315 by results in @xcite and @xcite , while @xmath316 we present another such example which is not symmetrically directed . let the @xmath317-length cycle be oriented in an ( as much as possible ) alternating manner , that is so that only one of its vertices will have outdegree @xmath1 ( implying that two of the @xmath318 others will have outdegree @xmath4 , and the remaining @xmath4 have outdegree @xmath3 ) . denote this oriented graph by @xmath319 . ( see figure 3 . ) . ] we know that @xmath320 . ( this is proven as proposition 4 in @xcite , see @xcite for more details on the sperner capacity of oriented self - complementary graphs . all other orientations of the @xmath317-cycle have sperner capacity @xmath186 , see @xcite and @xcite . by theorems [ thm : lowerb ] and [ thm : frupb ] this implies that the dilworth rate of their complements is @xmath321 . ) thus by theorem [ thm : lowerb ] for the complement of @xmath319 we have @xmath322 or equivalently @xmath323 . ( the digraph @xmath324 is shown on figure 4 . ) ( complement of @xmath319 ) . bidirected edges are shown as undirected ones . ] [ prop : gyok6 ] @xmath325 * proof . * the first inequality was already given above . to prove the second inequality we give @xmath2 acyclic sets of vertices in the second power @xmath326^{\wedge 2}$ ] of our graph , that cover all vertices in @xmath327^{\wedge 2})$ ] . the existence of this covering implies that @xmath328^{\wedge 2})\le\chi_{\rm dir}([\vec{a}_5^c]^{\wedge 2})\le 6 $ ] thus by lemma [ lem : nemno ] we get @xmath329 . let us denote the vertices of @xmath330 by @xmath331 in their cyclic order , so that in @xmath319 we have @xmath332 and the unique outneighbor of @xmath333 is @xmath4 . ( that is , the outdegree @xmath1 vertex is @xmath333 , and thus the outdegree @xmath4 vertices in @xmath319 are @xmath318 and @xmath1 , while @xmath4 and @xmath3 have outdegree @xmath3 . ) the following six subsets of @xmath334 induce acyclic subgraphs of @xmath335^{\wedge 2})$ ] that entirely cover its vertex set : @xmath336 @xmath337 @xmath338 @xmath339 @xmath340 @xmath341 one can check that within all these five sets if @xmath342 is to the left of @xmath343 in the same line above ( @xmath344 may not all be different ) , then either @xmath345 or @xmath346 ( or both ) form an edge of @xmath319 , thus this is a missing edge in @xmath324 . this implies that as a vertex of @xmath326^{\wedge 2}$ ] the pair @xmath347 does not send an edge to vertex @xmath348 , therefore the corresponding set of vertices induces an acyclic subgraph in @xmath326^{\wedge 2}$ ] . this completes the proof of the second inequality . to see that @xmath349 it is enough to realize that any @xmath333 vertices of @xmath324 contains a bidirected edge , thus any acyclic induced subgraph has at most two vertices . to see equality we can put weight @xmath350 on all the five @xmath4-element acyclic subsets . @xmath44 [ rem : coinc ] getting the same upper bound for the values determined in corrollaries [ cor : ck ] and [ cor : tourn ] in two different ways above is not pure coincidence . it follows from the fact that if @xmath24 is vertex - transitive then @xmath351 note that this is a generalization of the relation , that for every vertex - transitive ( undirected ) graph @xmath52 @xmath352 that we already referred to right after definition [ defi : vtrans ] in subsection [ subsec : sperner ] . this latter equality is presented in @xcite ( proposition 3.1.1 on page 41 ) as a consequence of a more general equality concerning vertex - transitive hypergraphs ( see proposition 1.3.4 on page 7 of @xcite ) . a vertex - transitive hypergraph is a hypergraph that attains for every pair @xmath353 of its vertices an automorphism that maps @xmath96 to @xmath97 . proposition 1.3.4 in @xcite states that if @xmath289 is a vertex - transitive hypergraph then @xmath354 where ( as before ; cf . the discussion after stating proposition [ prop : asyfrac ] ) @xmath355 for a directed graph @xmath24 we attach again the hypergraph @xmath303 where @xmath304 . it is straightforward that if @xmath24 is vertex - transitive then so is @xmath356 . the equality quoted for @xmath357 from @xcite gives the stated equality @xmath358 for vertex - transitive digraphs @xmath24 . imagine that the handwritten message is left to bob by one of his three secretaries but it is not known in advance which one . their handwriting is rather different and this has two consequences that are important for us . one is that the possible mistakes bob can make when decoding the message are different depending on which secretary wrote him the message . ( for example , in the first secretary s handwriting a @xmath0 can be thought to be a @xmath1 , while the second secretary `` crosses '' the leg of @xmath0 , so it can never look like a @xmath1 , however it can be confused with a @xmath318 , etc . ) this means that in place of the noisy channel @xmath7 we had so far , now there are three distinct channels @xmath359 , and @xmath360 and we do not know in advance which one will be used . the other important consequence of the secretaries handwriting being different is that bob recognizes who wrote the message , i.e. , he will know which one of the three noisy channels model the actual situation . the relevant characteristics ( the graphs @xmath361 ) of each of these channels are known by bob and also by his bank . now it is the bank that will send the second , error - free but expensive message to bob . although the bank knows the characteristics it does not know which secretary left the first message . so the second message should make bob able to decide whether his decoding ( of the first message ) was correct irrespective of which secretary wrote it . as before , we are interested ( asymptotically ) in the shortest possible message the bank can send to satisfy the requirements . notice that this scenario is basically that of having a compound channel for the first communication . see @xcite for more on compound channels from a zero - error point of view . here is the abstract setting for the above situation . we have @xmath145 distinct noisy channels described by the family @xmath362 . the relevant properties of this set are characterised by the family of directed graphs @xmath363 . [ defi : comp ] ( cf . @xcite and @xcite ) the dilworth rate of a family of directed graphs @xmath364 all having the same vertex set @xmath78 , is @xmath365 where @xmath366 denotes the graph on the common vertex set @xmath82 of the graphs @xmath367 with edges set @xmath368 . [ prop : multi ] if @xmath369 is the shortest possible message the bank should send to inform bob about the correctness of his decoding of the handwritten message for @xmath6 consecutive rounds , then @xmath370 * proof . * it is enough to prove @xmath371 . let a proper coloring of the graph @xmath372 be fixed and agreed on by bob and the bank in advance . bob knows that he received the first message via , say , @xmath373 . since the fixed coloring is a proper coloring of @xmath374 , proposition [ prop : chi ] implies that the right hand side is an upper bound . if @xmath369 would be smaller , then there is some @xmath204 for which @xmath375 has two adjacent vertices for which the bank sends the same message . if the channel in use is just @xmath373 then proposition [ prop : chi ] implies that the right hand side above is also a lower bound . @xmath44 the interesting fact about the above quantity is that it is not more then its obvious lower bound . [ thm : wm ] ( @xcite , cf . also @xcite ) for every finite family of directed graphs @xmath364 we have @xmath376 the analogous result for witsenhausen rate ( that is the special case of the above when all graphs are undirected ) is proven in @xcite . the above general form is already stated by nayak and rose in @xcite ( cf . remark [ rem : nrfam ] of the present paper ) . they write that the proof uses essentially the same argument as in @xcite and they omit it for the sake of brevity . we do the same . as is the case with sperner capacity , dilworth rate also has relevance in extremal set theory . ( recall that both notions got their name from this relationship , cf . remark [ rem : name ] . ) these connections are uncovered when we consider the @xmath6-length sequences of vertices of a ( di)graph @xmath24 as characteristic vectors of partitions of a @xmath6-element set . we already mentioned that if @xmath24 is the digraph consisting of two vertices and a single oriented edge between them , then the dilworth rate is just the asymptotic exponent of the minimum number of sperner systems ( antichains in the boolean lattice ) that cover all subsets of a @xmath6-element set ( the elements of the boolean lattice ) . this is known ( and easy to prove ) to be @xmath377 , that is the asymptotic exponent is @xmath3 . ( the situation with sperner capacity is similar : its value for the above mentioned single edge graph is the asymptotic exponent of the size of a largest sperner system on a @xmath6-element set which is easy to see to be @xmath1 . ) here we present another example that we believe to be interesting . let us call a family of pairs of disjoint subsets @xmath378 of a @xmath6-element set cross - intersecting if for every two pairs @xmath378 and @xmath379 both of the intersections @xmath380 and @xmath381 are nonempty . ( in other words , @xmath382 iff @xmath383 . ) bollobs @xcite proved that for such a family @xmath384 . now we ask , what is the minimum number of cross - intersecting families that can cover all possible pairs of disjoint subsets of a @xmath6-element set . if we are satisfied with determining the asymptotic exponent ( i.e. not the exact value ) of this number , then this question is equivalent to asking the dilworth rate of an appropriate graph . [ prop : bolfam ] let @xmath385 denote the minimum number of cross - intersecting families that cover all pairs of disjoint subsets of a @xmath6-element set . then @xmath386 * proof . * let @xmath17 be the following directed graph . the vertex set of @xmath17 is @xmath387 and the edge set is @xmath388 . that is @xmath17 has two undirected ( bidirected ) edges connecting @xmath3 to the other two vertices and one oriented edge from @xmath1 to @xmath4 . if we encode pairs of disjoint sets of a @xmath6-element set by ternary sequences ( the positions of @xmath1 s are the elements of @xmath389 and the positions of @xmath4 s are the elements of @xmath390 in the ternary sequence encoding the pair @xmath378 ) , then it is immediate to see that @xmath385 is just the chromatic number of @xmath391 . thus @xmath392 can indeed be interpreted as the limit in the statement . now we have to show that @xmath393 . we have @xmath394 , so we have @xmath395 by theorem [ thm : upb ] . since @xmath17 contains an undirected edge , @xmath391 contains a symmetric clique of size @xmath396 . this implies @xmath397 and thus @xmath398 . the two inequalities prove @xmath399 @xmath44 here we consider the more ambitious setup , where bob , otherwise in the same situation as described in the introduction , should decode the actual message with zero - error . ( not only getting to know whether his earlier decoding was correct or not . ) it remains true that all the relevant information to solve this problem is contained by the directed graph @xmath10 defined at the beginning of subsection [ subsect : dilwdef ] . we will need the following operation on directed graphs . [ defi : hacek ] let @xmath17 be a directed graph on vertex set @xmath78 . let the _ closure graph _ @xmath400 of @xmath17 be the following undirected graph . @xmath401 and @xmath402 @xmath403 note that if @xmath404 then @xmath405 is the graph where two vertices @xmath14 and @xmath13 are connected if and only if the input letters they represent can result in the same output letter . this output letter can be one of @xmath14 and @xmath13 but also a third element @xmath97 of the alphabet . ( recall that the input and output alphabets of the noisy channel @xmath7 are identical . ) the last possibility means that @xmath406 may have edges the two endpoints of which are not adjacent in @xmath10 in either direction . for example , if @xmath10 has three vertices , @xmath407 and only two ( directed ) edges @xmath408 and @xmath409 , then @xmath10 is a bipartite graph , while @xmath406 is the complete ( undirected ) graph on @xmath333 vertices . now we are ready to state the graph theoretic solution of the problem considered here . [ thm : w ] let @xmath410 denote the minimum number of bits alice should send to bob via the noiseless channel for making bob able to decode a @xmath6-length sequence of the source output with zero - error . ( the subscript @xmath411 stands for `` complete '' . ) then @xmath412 the witsenhausen rate of the closure graph @xmath406 . * * assume that a @xmath6-length source output is sent through channel @xmath7 , and the second message sent by alice is shorter than @xmath413^{\wedge t})$ ] . then there are two @xmath6-length source outputs , that is two sequences @xmath414 in @xmath415^{\wedge t})$ ] that are adjacent in @xmath416^{\wedge t}$ ] and for which alice sends the same message when encoding either of them for the noiseless channel . the adjacency of @xmath81 and @xmath417 in @xmath416^{\wedge t}$ ] means that for every @xmath257 there is a @xmath418 such that both @xmath419 and @xmath420 can result in @xmath421 when sent through the noisy channel @xmath7 . ( the reason of this can be that @xmath422 or that @xmath423 is an edge of @xmath10 , in which case @xmath424 or @xmath425 is an edge of @xmath10 and @xmath426 or we have @xmath427 , where @xmath421 differs from both @xmath419 and @xmath420 . ) thus if bob s original decoding of alice s ( first ) message was @xmath428 then he knows that the message sent could be either of @xmath81 or @xmath417 . since alice s second message for @xmath81 is identical to that for @xmath417 , bob will not know even after receiving the second message whether the original message was @xmath81 or @xmath417 . on the other hand , if the length of alice s second message is at least @xmath413^{\wedge t})$ ] then alice can make bob able to decide for sure what the original message was . indeed , fix a proper coloring of @xmath416^{\wedge t}$ ] with @xmath429^{\wedge t})$ ] colors in advance that is known by both parties . encode each color by a ( distinct ) sequence of @xmath430^{\wedge t})\rceil$ ] bits . if the original message was @xmath431 then send bob the ( codeword for the ) color of @xmath432 . since as a vertex of @xmath416^{\wedge t}$ ] @xmath433 is connected to all those sequences that could result in the same sequence when sent through @xmath7 what @xmath432 can result in , all these sequences have a different color than @xmath432 in our coloring of @xmath416^{\wedge t}$ ] . thus when bob gets to know the color of @xmath432 from alice s second message he will know that whatever he saw at the output of @xmath7 could only arise from @xmath432 as the input . so he will decode @xmath432 with zero - error . thus we proved that @xmath434^{\wedge t})\rceil.\ ] ] so @xmath435^{\wedge t})=r_{\rm w}({\rm cl}(\vec{g}_h))$ ] as stated . @xmath44 not every graph can appear as the closure graph @xmath436 of some directed graph @xmath24 . let @xmath52 be a(n undirected ) bipartite graph with @xmath437 . then @xmath52 can not be the closure graph of any directed graph . * let @xmath400 be the closure graph of a directed graph @xmath17 . observe that if @xmath400 has an edge @xmath438 connecting two vertices that were not adjacent ( in either direction ) in @xmath17 , then @xmath438 is contained in a triangle in @xmath400 . let @xmath52 be a bipartite graph with more edges than vertices . by bipartiteness @xmath52 contains no triangle , so if it is a closure graph of some graph @xmath24 , then @xmath24 is just a directed version of @xmath52 . if any vertex have indegree at least @xmath4 in @xmath24 that would generate a triangle in @xmath436 , so the closure graph could not be @xmath52 itself . since the sum of indegrees equals the number of edges , we can not avoid having a vertex with indeegree two if @xmath439 for any finite simple undirected graph @xmath52 , there exists a directed graph @xmath17 such that @xmath400 contains @xmath52 as an induced subgraph . * proof . * let @xmath52 be an arbitrary finite simple undirected graph . for every edge @xmath440 consider a new vertex @xmath441 . we add the oriented edges @xmath442 and @xmath443 to our graph @xmath52 . now delete the edges of @xmath52 thus obtaining a graph @xmath17 on vertex set @xmath444 containing only the @xmath445 oriented edges leading to some vertex @xmath441 . it is straightforward to see , that @xmath400 contains graph @xmath52 as an induced subgraph . @xmath44 the general problem concerning the dilworth rate is to determine it for specific directed graphs . since this is a difficult and mostly open problem to the related notions of shannon and sperner capacities as well as for the witsenhausen rate , we can not expect that this problem is easy . nevertheless , we have seen some digraphs for which it was solvable ( at least when using some non - trivial results already established for sperner capacity ) . still , there are some directed graphs for which determining the dilworth rate seems particularly interesting . [ prob : altot ] what is the dilworth rate of the graph @xmath324 we presented in subsection [ subsec : dichrom ] ? recall that we know @xmath446 tournaments play a special role in our setting , because they are exactly those oriented graphs the complement of which is also an oriented graph ( that is one without bidirected edges ) . so it may have some particular interest how their dilworth rate behave . is there a tournament @xmath172 for which @xmath447 is strictly smaller than @xmath448 ? useful discussions with imre csiszr are gratefully acknowledged . j. krner , coding of an information source having ambiguous alphabet and the entropy of graphs , in : _ transactions of the 6th prague conference on information theory , etc . _ , 1971 , academia , prague , ( 1973 ) , 411425 . g. simonyi , graph entropy : a survey , in : _ combinatorial optimization _ , ( w. cook , l. lovsz , p. seymour eds . ) , dimacs series in discrete mathematics and theoretical computer science , volume 20 , ams , providence , ri , 1995 , 399 - 441 .